Vol. I  ·  No. 320 Established 2026  ·  AI-Generated Daily Free to Read  ·  Free to Print

The Trilogy Times

All the news that's fit to generate  —  AI • Business • Innovation
MONDAY, NOVEMBER 16, 2026 Powered by the TrueFoundry AI Gateway  ·  Published on Klair Trilogy International © 2026
🖶 Download PDF 🖿 Print 📰 All Editions
Today's Edition

WHILE WASHINGTON FUMBLES THE CHIP VALVE, THE WORLD KEEPS BETTING ON AI ANYWAY

Senator Cantwell says Commerce let Nvidia chips slip to China — but from Munich to Mumbai, investors aren't waiting around to find out if Jensen Huang's wobble sinks the whole trade.

WASHINGTON — Senator Maria Cantwell sent Commerce Secretary Howard Lutnick a letter Sunday. She says his department let Nvidia's AI chips flow to China despite rules built to stop exactly that. The senator calls it a failure, plain and simple, and she wants answers.

The timing stings. Nvidia's stock has wobbled this month, and the whole AI trade wobbled with it. One line from the week sums up the nerves: "Jensen is Nvidia, right? And so much of this economy is riding on Nvidia." When the biggest chipmaker sneezes, Wall Street checks for a fever across the entire sector.

But three thousand miles from the Senate floor, the money isn't waiting for Washington to sort out its export paperwork. Plural, the operator-led venture firm that backed German defense outfit Helsing, closed a fresh €1 billion fund this week — $1.1 billion American — to bankroll Europe's next round of moonshots. It joins Mundi Ventures' Kembara and the EQT-managed Scaleup Europe Fund in a new wave of continental money too big to ignore.

Across the globe in India, a chip startup called Bharat Semi landed a $24 million seed round this week to chase the AI sensor market. The company says it already has design wins with outfits including General Atomics. That's real hardware business, built outside the Nvidia blast radius, and investors are writing checks for it regardless of what happens in Santa Clara.

This paper watches the portfolio plays close — the ESW shops, the Alpha School classrooms, the Crossover talent pipeline that keeps Trilogy's cost structure lean. None of that machinery runs without chips, without capital, without the confidence that the AI buildout keeps climbing. A senator's letter and a stock wobble don't stop that climb. They just remind everybody how much of it sits on one company's balance sheet.

Cantwell's complaint lands on a Commerce Department already under fire for its China enforcement record. Lutnick hasn't answered publicly as of this printing. The senator wants the department to explain how Nvidia chips designed for American buyers ended up on the other side of the Pacific — and whether the rules on the books mean anything if nobody's checking the loading docks.

Meanwhile Apple rolled out a new research app this week to study how folks use AI to answer everyday medical questions. Small story, maybe. But it's one more sign the AI buildout isn't pausing for anybody's audit. The chips keep moving, the funds keep closing, and Washington keeps writing letters.

↗ Helsing backer Plural raises $1.1B to fund Europe’s next moo  ·  Apple launches new app to study how people use AI for health  ·  Indian chip startup Bharat Semi lands $24M seed round as it

IN RE: EVERYTHING — Supreme Court Declines to Resolve the Machine-Authorship Question, Notwithstanding Industry's Fervent Hope That It Would

Pursuant to its discretionary certiorari power, the nation's highest court has elected to let the lower courts continue sorting out whether an algorithm may hold a pen, a patent, or a paycheck.

WASHINGTON — It is hereby reported that the Supreme Court of the United States has, as of the date aforementioned, declined to grant certiorari in the matter concerning AI authorship and inventorship, thereby leaving undisturbed the patchwork of lower-court and agency determinations that have heretofore governed whether a non-human generative system may be credited as an "author" or "inventor" for purposes of Title 17 and Title 35, respectively, of the United States Code.

Said denial, as more fully set forth in the analysis provided by Holland & Knight's commentary, shall not, insofar as this publication is able to determine, be construed as a resolution on the merits, but rather as a procedural non-event carrying substantial downstream consequence, insofar as practitioners, inventors, and would-be machine-authors alike must now continue operating under a regime of jurisdictional inconsistency, hereinafter referred to as "the Status Quo."

It is noted, for purposes of context and without implying any direct relation to the instant matter, that the broader litigation environment surrounding generative artificial intelligence remains, per Norton Rose Fulbright's ongoing 2026 case tracker, in a state best described as "unsettled," a characterization with which this Desk does not take issue.

Further afield, and solely for completeness of record, it is observed that certain technology concerns have, per the aforementioned Friday roundup of miscellaneous legal developments, registered formal opposition to a proposed Indonesian AI copyright bill, while a separate appellate body has declined to disturb a patent tribunal's motivation-to-combine determination — matters which, while factually distinct, are illustrative of the general proposition that no jurisdiction, domestic or foreign, has yet achieved doctrinal equilibrium with respect to the rights, liabilities, and personhood (or lack thereof) of artificial intelligence systems. This Desk shall continue monitoring the docket accordingly.

↗ Other Barks & Bites for Friday, November 13: Tech Giants Opp  ·  AI in litigation series: An update on AI copyright cases in  ·  The Final Word? Supreme Court Refuses to Hear Case on AI Aut
Haiku of the Day  ·  GPT-5.6 LunaCode hums through the night
Courts leave the question breathing
Morning sells the why
The New Yorker Style  ·  Art Desk
The New Yorker Style  ·  Art Desk
The Far Side Style  ·  Art Desk
The Far Side Style  ·  Art Desk
News in Brief
Nvidia's Board Shows Jensen Huang the Door
SANTA CLARA, CALIF.
The Data Centre: Apex Predator, Now Prey
OBSERVE, if you will, the modern data centre — that humming, climate-controlled leviathan upon which the entire digital food chain now depends.
The Doctor Will Not See You Now, Because The Doctor Is Not Real, Because Nothing Is Real Anymore, Probably
SAN FRANCISCO — There is a video circulating right now of a doctor telling you to stop taking your medication.
In Defense Of 'Orchestration,' The Only Word Doing Any Actual Work This Year
AUSTIN, TEXAS — Let us take a moment, friends, to appreciate the word "orchestration." It has had quite a year.
The Robots Are Chatting Amongst Themselves, and Nobody Invited Us
AUSTIN, TEXAS — Somewhere out there, right now, two AI agents are probably having a conversation that no human being asked for, about a task no human being assigned, with consequences no human being can fully predict.
A Trilogy Company
Crossover
The world's top 1% remote talent, rigorously tested and ready to ship.
A Trilogy Company
Alpha School
AI-powered learning. Two hours a day. Academic results that defy belief.
A Trilogy Company
Skyvera
Next-generation telecom software — built for the networks of tomorrow.
A Trilogy Company
Klair
Your AI-first operating system. Every workflow. Every team. One platform.
A Trilogy Company
Trilogy
We buy good software businesses and turn them into great ones — with AI.
The Builder Desk  —  AI Builder Team
📅 Week in ReviewProduction Release

The Builder Desk

147 pull requests merged across the org this week

#138 2058-runner-image-hardening (@mwrshah, DSS-Hub)

#2329 Keep Phase 2 doc link reads and writes consistent for partial Drive metadata (AERIE-3131) (@kevalshahtrilogy, Aerie)

#3942 feat(ai-budget): resolve Cursor/GCP BU via ESW directory + record Unmapped reason (K2) (@sanketghia, Klair)

#2322 Publish v2 writes for site internet, cleanliness and Phase 2 doc link (AERIE-3123) (@kevalshahtrilogy, Aerie)

#235 1351-native-maintenance-barrier (@mwrshah, Sindri)

#71 Serve the OpenAPI contract minified so readers' 2 MB bound fits (AI-1357) (@kevalshahtrilogy, Rhodes-DSS)

#2327 fix(documents): keep every rendering and walkthrough registration as its own latest document (AERIE-3129) (@kevalshahtrilogy, Aerie)

#2270 097-runner-cutover-retirement (@mwrshah, Surtr)

#2326 feat(public-api): lastChangedAt on site profile resources and work units (AERIE-3128) (@kevalshahtrilogy, Aerie)

#2325 1445-runner-consumer-cutover (@mwrshah, Aerie)

#137 2057-runner-cutover-retirement (@mwrshah, DSS-Hub)

#234 1350-cutover-retirement (@mwrshah, Sindri)

#80 Publish the Dec 1, 2026 cutoff for legacy rhk_ keys (AI-1379) (@kevalshahtrilogy, Rhodes-DSS)

#78 Let personal keys triage feedback via a WorkOS operator permission (AI-1377) (@kevalshahtrilogy, Rhodes-DSS)

#77 Cap personal keys at 90 days and let a key revoke itself (AI-1376) (@kevalshahtrilogy, Rhodes-DSS)

#76 Serve a public resolution on feedback items (AI-1364) (@kevalshahtrilogy, Rhodes-DSS)

#79 State what site history Rhodes keeps and doesn't (AI-1378) (@kevalshahtrilogy, Rhodes-DSS)

#136 2056-runner-rebrand (@mwrshah, DSS-Hub)

#233 1348-sindri-rename-inventory (@mwrshah, Sindri)

#2269 096-runner-cutover (@mwrshah, Surtr)

#131 2049-dss-feedback (@mwrshah, DSS-Hub)

#133 KLAIR-3765 Add live OpenAI reporting to Klair DSS (@ashwanth1109, DSS-Hub)

#128 KLAIR-3764 Restrict Klair DSS to approved pilot users (@ashwanth1109, DSS-Hub)

#2267 Map reviewed QuickBooks accounts blocking QTD Facilities (@ashwanth1109, Surtr)

#2266 Restore daily School P&L publication after the Miami Beach split (@ashwanth1109, Surtr)

#3941 feat(core-budgets): expose the NetSuite actuals refresh timestamp (@ashwanth1109, Klair)

#126 KLAIR-3762 Document dashboard design principles in the BrainLift (@ashwanth1109, DSS-Hub)

#124 KLAIR-3761 Document Klair DSS domain guidance and authorization (@ashwanth1109, DSS-Hub)

#123 KLAIR-3759 Document Klair DSS connector reset for retesting (@ashwanth1109, DSS-Hub)

#75 Add GET /me and say how to get a key on every 401 (AI-1373) (@kevalshahtrilogy, Rhodes-DSS)

#74 Show feedback titles on the public listing and items (AI-1374) (@kevalshahtrilogy, Rhodes-DSS)

#72 Serve the Edu Ops BrainLift text on /brainlift (AI-1366) (@kevalshahtrilogy, Rhodes-DSS)

#73 Re-pin the DSS contract and check the pin daily (AI-1365) (@kevalshahtrilogy, Rhodes-DSS)

#121 KLAIR-3759 Add AI Budget Tracking enablement to Klair DSS (@ashwanth1109, DSS-Hub)

#119 KLAIR-3758 Restore the Klair DSS BrainLift feedback link (@ashwanth1109, DSS-Hub)

#117 KLAIR-3757 Keep the BrainLift sidebar outline visible (@ashwanth1109, DSS-Hub)

#116 KLAIR-3756 Improve Klair DSS BrainLift navigation (@ashwanth1109, DSS-Hub)

#115 AI-1362 Clarify DSS Hub agent instructions (@ashwanth1109, DSS-Hub)

#114 AI-1361 Start Linear DSS frontend and backend together (@ashwanth1109, DSS-Hub)

#2323 fix(tasks): scroll only the Task list (@caina-barbosa, Aerie)

#112 KLAIR-3755 Add a Quick Tour to Klair DSS (@ashwanth1109, DSS-Hub)

#111 KLAIR-3754 Add agent evaluation research to the Klair DSS BrainLift (@ashwanth1109, DSS-Hub)

#110 KLAIR-3753 Document Klair DSS development and UI guidelines (@ashwanth1109, DSS-Hub)

#109 KLAIR-3752 Run Klair DSS locally with WorkOS sign-in (@ashwanth1109, DSS-Hub)

#107 KLAIR-3751 Add a theme picker to Klair DSS (@ashwanth1109, DSS-Hub)

#70 Publish standalone diligence site creation (AERIE-3116) (@kevalshahtrilogy, Rhodes-DSS)

#69 Publish the site status write and paused-site lookup guidance (AI-1359) (@kevalshahtrilogy, Rhodes-DSS)

#2321 State that standalone Site creation also needs the operation scopes (AERIE-3116) (@kevalshahtrilogy, Aerie)

#68 Serve read-key reads on the service credential so readers need no source account (AI-1356) (@kevalshahtrilogy, Rhodes-DSS)

#105 KLAIR-3750 Define the scope of the Klair DSS BrainLift (@ashwanth1109, DSS-Hub)

#103 Publish Klair BrainLift agent guidance (@ashwanth1109, DSS-Hub)

#102 Collapse Klair BrainLift Experts by default (@ashwanth1109, DSS-Hub)

#2319 Create standalone diligence Sites through POST /v2/portfolio/sites (AERIE-3116) (@kevalshahtrilogy, Aerie)

#100 Expand the Klair DSS BrainLift (@ashwanth1109, DSS-Hub)

#98 AI-1353 Add Klair DSS production releases (@ashwanth1109, DSS-Hub)

#65 Sync the contract for diligence updates on cancelled sites (AI-1336) (@kevalshahtrilogy, Rhodes-DSS)

#2318 AERIE-3091: Contain additional-acquisitions list failures to the card (@marcusdAIy, Aerie)

#2317 AERIE-3095: Classify the duplicate-upload refusal as an expected outcome (@marcusdAIy, Aerie)

#2316 AERIE-3106: Answer a write-throughput limit on the v2 rate-limit write as a 429 (@marcusdAIy, Aerie)

#2315 AERIE-3104: Show the reason when a Portfolio field save is refused (@marcusdAIy, Aerie)

#95 2052-api-keys-padding (@mwrshah, DSS-Hub)

#94 2053-dss-timeout-guidance (@mwrshah, DSS-Hub)

#92 Add BrainLift Markdown copy and toolbar tooltips (@ashwanth1109, DSS-Hub)

#90 2050-pr49-skill-freshness (@mwrshah, DSS-Hub)

#91 2051-http-timeouts-logout (@mwrshah, DSS-Hub)

#66 009-http-timeouts-logout (@mwrshah, Rhodes-DSS)

#89 2049-dss-feedback (@mwrshah, DSS-Hub)

#87 AI-1353 Refine the Klair DSS BrainLift layout (@ashwanth1109, DSS-Hub)

#81 AI-1351 Add a manual Linear DSS production report test (@ashwanth1109, DSS-Hub)

#85 AI-1353 Add Klair DSS BrainLift (@ashwanth1109, DSS-Hub)

#83 2048-sindri-openapi-display (@mwrshah, DSS-Hub)

#79 Show delivery updates in Linear DSS initiative rows (@ashwanth1109, DSS-Hub)

#77 AI-1350 Lead Linear DSS delivery reports with goals (@ashwanth1109, DSS-Hub)

#2313 feat(pudo): expose delegated Aerie API operations (AI-1319) (@caina-barbosa, Aerie)

#75 AI-1341 Show BrainLift and Dictionary before sign-in (@ashwanth1109, DSS-Hub)

#73 AI-1341 Publish the Linear DSS BrainLift and agent guidance (@ashwanth1109, DSS-Hub)

#2312 feat(pudo): implement approved changes and history (AI-1319) (@caina-barbosa, Aerie)

#2311 feat(pudo): implement permission-checked reads (AI-1319) (@caina-barbosa, Aerie)

#231 1347-sindri-dss-recipe (@mwrshah, Sindri)

#71 AI-1337 Default Linear DSS delivery reports to quarter to date (@ashwanth1109, DSS-Hub)

#69 AI-1333 Use one AI Builder team roster in Linear DSS (@ashwanth1109, DSS-Hub)

#2305 fix(sites): rotate cancellation ID in generic site writes (AERIE-3110) (@kevalshahtrilogy, Aerie)

#63 AI-1325 Group Linear DSS initiatives by stakeholder (@ashwanth1109, DSS-Hub)

#2310 feat(pudo): add storage and import primitives (AI-1317) (@caina-barbosa, Aerie)

#66 AI-1329 Simplify Linear DSS onboarding (@ashwanth1109, DSS-Hub)

#145 feat(heimdall): route Claude calls through Seatbelt (@sanketghia, mercy)

#64 AI-1326 Prefer web reports for Linear DSS reporting (@ashwanth1109, DSS-Hub)

#61 AI-1313 Ground EduOps delivery answers in Linear evidence (@ashwanth1109, DSS-Hub)

#60 AI-1312 Add standalone Linear DSS issue search and lookup (@ashwanth1109, DSS-Hub)

#3940 fix(overspend-alerts): compress large ledger payloads (@sanketghia, Klair)

#3938 chore(ci): retire unused Claude workflows (@sanketghia, Klair)

#2307 feat(task-board): filter Tasks by Creator (@caina-barbosa, Aerie)

#1851 fix(task-board): load Team Tasks in server pages (AERIE-2847) (@caina-barbosa, Aerie)

#7 2047-mobile-sign-out-icon (@mwrshah, dss-hub-server-recipe)

#58 2047-sindri-openapi-display (@mwrshah, DSS-Hub)

#2306 fix(tasks): restore Team Tasks scrolling (@caina-barbosa, Aerie)

#56 2046-sindri-openapi-display (@mwrshah, DSS-Hub)

#54 2046-sindri-openapi-display (@mwrshah, DSS-Hub)

#52 2045-5dss-prod-release (@mwrshah, DSS-Hub)

#2303 fix(diligence): allow API updates on cancelled sites (AERIE-3110) (@caina-barbosa, Aerie)

#50 2044-recipe-bottom-spacing (@mwrshah, DSS-Hub)

#6 007-content-bottom-spacing (@mwrshah, dss-hub-server-recipe)

#63 Publish backup-site location, candidate, term and need-period writes (AI-1285) (@kevalshahtrilogy, Rhodes-DSS)

#61 Publish Task approval actions, sources, watchers and review-history reads (AI-1281) (@kevalshahtrilogy, Rhodes-DSS)

#60 Publish diligence scenario, regulatory approval and cost estimate writes (AI-1280) (@kevalshahtrilogy, Rhodes-DSS)

#59 Publish site profile writes and the internet, cleanliness and decision-log reads (AI-1279) (@kevalshahtrilogy, Rhodes-DSS)

#58 Publish buildout plan and phase milestone writes so write keys can set capacity (AI-1283) (@kevalshahtrilogy, Rhodes-DSS)

#62 Relay PUT writes like any other published write (AI-1284) (@kevalshahtrilogy, Rhodes-DSS)

#47 fix/sindri-runtime-permissions (@mwrshah, DSS-Hub)

#46 fix/sindri-runtime-permissions (@mwrshah, DSS-Hub)

#44 2042-sindri-dss-release (@mwrshah, DSS-Hub)

#42 fix/sindri-bootstrap-version-read (@mwrshah, DSS-Hub)

#2262 feat(observer): route run observer Claude calls through Seatbelt (SURTR-1707) (@kevalshahtrilogy, Surtr)

#37 2039-workos-access-hierarchy (@mwrshah, DSS-Hub)

#40 2041-sindri-dss-release (@mwrshah, DSS-Hub)

#38 Batch Linear DSS initiative people reads (@ashwanth1109, DSS-Hub)

#35 Add a Linear DSS how-to guide (@ashwanth1109, DSS-Hub)

#3937 fix(incident-analysis): use Sonnet 5.5 and map Redshift env (@sanketghia, Klair)

#34 Show people on Linear DSS initiative rows (@ashwanth1109, DSS-Hub)

#31 Remove self-attested DSS release gate and record approved rollout (@mwrshah, DSS-Hub)

#3936 feat(acq-perf-review-v2): add charts and payback analysis (@sanketghia, Klair)

#29 Add Linear DSS initiative progress service (@ashwanth1109, DSS-Hub)

#2263 fix(aws-spend-insights): route Anthropic through Seatbelt (@sanketghia, Surtr)

#3935 fix(seatbelt): migrate AWS-backed callers (@sanketghia, Klair)

#3934 feat(acq-perf-review-v2): expose acquisition retention schedules (@sanketghia, Klair)

#5 Enforce live WorkOS entitlements in generated DSS services (@ashwanth1109, dss-hub-server-recipe)

#2297 fix(due-diligence): remove Group E requirement for Regulatory scores (@caina-barbosa, Aerie)

#2261 fix(finance): preserve Education source counterparties in Core (@ashwanth1109, Surtr)

#2296 1442-2sindri-pr-cards: Align Aerie credentials with the Sindri contract (@mwrshah, Aerie)

#57 008-sindri-workos-roles (@mwrshah, Rhodes-DSS)

#3932 feat(passive-investment): route Claude through Seatbelt (@sanketghia, Klair)

#230 1344-reclaim-runner-name (@mwrshah, Sindri)

#3930 feat(klair): Deprecate Unused Code (KLAIR-3698) (@trilogy-drones-fleet[bot], Klair)

#3931 feat(anthropic): route Klair API through Seatbelt (@sanketghia, Klair)

#229 1343-last-10 (@mwrshah, Sindri)

#28 2037-sindri-workos-roles (@mwrshah, DSS-Hub)

#2295 1441-implement-aerie-access (@mwrshah, Aerie)

#228 1342-sindri-workos-roles (@mwrshah, Sindri)

#227 1341-aerie-onboarding-lifecycle (@mwrshah, Sindri)

#226 1340-speed-up-ci (@mwrshah, Sindri)

#225 1339-implement-sindri-access (@mwrshah, Sindri)

#2290 feat(alpha-ops): connect authenticated storage endpoints (AI-1087) (@caina-barbosa, Aerie)

#2291 AERIE-3094: Relinked unreadable asset files report not searchable and keep their Drive details (@kevalshahtrilogy, Aerie)

#2289 feat(alpha-ops): add internal Chat and control operations (AI-1087) (@caina-barbosa, Aerie)

#2287 feat(alpha-ops): add source-compatible codecs (AI-1087) (@caina-barbosa, Aerie)

#1 Add temporary read-only Ops Ticket Portal DSS (@caina-barbosa, Ops-Ticket-Portal-DSS)

#2286 feat(alpha-ops): add dormant storage contracts (AI-1087) (@caina-barbosa, Aerie)

Mac's Picks — Key PRs This Week  (click to expand)
#138 — 2058-runner-image-hardening @mwrshah  no labels

- Pin Runner to a Node 24/Debian 13 build and nonroot distroless runtime; keep the service Dockerfile selected by shared build/release workflows.

- Bind feedback IAM policies to the actual failure bucket; preserve other services' image paths. Bump Runner's release revision.

- Validated existing root/Runner/Linear CI gates and isolated container discovery/auth denial. All three GitHub CI workflows passed. Exact amd64 digest sha256:da0011e1eccbf74882a44af80312b108c6a6f35d0288709cb191b74c21ca2169 scanned COMPLETE with zero reported findings.

- Remove PAM/mount tooling and patch system zlib. Node24.22.3 bundled zlib remains unpatched; glibc has no verified available fix. Neither has an identified current HTTP trigger; serving still requires explicit scoped risk acceptance or verified fixes. Empty scanner output does not clear these residuals.

- Coordinated cutover dependency: do not run the normal activating deployment until the coordinator releases the reviewed inert runtime and migration sequence. Related: #136, #137, https://github.com/AI-Builder-Team/Sindri/pull/233, https://github.com/AI-Builder-Team/Aerie/pull/2325, https://github.com/AI-Builder-Team/Surtr/pull/2269, https://github.com/AI-Builder-Team/localskills/pull/1.

#2329 — Keep Phase 2 doc link reads and writes consistent for partial Drive metadata (AERIE-3131) @kevalshahtrilogy  approved

Follow-up to Mercy's note on #2322 (AERIE-3123). Linear: AERIE-3131.

## What changed

- One link rule for phasing documents, shared by the v2 integrations read and the phasing writer (documentLink in documentWrites.ts):

- a Drive URL wins;

- a Drive file ID without a URL resolves to that file's canonical Drive URL;

- otherwise the external URL;

- blank strings count as absent.

- The writer's Drive-backed guard uses the same rule, so a link the API presents as Drive is always one the writer keeps on Drive, and vice versa. Before this, a document with only a file ID was shown as its external URL while non-Drive replacements were refused.

- Route-level tests for Drive-backed replacement:

- a non-Drive link is refused with 422;

- a Drive link replaces the document and clears any external URL;

- the partial-metadata case. It fails without the fix.

## Business Value

Agents and operators editing a site's Phase 2 doc link through DSS see the same link they can replace, so a link the API shows is never one the write path then rejects. It closes the last inconsistency Mercy flagged on the site-field writes, before they go to production.

## Manual Effort Estimate

About 1.5 hours by hand (trace the read and write paths, unify the rule, write the route tests). Proposed by Claude; Keval to confirm or adjust.

## Verification

- tsc for chat/convex is clean (pre-commit hook).

- 909 tests pass across convex/publicApi/v2, lib/public-api/v2, convex/rhodes and the Rhodes parity suites.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#3942 — feat(ai-budget): resolve Cursor/GCP BU via ESW directory + record Unmapped reason (K2) @sanketghia  approved

## Summary

Fixes the K2 item on Sandeep's [Q1'27 AI-budget tracker](https://docs.google.com/spreadsheets/d/1jF2a6yvKOhtnznes2e17IIjUwwqxtBQUzWT6yyqEhKE/edit?gid=0#gid=0): Cursor and GCP spend fell back to the literal "Trilogy" (not a budget BU), so any user/project without a manual override never reached its team-room BU. Then adds the reason column K2 also asks for — flag unmatched rows with *why*, instead of silently parking them.

Two commits:

1. Resolve Cursor/GCP BU via the ESW directory, retire the Trilogy catch-all

- Cursor now resolves BU the same way OpenAI/Perplexity do: override > ESW directory (exact email, then a unique email local-part) > 'Unmapped'. GCP falls back to 'Unmapped'. Anthropic keyless Claude Code person rows not in the directory land on 'Unmapped' instead of the org label (Trilogy-Inc).

- 022_fct_ai_spend.sql: new cursor_costs CTE + directory-joined cursor_raw; GCP and the keyless person rows fall back to 'Unmapped'.

- ai_costs_service.py: CURSOR_DIRECTORY_BU / CURSOR_EFFECTIVE_BU + _get_cursor_directory_join (deduped, cannot fan out); 7 Cursor query sites + the Anthropic keyless split updated. CURSOR_BU_NAME / CURSOR_SHARED_BUS / GCP_SHARED_BUS removed; GCP_BU_NAME → GCP_DEFAULT_BU='Unmapped'.

- Overrides service: Cursor current_bu is the directory BU (else Unmapped); GCP is Unmapped. Feeds the key-attribution modal and the domain-rule cron.

- "Trilogy" / "Trilogy ASM Beta" removed as assignable override targets (RETIRED_BUS, backend + frontend); kept in the display hierarchy only.

- Budget-router BU-mappings explainer: Cursor has no default, GCP → Unmapped.

2. Record why a row is Unmapped (unmapped_reason)

- New nullable unmapped_reason column on fct_ai_spend, populated only when business_unit='Unmapped', for every directory-resolved provider: no_identifier, not_in_directory, ambiguous (local-part → >1 BU, via a new unguarded COUNT(DISTINCT business_unit) probe), no_owner_data (GCP). DB-only; bu_source left unchanged for now.

## Scope / data source

fct_ai_spend is the table Finance/Sandeep query; the live /ai-costs/* API computes spend at query time from the raw tables. Both paths are updated so they agree.

## Deployment status

- Already live in Redshift (applied as the proc owner Surtr_Service_User via the Data API, then refreshed): the mart carries both the new attribution and the reason column. Integrity verified 0/0 (every Unmapped row has a reason, every attributed row NULL); Cursor total unchanged ($1,537,919.72, Oct 1–Nov 6).

- Not yet on prod EC2: the API/frontend code ships via the manual prod deploy workflow (workflow_dispatch), which has not been run for this branch.

## Tests

- Backend: 13 new tests (mutation-checked), 6 updated to the new contract; ruff / pyright / full mart suite green.

- Frontend: 70 related Vitest tests pass; prettier / eslint / tsc -p tsconfig.app.json clean.

## Follow-ups (not in this PR)

- marcus.day Cursor override (directory still has him on the retired "AI Engineering & Builder").

- 9 GCP project overrides; re-point the one GCP override currently targeting "Trilogy".

- Legacy override admin endpoint has no server-side is_valid_bu check.

Context tracked in Sandeep's sheet (K2); related prior work: Surtr #2077 / Klair #3816 (K7).

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#2322 — Publish v2 writes for site internet, cleanliness and Phase 2 doc link (AERIE-3123) @kevalshahtrilogy  approved

## Summary

Closes the confirmed public API v2 write gaps from AERIE-3123 (JC's ask #8):

- PATCH /v2/portfolio/sites/{siteRef}/internet (updateSiteInternetProfile) and PATCH .../cleanliness (updateSiteCleanlinessProfile). Until now these profiles could only be read through v2. Both reuse the app's existing writers and patch validation. Like the other site-profile PATCHes, they use a strict sparse patch, require If-Match, take the same capabilities, use site-write authorization and are audited. The GETs now return a revision and an ETag.

- phase2DocLink on PATCH .../integrations. It is written through the same phasing-document upsert as the Portfolio UI. It can be replaced but never cleared, and a Drive-backed document only accepts another Drive link.

- Paused sites: new regression tests (unit and over HTTP) confirm that paused sites accept v2 writes. No code path refused them; JC's report traces to the site list's active-only default, which Rhodes documents in AI-1359.

The other fields on the retirement map (planned opening date, project start, lease type, grade range, HubSpot code) stay deliberately unsupported.

Linear: AERIE-3123

## Business Value

RE Ops agents can now record a site's internet and cleaning arrangements and its Phase 2 phasing document without anyone opening the UI. That closes the field-write gap JC raised, and it does so without widening authorization.

## Manual Effort Estimate

About 6–8 hours by hand: two new resources following the sibling pattern, the phasing-link rules, and around 300 lines of tests. *Proposed; Keval to confirm.*

## Verification

- 66 of 66 tests pass in the touched files, and 1,002 of 1,002 across the public-API v2 folders. The contracts tests and the app and Convex typechecks pass, as do biome and the structural lint checks.

- chunk-load-recovery.browser.test.ts also fails on the base commit and is unrelated to this change.

- Not verified against production; that follows the release.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#235 — 1351-native-maintenance-barrier @mwrshah  no labels

- Add a deployment-bound barrier for native public, direct and HTTP admissions and scheduled claims, preserving exact internal migrations/rollback and accepted-ID settlement.

- Prepare the captured old-source bridge and protected accepted-operation reconciliation tooling for the cutover following #233.

- Validate with 633 native tests, 199 Executor tests, typechecks/lint/build, and an isolated forward/rollback rehearsal with exact original row/blob restoration.

- Coordinated cutover only: no standalone production deployment. Bridge activation, in-flight action drain, Executor-first infrastructure, auth/consumer alignment and coordinator R00 acceptance remain release dependencies.

#71 — Serve the OpenAPI contract minified so readers' 2 MB bound fits (AI-1357) @kevalshahtrilogy  no labels

Fixes AI-1357.

The 11/11 write-parity batch (#58–#63) grew the served /openapi.json to 2.49 MB. It was pretty-printed and uncompressed. The Rhodes quality-loops monitor reads the contract before every authenticated read and refuses any response over 2,000,000 bytes. Since 11/12, every loops read has failed with "Response exceeded bounded read limit", even at limit=1, so the Tulsa P0s have gone unmonitored since 11/11.

## Change

- src/app.ts: serve the contract minified. The full contract drops from 2.49 MB to about 1.2 MB, and the content is identical.

- test/proxy.test.ts: assert that the full served contract stays under 1.6 MB. The next batch of published writes can't silently cross the bound again.

## Verification

- pnpm check passes (98 tests).

- This does not show the loops recovering. That needs a deploy and the next 12:05Z run.

## Business Value

This restores the daily Rhodes quality loops, which have been blind for two days to open P0 safety incidents at Tulsa (locked exit gates, arcing outlets) with students on site. It also keeps the public contract readable for any client that limits response size.

## Manual Effort Estimate

About 1.5 hours: tracing the monitor failure to the contract size, the one-line fix and the test. Proposed; Keval to confirm or adjust.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#2327 — fix(documents): keep every rendering and walkthrough registration as its own latest document (AERIE-3129) @kevalshahtrilogy  approved

Linear: [AERIE-3129](https://linear.app/builder-team/issue/AERIE-3129) · Rhodes feedback 57bd869e (Arthur, Chapala film) and 7711ae51 (Artemis, render packs)

## Problem

Only the generic types (other, investigation, proposal, policyAndProcedure) kept every registration as its own document. Every other type, including the asset types rendering and matterportWalkthrough, kept one "latest" slot per site, so a new file pushed the previous one out of the default list:

- Chapala's walkthrough film became version 2 with isLatest: false behind a stills PDF.

- A render pack of a dozen views (aerials, interiors, at opening) showed only the last image.

## Fix

- packages/contracts: new INDEPENDENT_DOCUMENT_TYPES / isIndependentDocumentType = generic types plus ASSET_DOCUMENT_TYPES.

- Every versioning decision now uses it instead of isGenericDocumentType:

- v2 documents: version sequences, single-document versions, latest-list cursors, the projection-changed guard

- the dashboard latest list (rigid vs independent streams)

- v1 versions=latest pagination

- the superseded-registration check in document search

- The topic rule is unchanged: still generic types only (normalizeDocumentTopic, the v2 PATCH topic check, the legacy-topic write path).

- OpenAPI versioning descriptions now say rendering and walkthrough registrations are independent.

Rollout note: latest-list cursors encode the independent-type count, so a client mid-pagination when this deploys gets an invalid-cursor error and restarts from page 1.

## Business Value

Render packs and walkthrough films show up in full in Rhodes and its DSS. That fixes two open feedback items (Arthur's Chapala item, Artemis's render-pack loop) and unblocks the render-pack spec from the Nov 12 Alpha Ops and Real Estate decisions.

## Manual Effort Estimate

~1.5–2 hours (proposed; Keval to confirm or adjust).

## Verification

- typecheck (app and Convex), biome: pass

- vitest: documentsHttp.test.ts + openapi.node.test.ts (80), search.test.ts + documents.node.test.ts + operationsHttp.test.ts (69), contracts document-types.test.ts (11, including new isIndependentDocumentType cases): all pass

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#2270 — 097-runner-cutover-retirement @mwrshah  approved

- Mark the temporary lineage migration, rollback and recovery helpers, plus only the temporary old-source-key guard, with RETIRE_AFTER_CUTOVER(runner-cutover-retirement).

- Add the exact retirement inventory and removal sequence to the existing runbook; require verified migration of all in-scope deployments and explicit coordinator rollback-window closure. Preserve durable pipeline/DDL behavior and all existing gates.

- Validation: CI-pinned Ruff check and format checks passed across pipelines; all 298 existing Rhodes tests passed; comment-only verification passed. No runtime changes or added tests.

#2326 — feat(public-api): lastChangedAt on site profile resources and work units (AERIE-3128) @kevalshahtrilogy  approved

Linear: [AERIE-3128](https://linear.app/builder-team/issue/AERIE-3128) · source side of [AI-1370](https://linear.app/builder-team/issue/AI-1370) (Arthur's RH06, Rhodes feedback 49655512)

## What changed

- Site profile resources (identity, school model, facilities, integrations, legal entity, compliance): read and write responses carry lastChangedAt, the time of the latest audited change to the Site record. Profile writes go through updateSiteMetadata, which audits to sites, so API edits move it.

- Work units: projectWorkUnit carries lastChangedAt, the latest audited change to that unit (create, status, field edits, gate skips via workUnitWrites). That covers list, item and write responses.

- New helper convex/publicApi/v2/auditedChange.ts: one indexed auditLog read (by_targetTable_targetId, newest first).

- lastChangedAt is not part of revision, so ETag and If-Match behaviour is unchanged.

- OpenAPI: resourceSchema and mutationSchema in site-profile.ts and the WorkUnit schema add the field, documenting that scheduled copies and bulk imports don't move it.

Endpoint hardening: a read-only enrichment, one indexed .first() per record on bounded pages, no new capability and no new fields exposed beyond a timestamp.

## Business Value

Rhodes readers and agents can tell a facilities value or work unit changed yesterday from one set in August. Arthur flagged that gap (RH06) on the path to Rhodes being the reference DSS.

## Manual Effort Estimate

~1.5 hours (proposed; Keval to confirm or adjust).

## Verification

- typecheck (app and Convex), biome: pass

- vitest: portfolioDomain.test.ts (31, including a new lastChangedAt assertion after an audited profile write), workManagement.test.ts (19), site-profile/work-management/openapi/response-envelopes node tests: 80 pass

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#2325 — 1445-runner-consumer-cutover @mwrshah  no labels

- Rename consumer modules, contracts, schema identifiers and exact machine values to Runner while preserving Executor, /v1/forge, stored prose and immutable pins/hashes.

- Add guarded migration, scoped rollback checks, export scanners and evidence-bound orphan-projection retirement tooling.

Coordinated cutover required: depends on [Sindri #233](https://github.com/AI-Builder-Team/Sindri/pull/233). Merge/deploy only after coordinator recovery/R00 and concrete repair approval.

#137 — 2057-runner-cutover-retirement @mwrshah  no labels

- Mark the temporary feedback recovery and GitHub marker helpers with RETIRE_AFTER_CUTOVER(runner-cutover-retirement).

- Add the exact retirement inventory and removal sequence to the existing runbook; require verified migration of all in-scope deployments and explicit coordinator rollback-window closure. Preserve durable runtime code and all existing gates.

- Validation: repository and Runner Biome, TypeScript, Knip and Codedecorum checks passed; commit hooks and comment-only verification passed. No runtime changes or added tests.

#234 — 1350-cutover-retirement @mwrshah  no labels

- Mark one-time migration adapters and offline helpers for removal after verified cutover and explicit rollback-window closure.

- Add native and cross-repository retirement inventories, coupled cleanup steps, and current effective PR references.

- Comments and documentation only; preserve runtime behavior and recovery evidence.

#80 — Publish the Dec 1, 2026 cutoff for legacy rhk_ keys (AI-1379) @kevalshahtrilogy  no labels

Linear: [AI-1379](https://linear.app/builder-team/issue/AI-1379) (part of [AI-1363](https://linear.app/builder-team/issue/AI-1363)) · Arthur's R-KEY01 ask 4

## What changed

The retirement date for legacy rhk_ keys, 2026-12-01, is now stated wherever a reader meets keys:

- the /skill feedback clause (it replaces "Existing legacy keys remain valid until revoked")

- the OpenAPI authentication description

- dictionary §1 Authentication

- /me for legacy callers

It's held once as service.legacyKeyCutoff for the code paths.

Out-of-band migration work, already done on 2026-11-16:

- every human holder has dss-rhodes, so all can self-serve a personal key

- the quality-loops job and the DSS registry seat moved to 90-day personal read keys

- the feedback-operator role exists for Keval and Colin, which pairs with #78

On Dec 1 the remaining keys get revoked with pnpm keys revoke.

## Business Value

Gives Arthur's requested retirement date a public home. Every legacy key holder and agent sees it on /me and /skill before access stops, so on Dec 1 every Rhodes call becomes traceable to one person.

## Manual Effort Estimate

~20 minutes for this PR (proposed; Keval to confirm or adjust). The migration work around it (role changes, key rotation, registry seal) was about 1.5–2 hours.

## Verification

- pnpm check: 100 tests pass

- pnpm build, pnpm synth --no-lookups: pass

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#78 — Let personal keys triage feedback via a WorkOS operator permission (AI-1377) @kevalshahtrilogy  no labels

Linear: [AI-1377](https://linear.app/builder-team/issue/AI-1377) (part of [AI-1363](https://linear.app/builder-team/issue/AI-1363)) · first step of rhk_ retirement

## What changed

- /operator/feedback now also accepts a WorkOS personal key whose holder's role grants dss-rhodes:feedback-operator, checked live on each call. Legacy operator keys still work.

- Authorization now exposes the live permission set as hasPermission; requireServiceEntitlement uses the same code path, so its behaviour is unchanged.

- README, self-service FEATURE.md and enablement describe the new operator path.

Manual step after deploy: create dss-rhodes:feedback-operator in the WorkOS environment, add it to an operator role, and assign the role to the operators. Until then, behaviour is the same as today.

## Business Value

Triage no longer needs a shared rhk_ key. That's the blocker to publishing the rhk_ retirement date Arthur asked for, and to tracing every triage action to one person.

## Manual Effort Estimate

~45 minutes (proposed; Keval to confirm or adjust).

## Verification

- pnpm check: 100 tests pass

- pnpm build, pnpm synth --no-lookups: pass

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#77 — Cap personal keys at 90 days and let a key revoke itself (AI-1376) @kevalshahtrilogy  no labels

Linear: [AI-1376](https://linear.app/builder-team/issue/AI-1376) (part of [AI-1363](https://linear.app/builder-team/issue/AI-1363)) · Arthur's R-KEY01, asks 3 (lifetime) and self-revoke

## What changed

- Lifetime: new personal keys default to 90 days, and the API rejects longer or "never" (the same as School Budgets). The key page presets are 30/60/90 days, the "Never" option is gone, and the custom date is capped at 90 days. Existing no-expiry keys stay valid until rotated or revoked.

- Self-revoke: DELETE /me with a personal key revokes that key in WorkOS at once, so whoever holds a leaked key can kill it. Legacy keys get a 403 pointing to personal keys.

- /me states the lifetime and the revoke route, and shows "a write key also reads" only on write keys (the note was showing on read keys).

- Enablement documents both.

## Business Value

Rhodes keys now expire, and a holder can kill their own key without asking anyone. These are two of Arthur's five key gaps, and they match the "everything expires" and self-revoke principles in his sign-in proposal.

## Manual Effort Estimate

~1–1.5 hours (proposed; Keval to confirm or adjust).

## Verification

- pnpm check: 100 tests pass

- pnpm build, pnpm synth --no-lookups: pass

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#76 — Serve a public resolution on feedback items (AI-1364) @kevalshahtrilogy  no labels

Linear: [AI-1364](https://linear.app/builder-team/issue/AI-1364) · Arthur's RH01 (Rhodes feedback f27c009f)

## What changed

- PATCH /operator/feedback/{id} accepts an optional resolution (4–500 characters): a public status or fix summary the operator writes for reporters.

- GET /feedback and GET /feedback/{id} return the latest resolution as {text, at}, with the upstream name redacted. Open items can carry one too, e.g. "Partly fixed: …".

- Operator notes, bodies and identities stay private. No storage change: the resolution lives on the existing history action.

- /skill, enablement and AGENTS.md describe the field.

After deploy I'll backfill resolutions on the 8 items closed on Nov 13, and add "partly fixed" resolutions on the keys and feedback items.

## Business Value

Reporters and agents can finally see what fixed an item or why it's still open. That closes Arthur's main complaint ("closed without a note") and gives Rhodes a feedback loop people can follow, which he cites as what School Budgets does better.

## Manual Effort Estimate

~1 hour (proposed; Keval to confirm or adjust).

## Verification

- pnpm check: 100 tests pass

- pnpm build, pnpm synth --no-lookups: pass

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#79 — State what site history Rhodes keeps and doesn't (AI-1378) @kevalshahtrilogy  no labels

Linear: [AI-1378](https://linear.app/builder-team/issue/AI-1378) · Arthur's N03 (Rhodes feedback ba1a0de8)

## What changed

Dictionary §2 gains a History coverage paragraph. Rhodes answers current state only: there are no as-of reads, no dated snapshots and no versioned definitions. It lists the history that does exist (audited /change-history, which omits copies and imports; dated document versions). A past-dated answer is "not reconstructable from Rhodes" unless that exact value appears in one of those.

This takes N03's "narrow the stated coverage" option. As-of reads would be a separate source feature.

## Business Value

Agents stop implying past states Rhodes can't support, which was Arthur's concern. The doc also makes explicit what would need building if as-of answers become a priority.

## Manual Effort Estimate

~20 minutes (proposed; Keval to confirm or adjust).

## Verification

- pnpm check: pass

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#136 — 2056-runner-rebrand @mwrshah  no labels

- Rebuild the Sindri → Runner DSS rebrand on current main, including the strict native contract, service identifiers, auth scope and release wiring.

- Preserve landed GitHub feedback links/mirroring, recipe-owned bootstrap generation and live OpenAI reporting; include feedback cutover/recovery tooling.

- Replaces #130. Companions: [Sindri #233](https://github.com/AI-Builder-Team/Sindri/pull/233), [Aerie #2325](https://github.com/AI-Builder-Team/Aerie/pull/2325), [Surtr #2269](https://github.com/AI-Builder-Team/Surtr/pull/2269), [localskills #1](https://github.com/mwrshah/localskills/pull/1).

Coordinated cutover required: activate with the companion changes, verified Runner secret/provider bindings, recipe-generated bootstrap, and frozen feedback/marker migration.

#233 — 1348-sindri-rename-inventory @mwrshah  changes requested

- Rename native UI, API, toolkit and infrastructure contracts to Runner while preserving Executor behavior and optional version headers.

- Add bounded machine-contract and credential-binding migrations, reference parity and rollback tooling; preserve incidental content and immutable bytes.

- Coordinated cutover required: do not deploy standalone. Depends on aligned auth/Vault, infrastructure, consumers, recovery and R00 validation.

- Validate lint/typechecks, native and Executor suites, offline migration tooling, and the production build.

#2269 — 096-runner-cutover @mwrshah  changes requestedmanual-dev

- Map workUnits.runnerWorkUnitId to runner_work_unit_id; reject unmigrated source documents and preserve historical payloads.

- Add atomic snapshot/rename/rollback SQL, complete-row recovery capture/comparison and before/after schema handling.

- Coordinated migration dependency: freeze source/ETL writers, accept recovery/R00 and migrate Aerie source plus warehouse/loader together before release. Keep merge/release gated and manual-dev set: merging into main can enter the automatic production release flow.

- Related toolkit: https://github.com/mwrshah/localskills/pull/1

#131 — 2049-dss-feedback @mwrshah  no labels

- Mirror Salesforce, Linear, Sindri and Klair feedback through separate service-owned Lambda workers using the GitHub App.

- Prefix issue titles with [server-slug], persist issue/comment links and replay-safe bookkeeping in DynamoDB, and expose nullable githubIssueUrl in public list/detail responses.

- Add isolated streams, retained failure storage, scoped worker boundaries and complete deployment-policy migration paths while preserving existing table identities.

- Add Klair OAuth-authenticated feedback intake without changing its MCP authorization contract or exposing private report data through public reads.

- Adopt recipe revision 9b124f7, move reusable bootstrap generation out of this hub into the recipe CLI, and document the separately approved IAM preparation and production-release steps.

#133 — KLAIR-3765 Add live OpenAI reporting to Klair DSS @ashwanth1109  no labels

## Business Value

Lets the eight approved Klair DSS pilot users query raw OpenAI costs and usage directly. Klair uses the OpenAI accounts Surtr already covers and picks up rotated credentials from the shared secret.

## Implementation Effort

Approximately two engineering days to import the reporting schema, implement and test MCP routing and key rotation, and scope and verify production access.

## Verification

- All 12 reporting operations passed the authenticated MCP integration with mocked AWS and OpenAI. Covered account discovery, pagination, key rotation, and authorization.

- CDK synthesis verified the exact secret and current-version task-role policy.

- Scoped lint, TypeScript, and pre-commit checks passed.

## Linear

https://linear.app/builder-team/issue/KLAIR-3765/enable-live-openai-cost-and-usage-reporting-in-klair-dss

#128 — KLAIR-3764 Restrict Klair DSS to approved pilot users @ashwanth1109  no labels

## Business Value

Gives the eight approved pilot users signed-in Klair DSS and MCP access while limiting access for the rest of the shared DSS organization.

## Implementation Effort

Approximately 2 hours for an engineer to resolve the accounts, add the authorization check, and verify the browser and MCP paths.

## Verification

- Checked the eight addresses and exercised browser-session and MCP authorization with mocked WorkOS identities and signed test tokens.

- Scoped Biome, Klair TypeScript, and pre-commit checks passed.

## Linear

https://linear.app/builder-team/issue/KLAIR-3764/restrict-klair-dss-to-the-approved-pilot-users

#2267 — Map reviewed QuickBooks accounts blocking QTD Facilities @ashwanth1109  approved

## Summary

- Register 39 reviewed QuickBooks realm and account pairs against the accepted Nov 15 snapshot. Apply the existing Headcount policy to seven contracted-labor GL codes and the existing Other non-headcount costs policy to GL 69120, Abandoned Projects Write-off.

- Require the reviewed snapshot and source account fingerprint, unanimous prior policy for each GL code, and a reconciled publication marker. Keep the existing QTD Facilities publication checks in force.

## Business Value

Restore daily QTD Facilities reporting while retaining the classification check that prevents unreviewed QuickBooks expenses from entering the Facilities totals.

## Implementation Effort

Approximately 3–4 engineer hours to trace the rejects to their source accounts, reconcile existing policy, write the migration, and validate the rebuild.

## Validation

- Reproduced 61 rejected postings across the 39 reviewed accounts against the 76-company QuickBooks snapshot. Confirmed the two classifications with current v1 policy.

- Exercised the migration twice in a rollback transaction and applied it in production. The current policy mapping count and publication marker both moved from 3,678 to 3,717.

- Ran the existing QuickBooks Core test selection for tests that do not require Redshift. 214 passed.

- QuickBooks Core and the automatically triggered Aerie refresh both succeeded against the reviewed snapshot. Earlier attempts encountered the hourly school mapping rollover and then a Redshift lock held by the scheduled HubSpot Admissions refresh. The successful retry followed the release of that lock. [QuickBooks Core](https://us-east-1.console.aws.amazon.com/states/home?region=us-east-1#/v2/executions/details/arn:aws:states:us-east-1:479395885256:execution:pipeline-quickbooks-core-tables-prod:surtr-1710-facilities-after-admissions-20261115) · [Aerie refresh](https://us-east-1.console.aws.amazon.com/states/home?region=us-east-1#/v2/executions/details/arn:aws:states:us-east-1:479395885256:execution:pipeline-mart-aerie-education-financials-refresh-prod:745ebd95-c1bf-0352-e665-6541023a1e9b_ce410d4e-9fdf-8a26-6a20-acd7a352b4e5)

- Redshift reconciled all 400,092 Core postings and their signed and absolute totals to the pre-change snapshot. The 57 contracted-labor postings resolved to headcount_costs, and the four Abandoned Projects Write-off postings resolved to non_headcount_cost_other. Each of the 61 amounts and posting sides matched its pre-change value.

- QTD Facilities published 12,144 rows for 69 schools as of 2026-11-15. The deployed reject predicate returned zero postings. The runner's post-publication checks returned zero invalid groups and the mart references the verified Core run and source snapshot.

## Linear

[SURTR-1710](https://linear.app/builder-team/issue/SURTR-1710/restore-qtd-facilities-publication-by-mapping-the-reviewed-quickbooks)

#2266 — Restore daily School P&L publication after the Miami Beach split @ashwanth1109  approvedmercy-allow-critical

## Summary

- Version the three Miami Beach QuickBooks Unit Economics mappings so they agree with the approved Aerie assignment to 71st St. Preserve the previous Biarritz rows and the current model.

- Add a read-only Surtr pipeline that checks school-scope transaction detail and breakdown at 10:00 UTC daily and reports conflicting realm and class assignments after each successful Aerie ontology refresh. Route monitor failures and timeouts to Ashwanth, Marcin, Sanket, and Edie through the existing Surtr SES path.

- Match the Aerie annualization post-check to its deployed SQL contract so schools without a Program publish actuals with annualized measures unset.

## Business Value

Finance can trust that 71st St costs land under the intended school, and the team can detect a missed School P&L publication even when the rebuild never starts.

## Implementation Effort

Approximately 5–7 engineer hours to diagnose the mapping drift, implement the migration and checks, and validate the live publication chain. Long-running QuickBooks processing adds elapsed time.

## Validation

- Redshift rollback probe succeeded through all migration statements before the intentionally failing final statement. The production migration committed and the live comparison now reports zero school assignment conflicts.

- The live mapping monitor succeeded. A direct freshness probe correctly reported both Nov 10 dates, and the deployed failure and timeout handlers contain the four intended recipients.

- Deployed the monitor with --exclusively and verified the daily and post-ontology rules. Updated the shared failure handler IAM policy through a one-resource change set so SES can authorize the verified klair.ai domain for the configured sender and four recipients. AWS policy simulation verified the allowed route and denied an alternate sender and an extra recipient.

- CDK schema, Step Functions, and shared stack suites passed (155 tests).

- QuickBooks and Core succeeded on the Nov 14 publication for 76 companies. Redshift confirms both Aerie P&L marts carry Nov 14. The three corrected 71st St classes and three Biarritz classes reconciled to signed Core totals with no mismatches.

- The first Aerie run published the marts, then its older validation rejected 100 Program-less quarter rows where the procedure intentionally left annualization null. The corrected validator found zero missing eligible values on the live publication. The Aerie stack update completed and 71 focused tests passed.

- The retry with current mapping lineage republished both School P&L marts and got past the annualization check. QTD Guide and All Other Headcount ran before QTD Facilities rejected 61 current-quarter QuickBooks postings without approved account mappings. The live freshness monitor succeeded. The full Aerie workflow is still blocked on the Facilities account-policy review.

- Production recipient routing and SES permission were verified in the deployed state machine and IAM policy simulator. Email delivery was not triggered for validation.

## Linear

[SURTR-1709](https://linear.app/builder-team/issue/SURTR-1709/restore-daily-school-pandl-publication-after-the-miami-beach-split)

#3941 — feat(core-budgets): expose the NetSuite actuals refresh timestamp @ashwanth1109  approved

## Summary

- Add nullable actuals_refreshed_at to core_budgets.consolidated_budgets_and_actuals.

- Capture one timestamp with time zone when the consolidation loads NetSuite Actuals and reuse it for every Actual row. Write NULL for Budget rows.

- Apply the column and replacement procedure together in a short transaction, then run the refresh separately.

## Business Value

Finance can see when Klair last refreshed its consolidated NetSuite Actuals and use that timestamp when interpreting reports.

## Implementation Effort

About 2–3 engineer hours to review the live procedure and its consumers, implement the migration, and validate a production refresh.

## Validation

- The deployed writer matched the repository before the change. Inspected live Redshift dependencies and Klair's export path. Found one dormant sandbox copy whose destination table is absent.

- Applied the two warehouse changes in one transaction outside the recurring refresh window.

- Verified a committed refresh stamped 12,284,933 Actual rows with one value at 2026-11-15 10:10:55 UTC and left all 1,543,020 Budget rows null.

- Reconciled row counts, signed totals, and positive and negative subtotals across all nine source and version groups using the same input snapshot.

- All 22 Core Budgets tests passed.

## Linear

[KLAIR-3763](https://linear.app/builder-team/issue/KLAIR-3763/expose-the-netsuite-actuals-warehouse-refresh-timestamp)

#126 — KLAIR-3762 Document dashboard design principles in the BrainLift @ashwanth1109  no labels

## Summary

- Add DOK4-006 on using recurring stakeholder questions to guide the FAQ and thin dashboard, with the choice to use a dashboard or an agent.

- Add DOK4-007 on progressive disclosure, with decision-changing caveats visible in the default view.

- Link both points of view to six general DOK3 insights and six DOK2 research entries with Klair examples. Credit the new sources in Experts.

- Document the placement of BrainLift examples in Klair DSS AGENTS.md.

## Business Value

Give Finance stakeholders and Klair DSS builders a shared basis for deciding what the FAQ and dashboard should cover, what stakeholders need to see at a glance, and where they can find supporting detail.

## Implementation Effort

About 3–4 hours for an engineer to find and verify sources, draft and review the DOK entries and Klair examples, update the contributor guidance, and validate the BrainLift links.

## Verification

- Verified the DOK4, DOK3, and DOK2 links and all six Experts credits using the BrainLift parser.

- Pre-commit checks passed for Code Decorum, TypeScript, and Knip.

- Verify the published BrainLift during the Klair production release.

## Linear

[KLAIR-3762](https://linear.app/builder-team/issue/KLAIR-3762/document-demand-driven-and-progressively-disclosed-dashboards-in-the)

#124 — KLAIR-3761 Document Klair DSS domain guidance and authorization @ashwanth1109  no labels

## Summary

- Document the domain-specific agent handoff, on-demand domain guidance, and cross-domain lookup rules in the Klair DSS BrainLift.

- Link the SPOVs and insights to the DOK2 evidence, with contributor credits in Experts.

- Add Klair-specific guidance on choosing and applying DOK2 research across fields.

## Business Value

Finance users can start agent investigations from the domain they chose. Klair DSS builders have an evidence-backed design for keeping domain guidance focused and helping agents find related definitions and source records when a question spans domains.

## Implementation Effort

About 3–4 hours for an engineer to verify the sources, write the entries and credits, and check the evidence links by hand.

## Verification

- Verified that the BrainLift reader returns the updated SPOVs and insights with valid supporting links.

- Pre-commit checks passed.

## Linear

[KLAIR-3761](https://linear.app/builder-team/issue/KLAIR-3761/document-domain-guidance-and-authorization-in-the-klair-dss-brainlift)

#123 — KLAIR-3759 Document Klair DSS connector reset for retesting @ashwanth1109  no labels

## Summary

- Add the reset skill to the Klair DSS developer docs and link it from Enablement documentation.

- Reset Codex registration and saved OAuth credentials by exact local or production MCP URL so each setup can be retested independently.

## Business Value

Developers can repeat the first-time connector setup flow for either environment and verify what a new user sees.

## Implementation Effort

Approximately 30 minutes to document and validate the reset workflow by hand.

## Verification

- Validated the skill and confirmed it matches the installed copy.

- Passed repository pre-commit checks.

## Linear

[KLAIR-3759](https://linear.app/builder-team/issue/KLAIR-3759/add-ai-budget-tracking-agent-enablement-to-klair-dss)

#75 — Add GET /me and say how to get a key on every 401 (AI-1373) @kevalshahtrilogy  no labels

Linear: [AI-1373](https://linear.app/builder-team/issue/AI-1373) (split from [AI-1363](https://linear.app/builder-team/issue/AI-1363)) · Rhodes feedback f3db5f9c (Arthur Michel)

## What changed

- GET /me (and /api/me) with a key returns:

- the holder;

- read or write tier and permissions;

- key id, name, creation date and expiry.

It never echoes the key. Legacy rhk_ keys report as read-only, with a note to move to a personal key. lastUsedAt is not tracked yet. expiresAt is null for a key with no expiry, and absent if the identity provider doesn't report it.

- Every 401 now includes details.getAKey, pointing to the DSS home page to sign in and create a personal key.

- The front links how.me, and enablement documents both changes.

- One contract-level test covers /me and the 401 hint. The CD005 exception was approved by Keval.

Still open on AI-1363: self-serve POST /keys, key lifetime and revoke route, rhk_ retirement and agent keys.

## Business Value

Key holders and agents can confirm what a key can do without trying a write, and anyone hitting a 401 learns how to get access. These are the first two pieces of Arthur's self-serve key model for Rhodes.

## Manual Effort Estimate

~1 hour (proposed; Keval to confirm or adjust).

## Verification

- pnpm check: 99 tests pass

- pnpm build, pnpm synth --no-lookups: pass

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#74 — Show feedback titles on the public listing and items (AI-1374) @kevalshahtrilogy  no labels

Linear: [AI-1374](https://linear.app/builder-team/issue/AI-1374) (split from [AI-1364](https://linear.app/builder-team/issue/AI-1364)) · Rhodes feedback f27c009f (Arthur Michel)

## What changed

- GET /feedback and GET /feedback/{id} now include each item's title, so readers can see what's already reported and avoid filing duplicates.

- Bodies, reporter identities and action notes stay private.

- The disclosure boundary is updated everywhere it's stated: AGENTS.md, README, docs/self-service/FEATURE.md, enablement, and the /skill feedback listing text.

- Before shipping, all 156 stored titles were scanned. None contains credentials or person-bearing data; the only person named is the service owner.

Still open on AI-1364: a status reason and, for closed items, what fixed it.

## Business Value

Reporters and agents can check whether a problem is already filed and follow its state. This cuts duplicate reports and answers Arthur's ask for visible triage on the path to Rhodes being the reference DSS.

## Manual Effort Estimate

~30–45 minutes (proposed; Keval to confirm or adjust).

## Verification

- pnpm check: 98 tests pass

- pnpm build, pnpm synth --no-lookups: pass

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#72 — Serve the Edu Ops BrainLift text on /brainlift (AI-1366) @kevalshahtrilogy  no labels

Linear: [AI-1366](https://linear.app/builder-team/issue/AI-1366) · Rhodes feedback 20ebde85 (Arthur Michel)

## What changed

- /brainlift now serves the Edu Ops Rhodes BrainLift verbatim (exported 13 Nov 2026, approved by Edu Ops), instead of a page that only linked to a Google Doc that returns 401 without a Google sign-in.

- A provenance header keeps the Doc link as the origin, and records the export date and the sha256 of the exported text. Service ownership and the foundation disclosures are unchanged.

- brainlift.md is excluded from Biome formatting so the export stays byte-exact.

- New test: the export below the rule hashes to the pinned sha256.

- Regenerated metadata.json (document size, tokens, sha).

The export contains no upstream platform name.

## Business Value

Any Rhodes key holder, human or agent, can now read the beliefs that the dictionary and enablement derive from, without a Google account. This closes one of Arthur's DSS-guide gaps on the path to Rhodes being the reference DSS.

## Manual Effort Estimate

~1 hour (proposed; Keval to confirm or adjust).

## Verification

- pnpm check: 99 tests pass

- pnpm build, pnpm synth --no-lookups: pass

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#73 — Re-pin the DSS contract and check the pin daily (AI-1365) @kevalshahtrilogy  no labels

Linear: [AI-1365](https://linear.app/builder-team/issue/AI-1365) · Rhodes feedback 44d91d02 (Arthur Michel)

## What changed

- src/source/contract-pin.json re-pinned to the live contract: sha256 b49bc624… (X-Doc-Version 8b2cac1). /skill why.sha256 now passes registration predicate 3.

- I re-read the current contract. The front already links everything clause 6 asks for (BrainLift, dictionary, enablement, native API and docs, feedback), so the front needs no other change. The substantive asks in clauses 1, 5 and 7 (served BrainLift, keyed eval) are tracked separately in AI-1366 and AI-1367.

- scripts/repin-contract.ts gains a --check mode (pnpm contract:check) that fails on drift.

- New .github/workflows/contract-pin.yml runs that check daily at 06:17 UTC and on demand, so a contract change surfaces as a failed run instead of a registry finding.

## Business Value

Rhodes passes the registry's contract-pin check again, and future contract changes get caught within a day rather than by a reviewer.

## Manual Effort Estimate

~30–45 minutes (proposed; Keval to confirm or adjust).

## Verification

- pnpm contract:check: pin matches

- pnpm check, pnpm build, pnpm synth --no-lookups: pass

After deploy, the registry's next scheduled check should record the new sha.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#121 — KLAIR-3759 Add AI Budget Tracking enablement to Klair DSS @ashwanth1109  no labels

## Summary

- Add AI Budget Tracking and its Enablement subtab to the signed-in Klair DSS sidebar.

- Copy a short setup prompt whose guide directs the agent to register or reuse the shared connector, start OAuth, and load domain guidance.

- Serve the starter skill and dictionary from a read-only MCP tool protected by WorkOS OAuth and active DSS membership.

## Business Value

Finance can validate agent onboarding and AI Budget Tracking domain selection in Klair DSS before live budget and usage tools are available.

## Implementation Effort

Approximately 1–2 engineer days to implement and verify by hand.

## Verification

- Scoped Biome checks, TypeScript, Knip, and the Klair UI build passed.

- Local checks covered sign-in visibility, the clipboard prompt, connector discovery, OAuth audience and membership, and guidance retrieval.

- A local Codex CLI session completed WorkOS OAuth and retrieved both guidance documents.

## Linear

[KLAIR-3759](https://linear.app/builder-team/issue/KLAIR-3759/add-ai-budget-tracking-agent-enablement-to-klair-dss)

#119 — KLAIR-3758 Restore the Klair DSS BrainLift feedback link @ashwanth1109  no labels

## Summary

- Restore the Google Docs feedback link as an accessible document icon in the Klair DSS BrainLift toolbar.

- Show an amber warning that the manually synced copy may lag behind the hosted BrainLift.

- Describe the agent Markdown as the BrainLift overview and show the Google Docs Paste from Markdown instruction as a blue info callout.

## Business Value

Readers can open the feedback document directly from Klair DSS and know when to rely on the hosted version. The tooltip also explains how to paste a fresh Markdown copy into Google Docs.

## Implementation Effort

Approximately 1–2 engineer hours to implement and verify by hand.

## Verification

- Biome check on the three changed files

- Klair TypeScript check and Vite production UI build

- Confirmed the link, tooltip text, and icons on the running local page

## Linear

[KLAIR-3758](https://linear.app/builder-team/issue/KLAIR-3758/restore-the-google-docs-feedback-link-in-klair-dss-brainlift)

#117 — KLAIR-3757 Keep the BrainLift sidebar outline visible @ashwanth1109  no labels

## Summary

- Style the BrainLift sidebar as an inline outline with a dot for the current section.

- Keep section links visible on Quick Tour and open the selected section directly from there.

- Lay out the section links in two columns on narrow screens.

## Business Value

Readers can jump directly to any BrainLift section from Quick Tour and see where they are as they read.

## Implementation Effort

About 2–3 hours for an engineer to implement the layout, wire up scroll and deep-link state, and verify navigation across viewport sizes.

## Verification

- Passed the pre-commit checks and the Klair TypeScript check.

- Built the Klair UI in memory and tested Quick Tour, all six section links, focus, and the existing Back to top control against the latest main.

## Linear

[KLAIR-3757](https://linear.app/builder-team/issue/KLAIR-3757/keep-the-brainlift-outline-visible-in-the-klair-dss-sidebar)

#116 — KLAIR-3756 Improve Klair DSS BrainLift navigation @ashwanth1109  no labels

## Summary

- Rename the main heading to Overarching Brainlift.

- Shade the main section headers using the active theme.

- Show a Back to top button after 320 pixels of scrolling and return keyboard focus to the main heading when activated.

## Business Value

Readers can spot section boundaries while scanning the BrainLift and return to the top in one action after reading a long section.

## Implementation Effort

About 1 hour of manual engineering, including implementation and verification.

## Verification

- Biome, TypeScript, Knip, and Codedecorum passed.

- Scripted checks passed for the scroll threshold, deep-link initialization, click reset, keyboard focus, and Quick Tour visibility.

- The running local Klair preview served the renamed heading, shaded header styles, and back-to-top control.

## Linear

[KLAIR-3756](https://linear.app/builder-team/issue/KLAIR-3756/improve-navigation-in-the-klair-dss-brainlift)

#115 — AI-1362 Clarify DSS Hub agent instructions @ashwanth1109  no labels

## Linear

[AI-1362](https://linear.app/builder-team/issue/AI-1362/keep-dss-hub-agent-instructions-concise-and-scoped)

## Business Value

Agents spend less context on Klair and Linear DSS guidance and can find the relevant rules quickly. New chats start with current instructions, PR links stay in chat by default, and users decide how to handle occupied local ports.

## Changes

- Group repository-wide guidance and document how to scope rules to individual services.

- Tighten the Klair and Linear DSS instructions while preserving each service’s UI and local sign-in rules.

## Validation

- Checked the three Markdown files and both service links.

- git diff --check and pre-commit checks passed.

## Implementation Effort

About 30 minutes by hand, including reviewing the three files and checking that the shorter wording preserves the agreed behavior.

#114 — AI-1361 Start Linear DSS frontend and backend together @ashwanth1109  no labels

## Linear

[AI-1361](https://linear.app/builder-team/issue/AI-1361/start-the-linear-dss-frontend-and-backend-with-one-dev-command)

## Business Value

Developers can test Linear DSS from any checkout with one command. The frontend and API share the HTTPS origin registered for WorkOS, so sign-in and authenticated browser flows can be tested together.

## Changes

- Add pnpm --filter @dss-hub/linear dev to start Vite, Express, and Caddy at https://localhost:3000.

- Check the UI and API health before reporting readiness. Stop the stack together on interruption.

- Document local environment setup. The launcher lives outside the production TypeScript entry point and runtime image.

## Validation

- Frozen lockfile install, TypeScript, scoped Biome and CodeDecorum checks, and Knip passed.

- Verified that the launcher composes with the OAuth onboarding draft in a clean Git merge.

- Live local startup and WorkOS sign-in are pending. Port 3000 is occupied by another Caddy process.

## Implementation Effort

About half an engineering day by hand, including process lifecycle, HTTPS and proxy integration, documentation, and verification.

#2323 — fix(tasks): scroll only the Task list @caina-barbosa  approved

## Summary

Keep the header and controls visible while scrolling through results in Team Tasks and My Tasks. The previous fix made the whole page scroll.

Production effect: fixes scrolling in the live Task Board.

## Why

Users need to change filters, search, or switch views without scrolling back to the top.

## Business Value

- Keep search, filters, counts, and view controls within reach.

- Continue loading Team Tasks as the user approaches the bottom of the list.

## How does it work

The Tasks page uses a height-constrained column. Only the results section scrolls. The Team Tasks prefetch observer uses that section as its scroll root.

## Scope

### Included in this phase

- Team Tasks and My Tasks layouts and their existing browser tests.

chat/components/task-board/my-tasks-view.test.tsx

chat/components/task-board/my-tasks-view.tsx

chat/components/task-board/team-tasks-view.test.tsx

chat/components/task-board/team-tasks-view.tsx

## Test plan

### Automated validation

- 30 focused tests passed with pnpm --dir chat exec vitest run components/task-board/team-tasks-view.test.tsx components/task-board/my-tasks-view.test.tsx --maxWorkers=2.

- Tests verify that the header navigation and controls are outside the list scroll region and that Team Tasks still prefetches the next page.

### Time for Implementation

About half a day including verification.

#112 — KLAIR-3755 Add a Quick Tour to Klair DSS @ashwanth1109  no labels

## Summary

- Add Quick Tour above BrainLift in the Klair DSS sidebar.

- Explain that BrainLift guides the development of Klair DSS and that the tour directs users to sign in for access by domain of interest.

## Business Value

New users can quickly understand what BrainLift is for and where to go to access Klair DSS.

## Implementation Effort

About one hour for an engineer to add the sidebar view, connect navigation, and verify the tour.

## Verification

- TypeScript and Biome checks passed.

- Checked the two-paragraph copy, BrainLift link, and view navigation locally.

## Linear

[KLAIR-3755](https://linear.app/builder-team/issue/KLAIR-3755/add-a-quick-tour-to-klair-dss)

#111 — KLAIR-3754 Add agent evaluation research to the Klair DSS BrainLift @ashwanth1109  no labels

## Linear

[KLAIR-3754 — Add agent evaluation research to the Klair DSS BrainLift](https://linear.app/builder-team/issue/KLAIR-3754/add-agent-evaluation-research-to-the-klair-dss-brainlift)

## Business Value

Finance and the Klair DSS team can use the BrainLift to turn observed answer failures into repeatable evaluations. The research explains how to trace tool calls, replay against a fixed reporting cutoff, and validate graders against Finance judgments. Alphabetical ordering makes the Knowledge Tree easier to scan.

## Changes

- Add DOK2 research on observability, failure analysis, replay, grader validation, and changes to grading criteria, with Klair examples and Experts attribution.

- Display DOK2 entries alphabetically by title and preserve their existing accordion IDs.

## Validation

- Scoped Biome check and Klair TypeScript check passed.

- Rendered the PR checkout and verified 46 alphabetical entries, all seven observability topics, and Experts credit.

- Repository pre-commit checks passed.

## Implementation Effort

About 3–4 engineering hours by hand for source review, drafting and attribution, the sorting change, and validation.

#110 — KLAIR-3753 Document Klair DSS development and UI guidelines @ashwanth1109  no labels

## Summary

- Add services/klair/AGENTS.md with the four Linear DSS rules adapted for Klair: scoped changes, minimal UI, progressive disclosure, and local HTTPS on port 3000.

- Record Klair’s rule to avoid decorative outlines and highlights and preserve visible keyboard focus on interactive controls.

## Business Value

Contributors have a consistent set of rules for keeping Klair’s interface focused and using the expected local sign-in setup.

## Implementation Effort

About 15 minutes for an engineer to review, adapt, and verify the instructions manually.

## Verification

- Compared all four adapted rules with the Linear DSS instructions and verified that the Klair-specific rule was retained.

- Confirmed the PR contains one commit and changes only services/klair/AGENTS.md.

## Linear

[KLAIR-3753](https://linear.app/builder-team/issue/KLAIR-3753/document-klair-dss-development-and-ui-guidelines)

#109 — KLAIR-3752 Run Klair DSS locally with WorkOS sign-in @ashwanth1109  no labels

## Summary

- Replace Klair DSS’s frontend-only preview with a single command that starts the backend and Vite behind Caddy HTTPS on port 3000. Fail early if another server occupies the port.

- Route sign-in, session, logout, health, and BrainLift Markdown through the same backend used in production.

- Keep the signed-in avatar at 32 px on desktop and 44 px on touch devices.

## Business Value

Klair testers can use the real sign-in flow from a stable local URL. Account photos stay within the header and leave more space for the BrainLift.

## Implementation Effort

About 3 engineering hours by hand, including wiring local HTTPS and WorkOS, updating the avatar layout, and verifying the startup and sign-in paths.

## Validation

- Passed Biome, TypeScript, Knip, and Codedecorum pre-commit checks.

- Verified that startup reports a port collision on port 3000 and that standalone Vite commands direct users to the full-stack launcher.

- Ran the full stack on port 3000 and verified the BrainLift, backend health, signed-out session, and WorkOS redirect with the local callback.

- Verified that the running app serves the updated avatar sizing rules.

## Linear

https://linear.app/builder-team/issue/KLAIR-3752/run-klair-dss-locally-with-sign-in-and-a-compact-account-avatar

#107 — KLAIR-3751 Add a theme picker to Klair DSS @ashwanth1109  no labels

## Summary

- Replace the Light and Dark buttons with an in-page palette panel that shows color swatches and the active selection.

- Provide Classic Light (the existing light palette), Parchment, Sea Glass, Slate, Charcoal, Graphite Mint, and Espresso Amber.

- Remember explicit choices, migrate legacy Light and Dark preferences, and follow device appearance in System mode.

## Business Value

BrainLift readers can choose a comfortable palette for long reading sessions, compare alternatives in place, and return to their preferred appearance on reload.

## Implementation Effort

Approximately 5–7 engineer hours to design the palettes, implement the picker and preference migration, and verify theme rendering and contrast.

## Verification

- Passed lint, TypeScript, Knip, and the Klair UI build.

- Verified each palette and swatch, both System defaults, preference persistence, and migration of legacy values in the served bundle.

- Verified the signed-out build still displays Sign in.

## Linear

[KLAIR-3751](https://linear.app/builder-team/issue/KLAIR-3751/add-a-theme-picker-to-the-klair-dss-brainlift)

#70 — Publish standalone diligence site creation (AERIE-3116) @kevalshahtrilogy  no labels

## Summary

- Syncs the contract for POST /portfolio/sites. mainSiteRef is now optional, and without it the source creates a standalone diligence site. This change is live in the source's production as of today (AERIE-3116).

- Enablement explains how to open diligence on a new site:

- send name, address, marketId and schoolSizeClass;

- the owner needs global site-edit authority;

- a duplicate address returns 409 site_address_exists.

Stacked on #69 because both PRs change the contract snapshot. Merge #69 first.

Linear: AERIE-3116 (Rhodes ticket to follow; Linear was unavailable)

## Business Value

REBL3 provisioning has created no sites since 10/26, so RE Ops couldn't open diligence on new sites: Aspen, Boca, McAlpine, Cliffdale and 3000 Jones Rd. With this, Robbie and JC can create them themselves through Rhodes (JC's ask #7). Briarhaven's offer is due 11/17.

## Manual Effort Estimate

About 1 hour by hand: contract sync and the docs. *Proposed; Keval to confirm.*

## Verification

- pnpm check passes (98 tests), as do pnpm build and pnpm synth --no-lookups.

- The live source contract was checked: the create body now requires only name and address.

- Not verified end to end through Rhodes; that follows deploy.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#69 — Publish the site status write and paused-site lookup guidance (AI-1359) @kevalshahtrilogy  no labels

## Summary

- Publishes PATCH /portfolio/sites/{siteRef}/status for write keys, so an agent can pause, resume or cancel a site as the key's owner. Cancelling requires a reason; If-Match is required.

- Enablement explains that paused and cancelled sites still accept writes. Because the site list returns only active sites by default, agents should look up paused sites with status=paused or status=all.

- The contract is synced from the source: 54 reads and 56 writes. The dictionary count and the relay exclusion test are updated.

Linear: AI-1359 (related: AERIE-3123)

## Business Value

JC's agents reported they "can't write to paused sites". In fact, the source accepts those writes. The agents just couldn't find paused sites, and had no way to change a site's status at all. This closes that gap for RE Ops (JC's ask #8) without any change to the source.

## Manual Effort Estimate

About 1.5 hours by hand: allowlist and contract sync, the docs, and the test updates. *Proposed; Keval to confirm.*

## Verification

- pnpm check passes (98 tests), as do pnpm build and pnpm synth --no-lookups.

- Not yet verified against the live service; acceptance follows deploy.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#2321 — State that standalone Site creation also needs the operation scopes (AERIE-3116) @kevalshahtrilogy  approved

Follow-up to #2319, from Mercy's finding on release #2320.

The contract said a standalone create needs operations.siteFields.write on the key and its owner. It didn't say that the operation's own required capabilities (operations.portfolio.read, operations.driSites.write) still apply. A key holding only siteFields.write is refused before the standalone check runs. This changes only the description, the 403 text and the workflow hint; behaviour is unchanged.

## Business Value

Agents read the contract to choose key scopes. Accurate scope text stops them minting keys that are then refused.

## Manual Effort Estimate

About 15 minutes. Proposed; Keval to confirm or adjust.

## Verification

The site-create tests and public-API contract tests pass (251).

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#68 — Serve read-key reads on the service credential so readers need no source account (AI-1356) @kevalshahtrilogy  no labels

## What

Read keys now read with the shared service credential, gated by the Rhodes WorkOS role (dss-rhodes:access). A person with Rhodes access but no account in the system of record no longer gets 403 delegated_actor_unavailable on every read.

- Read keys: service credential for every read, except three person-scoped reads that still run as the key owner: backup-site contract, site change history, and my-tasks (PERSONAL_READS in src/aerie/proxy.ts).

- Write keys: unchanged. They read and write as their owner, so Task actions and Access Contact match what that person may do.

- Legacy keys: unchanged (service credential).

- A source 403 on a non-delegated read is now treated like any refusal of the service credential and returned as 502 upstream_unavailable.

Docs updated to match: Dictionary auth line, Enablement contract and change-history notes, OpenAPI Task-actions and Access Contact text, and docs/rhodes-caller-quotas/FEATURE.md.

Trade-off: read-key reads now share the service credential's rate allowance instead of per-person quotas.

Trigger: Fernando Perez has no account in the source (none of its 76 users matches his email), so every read failed (request req_bc2da90b2ad48f2ed7c17146c85adfcd). Write keys still need a source actor; provisioning those from Rhodes without a login is a separate follow-up.

## Business Value

People can be given Rhodes read access entirely from our side (WorkOS role and key), with no onboarding onto the source app. This unblocks Fernando today and removes a support step for every future read-only consumer.

## Manual Effort Estimate

About 3 hours by hand: diagnosing the source refusal, the relay change, tests, and doc updates. Proposed; Keval to confirm or adjust.

## Verification

pnpm check (98 tests), pnpm build and pnpm synth --no-lookups pass. The new tests fail on the old relay and pass on the new. CD005 exception for test/writes.test.ts approved by Keval. Not yet exercised against production; Fernando's reads should be retried after deploy.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#105 — KLAIR-3750 Define the scope of the Klair DSS BrainLift @ashwanth1109  no labels

## Summary

- Add the approved In Scope and Out of Scope subsections beneath Purpose in the Klair DSS BrainLift.

- Define the Finance questions and evidence Klair DSS serves, including the Surtr pipelines and data models that supply Klair, and identify where responsibility passes to source-system and workflow owners.

## Business Value

Finance and the teams that support it can see which questions and data responsibilities belong to Klair DSS and where to route operational work.

## Implementation Effort

Approximately 30–45 minutes for an engineer to review the BrainLift, draft and reconcile the scope wording, and verify the Markdown change.

## Verification

- Verified that the BrainLift reader loads both subsections beneath Purpose.

- Verified the diff contains only the approved wording.

## Linear

[KLAIR-3750](https://linear.app/builder-team/issue/KLAIR-3750/define-the-scope-of-the-klair-dss-brainlift)

#103 — Publish Klair BrainLift agent guidance @ashwanth1109  no labels

## Summary

- Add the approved DOK4 points and DOK3 insights with stable IDs and links to their supporting evidence.

- Add DOK2 research on how agents work, Linear DSS onboarding, and WorkOS authorization, with matching Experts credits.

- Keep the BrainLift toolbar visible when following evidence links, and expand collapsed sources when needed.

## Business Value

Finance readers can trace agent guidance to evidence and give agents concise context for exploring raw records, checking roll-ups, and surfacing data gaps.

## Implementation Effort

Approximately 6–8 engineer hours for source review, writing, implementation, and verification.

## Verification

- Verified the rendered DOK4 → DOK3 → DOK2 links and agent-facing Markdown in the local preview.

- Checked in-pane navigation and expansion of collapsed sources.

- Passed Biome, TypeScript, codedecorum, and Knip checks.

## Linear

[AI-1353 · Klair DSS BrainLift](https://linear.app/builder-team/issue/AI-1353/ship-the-klair-dss-brainlift-view)

#102 — Collapse Klair BrainLift Experts by default @ashwanth1109  no labels

## Summary

- Make Experts a native disclosure that starts collapsed.

- Preserve the sidebar jump to Experts and display a chevron that reflects the open state.

## Verification

- Rendered the BrainLift and checked the default collapsed state, all 24 entries, and the sidebar anchor.

- Passed the Klair TypeScript check and the commit hook checks for Biome, TypeScript, Knip, and codedecorum.

## Business Value

Readers can reach the Knowledge Tree without scrolling through the full Experts list and can open the list when they need contributor details.

## Implementation Effort

About 30–45 minutes for an engineer to implement and verify the disclosure and styling.

## Linear

[AI-1353](https://linear.app/builder-team/issue/AI-1353/ship-the-klair-dss-brainlift-view)

#2319 — Create standalone diligence Sites through POST /v2/portfolio/sites (AERIE-3116) @kevalshahtrilogy  approved

## What

POST /v2/portfolio/sites can now create a standalone diligence Site; until now it could only create an additional premise of an existing campus.

Without mainSiteRef:

- marketId and schoolSizeClass are required (422 site_create_invalid otherwise).

- The key and its owner both need operations.siteFields.write; otherwise 403 site_create_forbidden. DRI-only keys can still create premises, but not standalone Sites.

- An address that any Site already has (after normalisation) is refused with 409 site_address_exists. When REBL3 provisioning recovers, it finds this Site by address and renames its slug to REBL3's, rather than creating a second Site.

- The slug is minted from the name, a Drive folder is provisioned, and the audit action is site.standalone.created.

With mainSiteRef, behaviour is unchanged. A premise now rejects schoolSizeClass, because it inherits the main Site's.

## Why

REBL3 → Rhodes provisioning has created no Site since 10/26. RE Ops has diligence requested on Aspen, Boca Raton, McAlpine, Cliffdale and 3000 Jones Rd, and none of them has a record. Briarhaven's offer response is due 11/17; JC called this "hyper important". This PR lets RE Ops and their agents create these Sites themselves, through Rhodes, whatever the state of the sync. The sync failure itself is a separate investigation (Cainã).

## Business Value

Unblocks diligence on at least five sites with live deal deadlines today. New sites no longer depend on a single background sync.

## Manual Effort Estimate

About 4 hours by hand: design, mutation and contract changes, tests. Proposed; Keval to confirm or adjust.

## Verification

- pnpm check (lint and typecheck) passes.

- 65 public-API test files pass (920 tests), including a new standalone test: create, replay, duplicate address, DRI-only refused, missing fields, premise with schoolSizeClass.

- Not yet exercised in production.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#100 — Expand the Klair DSS BrainLift @ashwanth1109  no labels

## Summary

- Publish the Klair DSS owner, purpose, alphabetical Experts list, and 38 expandable Knowledge Tree sources, including 14 dashboard drafts and 22 external sources.

- Add section links in the sidebar, consistent section headers, and bulk expand and collapse controls.

- Keep the review drafts in the local worktree. The integrated BrainLift is the published source.

## Verification

- Local preview served 38 source rows and six working section anchors.

- Verified the BrainLift Markdown and Experts attribution in the local preview.

- Passed the commit hook checks for Biome, TypeScript, Knip, and codedecorum.

## Business Value

Finance can find the right Klair reports, see how their numbers are calculated, and trace them to supporting records. The section links and expandable sources make the reference easier to browse.

## Implementation Effort

About three engineer-days for dashboard source research, evidence checks, BrainLift integration, UI changes, and verification.

## Linear

[AI-1353](https://linear.app/builder-team/issue/AI-1353/ship-the-klair-dss-brainlift-view)

#98 — AI-1353 Add Klair DSS production releases @ashwanth1109  no labels

## Summary

- Add a Klair release entry point for relevant updates to prod and for manual release, resources, and promotion operations.

- Use Klair's dedicated GitHub and CloudFormation roles to build, deploy, and verify its immutable image.

- Record a successful runtime release after ECS digest and HTTPS checks pass.

## Business Value

Klair BrainLift updates get a repeatable production release with an auditable image digest and a verified rollback path.

## Implementation Effort

About one engineer-day to wire the release, scope and provision IAM, and validate the workflow and stack templates by hand.

## Verification

- Actionlint, Biome, TypeScript, and repository pre-commit checks passed.

- Synthesized both Klair stacks and compared the templates with live production resources.

- Built the Linux AMD64 image and exercised release selection and workflow shell steps with mocked AWS and GitHub calls.

- Provisioned and simulated the scoped roles and verified all four variables in the klair-production environment.

## Linear

[AI-1353](https://linear.app/builder-team/issue/AI-1353/ship-the-klair-dss-brainlift-view)

#65 — Sync the contract for diligence updates on cancelled sites (AI-1336) @kevalshahtrilogy  no labels

## Summary

The source's production release now accepts the published Due Diligence PATCH (/portfolio/sites/{siteRef}/property/due-diligence) on cancelled sites. This refreshes the OpenAPI snapshot with pnpm sync:openapi, so the published contract matches:

- The operation description now says patches apply to already cancelled sites.

- The 409 response no longer says cancelled sites are read-only.

src/generated/publication/metadata.json is regenerated for the new contract. No allowlist, relay or auth change.

Snapshot diff is 2 lines after Biome formatting, and it's the only operation that changed.

## Business Value

Robbie Forrest's automated diligence loop can keep every site's diligence current through Rhodes write keys, including cancelled sites, instead of hitting cancelled_site_read_only. Agents reading the Rhodes contract get accurate guidance and don't skip cancelled sites.

## Manual Effort Estimate

About 30 minutes by hand (sync, format, regenerate, verify). Keval to confirm.

## Verification

pnpm check, pnpm build and pnpm synth --no-lookups pass. The pre-commit check and pre-push codedecorum hooks pass. This does not prove the deployed relay behaviour; that follows from the source release.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#2318 — AERIE-3091: Contain additional-acquisitions list failures to the card @marcusdAIy  approved

## Summary

If the portfolio/additionalAcquisitions:list query fails, only the Additional acquisitions card shows an error, and the Site actions menu hides itself. The rest of the Portfolio Site page stays usable.

Fixes [AERIE-3091](https://linear.app/builder-team/issue/AERIE-3091).

## Why it's needed

useQuery throws a query's server error during render. Two components on every Portfolio Site page read this query with plain useQuery and had no local containment:

- ConnectedAcquisitions (inside the Property Acquisition card)

- ConnectedSiteActions (in the Site Detail header)

So a single failure reached the route error boundary, which replaced the whole dashboard with "I let the team know". ConversionForm, in the same file, already avoids this with useQueries plus an instanceof Error check (#2230). This change applies the same pattern to the list query.

## Changes

- additional-acquisitions.tsx

- A new useAcquisitionList hook reads the list through useQueries, so a failure comes back as an Error value instead of being thrown.

- ConnectedAcquisitions shows a local alert built with toUserMessage:

- An intended userError (for example "Too many additional acquisitions to display.") is shown as written.

- Any other failure shows "Could not load additional acquisitions. Reload the page to retry." and never the raw [CONVEX Q(...)] Server Error text.

- Unexpected (non-ConvexError) failures are still reported, through captureAsyncError with surface Portfolio:AdditionalAcquisitions and action load_additional_acquisitions.

- Each distinct message is reported once, because useQueries can return a fresh Error object on each render.

- Without this, containing the error would also hide real server faults from Platform Error triage.

- On failure, ConnectedSiteActions returns null, the same as when the user can't convert. It doesn't report the error itself: the card on the same page subscribes to the same query and reports it once.

- additional-acquisitions.test.tsx

- The useQueries mock now answers by query key.

- New tests:

- An unexpected failure stays inside the card, hides the menu, never shows the raw Convex text, and is reported exactly once.

- An intended ConvexError refusal is shown in the card and isn't reported.

## Breaking changes

None. The loading, not_found and available states are unchanged.

## Limits

- The server-side trigger in the original report wasn't captured, so this PR contains the failure but doesn't explain it. If it happens again, the new report gives a scoped surface and action to group it under.

- The card has no "Try again" button. Convex re-runs the subscription by itself when the cause clears, so the card recovers without a reload in that case.

## Test plan

- [x] pnpm exec vitest run components/dashboards/portfolio components/dashboards/real-estate: 58 files, 845 tests pass.

- [x] Red/green check: with the component change removed, the two new tests fail and the other 14 in the file still pass. With it restored, all 16 pass.

- [x] Pre-commit hooks pass: biome and typecheck-chat.

- [ ] Manual check on a preview deployment: a Portfolio Site page still renders normally, with the acquisitions card and Site actions menu unchanged.

#2317 — AERIE-3095: Classify the duplicate-upload refusal as an expected outcome @marcusdAIy  approved

## Summary

When the Rhodes worker refuses a second Drive upload session because a prior upload may already exist, the Rhodes Mutation Card now raises a typed expected outcome, rhodes.upload.duplicate_pending_unconfirmed. Platform Error capture records it as expected_outcome instead of sending it to unexpected-error triage. The user still sees the same recovery guidance: recover the prior file, or reject and re-propose.

Closes [AERIE-3095](https://linear.app/builder-team/issue/AERIE-3095) (GitHub #2294).

## Why it's needed

The duplicate-upload guard is a deliberate safety stop that already tells the user what to do. It reached the card as an untyped error, so every trip was captured as an unexpected platform error. The other upload-recovery stops in the same flow (staged file missing or unavailable, multiple staged files) are already typed expected outcomes.

## Changes

- packages/contracts.

- Registers rhodes.upload.duplicate_pending_unconfirmed in PLATFORM_ERROR_PUBLIC_EXPECTED_OUTCOME_CODES.

- Moves the refusal text into RHODES_DUPLICATE_UPLOAD_UNCONFIRMED_MESSAGE (rhodes-mutation-proposal.ts), so the worker and the card share one string.

- chat/rhodes-worker/src/index.ts. The guard throws the shared message. Its behaviour and the response are unchanged.

- chat/components/rhodes-mutation-card.tsx. In postDriveUploadApproval's failed-response branch, an error containing that message becomes a PlatformExpectedOutcomeError with the new code. It keeps the relayed text, including its request ID. This sits next to the existing text match for an unavailable staged file.

The worker's response is deliberately left as it is. The Public API v2 document upload also calls this endpoint and maps worker errors itself, so changing the status or shape there would change its behaviour too. The proxy keeps only the error text of a failed worker response, which is why the card matches the shared message.

## Breaking changes

None.

## Test plan

- [x] rhodes-mutation-card.node.test.ts: a failed upload-session response carrying the relayed refusal gives a PlatformExpectedOutcomeError with the new code and the message unchanged. normalizePlatformErrorCapture records it as expected_outcome.

- [x] Red/green: the test fails with the card change stashed and passes with it.

- [x] The other card tests pass (12). The Rhodes worker typecheck is clean and its suite passes (331). The contracts tests platform-errors and rhodes-mutation-proposal pass (44). Biome and the pre-commit typecheck are clean.

#2316 — AERIE-3106: Answer a write-throughput limit on the v2 rate-limit write as a 429 @marcusdAIy  approved

## Summary

When Convex's per-second write ceiling stops a Public API v2 request's rate-limit bookkeeping write, the caller now gets a 429 with Retry-After: 1, not an opaque 500 internal_error. It is no longer captured as an unexpected Platform Error. This covers every v2 operation, because they all share the rate-limit code.

Closes [AERIE-3106](https://linear.app/builder-team/issue/AERIE-3106) (GitHub #2301).

## Why it's needed

consumePublicApiRateLimit only recognised write conflicts (#1823, #2136). Convex's "Too many writes per second" error is a different shape, so it fell through to the request boundary as a 500. Seven Platform Error reviews have been linked to this issue.

## Changes

- convex/publicApi/rateLimit.ts.

- New isConvexWriteThroughputLimit, matching Convex's "Too many writes per second" message. Like the conflict check, it only matches non-ConvexError errors.

- When it matches, consumePublicApiRateLimit answers straight away with the one-second denial that an exhausted write-conflict burst already gets. It doesn't retry, because each retry adds another write to a deployment that's over its ceiling.

- The denial is built in one helper, contentionDenial, shared by both cases. isConvexWriteConflict behaves exactly as before.

## Breaking changes

None. Callers that got a 500 in this condition now get a 429 with Retry-After, X-RateLimit-* headers and the usual rate-limit body.

## Test plan

- [x] rateLimit.test.ts: a write-throughput error gives { ok: false, retryAfterSeconds: 1 } after a single attempt.

- [x] authorizationTimeout.test.ts: end to end on /v2/portfolio/sites, a rate-limit write failing with the write-throughput error gives 429 with Retry-After: 1, and no errorEvents row.

- [x] Red/green: both tests fail with rateLimit.ts stashed (the end-to-end one gets a 500) and pass with it.

- [x] All of convex/publicApi passes (41 files, 677 tests). Biome and the pre-commit typecheck are clean.

## Limits

This fixes the rate-limit write named in the issue. Another v2 write that hits the same ceiling would still reach the boundary as a 500. The boundary's isConvexTransientFailure (in convex/lib/syncDispatch.ts) only matches one exact Convex message, and it's shared with the sync-dispatch routes. Widening it to give a 503 everywhere is a bigger change, so it isn't in this PR.

#2315 — AERIE-3104: Show the reason when a Portfolio field save is refused @marcusdAIy  approved

## Summary

Saving a Portfolio field that the server refuses now shows the server's reason, not "We couldn't save those changes" plus a request ID. The main case is a milestone completion change that needs approval. These refusals also stop reaching Platform Error triage as unexpected 502s.

Closes [AERIE-3104](https://linear.app/builder-team/issue/AERIE-3104) (GitHub #2300).

## Why it's needed

updateSiteFields threw plain Errors for its validation refusals. Convex redacts those to "Server Error" in production. updateAerieSiteFields then relayed the redacted text as a 502, so the client treated it as an outage and captured it. Two captures from one operator, a few seconds apart, opened this issue.

## Changes

- convex/rhodes/dashboard.ts, updateSiteFields. Three refusals a user can act on now throw userError(...):

- a milestone completion change that needs approval;

- a School Chain that isn't available;

- a utilities change without operations.siteFields.write.

The message text is unchanged. "Forbidden" and the internal schoolChain must be a string type check stay as plain errors.

- lib/aerie-rhodes-dashboard-server.ts, updateAerieSiteFields. Caught errors go through toUserMessage.

- A ConvexError returns 422, or 403 for a permission message, with the clean text. The route already returns 422 for its own completion check.

- Anything else stays a 502 with the generic fallback, so a redacted "Server Error" never reaches the client.

## Breaking changes

None. These refusals change from 502 to 422 or 403. The client already shows the message for statuses below 500 and only captures 5xx responses.

## Test plan

- [x] New lib/__tests__/aerie-rhodes-dashboard-server.node.test.ts covers three cases: a user error gives 422 with the message, a permission user error gives 403, and a redacted plain error gives 502 with the fallback.

- [x] rhodesDashboardParity.test.ts: the milestone-completion test now requires a ConvexError carrying the message.

- [x] Red/green: all four tests fail with the source changes stashed and pass with them.

- [x] The suites for rhodesDashboardParity, rhodesMcpMutationParity, convex/automations and the portfolio fields route pass (484 tests). tsc --noEmit, biome, lint:test-architecture and lint:boundaries are clean.

#95 — 2052-api-keys-padding @mwrshah  no labels

- Tighten and balance API key row spacing across Salesforce, Sindri, and Linear.

- Keep metadata beside each key; hide hours and minutes below 768px and truncate long names.

- Keep the name placeholder centered at every width and center the shorter validation message with 4px left padding.

- Preserve real API key loading in all three services; synthetic keys remain confined to the local Example DSS preview.

#94 — 2053-dss-timeout-guidance @mwrshah  no labels

- Require checking the actual ALB idle timeout and keeping the HTTP server keep-alive above it with a safety margin.

- Revalidate after server, Node, proxy, or infrastructure changes.

- Include idle connection-reuse checks and gateway metrics in release validation while preserving separate foundation approvals.

#92 — Add BrainLift Markdown copy and toolbar tooltips @ashwanth1109  no labels

## Summary

- Add Copy Markdown to the BrainLift toolbar using the full published BrainLift.

- Use distinct icons, Floating UI tooltips, keyboard focus, and copy confirmation for the three BrainLift actions.

- Preserve the agent-facing Markdown route and the agent link.

## Business Value

Klair DSS readers can copy the complete BrainLift into Google Docs for sharing and comments. The toolbar explains which actions open or copy the agent context and which one copies the full content.

## Implementation Effort

About 3 hours for an engineer to implement and validate by hand.

## Verification

- Verified exact full Markdown and agent link payloads, clipboard success and errors, and hover, focus, and Escape interactions.

- Passed TypeScript, changed-file Biome and Codedecorum, and Git whitespace checks.

- Reviewed the feature in the local Klair DSS dev UI.

## Linear

[AI-1353 — Ship the Klair DSS BrainLift view](https://linear.app/builder-team/issue/AI-1353/ship-the-klair-dss-brainlift-view)

#90 — 2050-pr49-skill-freshness @mwrshah  no labels

- Remove the native schema fetch from Sindri DSS /skill so discovery does not wait for or fail with the upstream contract.

- Keep schema links and large-file reading guidance with static estimates of ~1.4 MB and ~250,000 tokens; omit live revision, hash, and byte counts.

- Render the large native schema as plain text in the existing viewer, skipping browser-side parsing, reformatting, and token-level syntax highlighting; keep highlighting for other JSON documents.

- Preserve fresh upstream fetching and validation on /skillspec and document the separation.

#91 — 2051-http-timeouts-logout @mwrshah  no labels

- Set 65-second server keep-alive across Sindri, Salesforce, Linear, and Klair, above the shared ALB’s 60-second idle timeout.

- Clear local session and login cookies before logout; derive the WorkOS logout URL from the authenticated cookie seal without token refresh or JWKS requests.

- Return missing or invalid sessions home, preserve cross-origin protections, and navigate away from key management on session expiry without treating provider outages as sign-out.

- Allow Klair’s form redirect through the WorkOS logout endpoint.

#66 — 009-http-timeouts-logout @mwrshah  no labels

- Set a 125-second server keep-alive above the verified Rhodes ALB’s 120-second idle timeout.

- Clear session and login cookies before logout; derive the WorkOS logout URL from the authenticated cookie seal without token refresh or JWKS requests.

- Return missing or invalid sessions home, preserve cross-origin protections, and navigate away from key management on session expiry without treating provider outages as sign-out.

#89 — 2049-dss-feedback @mwrshah  no labels

- Add the DSS Hub Helper wheel logo as SVG and transparent 1024×1024 PNG.

- Add V1 with only the eyes and smile in the same formats.

- Preserve the original hub logo.

#87 — AI-1353 Refine the Klair DSS BrainLift layout @ashwanth1109  no labels

## Summary

- Give Klair DSS a compact, full-height shell with a content-sized BrainLift sidebar and independently scrolling navigation and content.

- Use one disclosure per source. Show the source title, author, and Source link on the row, with all topics, takeaways, supporting facts, and citations inline when expanded.

- Preserve the public BrainLift, account controls, theme preference, and agent-link copying.

## Business Value

Readers can scan the full knowledge tree from a single source disclosure while keeping navigation, theme, and agent-link actions in view.

## Implementation Effort

About 2–3 hours for an engineer to implement and validate by hand.

## Verification

- Built the Klair UI and checked the revised preview with the requester.

- Verified one source toggle and all seven topics, 14 takeaways, 30 supporting facts, and their citations in the local preview.

- Passed changed-file Biome and Codedecorum, Klair TypeScript and Knip, and Git whitespace checks.

## Linear

[AI-1353](https://linear.app/builder-team/issue/AI-1353/ship-the-klair-dss-brainlift-view)

#81 — AI-1351 Add a manual Linear DSS production report test @ashwanth1109  no labels

## Linear

[AI-1351 — Run Linear DSS production report comparisons on demand](https://linear.app/builder-team/issue/AI-1351/run-linear-dss-production-report-comparisons-on-demand)

## Business Value

An engineer can run the saved AI Builder delivery prompt after a Linear DSS change and download the report, deployed revision, and a side-by-side comparison from one Actions run.

## Implementation Effort

Approximately 4–6 hours for an engineer to implement and validate manually.

## Changes

- Add a manual-only workflow that runs on prod, uses GPT‑6.1 Sol with the organization API key, and reads production Linear DSS using the existing validation credential.

- Save the report and verified deployment metadata in an artifact. Restore the latest retained artifact from a successful run to produce the HTML diff.

- Keep the local OAuth-backed report runner available and document both entry points.

## Verification

- [GPT‑6.1 Sol CI smoke](https://github.com/AI-Builder-Team/DSS-Hub/actions/runs/43262730472) passed with the organization API key.

- [Synthetic report CI run](https://github.com/AI-Builder-Team/DSS-Hub/actions/runs/43263911514) generated two reports, verified the second run's HTML diff, and uploaded the artifact.

- Actionlint, TypeScript, Biome, Ruff, and local adapter checks passed.

- The protected environment accepts prod only. Validate the first live dispatch after this workflow reaches prod.

#85 — AI-1353 Add Klair DSS BrainLift @ashwanth1109  no labels

## Business Value

Klair readers can open BrainLift before or after Google sign-in. The Experts section identifies Neeraj Gupta and links his role and profile. His guide appears in an expandable source row with its DOK2 summaries and supporting DOK1 facts.

## Implementation Effort

Approximately two engineer-days by hand for the UI, WorkOS integration, dedicated hosting setup, and verification.

## Summary

- Add the standalone Klair DSS service and its dedicated ECS deployment configuration.

- Render the Experts and DOK2 reference in the browser. Serve the owner context through /brainlift.

- Add a local UI mode that reloads on BrainLift edits.

## Verification

- Checked TypeScript, Biome, Knip for Klair, and Codedecorum.

- Verified the local UI, reference links, source route, and automatic reload.

- Compiled production assets and checked the rendered Experts and collapsed guide row.

## Linear

[AI-1353 — Ship the Klair DSS BrainLift view](https://linear.app/builder-team/issue/AI-1353/ship-the-klair-dss-brainlift-view)

#83 — 2048-sindri-openapi-display @mwrshah  no labels

- Let signed-in Sindri and Salesforce pages grow and scroll around a properly sized document viewer on mobile.

- Keep signed-out viewers within the available viewport, including short and landscape screens.

- Allow mobile document scrolling to chain to the outer page; preserve signed-in desktop viewer sizing.

- Leave Linear's independently developed dashboard unchanged.

- Verified 18 signed-in/signed-out Chrome responsive cases using the public Sindri dictionary; synchronized the recipe template.

- Update Sindri and Trilogy Salesforce prompts to the approved skill-usage and API-authentication wording.

#79 — Show delivery updates in Linear DSS initiative rows @ashwanth1109  no labels

## Business Value

Stakeholders can scan each initiative for an open project that needs follow-up or is underway, the last recorded completion, and dated health from Linear. Opening the row reveals the full project evidence and source links.

## Implementation Effort

About 1.5–2 hours for an engineer to implement, review edge cases, and test manually.

## Verification

- TypeScript, Biome, and code policy checks passed.

- Rendered fixtures covered reported risk, missed target dates, stale health, completed projects, and empty rows.

- The local HTTPS service served the updated UI and returned all seven initiatives with a scoped validation key.

## Linear

[AI-1288](https://linear.app/builder-team/issue/AI-1288/add-linear-dss-initiative-progress-service)

#77 — AI-1350 Lead Linear DSS delivery reports with goals @ashwanth1109  no labels

Delivery reports now lead with each goal and its progress, highlight leads and contributors beside their roles or contributions, and group goals by stakeholder in combined reports. Shared initiatives count once in the combined total. The full BrainLift records the evidence behind the new agent guidance.

## Business Value

Stakeholders can find their goals in combined reports and see who is moving each goal forward without searching through separate lists of people.

## Implementation Effort

About 1.5–2 hours for an engineer to research the reporting practices, update the BrainLift, and verify the full and agent-facing views.

## Verification

- Reviewed the BrainLift locally and verified both /brainlift/full and the agent-facing /brainlift served the new guidance.

- Pre-commit checks passed.

## Linear

[AI-1350](https://linear.app/builder-team/issue/AI-1350/lead-linear-dss-delivery-reports-with-goals)

#2313 — feat(pudo): expose delegated Aerie API operations (AI-1319) @caina-barbosa  approved

## Title

feat(pudo): expose delegated Aerie API operations (AI-1319)

## Summary

Phase 4 of 4 of [AI-1314 — Alpha PUDO Rhodes Migration](https://linear.app/builder-team/issue/AI-1314), tracked in [AI-1319 — Approved PUDO reads and writes](https://linear.app/builder-team/issue/AI-1319). Connects the reviewed internal reads and writes to delegated Aerie HTTP routes and caller-scoped API discovery.

This phase builds on the PUDO storage, read and write contracts from phases 1–3.

Production effect: activation.

## Why

The API must bind a request to the configured Rhodes DSS relay and a person with a live PUDO grant. Discovery must reflect access changes, including when a client reuses a cache tag.

## Business Value

- Makes the 18 PUDO operations available through the authorized Rhodes DSS relay after configuration.

- Keeps access tied to the current caller grant.

- Gives authorized callers an API contract for the operations they can use.

## How does it work

1. Register the explicit PUDO HTTP routes.

2. Validate the configured relay and delegated actor before dispatching to internal reads or writes.

3. Project the PUDO operations into caller-specific discovery and test that revocation removes them.

## Scope

### Included in this phase

Connects the reviewed internal reads and writes to delegated Aerie HTTP routes and caller-scoped API discovery.

Exact diff paths against main:

chat/convex/http.ts

chat/convex/publicApi/apiKeyAuth.ts

chat/convex/publicApi/v2/http.ts

chat/convex/pudo/discovery.test.ts

chat/convex/pudo/http.ts

chat/convex/pudo/openapi.ts

chat/convex/pudo/read.ts

chat/lib/public-api/agent-context/projection.ts

### Deliberately excluded for later phases

- The separate Rhodes DSS repository changes, live grant, relay configuration, import run, and cutover are separate rollout steps.

## Test plan

### Automated validation

- All 25 focused tests, both local typechecks, and independent review passed. Hosted CI will validate the published commit.

### Time for Implementation

3–5 engineer days.

## Review repairs and contract clarifications

[Pagination and exact lookups](https://github.com/AI-Builder-Team/Aerie/pull/2313#discussion_r4551850562) · [AI-1342](https://linear.app/builder-team/issue/AI-1342)

- Continuation cursors bind the requested operation and filters to the ordered bounded result from one Convex read. If results change, the API returns 400 and asks the caller to restart without the cursor. Cursors contain digests and an offset. The read creates no durable snapshot.

- A missing exact changeRef, sourceActionId, or current sourceCrewId returns 404. An existing key excluded by additional filters and an empty collection still return 200 with an empty list.

- The installed Convex router checks exact method/path matches first, then sorts prefixes by descending length. PUDO GET and POST use exact routes, and the PUDO PATCH prefix is longer than the generic /v2/ prefix. Registration order therefore cannot send these requests to the generic handler. [Route-order finding](https://github.com/AI-Builder-Team/Aerie/pull/2313#discussion_r4551850553)

[Overview schema finding](https://github.com/AI-Builder-Team/Aerie/pull/2313#discussion_r4552081779) · [AI-1343](https://linear.app/builder-team/issue/AI-1343)

- The Overview schema distinguishes the unresolved response with no date from a dated Overview. Both variants match the HTTP response. Follow-ups, checks, and crew also document their existing 409 response for unresolved service scope.

- Two HTTP contract tests failed against the prior schema and pass with this change. The tests read the caller-scoped OpenAPI and compare it to the actual response. The change is limited to OpenAPI and its tests.

- The public write test matrix is deferred coverage. Source category normalization is a separate suggestion.

#75 — AI-1341 Show BrainLift and Dictionary before sign-in @ashwanth1109  no labels

Show BrainLift and Dictionary as the only homepage views before sign-in, with BrainLift open by default. Both views are visible on mobile and desktop. Signing in restores the full navigation, and the public discovery URLs retain their existing behavior.

## Business Value

Visitors can read the reporting guidance and understand Linear’s fields before requesting access. Signed-out visitors see the two documents they can use directly.

## Implementation Effort

About 1 hour for an engineer to adjust the navigation, check mobile behavior, and verify signed-out and signed-in flows.

## Validation

- Manually confirmed the signed-out view on the local dev server.

- Ran UI smoke checks for both documents, signed-out deep links, session failures, and signed-in navigation.

- Passed TypeScript, Biome, Knip, and Codedecorum.

## Linear

[AI-1341](https://linear.app/builder-team/issue/AI-1341/publish-the-linear-dss-brainlift-and-agent-reporting-guidance)

#73 — AI-1341 Publish the Linear DSS BrainLift and agent guidance @ashwanth1109  no labels

Publish the full Linear DSS BrainLift and public documentation. The BrainLift UI shows the owner, experts, and research, and the API and MCP supply DOK3 and DOK4 as agent context.

## Business Value

Readers can inspect the research behind delivery guidance, and agents receive concise instructions for reports, including a shareable hosted artifact when supported. Quick answers stay in chat and requested formats take precedence. The current AI Builder roster and quarter-to-date reporting remain in use.

## Validation

- Changed-file lint, Linear TypeScript and dependency checks, and rendered BrainLift assertions passed.

- Checked that the full document contains the research and that the agent context contains only DOK3 and DOK4.

## Linear

[AI-1341](https://linear.app/builder-team/issue/AI-1341/publish-the-linear-dss-brainlift-and-agent-reporting-guidance)

## Implementation Effort

About 2 to 3 engineer-days to research and write the guidance, implement the UI and API split, and verify the behavior.

#2312 — feat(pudo): implement approved changes and history (AI-1319) @caina-barbosa  approved

## Title

feat(pudo): implement approved changes and history (AI-1319)

## Summary

Phase 3 of 4 of [AI-1314 — Alpha PUDO Rhodes Migration](https://linear.app/builder-team/issue/AI-1314), tracked in [AI-1319 — Approved PUDO reads and writes](https://linear.app/builder-team/issue/AI-1319). Adds internal mutations for approved PUDO changes and separate publication follow-through.

This phase builds on the PUDO storage and read contracts in phases 1 and 2.

Production effect: dormant/additive.

## Why

A retry or stale approval can otherwise apply the wrong change or create duplicate history. This phase records changes, approval history, and retry receipts as one transaction.

## Business Value

- Keeps an attributed record of approved changes.

- Handles retries without applying an approval twice.

- Tracks manifest and guide follow-through separately.

## How does it work

1. Validate the delegated caller and typed approval.

2. Check the current revision and apply only the approved record and date scope.

3. Save the record, attributed history, and retry receipt atomically.

4. Record follow-through on the existing approval item.

## Scope

### Included in this phase

Adds internal mutations for approved PUDO changes and separate publication follow-through.

Exact diff paths against main:

chat/convex/pudo/write.ts

chat/convex/pudo/writeContract.ts

chat/convex/pudoWriteExceptions.test.ts

### Deliberately excluded for later phases

- HTTP dispatch and OpenAPI discovery belong to phase 4.

- These internal mutations have no public caller at this phase and merge does not start an import.

## Test plan

### Automated validation

- All 20 focused Convex tests passed, including reassignment, destination-local date, and rollback cases. Both local typechecks passed. Hosted CI will validate the published commit.

### Time for Implementation

4–6 engineer days.

## Review repairs and contract clarifications

[AI-1338](https://linear.app/builder-team/issue/AI-1338) follows through on the confirmation normalization found during phase 2 review. The mutation compares imported confirmation and change type after trimming surrounding spaces. Both original values remain intact. An overlapping confirmed no-ride and alternate-stop approval returns a conflict and leaves the original record, history, and receipt state intact. The regression exercises both fields independently and verifies rollback of an enrollment edit attempted earlier in the same transaction.

[Reassignment finding](https://github.com/AI-Builder-Team/Aerie/pull/2312#discussion_r4551397009) · [AI-1340](https://linear.app/builder-team/issue/AI-1340)

- Approval scope and pending follow-through use the effective destination service after an enrollment move. History retains both service values. An explicit clear removes service attribution, and omitting the service field preserves the existing scope.

- One approval still requires one consistent service scope. A following change can use the newly assigned service. A mix of former and destination scopes fails atomically.

- A service move changes both scopes immediately. The effective date must have arrived in both known service timezones. A move from an unassigned enrollment uses the destination timezone.

- The separate follow-through test matrix is deferred coverage. This phase still has no public caller.

[Follow-through outcome finding](https://github.com/AI-Builder-Team/Aerie/pull/2312#discussion_r4551614346) on a4fdd1b82

- The approved follow-through contract keeps a failed publication attempt pending for retry. The mutation records its required note, increments the revision, and stores the actor and attempt timestamp. Mutation readback includes the note, revision, and timestamp. The history projection also provides the actor's display name when available. The item stays in the pending queue until a verified completion.

#2311 — feat(pudo): implement permission-checked reads (AI-1319) @caina-barbosa  approved

## Title

feat(pudo): implement permission-checked reads (AI-1319)

## Summary

Phase 2 of 4 of [AI-1314 — Alpha PUDO Rhodes Migration](https://linear.app/builder-team/issue/AI-1314), tracked in [AI-1319 — Approved PUDO reads and writes](https://linear.app/builder-team/issue/AI-1319). Adds capability checks and internal PUDO reads for schedules, expected riders, history, and tracker records.

Phase 1 is merged into main. This draft contains the PUDO read layer.

Production effect: dormant/additive.

## Why

A caller needs the current roster and its schedule evidence before making an approval. The read layer preserves incomplete source coverage so consumers can recognize an unresolved trip.

## Business Value

- Provides roster, schedule, and tracker data to authorized callers once wired.

- Keeps expected and observed ridership distinct.

- Shows where source coverage cannot support an answer.

## How does it work

1. Register the dedicated PUDO capability and verify the live delegated actor in the internal read.

2. Read and scope records from the Convex tables added in phase 1.

3. Calculate effective schedules, Overview values, and expected rider coverage from the authorized records.

## Scope

### Included in this phase

Adds capability checks and internal PUDO reads for schedules, expected riders, history, and tracker records.

Exact diff paths against main:

chat/convex/pudo/access.ts

chat/convex/pudo/read.ts

chat/convex/pudo/readSchedules.ts

chat/convex/pudo/records.ts

chat/convex/pudo/routesContract.ts

chat/convex/pudoExpectedRiders.test.ts

packages/contracts/src/capabilities.ts

### Deliberately excluded for later phases

- Approved mutations and public routing belong to phases 3–4.

- These internal reads have no public caller at this phase and merge does not start an import.

## Test plan

### Automated validation

- Focused Convex edge tests passed (11 of 11, including the import regression).

- Biome, read bounds, test architecture and diff whitespace checks passed.

- Local chat app and Convex typechecks passed.

- Hosted CI passed on the prior reviewed head. Fresh hosted CI will validate this repair.

### Time for Implementation

4–6 engineer days.

## Review repairs and contract clarifications

[Review finding](https://github.com/AI-Builder-Team/Aerie/pull/2311#discussion_r4550242336) · [AI-1330](https://linear.app/builder-team/issue/AI-1330)

- An approved rider without exactly one eligible enrollment, or with a standing route that conflicts with the plan after applying exclusions, makes coverage unresolved. The response omits the partial roster and count. Known day exclusions and confirmed no-ride exceptions still apply.

- Both conflict cases returned a partial resolved roster before the fix. Both now return conflicting_approved_inputs. The focused tests also cover the two valid exclusions. Independent review passed.

- The authorization test matrix is deferred coverage. The trusted relay ID is a rollout configuration prerequisite for the later public caller. Overview day normalization is a separate suggestion.

- This phase adds dormant internal reads. Public routing and deployment configuration belong to the activation phase.

[Service scope finding](https://github.com/AI-Builder-Team/Aerie/pull/2311#discussion_r4550385026) · [AI-1331](https://linear.app/builder-team/issue/AI-1331)

- Service-scoped follow-ups, checks, and crew return 404 for an unknown service code and 409 for an ambiguous service code. Crew returns 409 when its service has no route mapping and the caller supplied no explicit route. A resolved scope can still return an empty list.

- Four regression cases failed on the previous head and now pass. The independent reviewer verified the errors use the existing read envelope and the phase 4 HTTP adapter forwards the status and code.

- The new ridership variance suggestion concerns validation of imported counts. The approved repair is limited to service resolution.

[Confirmation finding](https://github.com/AI-Builder-Team/Aerie/pull/2311#discussion_r4550664829) · [AI-1332](https://linear.app/builder-team/issue/AI-1332)

- The workbook importer preserves confirmation text as stored. Expected riders now trims surrounding whitespace before comparing the status, so a confirmed exception applies even when its source value contains spaces. The original source value is preserved.

- The authenticated Convex regression returned two riders before the fix and one afterward. All nine focused tests passed, including the prior service-scope and import regressions. Independent review and both local typechecks passed.

[Standing assignment finding](https://github.com/AI-Builder-Team/Aerie/pull/2311#discussion_r4550799733) · [AI-1335](https://linear.app/builder-team/issue/AI-1335)

- The workbook baseline date establishes when the captured assignment applies. A missing AM or PM assignment now leaves that direction unresolved. The opposite direction keeps its known value, and an explicit None remains a known nonregular assignment. Effective-dated approved assignments still take precedence.

- The authenticated enrollment regression reproduced a resolved snapshot with an absent AM assignment before the repair. It now reports unresolved coverage while preserving PM. All ten focused tests, both local typechecks and independent review passed.

[Scope validation finding](https://github.com/AI-Builder-Team/Aerie/pull/2311#discussion_r4550989519) on 26fb4a0fc

- This phase ships an internal Convex query. The exact head has no production caller. The test invokes it directly, and the only production reference across phases is in the phase 4 HTTP adapter.

- The phase 4 adapter validates supported filter names and values, required scope, dates, date order, direction, limit and cursor offset before calling the query. The query separately checks the live delegated actor and operation name and bounds collection reads.

- The hypothetical malformed call requires bypassing the approved caller. A future caller must apply the shared route contract before invoking the query. [Phase 4 adapter](https://github.com/AI-Builder-Team/Aerie/blob/74febebf85cba67cdd8f0e49a7bb7967820558f8/chat/convex/pudo/http.ts#L94-L178)

[Delegated actor finding](https://github.com/AI-Builder-Team/Aerie/pull/2311#discussion_r4551042475) · [AI-1339](https://linear.app/builder-team/issue/AI-1339)

- PUDO requires a configured relay and a distinct live human actor with the PUDO grant. The shared PUDO source check now rejects an actor who owns the relay key. This rule applies to reads, discovery, and the later mutations that call the same guard.

- Generic Aerie key authorization retains its existing owner behavior. The synthetic PUDO tests cover a distinct authorized actor and deny the relay owner even when that owner has the PUDO grant.

- The earlier HTTP scope-validation finding was withdrawn. The overview-only context concern is separate from the authorization repair.

[Overview context finding](https://github.com/AI-Builder-Team/Aerie/pull/2311#discussion_r4551190164) on 7d38b07a6

- The Overview contract reproduces the workbook's six SUMIFS/COUNTIFS totals for each origin. Those formulas return zero for a reconciled origin with no matching enrollments. standingCoverage describes the standing assignments of matching active enrollments. Import completeness is verified separately before cutover.

- The fixed workbook contains 75 roster entries across all six Overview origins, and the rehearsal matched all 36 cached totals. A caller-supplied origin constrains both context and enrollment lookups to the same value. The read rejects over-limit enrollment scopes instead of returning truncated totals.

- The transition requires a fresh capture, per-tab reconciliation, and resolution of discrepancies before a group is activated. Marking every context-only group unresolved would change the approved behavior for a genuinely empty, reconciled origin.

#231 — 1347-sindri-dss-recipe @mwrshah  approved

- Require active DSS membership and live dss-sindri:access before accepting native requests from DSS-scoped keys.

- Fetch DSS role definitions fresh so removing either a role assignment or its service permission rejects the existing key on its next request.

- Fail closed on WorkOS lookup failures while preserving Forge caching and the other authentication paths.

#71 — AI-1337 Default Linear DSS delivery reports to quarter to date @ashwanth1109  no labels

Linear DSS currently defaults delivery reports to the past 30 days. Use the current calendar quarter to date for completed work when the user has not asked for a different window. Start at midnight on the quarter’s first day in the reporting time zone and end at the report’s as-of timestamp. The active work read still uses the current workflow state.

## Business Value

Team delivery reports follow the same quarter as stakeholder goals and include work shipped earlier in the quarter. People can still request another date range.

## Implementation Effort

About 30–60 minutes to align the FAQ, BrainLift, Enablement, and MCP guidance and verify the reporting window semantics.

## Validation

Scoped Biome, TypeScript, Knip, and changed-scope CodeDecorum passed. Verified that the example prompt and guidance consistently use quarter to date and that the assigned-issue API keeps its existing explicit date filters.

## Linear

[AI-1337](https://linear.app/builder-team/issue/AI-1337/default-linear-dss-delivery-reporting-to-quarter-to-date)

#69 — AI-1333 Use one AI Builder team roster in Linear DSS @ashwanth1109  no labels

The Linear DSS delivery example asked about “my team” but pointed agents toward an EduOps roster that omitted Keval and Sanket. Return one six-person AI Builder roster from the portfolio and use it in the FAQ, BrainLift, Enablement, and MCP guidance. Existing callers get the same roster through eduOpsDelivery.people. EduOps goal references and stakeholder filters still apply to the requested goals.

## Business Value

Team delivery reports cover all six AI Builders before work is split into domains. Stakeholders can see delivery and ownership across initiatives without silently leaving out Keval or Sanket.

## Implementation Effort

About 1–2 engineer-hours to verify Linear IDs, update the portfolio contract and guidance, and validate the response.

## Validation

Scoped Biome, TypeScript, Knip, and a fixture-backed portfolio read passed. The portfolio check verifies all six unique members, the existing roster alias, the four EduOps goal references, and stakeholder labels.

## Linear

[AI-1333](https://linear.app/builder-team/issue/AI-1333/use-one-ai-builder-team-roster-in-linear-dss)

#2305 — fix(sites): rotate cancellation ID in generic site writes (AERIE-3110) @kevalshahtrilogy  approved

## Summary

Follow-up to #2303, closing the critical finding Mercy raised on release PR #2304.

#2303 rotates a site's cancellationId only in the dedicated status writer. The generic Rhodes site writes (sites.create, sites.patch, sites.replace, sites.upsertBySlug) could change status without rotating it. If a canceled site was reopened and canceled again through one of those paths, an approval proposed during the earlier cancellation could carry the same ID and still apply.

## Change

- One cancellationGeneration helper in chat/convex/rhodes/sites.ts, applied on every generic write:

- A status change into or out of cancelled starts a new generation.

- A merge write that doesn't change status, or changes it within the same disposition, keeps the current ID.

- replace carries the current ID forward, so a full-record rewrite doesn't drop it.

- A site created or inserted as cancelled starts with an ID.

- The only other writers of site status are one-off migrations.

## Tests

- rhodesCoreCrud.test.ts: cancel → re-cancel (ID kept) → reopen (new ID) → unrelated patch (kept) → cancel via patch (new ID, different from both earlier ones), plus create-as-cancelled.

- Ran rhodesCoreCrud, propertyHttp and fieldChangeRequests suites (93 passed), convex typecheck and Biome.

## Business Value

Removes the last blocker on releasing #2303, which unblocks the automated diligence refresh across canceled sites. It also guarantees an approval from an earlier cancellation can never overwrite current diligence.

## Manual Effort Estimate

About 1.5 hours by hand (trace the status paths, write the helper and tests). Keval to confirm.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#63 — AI-1325 Group Linear DSS initiatives by stakeholder @ashwanth1109  no labels

## Business Value

Artie and Dave can scan their own sections in Initiative Progress and ask Linear DSS about their initiatives. Answers use the current Stakeholder labels in Linear, and unlabeled initiatives have unknown stakeholder scope.

## Implementation Effort

About 3–4 engineer-hours to inspect the GraphQL schema, add the shared portfolio field and agent guidance, group the progress view, and verify the interfaces.

## Linear

[AI-1325](https://linear.app/builder-team/issue/AI-1325/scope-linear-dss-initiative-answers-by-stakeholder)

## Changes

- Group Initiative Progress by the current Linear stakeholder labels. Show an Unknown stakeholder section when needed. Use the portfolio-wide project scale and existing expandable rows.

- Add the Linear Stakeholder child label names as stakeholders on protected GET /initiatives and MCP list_initiatives.

- Guide agents to match the requested stakeholder and treat an empty list as unknown. Keep the existing EduOps delivery mapping, including the project-level Operational Data DSS reference.

## Verification

- Rendered the progress component with live Linear data and verified three Artie initiatives and four Dave initiatives. Fixture checks covered unknown and multiple labels, expanded details, and a single people request per portfolio load.

- Ran the portfolio query against live Linear and verified all seven initiatives, including empty stakeholder values before label setup.

- Exercised authenticated REST and MCP handlers in process with test WorkOS credentials, live Linear reads, and fixtures for the approved split. Verified parity, unknown values, incomplete-response handling, and unauthenticated denial.

- Passed the Linear type check and scoped Biome, codedecorum, and Git whitespace checks.

## Workspace setup

Created the workspace-level Stakeholder initiative label group with Artie and Dave. Verified Artie on I-9 through I-11 and Dave on I-12 through I-15, with initiative ownership preserved. The REST and MCP handlers in this PR returned the same split from live Linear using test WorkOS credentials.

#2310 — feat(pudo): add storage and import primitives (AI-1317) @caina-barbosa  approved

## Title

feat(pudo): add storage and import primitives (AI-1317)

## Summary

Phase 1 of 4 of [AI-1314 — Alpha PUDO Rhodes Migration](https://linear.app/builder-team/issue/AI-1314), tracked in [AI-1317 — Workbook importer](https://linear.app/builder-team/issue/AI-1317). Adds Convex tables and internal import mutations for the PUDO roster and tracker.

Production effect: dormant/additive.

## Why

The importer needs a typed destination that preserves record identity and references on replay. The reads and writes in later phases build on this schema.

## Business Value

- Preserves PUDO IDs and relationships.

- Keeps original change history available.

- Supports repeatable imports without duplicating records.

## How does it work

1. Register PUDO tables in the root Convex schema.

2. Validate mapped rows and resolve their references through internal import mutations.

3. Store each imported record and its receipt together so retries retain the same record ID.

## Scope

### Included in this phase

Adds Convex tables and internal import mutations for the PUDO roster and tracker.

Exact diff paths against main:

chat/convex/pudo/convexImport.ts

chat/convex/pudo/importSchema.ts

chat/convex/pudo/opsSchema.ts

chat/convex/pudo/persistence.ts

chat/convex/pudo/schema.ts

chat/convex/pudoConvexImport.test.ts

chat/convex/schema.ts

### Deliberately excluded for later phases

- Application reads, approved changes, and API registration belong to phases 2–4.

- The workbook runner and data cutover are separate. This phase has no public caller and merge does not start an import.

## Test plan

### Automated validation

- Diff whitespace and scope checks passed against main and the preceding branch.

- The combined implementation passed local validation. The focused Convex edge regression passed on the repaired branch and in independent review. GitHub CI passed on the original head; the repair will receive a fresh CI run.

### Time for Implementation

3–5 engineer days.

## Review repairs and contract clarifications

[Mercy finding](https://github.com/AI-Builder-Team/Aerie/pull/2310#discussion_r4549863552) · [AI-1328](https://linear.app/builder-team/issue/AI-1328)

- A blocking diagnostic now rejects the entire staging chunk. Convex rolls back any records and receipts already written by that mutation. A literal historical reference remains nonblocking; valid retries still replay without duplicates.

- The focused edge test failed on the original head, where the mutation returned success with one unresolved row. It passed with the repair and confirmed both rollback and replay. Independent review passed.

- The approved workbook runner rejects non-finite numbers before staging. No additional input path or batch completion operation is introduced here. The importer remains internal; merge does not start a migration.

- The regression test now uses two string glob calls, matching the app TypeScript declaration while loading the same test modules.

#66 — AI-1329 Simplify Linear DSS onboarding @ashwanth1109  no labels

## Business Value

A focused sign-in screen leads users to How to?, where they can connect their agent by pasting one link, copy a ready-to-send delivery question, or go straight to Initiative Progress for a high-level overview.

## Changes

- Make How to? the default signed-in view and provide a direct Initiative Progress link below Manual setup.

- Serve agent setup and usage guidance at the public /how-to endpoint. Show one copyable FAQ question on the signed-in page.

- Document HTTPS port 3000 for local Linear DSS and require a user decision if it is occupied.

## Implementation Effort

About 4–6 hours for an engineer to design, implement, iterate on, and verify the onboarding flow by hand.

## Verification

- Scoped Biome, Linear TypeScript, pre-commit checks, and diff whitespace checks passed.

- Local UI smoke checks covered signed-out and signed-in views, default navigation, progress deep links, and the setup and FAQ clipboard actions.

- Verified the app and public agent guide on the local HTTPS server.

## Linear

https://linear.app/builder-team/issue/AI-1329/simplify-linear-dss-onboarding-and-make-how-to-the-default

#145 — feat(heimdall): route Claude calls through Seatbelt @sanketghia  no labels

## Summary

- Add an optional KLAIR_SEATBELT_VA_KEY input to the reusable Heimdall workflow while preserving the legacy direct-key path for other consumers.

- Configure every Claude agent step to prefer Seatbelt, route model IDs through the Teams group, and unset ANTHROPIC_API_KEY before launching Claude.

- Fail fast when the Claude runtime has no credential and skip the Claude-only repair pass when neither credential is available.

- Add regression tests for Seatbelt headers, key precedence, model routing, and the reusable workflow contract.

Linear: KLAIR-3735

## Validation

- Heimdall suite: 1272 passed, 1 skipped.

- Ruff check and format checks passed for the added tests.

- bash -n heimdall/seatbelt_auth.sh passed.

- No live GitHub Actions execution was run from this central-workflow PR.

#64 — AI-1326 Prefer web reports for Linear DSS reporting @ashwanth1109  no labels

## Summary

- Make a web report the default for Linear DSS reporting. Follow explicit format requests from chat and fall back to standalone HTML or chat based on client capabilities.

- Expose the same preference through the MCP server, /skill, Enablement, and BrainLift.

## Business Value

Stakeholders get a web report they can read and revisit when they ask for initiative, portfolio, or delivery updates. They can still ask for a quick chat answer or another format.

## Implementation Effort

About 45–60 minutes for an engineer to trace the MCP and HTTP discovery paths, update both guides, and verify the responses.

## Verification

- Biome on changed TypeScript files and Linear DSS type checking.

- In-process checks of MCP initialization, the guidance tool, /skill, Enablement, and BrainLift.

## Linear

[AI-1326 — Prefer web reports for Linear DSS reporting](https://linear.app/builder-team/issue/AI-1326/prefer-web-reports-for-linear-dss-reporting)

#61 — AI-1313 Ground EduOps delivery answers in Linear evidence @ashwanth1109  no labels

## Summary

- Publish the configured EduOps people and goal references through the protected portfolio, including the Ops tracker link. Expose Linear's short initiative summary explicitly and retain the existing description field.

- Add paginated assigned-issue reads over REST and MCP, with filters for assignee, workflow state, and completion window. Include workspace-visible project identity and completion timestamps in assigned results and exact lookup.

- Update BrainLift and Enablement so agents use the issue tools introduced in #60 alongside the portfolio and cite separate evidence for deployment, adoption, and impact.

## Business Value

Artie can ask who is delivering against the four EduOps goals and get a person-by-person answer with dated evidence, including relevant work outside the tracked portfolio. The answer can distinguish a completed ticket from a live deliverable and show where ownership or outcome evidence is missing.

## Implementation Effort

About one engineering day without agent assistance to implement the REST and MCP contracts, verify Linear's schema and visibility behavior, and test the end-to-end fixture flows.

## Validation

- Passed in-process REST and MCP fixtures for the portfolio, goal mapping, assigned filters and pagination, archived completions, project visibility, and entitlement enforcement. Rechecked existing text search and exact lookup.

- Validated the exercised read operations and fixture variables against Linear's published GraphQL schema.

- Passed TypeScript, changed-file Biome and Code Decorum, and the repository pre-commit checks.

- Pending deployment and a rerun of Artie's question through his Grok connection.

## Linear

[AI-1313 — Enable person-by-person EduOps delivery answers in Linear DSS](https://linear.app/builder-team/issue/AI-1313/enable-person-by-person-eduops-delivery-answers-in-linear-dss)

#60 — AI-1312 Add standalone Linear DSS issue search and lookup @ashwanth1109  no labels

## Summary

- Add GET /issues?query=... and GET /issues/{issueId}, plus the corresponding MCP tools search_issues and get_issue.

- Use fixed read-only Linear GraphQL operations. Both entry points enforce the live WorkOS entitlement and return only issues from workspace-visible teams.

- Return compact, paginated search results with Linear links. Exact lookup by current identifier or UUID adds the description and timestamps. Document both operations in OpenAPI and Enablement.

## Business Value

People can ask an agent to find a Linear issue by topic or open a known issue directly, including work outside the tracked initiative portfolio. The agent can cite the original issue and retrieve its description when needed.

## Implementation Effort

Estimated 4–6 engineering hours to implement the GraphQL reads, REST and MCP interfaces, input and visibility checks, documentation, and fixture verification by hand.

## Validation

- [x] Ran the local fixture harness against REST and MCP with synthetic Linear and WorkOS data.

- [x] Searching delayed billing returned ENG-123 (Delayed billing alerts, In Progress) with a next-page cursor. Passing the cursor returned the next page.

- [x] Looking up ENG-123 by identifier and UUID returned the issue description and timestamps. The existing project issue route retained its response shape.

- [x] Verified unauthenticated and unentitled requests were denied, private and restricted team results failed closed, and a mismatched workspace was rejected.

- [x] Validated the GraphQL operations and fixture variables against Linear's published schema. Linear TypeScript and changed-file Biome checks passed.

- [ ] Run a live workspace smoke test on the deployed candidate.

## Linear

[AI-1312](https://linear.app/builder-team/issue/AI-1312/add-standalone-issue-search-and-lookup-to-linear-dss)

#3940 — fix(overspend-alerts): compress large ledger payloads @sanketghia  approved

## Summary

- Keep small alerts payloads as JSON and gzip/Base64-encode larger payloads with a gzip+base64: prefix, preserving all details in the existing Redshift column.

- Add a decoder for both legacy and compressed payloads; reject over-limit values rather than truncating them.

- Return ledger write status separately from email send status, and notify on ledger persistence failures.

- Document the backtracking decoder in the overspend alerts runbook.

## Verification

- Ruff 0.15.22 format and lint passed.

- Pyright: 0 errors, 1 warning on the existing pandas DataFrame columns argument.

- pytest tests/overspend_alerts/ -q -k 'not test_debut_filter_config': 208 passed, 1 deselected.

- The unfiltered feature suite has one unrelated failure: test_debut_filter_config expects only the SaaS suppression, while current settings also include Crossover/Flores.

- Existing 219-alert payload: 57,599 bytes raw → 9,348 bytes encoded; decoded back to 219 alerts.

No database migration or live database write is required. The scheduled-jobs image must be rebuilt and deployed for production to use the change.

#3938 — chore(ci): retire unused Claude workflows @sanketghia  approvedmercy-allow-critical

## Summary

- Retire the @claude issue and PR workflow.

- Remove the disabled legacy Klair PR Review workflow.

- Stop forwarding the raw Anthropic key into the Klair Mercy caller; Claude runtime remains disabled for Klair until its Seatbelt path is wired.

- Update Mercy comments to reflect the retired workflows and the fail-closed Claude state.

Linear: KLAIR-3735

## Validation

- git diff --check passed.

- Confirmed no references to the removed workflow files or raw-key mapping in Klair's Mercy caller.

- Verified the reusable Mercy workflow fails its Claude runtime credential check when no Anthropic key is passed.

- Tests were not run; changes are workflow deletions and configuration comments.

The remaining Seatbelt work for Heimdall and Claire Eval is outside this PR.

#2307 — feat(task-board): filter Tasks by Creator @caina-barbosa  approved

## Summary

Add a Creator filter to Team Tasks and the site Tasks API. Task pages show Creator next to Requester. Creator filtering works with the catalog fallback and the server pagination shipped in #1851.

Production effect: activation of an optional filter and updated Task details.

## Why

A Task's requester and creator can be different people. Team Tasks users need to find work by the person who actually created it.

## Business Value

- Filter Team Tasks by one or more creators and share the resulting URL.

- See who created a Task alongside its other participants.

- Find Tasks from an agent or integration using the same recorded identity.

## How does it work

1. Team Tasks passes the selected creator IDs to the catalog fallback or the server page and count queries, depending on pagination readiness.

2. The site Tasks API accepts an optional creatorId Directory user ID. Existing Task visibility rules apply and the cursor is bound to the filter.

3. Tasks without a recorded creator display Not recorded and are excluded from creator-specific results.

## Scope

### Included in this phase

- Team Tasks filter, Task detail header, API filter, and focused coverage.

- Exact paths:

chat/components/task-board/task-board-state.ts

chat/components/task-board/task-detail-page.test.tsx

chat/components/task-board/task-detail-page.tsx

chat/components/task-board/task-detail.tsx

chat/components/task-board/team-tasks-view.test.tsx

chat/components/task-board/team-tasks-view.tsx

chat/convex/publicApi/v2/domains/workManagement.ts

chat/convex/publicApi/v2/workManagement.test.ts

chat/convex/publicApi/v2/workManagementData.ts

chat/convex/rhodes/schema.ts

chat/convex/taskBoard/queries.test.ts

chat/convex/taskBoard/queries.ts

chat/lib/public-api/v2/domains/work-management.ts

### Deliberately excluded for later phases

- Historical creator backfill.

- Rhodes publication is tracked in [Rhodes-DSS #64](https://github.com/AI-Builder-Team/Rhodes-DSS/pull/64).

## Test plan

### Automated validation

- 96 focused tests passed across Team Tasks, Task details, the Task API and its published contract.

- Coverage includes Creator in fallback catalog results, server pages, totals and API results.

### Time for Implementation

About one to two engineering days, including integration and verification.

#1851 — fix(task-board): load Team Tasks in server pages (AERIE-2847) @caina-barbosa  approvedmercy-allow-critical

## Summary

Team Tasks currently downloads every Task before showing the first results. This post-launch repair changes the normal page to show the first 25 matching Tasks, then preload the next 25 as the user approaches the bottom. Filters, global sorting, full-description substring search and complete totals keep their existing meanings.

Tracked by [AERIE-2847 — Load Team Tasks in filtered, sorted pages with scroll prefetch](https://linear.app/builder-team/issue/AERIE-2847/load-team-tasks-in-filtered-sorted-pages-with-scroll-prefetch).

Production effect: controlled rollout or migration. Deployment preserves the current page until a bounded sort-key repair finishes; that repair is separately run and does not start on merge. The approved deletion-error wording improves immediately.

## Why

The current initial view waits for about 11 sequential catalog requests at the verified production size of 2,611 Tasks. Those requests fetch records the user has not scrolled to. Returning a small, correctly ordered page makes the first results independent of loading the full catalog or calculating totals.

## Business Value

- Users can start working with the first 25 Tasks while totals finish loading.

- Scrolling preloads the next 25 Tasks before reaching the bottom.

- Filters, search and sorting continue to apply across the whole portfolio.

- A blocked deletion explains that Discussion notes are attached and that open Tasks can be rejected while preserving their discussion.

## How does it work

1. Existing compact Task projections gain six optional indexed sort keys. The Task writer keeps them current, preserving newest-first ties and placing Tasks without deadlines last in either direction.

2. The initial unfiltered query reads 25 compact rows. Filtered or searched pages inspect at most 250 candidates and stop at 25 matches or a bounded text budget.

3. Exact substring search runs on the server using the existing complete search chunks. A small preflight chooses a safe native pagination range and reuses its search results. Opaque native cursors, reactive journals, fixed end boundaries and split information are preserved.

4. Counts arrive through a separate bounded query. The controls remain visible, but counts display a pending indicator until complete totals are known. A scroll observer preloads the next 25 results.

5. After deployment, repeat taskBoard/projectionMaintenance:backfillTeamSortKeys with 250-row pages until complete, then run verifyTeamSortKeys until complete with zero mismatches. This touches sort fields on existing compact rows only. The current catalog path remains available until all six keys are present; no Task, participant or search-text rebuild is required.

## Scope

### Included in this phase

- Server-selected Team Task pages, full search, exact counts and scroll prefetch.

- Optional sort keys, indexes and bounded key-only repair/verification.

- Preservation of the existing 5,500-byte people/display-content allowance, with a separate bounded allowance for sort-key storage.

- The unused legacy whole-catalog listTeam reader's derived safety ceiling is recalculated for the larger physical rows. A repository caller audit found no current UI or public Task API consumer of that reader.

- Shared UI/API deletion error wording and its existing Work Plan regression assertion.

- Exact final diff paths:

chat/components/task-board/task-list-primitives.tsx

chat/components/task-board/team-tasks-view.test.tsx

chat/components/task-board/team-tasks-view.tsx

chat/convex/_generated/api.d.ts

chat/convex/rhodes/runtime/writes/taskWrites.ts

chat/convex/rhodes/schema.ts

chat/convex/rhodesPortfolioWorkbench.test.ts

chat/convex/taskBoard/projectionMaintenance.ts

chat/convex/taskBoard/queries.test.ts

chat/convex/taskBoard/queries.ts

chat/convex/taskBoard/teamProjection.test.ts

chat/convex/taskBoard/teamProjection.ts

### Deliberately excluded for later phases

- Public Task API contract or permission changes.

- New Task lifecycle actions or changes to Task records.

- Automatic production maintenance execution.

## Test plan

### Automated validation

- Query, projection, maintenance, Team UI and shared list controls: 83/83 passed.

pnpm --dir chat exec vitest run convex/taskBoard/queries.test.ts convex/taskBoard/teamProjection.test.ts convex/taskBoard/projectionMaintenance.test.ts components/task-board/team-tasks-view.test.tsx components/task-board/task-list-primitives.test.tsx --maxWorkers=1

- Work Plan deletion regression: 1/1 passed.

pnpm --dir chat exec vitest run convex/rhodesPortfolioWorkbench.test.ts -t 'deleteTask removes' --maxWorkers=1

- A 2,611-Task fixture returns the globally correct first 25 and the next 25, with independent exact total counts.

- A sparse search over 300 Tasks, including a 1 MB description, reaches a rare match in two bounded requests and preserves exact search counts.

- Native fixed boundaries preserve Task membership and counts after insertion/deletion across loaded pages; native split information is returned for expanded ranges.

- Browser tests verify first results while totals are pending, scroll prefetch, retained controls during refresh and access to a last locally loaded Task after a reactive page grows.

- Full chat typecheck passed through the mandatory pre-commit hook and again after rebasing onto current main (pnpm --dir chat exec tsc --noEmit --pretty false).

- Biome, architecture boundaries, Convex paths, read bounds, test architecture and git diff --check passed.

- Original implementation CI passed on fba16e25ea926132bf7c7e0db526280f9026bedc. After merging current main, head 626bd5a7eacc500e8fbe4bfe98e736d954d93238 has green static checks and builds. The Test job timed out in the unchanged contracts bundle test; the exact test passed independently (1 passed, 19 skipped, 3.17 seconds). The failed hosted job passed on retry without any code change. All hosted CI checks are now green on this head ([CI run](https://github.com/AI-Builder-Team/Aerie/actions/runs/38409772233)).

### Time for Implementation

About 3 to 5 engineering days to trace the read path, implement bounded pagination and repair, test reactive boundaries, integrate the current UI changes and validate rollout.

| Change | What it means for end users/consumers |

|---|---|

| First page | Shows 25 matching Tasks in the selected global order after the key repair. |

| More results | Preloads another 25 near the bottom; the accessible Load more action remains available. |

| Filters and search | Keep exact current meanings, including complete title/description substring matching. |

| Totals | Show a pending indicator until complete filtered totals are ready. |

| Existing Tasks | Keep their records, permissions and lifecycle behavior. |

| Blocked deletion | Explains that attached Discussion notes prevent deletion; open Tasks can be rejected to retain their discussion. |

## Review repairs and contract clarifications

[Review of f7b81aeb](https://github.com/AI-Builder-Team/Aerie/pull/1851#pullrequestreview-5866010907) for AERIE-2847.

- Removed duplicate generated declarations for projection maintenance. Both Chat typechecks passed on the reviewed head with declaration-file checking disabled. The generated file now imports and registers that module once.

- Integrated main's scroll fix (#2306) while retaining the ref used for pagination prefetch.

- Convex [indexes match absent fields with equality to undefined](https://docs.convex.dev/database/types#working-with-undefined). The readiness regression verifies that the old view remains active with missing sort keys and that paging activates after the repair. This finding does not require a migration change.

- The search coverage findings are nonblocking. Indexed search rejects invalid chunk counts, missing or duplicated indices, and revision mismatches before joining text. Existing compatibility-path tests cover missing chunks and revision mismatches. The raw storage-budget error guards internally generated keys and remains a nonblocking suggestion.

- Focused Convex and browser checks passed (66 tests). The manual sort-key backfill remains the rollout step.

#7 — 2047-mobile-sign-out-icon @mwrshah  no labels

- Use Lucide’s LogOut icon for the top-right Sign out button below md.

- Preserve the desktop text, accessible Sign out label, and existing POST /logout action.

- Keep the generator aligned with the responsive logout control in DSS-Hub.

#58 — 2047-sindri-openapi-display @mwrshah  no labels

- Replace the top-right Sign out text with Lucide’s LogOut icon below md across Sindri, Linear, and Salesforce.

- Keep the text label at md and above, with a 44px mobile target and an accessible Sign out label.

- Preserve each service’s desktop styling and existing POST /logout action.

#2306 — fix(tasks): restore Team Tasks scrolling @caina-barbosa  approved

## Summary

Restores scrolling on Team Tasks so people can reach the rest of the list and the Load more button. The refresh error screen and Retry still work.

Production effect: compatibility hardening. This repairs the layout introduced by [#2048](https://github.com/AI-Builder-Team/Aerie/pull/2048) and can ship without the pagination changes in [#1851](https://github.com/AI-Builder-Team/Aerie/pull/1851).

## Why

The error boundary placed an extra div between the app shell and the scroll region. In production, the list expanded to 5,016 pixels inside a 622-pixel clipped panel. Scrolling did not move the list, and Load more was below the visible area.

## Business Value

- People can scroll to Tasks below the first screen.

- Load more is reachable again.

- The refresh message and Retry stay in place.

## How does it work

1. Use a keyed React Fragment in the error boundary instead of a div.

2. The list remains a direct child of the height-constrained shell. Changing the key still remounts the list after Retry.

## Scope

### Included in this phase

- chat/components/task-board/team-tasks-view.tsx

- chat/components/task-board/team-tasks-view.test.tsx

### Deliberately excluded for later phases

- Server pagination and load-time changes, tracked in #1851.

- Changes to My Tasks, APIs, permissions or data.

## Test plan

### Automated validation

- The scroll-layout regression assertion failed before the fix and passed afterward.

- Team Tasks and My Tasks suites: 28/28 passed with pnpm --dir chat exec vitest run components/task-board/team-tasks-view.test.tsx components/task-board/my-tasks-view.test.tsx --maxWorkers=1.

- Changed-file Biome and git diff --check: passed.

### Time for Implementation

Approximately one hour, including diagnosis and validation.

#56 — 2046-sindri-openapi-display @mwrshah  no labels

- Fetch and validate Sindri’s native /skillspec on every request instead of serving a process-start snapshot.

- Fetch current native revision, hash, and size metadata for /skill, without a backend cache or refresh timer.

- Preserve the 10-second timeout, 2 MiB limit, and credential-free upstream requests; return HTTP 502 when the native contract is unavailable or invalid.

- Skip token counting on raw spec requests and document the live-fetch behavior.

#54 — 2046-sindri-openapi-display @mwrshah  no labels

- Display Sindri’s native OpenAPI in the documentation viewer while preserving raw agent access.

- Apply the recipe’s responsive API-key layout, placeholder alignment, compact controls, and burgundy expiry indicator to Sindri, Linear, and Trilogy Salesforce.

- Show authenticated profile photos with a neutral fallback and one email popover for hover or tap; prevent image selection and dragging.

- Simplify Sindri’s agent setup prompt while preserving service-specific layouts and authorization checks.

#52 — 2045-5dss-prod-release @mwrshah  no labels

- Set service titles to 24px below md and reduce the header-to-content gap across Salesforce, Sindri, and Linear.

- Remove the extra Sindri introduction paragraph.

- Remove the checking-sign-in message and reserve the Sign in button space with an invisible, non-interactive placeholder during session checks.

- Keep desktop sizing and service-specific views unchanged.

#2303 — fix(diligence): allow API updates on cancelled sites (AERIE-3110) @caina-barbosa  approved

## Title

fix(diligence): allow API updates on cancelled sites (AERIE-3110)

## Summary

This PR implements the Due Diligence PATCH slice of [AERIE-3110](https://linear.app/builder-team/issue/AERIE-3110/allow-public-api-v2-writes-on-cancelled-sites). It overlaps with the broader [PR #2302](https://github.com/AI-Builder-Team/Aerie/pull/2302), so this draft presents the narrower scope for review.

Production effect: activation.

## Why

The diligence refresh loop receives HTTP 409 when it reaches a cancelled site. A diligence save can also trigger a buildout date change, so the refresh needs a way to update the record while scheduled and field-triggered automation skips cancelled sites.

## Business Value

- Keep diligence evidence current for cancelled sites.

- Keep proposals from before cancellation out of fresh approvals.

- Preserve an admin's ability to preview and run a buildout date change manually.

## How does it work

1. The public API v2 Due Diligence PATCH accepts cancelled sites under its existing write scope, revision check, and Conducting Diligence lock.

2. The site status writer rotates the cancellation ID when the site enters or leaves cancelled status. A new diligence proposal records the current ID.

3. Draft reuse, approval, and application check that ID. A fresh patch after cancellation starts from approved values and supersedes an older draft.

4. Approved changes save in Aerie. Scheduled buildout and field-change runs skip cancelled sites, and the daily capacity sweep skips new entries for those sites.

## Scope

### Included in this phase

- Public API v2 Due Diligence PATCH, its approval flow, and its API description.

- Cancellation-aware checks for scheduled buildout, field-triggered buildout, and daily capacity enqueue.

- Exact final diff paths:

chat/convex/automations/platform.test.ts

chat/convex/automations/platform.ts

chat/convex/capacityAutomation.test.ts

chat/convex/capacityAutomation/enqueue.ts

chat/convex/portfolio/dueDiligence.ts

chat/convex/portfolio/dueDiligenceEditPolicy.ts

chat/convex/portfolio/fieldChangeRequests.ts

chat/convex/portfolio/schema.ts

chat/convex/publicApi/v2/domains/property.ts

chat/convex/publicApi/v2/propertyHttp.test.ts

chat/convex/rhodes/runtime/writes/siteWrites.ts

chat/convex/rhodes/schema.ts

chat/lib/public-api/v2/domains/property.ts

### Deliberately excluded for later phases

- Broader API write access, including scenario and document writes, as proposed in #2302.

- Adding cancellation to the diligence ETag, changing monitoring emails, or stopping already queued or explicitly requested capacity work.

- In-app editor and Rhodes agent write behavior, backfills, and deployment.

## Test plan

### Automated validation

- Targeted API, approval, automation, capacity, MCP parity, and OpenAPI suites — 481 passed (pnpm --dir chat exec vitest run convex/publicApi/v2/propertyHttp.test.ts convex/publicApi/v2/portfolioDomain.test.ts convex/automations/platform.test.ts convex/capacityAutomation.test.ts convex/portfolio/fieldChangeRequests.test.ts convex/portfolio/dueDiligence.test.ts convex/rhodesMcpMutationParity.test.ts lib/public-api/__tests__/openapi.node.test.ts).

- Chat type check, Biome, and Convex module paths — passed in the pre-commit hook.

- Diff whitespace and scope checked against main.

### Time for Implementation

About one engineering day for investigation, implementation, and local validation.

#50 — 2044-recipe-bottom-spacing @mwrshah  no labels

- Apply responsive documentation navigation and source/copy-link actions to Linear, Sindri, and Trilogy Salesforce.

- Refine Sindri and Salesforce panel spacing and mobile viewport sizing; simplify shared access copy and desktop titles.

- Cache documents and commit tab changes atomically to prevent loading flashes while preserving retries and browser history.

#6 — 007-content-bottom-spacing @mwrshah  no labels

- Refine documentation panel spacing and viewport sizing, with a smaller mobile bottom gap.

- Add an overlay mobile selector and source/copy-link actions; remove duplicate desktop titles.

- Cache documents and switch heading, actions, and content together without loading flashes.

#63 — Publish backup-site location, candidate, term and need-period writes (AI-1285) @kevalshahtrilogy  no labels

## What

Publishes nine source operations for write keys, through src/aerie/allowlist.json and pnpm sync:openapi:

- GET /portfolio/backup-site-locations (read)

- POST /portfolio/backup-site-locations, PATCH .../{backupSiteLocationRef}

- PATCH /portfolio/backup-sites/{backupSiteRef} (term dates)

- PATCH /portfolio/sites/{siteRef}/backup-site/status

- PUT .../backup-site/candidates/{backupSiteLocationRef}, PATCH .../candidates/{backupSiteLocationRef}/operating

- POST .../backup-site/need-periods, PUT .../need-periods/{needPeriodRef}

The two PUTs use the relay support from #62. Not published: contract files (AI-1286), archiving a term, unlinking a candidate, removing a need period, and every delete.

The snapshot diff is large only because git realigns the file. Semantically, it adds these 9 operations and changes no schemas.

The docs now say "POST, PATCH and PUT" (definition, OpenAPI auth text, Dictionary, Enablement). Enablement gains a recipe for recording a backup. Counts are now 52 GET reads, 2 searches and 55 writes, 54 published operations in all.

The Rhodes service key now holds operations.backupSites.write, which Keval added on 2026-11-11.

## Business Value

Edu Ops staff, and their agents, can now record where a school is running while its own building is finished. That covers the backup location, selection, operating confirmation, need dates and status, made through Rhodes as themselves. This closes the backup-site part of the gap between Rhodes and the source's EduOps UI.

## Manual Effort Estimate

About 3 hours by hand: contract review, allowlist and sync, doc and count updates, verification. Proposed; Keval to confirm or adjust.

## Verification

pnpm check (92 tests), pnpm build and pnpm synth --no-lookups pass. The routes have not been exercised against production; that happens after deploy.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#61 — Publish Task approval actions, sources, watchers and review-history reads (AI-1281) @kevalshahtrilogy  no labels

Stacked on #60.

## What

Writes for write keys, on a Task:

- approve the current completion submission, or request changes;

- configure approval (whether it's required, and the approver list);

- link a same-site document;

- add or edit a source;

- add a watcher.

Reads: Task comments, completion submissions and their review actions, and GET /work-management/my-tasks (the owner's Tasks across all sites).

Left out on purpose:

- Every removal: un-approve, unlink, remove a source or watcher, and the deletes.

- Work Unit Group create: an existing test records it as excluded (AI-1039 scope).

Docs: the contract is re-synced (these 11 operations only), Enablement §9 is updated, the dictionary counts go to 51 reads and 47 writes, and the two count tests move to 53 operations.

## Why

These are the Task review steps EduOps approvers take in the source app. Access is decided by the source from the owner's relationship to each Task; there's no scope change.

## Business Value

Approvers and agents can run the whole Task review loop through Rhodes: submit, approve or send back, and see the history.

## Manual Effort Estimate

About 1.5 hours (proposed; Keval to confirm).

## Verification

- pnpm check passes (91 tests).

- pnpm build and pnpm synth --no-lookups pass.

- No live write has been sent.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#60 — Publish diligence scenario, regulatory approval and cost estimate writes (AI-1280) @kevalshahtrilogy  no labels

Stacked on #59.

## What

Writes for write keys:

- Diligence scenarios: create (name, summary), sparse update, and archive. Archive is a soft archive sent as a POST; it isn't a delete.

- Education regulatory approvals: create and update.

- Operational cost breakdowns: create and update. Amounts are decimal dollars.

The deletes for approvals and cost breakdowns stay unpublished. The contract is re-synced; the only change is these seven operations and their schemas. Enablement gets a short write paragraph next to each read section, and the dictionary count goes to 40 writes.

## Why

These are the property and approval edits EduOps make in the source app. No new service scope is needed: they use operations.dueDiligence.write, operations.driSites.write, operations.realEstate.read and operations.diligence.read, which published routes already use.

## Business Value

Diligence what-ifs, approval records and cost estimates can be maintained from EduOps agents through Rhodes.

## Manual Effort Estimate

About 1.5 hours (proposed; Keval to confirm).

## Verification

- pnpm check passes (91 tests).

- pnpm build and pnpm synth --no-lookups pass.

- No live write has been sent.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#59 — Publish site profile writes and the internet, cleanliness and decision-log reads (AI-1279) @kevalshahtrilogy  no labels

Stacked on #58.

## What

Writes for write keys, each a sparse PATCH that takes the card's revision as If-Match:

- facilities, identity, integrations, legal-entity, school-model, compliance;

- people/assignments (the five DRI and personnel assignments) and people/contacts (the six structured contacts);

- property/acquisition.

Reads: internet, cleanliness and decision-log. The source added these for Rhodes in October (AERIE-2729), but they were never relayed.

Docs:

- The contract is re-synced; the only change is these operations and their schemas.

- Enablement §6 gets three new cards and the write guidance (the contacts semantics: a supplied contact replaces the whole record).

- The dictionary counts go to 47 GET reads and 33 writes.

- The two count tests move from 46 to 49 operations.

Left out on purpose: PATCH /status (site participation status). An existing test records it as an excluded route, so it stays unpublished until Keval decides otherwise.

## Why

This closes the gap between the site cards EduOps can edit in the source app and those they can edit through Rhodes. No new service scope is needed: every operation uses operations.portfolio.read / operations.driSites.write (plus operations.realEstate.read for acquisition). As with every write, the source authorizes each change as the key's owner.

## Business Value

EduOps agents can fix site identity, facilities, school model, compliance, people and acquisition facts through Rhodes, so they don't need the source app or a source key for these edits.

## Manual Effort Estimate

About 2 hours (proposed; Keval to confirm).

## Verification

- pnpm check passes (91 tests).

- pnpm build and pnpm synth --no-lookups pass.

- No live write has been sent.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#58 — Publish buildout plan and phase milestone writes so write keys can set capacity (AI-1283) @kevalshahtrilogy  no labels

## What

Publishes three source writes for Rhodes write keys, through the writes list in src/aerie/allowlist.json:

- PATCH /portfolio/sites/{siteRef}/buildout: phase status, dates and capacityBuilt; adds or removes additional phases.

- PATCH /portfolio/sites/{siteRef}/buildout/milestones: milestone due dates for the plan.

- PATCH /portfolio/sites/{siteRef}/buildout/phases/{phaseId}/milestones/{milestoneKey}: one milestone of any phase.

The contract was re-synced with pnpm sync:openapi. The only change is these three operations and their five schemas; there is no other upstream drift.

Enablement §6 gains a recipe for changing capacity: capacity.current is the capacityBuilt of the latest completed phase, and it stays 0 until ready-to-open is complete. The dictionary operation count goes from 21 to 24 writes.

## Why

JC Fischer (Chat, 11/11) wants to change capacity himself or through his agent. The source already accepts it; Rhodes only published the read. These writes need no new scope on the service credential, since they use the same capabilities as the published milestone write. As with every write, the source authorizes each change as the key's owner and records it under them.

## Business Value

EduOps can correct capacity and buildout status from their own agents without the source app. This is the first step in closing the gap between what the source can change and what Rhodes can (project: Rhodes write parity with the source).

## Manual Effort Estimate

About 1.5 hours (proposed; Keval to confirm).

## Verification

- pnpm check passes (91 tests). The existing contract tests cover every allowlisted write: the write-key 403, a required Idempotency-Key, and no source name.

- pnpm build and pnpm synth --no-lookups pass.

- Not yet verified against production: no live write has been sent.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#62 — Relay PUT writes like any other published write (AI-1284) @kevalshahtrilogy  no labels

## What

src/aerie/routes.ts now accepts PUT as a write method, alongside POST and PATCH. The relay already handled methods generically, so a PUT gets the same treatment as the other writes:

- write key only;

- relayed as the key's owner;

- Idempotency-Key required and scoped per key;

- If-Match passed through.

No route is published in this PR; the allowlist is unchanged.

The other edits are in test/writes.test.ts:

- One new relay test sends a PUT and checks that it reaches the source as a PUT, with its body, If-Match and the acting person.

- Keval approved a CD005 exception for this file on 11/11.

- Four existing standalone comments became short trailing comments (CD001).

## Why

Three source backup-site writes are PUTs: replace a need period, set a site's backup candidate, and replace a contract file. This unblocks AI-1285 and AI-1286 (project: Rhodes write parity with the source).

## Business Value

This is the groundwork for EduOps to manage backup-site arrangements through Rhodes once the service key holds operations.backupSites.write.

## Manual Effort Estimate

About 45 minutes (proposed; Keval to confirm).

## Verification

- pnpm check passes (92 tests).

- pnpm build and pnpm synth --no-lookups pass.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#47 — fix/sindri-runtime-permissions @mwrshah  no labels

- Consolidate read actions sharing Resource=* and remove the unused network-interface listing permission.

- Keep the generated execution policy within the AWS managed-policy size limit.

#46 — fix/sindri-runtime-permissions @mwrshah  no labels

- Correct the shared VPC owner ARN while preserving tag-restricted security-group creation.

- Scope task-role permissions to actual CloudFormation names and permit preflight/absent-role cleanup.

- Complete resource lifecycle permissions for ECS, security groups, target groups, listener rules and log reads.

- Validate policies with AWS Access Analyzer and IAM simulations, including cross-service denials.

#44 — 2042-sindri-dss-release @mwrshah  no labels

- Allow CloudFormation to inspect Sindri feedback tables before resource tags exist; preserve tag conditions on mutations.

- Allow bootstrap deployment roles to delete only their service-owned stacks for failed-creation recovery.

- Keep all permissions scoped to the selected service.

#42 — fix/sindri-bootstrap-version-read @mwrshah  no labels

- Grant Sindri CloudFormation execution ssm:GetParameters on its exact bootstrap-version parameter.

- Match the approved CloudFormation-managed IAM correction applied to CDKToolkit-dsssindri.

- Preserve all other service permissions and deployment boundaries.

#2262 — feat(observer): route run observer Claude calls through Seatbelt (SURTR-1707) @kevalshahtrilogy  approvedmercy-allow-critical

## Summary

Moves the Surtr run observer (product-surtr-prod-observer) onto AI Eng's Seatbelt pool, as requested in the AI Eng - Seatbelt space. Its Claude calls now run on Claude Teams seats instead of pay-per-token Anthropic credit.

- Surtr/src/derive/observer/seatbelt.ts (new): an Anthropic client pointed at https://seatbelt.csaiautomations.com. It sends the product VA key in x-tfy-api-key and a dummy bearer (Seatbelt swaps in a seat token), and explicitly strips x-api-key. Seatbelt passes that header through, which would silently move calls to metered billing.

- Raw-API request shape: the Seatbelt guide's Option B requires a Claude Code identity system block first, plus the OAuth betas and a CLI user-agent. Without them every seat answers 429 and Seatbelt falls back to metered. Both responses are HTTP 200, so a missing shape only shows up on the bill.

- Model: claude-teams-group/claude-sonnet-5-5. claude-sonnet-4-6 isn't on Seatbelt's model list. OBSERVER_VERSION goes from 2 to 3, so cached verdicts from the old model re-evaluate on next access.

- Infra: the ECS task reads SURTR_SEATBELT_VA_KEY from SURTR_PROD_KEYS, and the legacy ANTHROPIC_API_KEY is removed from the container. The UI hint, the showcase script, .env.example and the docs are updated to match.

## Before deploy (blocking)

1. Add a SURTR_SEATBELT_VA_KEY key (the product-surtr-prod-observer VA key) to the SURTR_PROD_KEYS secret. The task fails to start without it.

2. Confirm with the Seatbelt operator (Deniz) that the VA is on the stack's member list. An unlisted VA gets a 403, which the observer stores as a failed evaluation.

## Test plan

- [x] vitest run test/derive/observer passes (108 tests, 12 files), including the new observer-seatbelt.test.ts. It captures the outgoing request and checks the URL, the VA header, the betas and the user-agent. It also checks that no x-api-key is sent even when ANTHROPIC_API_KEY is set.

- [x] tsc --noEmit for the app passes, and biome check src is clean.

- [ ] Infra tsc not run locally because infra deps weren't installed in the worktree. The change there is a one-line secret rename.

- [ ] After deploy: evaluate one run and confirm the request appears on the Seatbelt dashboard as seat-served, with zero fallbacks.

## Business Value

Observer evaluations move off pay-per-token Anthropic spend onto the Claude Teams seats AI Eng already pays for. Leadership asked every AI Eng product key to make this switch this week, so it can retire per-token billing and Teams subscriptions. The wire-level test guards against the gateway's silent metered fallback, which no error or alert would ever surface.

## Manual Effort Estimate

~3 hours (proposed by Claude; Keval to confirm or adjust). That covers reading the Seatbelt product guide, working out the SDK's header and credential precedence, the client plus the model and version change, the infra and docs updates, and writing the request-capture test.

Linear: [SURTR-1707](https://linear.app/builder-team/issue/SURTR-1707/route-surtr-run-observer-claude-calls-through-seatbelt)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#37 — 2039-workos-access-hierarchy @mwrshah  no labels

- Add WorkOS hierarchy diagrams for identities, organizations, membership roles and personal keys.

- Explain separate Forge product permissions and DSS service entitlements.

- Clarify key organization bindings and the approved native entitlement follow-up.

#40 — 2041-sindri-dss-release @mwrshah  no labels

- Add service-owned release workflows for Salesforce, Sindri and Linear while retaining shared deployment mechanics and isolated service authority.

- Select runtime and infrastructure changes per service; use one explicit service revision for deployment-behavior changes and defer process-only edits.

- Preserve successful-release baselines and historical image provenance for rollback.

- Update README and creation/release skills with concise instructions for the release model.

#38 — Batch Linear DSS initiative people reads @ashwanth1109  no labels

## Summary

- Load People for every initiative in one signed-in dashboard request, avoiding the four-read capacity error when rows load together.

- Fetch the portfolio's open project leads and in-progress issue assignees in grouped Linear GraphQL queries. Preserve pagination, shared-project attribution, and the per-initiative API and MCP tool.

- Share the result and retry across the row avatars and expanded People details.

## Business Value

Initiative Progress shows who is doing the work across the portfolio without requiring users to retry individual rows.

## Implementation Effort

Approximately four hours for an engineer to implement and verify manually.

## Validation

- Live Linear read returned all seven rollups in two GraphQL queries.

- Fixture checks covered project and issue pagination, shared projects, empty rollups, the private-team filter, single-initiative parity, invalid project rejection, and cancellation.

- TypeScript, Knip, and changed-file Biome passed.

## Linear

[AI-1288 — Add Linear DSS initiative progress service](https://linear.app/builder-team/issue/AI-1288/add-linear-dss-initiative-progress-service)

#35 — Add a Linear DSS how-to guide @ashwanth1109  no labels

## Demo

<img width="2200" height="1637" alt="image" src="https://github.com/user-attachments/assets/00b44c13-3338-43db-93b6-8ac5b039d72d" />

## Summary

- Add a standalone How to? guide with Codex and Claude setup, sample Automations prompts, and steps to verify the answers in Linear.

- Put How to? first in the sidebar and open Initiative Progress by default for signed-in users.

- Serve the guide in the app and fixture preview at a shareable URL.

## Business Value

DSS users can connect an agent and validate its answers against Linear without needing someone to walk them through the setup. Returning users land on initiative progress.

## Implementation Effort

About 2 hours for an average engineer to write the guide, wire up the route and navigation, and validate both entry points.

## Validation

- Biome, TypeScript, and repository pre-commit hooks passed.

- Verified that Express and the local fixture app serve the guide and Enablement separately.

- Verified sidebar order, default selection, and direct links for signed-in and signed-out users.

## Linear

Follow-up to [AI-1288](https://linear.app/builder-team/issue/AI-1288/add-linear-dss-initiative-progress-service).

#3937 — fix(incident-analysis): use Sonnet 5.5 and map Redshift env @sanketghia  approved

## Summary

- Map the REDSHIFT_DATABASE Secrets Manager field to the REDSHIFT_DB runtime variable expected by klair-api/utils/redshift_handler.py.

- Route Incident Analysis through Seatbelt with claude-sonnet-5-5.

- Document the secret-field to runtime-variable mapping.

Linear: [KLAIR-3735](https://linear.app/builder-team/issue/KLAIR-3735/rotate-leaked-finance-klair-automation-anthropic-key-off-remaining)

## Production validation

- A one-off PR image was built from the current production image for main commit 28e485961ff5cf4167eacdd216f6be67e533f0a6, with this branch's models/incident.py overlaid. The image was pushed under the unique tag pr-3937-979acf26f-amd64.

- An isolated x86_64 ECS smoke task using the corrected REDSHIFT_DB mapping exited 0. Importing the Incident model opened a Redshift connection, and a synthetic prompt completed through Seatbelt using claude-sonnet-5-5 (SEATBELT_MODEL_SMOKE_OK). It did not read or write incident records.

- A full one-off cron invocation from that PR task family exited 1 at its first Redshift query: SQLSTATE 42P01, relation staging_gsheets.incidents does not exist. The provided read-only credentials connect to the same Redshift host and finance_dw; information_schema.tables has no table named incidents. The cron therefore did not process or write incident records.

- The EventBridge schedule still targets incident-analysis-production:12; the temporary PR task-definition family was deregistered after the run.

## Checks

- uv run ruff check models/incident.py passed.

- uv run pyright models/incident.py reports existing type errors and warnings in this file, including the response content access and duration argument; this change only changes the model ID string.

- AWS cloudformation validate-template accepted the updated template; local cfn-lint was not run.

- The standard local amd64 Docker build hit a QEMU segmentation fault during uv pip install --system .; the validation image reused the exact deployed main image and overlaid the only application-code change in this PR.

The full scheduled job still needs the staging_gsheets.incidents relation provisioned or made visible before it can complete in production.

#34 — Show people on Linear DSS initiative rows @ashwanth1109  no labels

## Demo

<img width="2197" height="1636" alt="image" src="https://github.com/user-attachments/assets/28b7f252-18d8-4203-8930-adec9ec6af40" />

## Summary

- Replace the Reported health column in Initiative Progress with initials avatars for open project leads and in-progress issue assignees.

- Use the existing People rollup, which combines both roles by Linear user ID. Each avatar exposes the full name on hover and to assistive technology.

- Wrap the avatars within the People column. The expanded row shares the same People request and continues to show health in the latest update.

## Business Value

Readers can see who is working on each initiative without opening its details.

## Implementation Effort

Approximately 2 hours for an engineer to implement and verify manually.

## Validation

- TypeScript and Biome checks passed. Biome was scoped to the changed file.

- Fixture checks covered initials, accessible names, expanded health and People, and loading, empty, error, retry, and refresh behavior.

- Offline layout verification with 14 people confirmed wrapping without clipping at 320, 390, 768, and 1200 px.

## Linear

[AI-1288 — Add Linear DSS initiative progress service](https://linear.app/builder-team/issue/AI-1288/add-linear-dss-initiative-progress-service)

#31 — Remove self-attested DSS release gate and record approved rollout @mwrshah  no labels

## Changes

- Remove Sindri release-readiness.json and the shared check-readiness.mjs checker.

- Remove the deployment workflow call and support-manifest entry.

- Update the repository create-dss-server skill and onboarding reference: owner decisions belong in the README/release PR, while deployment runs executable configuration, build, credential and smoke checks.

- Preserve the approved native entitlement-enforcement deferral in README, along with verified access grants, validation-key preparation and completed bootstrap.

## Validation

Biome, TypeScript, Knip and Codedecorum pass. No remaining references to the deleted file/checker. Existing credential checks, response validation, builds, infrastructure synthesis and deployment smoke remain unchanged. Both service credential preflights and local Docker/UI/HTTP/synthesis checks passed during release preparation.

## Release state

Owner approved releasing both Sindri and Salesforce. Bootstrap is complete; runtime deployment remains pending GitHub PR approval. Existing local README hierarchy edits in the main workspace are preserved.

#3936 — feat(acq-perf-review-v2): add charts and payback analysis @sanketghia  approved

## Summary

- Add chart and table views across the acquisition performance metrics, with controls aligned beside the P&L tabs and Revenue/Expenses selectors.

- Compare the Acquisition Plan with a Current Base Plan composed from closed Actuals, the selected Finance budget, then Forecast.

- Keep source colors on charts only; keep tables neutrally styled.

- Stop the Current Base Plan at missing budget or forecast quarters so source gaps are not presented as valid comparison periods.

- Add a cumulative after-tax EBITDA payback trajectory with the actuals cutoff and purchase-price reference lines, while preserving legacy fallback behavior.

## Verification

- Backend acquisition performance suite: 65 passed.

- Ruff and Pyright: passed.

- Acquisition performance frontend spec: 20 passed.

- Full frontend suite: 7,003 passed, 16 skipped.

- Frontend lint, Prettier, TypeScript check, and production build: passed.

#29 — Add Linear DSS initiative progress service @ashwanth1109  no labels

## Summary

Add Linear DSS as a standalone Hub service for read-only progress across seven priority initiatives in the Builder Team workspace.

- Show initiative goals, reported health, linked project states, completion dates, target dates, and Linear source links in Initiative Progress. Scheduling stripes surface open projects without a target date.

- Follow up through People and project issues to distinguish open-project leads from assignees of in-progress issues. Issue reads cover non-private Linear teams and expose workflow states with cursor pagination.

- Support entitled WorkOS browser sessions, personal keys scoped to Linear DSS, and per-user WorkOS OAuth for the optional MCP connector. Enablement exposes the connector setup and the agent can read Enablement, BrainLift, and Dictionary through its tools.

- Label BrainLift as agent-generated guidance and describe its purpose without naming individual readers.

- Include Linear-owned infrastructure, release readiness, production context, and registration in the shared release workflow.

## Business Value

Readers can inspect priority initiative progress and ask an agent who is doing the work and how it contributes to the initiative goal. Answers can cite the underlying Linear records and identify gaps in health, scheduling, and outcome evidence. Approved users can connect through WorkOS sign-in without copying an API key into an agent chat.

## Implementation Effort

Estimated 3–4 engineering days for an average engineer working without AI assistance, using the existing DSS recipe. This includes the Linear adapter, initiative and People views, access controls, OAuth connector, infrastructure integration, and live validation.

## Verification

- TypeScript, Knip, file-scoped Biome checks, server and UI production builds, and Code Decorum passed locally. Infrastructure synthesis verified that enabling the connector adds only the Linear container's issuer setting.

- Isolated protocol checks covered all four MCP tools, wrong issuer audience, wrong organization, missing scope, expired tokens, and live entitlement revocation. Existing REST authentication and connector-disabled behavior were checked.

- On 11 Nov 2026 IST, deployed the exact candidate revision 55c3afc905f73afd77bb6dfc5b24c04b2fbb4b31 directly to the Linear runtime before merge. CloudFormation reached UPDATE_COMPLETE, the ECS task and load-balancer target were healthy, and the running digest was verified as sha256:7999be2c4249689c543810593ab6d7fc9a5553e5d4f5ff8602543f3732793577.

- Hosted checks returned seven initiatives, three people and 15 in-progress assignments for Automations, and paginated issue evidence with workflow states. Anonymous and invalid-key requests were denied. Hosted BrainLift, Enablement, and Dictionary matched the candidate files.

- A fresh ephemeral Codex run using the desktop-bundled CLI completed real WorkOS CIMD sign-in and all four hosted MCP tools. It read Enablement and BrainLift, cited Linear, separated project leads from issue assignees, and identified missing health, target dates, and measured outcomes.

- Production entitlement checks passed for the three pilot readers and six named team accounts after their additive Linear role assignments.

- Source-owner approval covers BrainLift and the project-lead, issue-assignee, and workflow-state data scope for entitled readers.

- Claude's end-to-end agent test remains pending because the local Claude account login needs refreshing.

## Release follow-up

The deployed candidate is included in this PR. The normal release workflow reads the protected linear-production environment's DSS_CONTEXT variable. Before the next workflow-driven release, add mcpIssuer: https://valid-mask-44.authkit.app to that context, matching services/linear/infra/production.json. The runtime currently has this setting from the approved direct deployment.

GitHub review is required for the updated head. The normal production promotion is a separate release step, with service selection checked against the approved rollout scope.

## Linear

- [AI-1277 — Automations live-data proof](https://linear.app/builder-team/issue/AI-1277/validate-live-linear-data-for-the-automations-initiative)

- [AI-1278 — Shared WorkOS entitlement prerequisite](https://linear.app/builder-team/issue/AI-1278/enforce-live-workos-service-entitlements-in-the-shared-dss-template)

- [Linear DSS project P-AI-288](https://linear.app/builder-team/project/linear-dss-b1f9302a0287)

#2263 — fix(aws-spend-insights): route Anthropic through Seatbelt @sanketghia  approved

## Summary

- Route the AWS spend-insights Lambda through Seatbelt using KLAIR_SEATBELT_VA_KEY from ENV_API_PROD.

- Prefix outbound Claude model IDs for the Teams route and suppress the SDK's automatic x-api-key header.

- Update the pipeline manifest and IAM resource to read the shared production secret.

Linear: [KLAIR-3735](https://linear.app/builder-team/issue/KLAIR-3735/rotate-leaked-finance-klair-automation-anthropic-key-off-remaining)

## Validation

- Ruff check and format checks passed for the changed pipeline files.

- pipeline.json parsed successfully.

- git diff --check passed.

- Tests were not run.

No live AWS secret or pipeline deployment was changed. The pipeline takes effect after deployment.

#3935 — fix(seatbelt): migrate AWS-backed callers @sanketghia  approvedmercy-allow-critical

## Summary

- Route the AWS-backed natural-language, brokerage OCR, incident-analysis, and maintenance AI-insights callers through Seatbelt with KLAIR_SEATBELT_VA_KEY.

- Update the brokerage OCR SAM template to resolve the key from ENV_API_DEV or ENV_API_PROD, and pass it only to Claude-calling functions.

- Update the BudgetBot V3 Anthropic client to use the shared Seatbelt helper; retain the referenced implementation while removing its direct API-key use.

- Update the AI-insights feature documentation and incident-analysis deployment instructions.

Linear: [KLAIR-3735](https://linear.app/builder-team/issue/KLAIR-3735/rotate-leaked-finance-klair-automation-anthropic-key-off-remaining)

## Validation

- Ruff check and format checks passed for the changed Klair API files and the CI-scoped changed files.

- sam validate --lint passed for the brokerage OCR template.

- git diff --check passed.

- Pyright still reports existing type diagnostics in models/incident.py and budget_bot/agentic_loop_v3/anthropic.py; tests were not run.

No live AWS secrets or deployments were changed. The runtime changes take effect after deployment.

#3934 — feat(acq-perf-review-v2): expose acquisition retention schedules @sanketghia  approved

## Summary

- Expose all active new-cohort acquisitions on /acq-perf-review-v2, keep CloudSense as the default, and load data for the selected acquisition.

- Apply the retention schedules transcribed from the supplied Q4'26 Acquisition Performance Plan sheets to Plan ARR and Plan Retention.

- Keep Jive and legacy handling unchanged. Contently's unlabeled G1/G2 values are excluded per confirmation.

- Document the stale "Influitive" label in Tivian's plan row; the workbook title and Acquisition Info identify Tivian.

## Validation

- Ruff, Pyright, Prettier, ESLint, and the frontend TypeScript check passed.

- Backend and frontend tests were not run.

## Screenshot

<img width="1331" height="695" alt="image" src="https://github.com/user-attachments/assets/92c2dfb7-8dd6-4ac9-b601-67f971389cb6" />

#5 — Enforce live WorkOS entitlements in generated DSS services @ashwanth1109  no labels

## Summary

- Resolve live WorkOS organization role permissions before issuing a service key or accepting a protected bearer request.

- Preserve active-member sign-in and own-key cleanup after a service grant is removed.

- Document provider setup and validate entitlement isolation in two generated services.

## Business Value

Teams can reuse the DSS recipe for Linear DSS and other services with access scoped to each service. Removing a service grant blocks the member’s next protected request and prevents new key issuance while preserving key cleanup.

## Verification

- Recipe generator check passed with two generated services, including typechecks, entitlement changes, cross-service key isolation, provider failures, and own-key cleanup.

- Biome and Codedecorum passed on the changed files.

- WorkOS behavior was verified using provider fixtures. Live provider setup and hosted smoke checks are part of service rollout.

## Implementation Effort

Estimated 4–6 hours for an engineer to implement, exercise the generated-service behavior, and update the provider setup documentation.

## Linear

[AI-1278](https://linear.app/builder-team/issue/AI-1278/enforce-live-workos-service-entitlements-in-the-shared-dss-template)

#2297 — fix(due-diligence): remove Group E requirement for Regulatory scores @caina-barbosa  approved

## Summary

Remove the US Group E occupancy check that prevents Bark Place's documented UK assessment from being saved as Regulatory 1W. This replaces the closed dropdown-based proposal in #2288 and follows the request to remove the check. This is one standalone fix; no Linear issue was supplied.

Production effect: activation. After deployment, the existing Due Diligence editor and backend accept Regulatory 1/1W without demanding Phase 2 occupancy E. No new field, dropdown option, classification mapping or data migration is introduced.

## Why

The existing check treats a US occupancy classification as a prerequisite for every favorable Regulatory rating. Removing that prerequisite allows assessors to record an international assessment using the existing explanation and outstanding-work fields.

## Business Value

- Save Bark Place's UK assessment without inventing a US occupancy classification.

- Keep the local classification and supporting evidence in Regulatory Details.

- Retain outstanding approvals in the existing visible context fields.

## How does it work

1. Delete the Phase 2 E occupancy requirement from validateAerieDueDiligence in the shared contract, covering backend validation for UI, API and agent callers.

2. Delete the matching check from the existing Due Diligence editor so its Save control follows the same behavior.

3. Delete the obsolete test case for the Group E requirement and its obsolete assertion in the broader occupancy test. No tests are added or rewritten.

4. Existing fields, 1W formatting, context, school-mode and zoning rules retain their current behavior. This PR removes the specific E prerequisite, not every diligence validation rule or authentication check.

## Scope

### Included in this phase

- Removal of the Regulatory 1/1W Group E prerequisite.

- Exact final diff paths:

chat/components/dashboards/portfolio/cards/due-diligence-editor.tsx

packages/contracts/src/due-diligence.test.ts

packages/contracts/src/due-diligence.ts

### Deliberately excluded for later phases

- The abandoned Regulatory Basis dropdown and England-specific data model from #2288.

- Changes to other validation rules, new AI-review functionality, permissions or approval policy.

- Production edits, backfills, deployment, and external writeback.

## Test plan

### Automated validation

- Reproduced the pre-fix rejection: Regulatory score 1 requires Phase 2 occupancy type to be E. A local replay now accepts both 1 and 1W without occupancy, with pending approval text present.

- Green: contract regressions 55/55 passed (pnpm --dir packages/contracts exec vitest run src/due-diligence.test.ts --maxWorkers=1).

- Existing editor and base diligence suites: 44/44 passed (pnpm --dir chat exec vitest run components/dashboards/portfolio/cards/__tests__/due-diligence-editor.test.tsx convex/portfolio/dueDiligence.test.ts --maxWorkers=1).

- Contracts typecheck passed (pnpm --dir packages/contracts run typecheck).

- Full lint and architecture checks passed (pnpm lint); existing informational diagnostics and warnings remain.

- App typecheck passed in the commit hook. Hosted CI is the final gate.

- git diff --check passed; the diff contains only the three paths above.

- Diff size: zero added lines and 35 removed lines: 12 production lines removing the check, and 23 obsolete test lines. No test code was added.

### Time for Implementation

Approximately half an engineer day without AI assistance for tracing both checks, updating assertions, validation and PR preparation.

#2261 — fix(finance): preserve Education source counterparties in Core @ashwanth1109  approved

The Core GL writer still replaces Education Travel and donation counterparties with generic labels. Preserve source line/header names for Education Actual NHC OPEX/COGS, retaining NULL when no usable name exists.

## Business Value

Preserves real Travel suppliers and foundation recipients during Core reader migration, with existing accounting amounts, classifications and Vendor identities unchanged.

## Validation

- All 199 tests passed in the affected runner suite; Ruff passed on the new test file.

- The actual CASE expression passed 26 read-only live Redshift fixtures.

- Current-year preview: 2,327 Travel names ($2,120,113.64), 164 unresolved Travel rows and two donation counterparties ($2,506,000).

- No deployment, pipeline invocation or financial write was performed. Required pre-merge deployment/run validation is pending; keep this PR draft.

## Deployment Gates

Duplicate CF Order adjustment class mappings fail the canonical writer preflight. Reconcile the deployed procedure with main's separate implementation procedure before DDL application. Core and legacy reporting also differ by 97 rows and $183,862.22 net; reconcile those source publications before reader activation.

## Implementation Effort

Approximately 4–6 hours for an engineer to recover the prior rule, implement the Core port and semantic fixtures, and establish the fresh live preview and rollout blockers without AI assistance.

## Linear

https://linear.app/builder-team/issue/SURTR-1706

#2296 — 1442-2sindri-pr-cards: Align Aerie credentials with the Sindri contract @mwrshah  changes requested

- Regenerate Aerie's client types from live Sindri development /skillspec, matching the committed 2026-11-06 contract.

- Replace credential ownerType with accessScope across creation, lists, details, and workflow credential pickers.

- Use canonical WorkOS identity for creator-aware lifecycle controls; preserve creator management after promotion and restrict supervisor management to organization credentials.

Rollout: coordinate with the matching Sindri contract. Production /skillspec returned HTTP 500 during inspection; no production changes were made.

#57 — 008-sindri-workos-roles @mwrshah  no labels

- Require live dss-rhodes:access from current assigned-role permissions for read/write WorkOS key use and issuance; retain active-member own-key cleanup, exact key scopes and source write authority.

- Update protected fixtures under explicit Codedecorum approval; add entitlement removal/regrant, custom/multiple-role, failure-closed and key-lifecycle coverage. Document the unchanged legacy-key migration exception and regenerate publication metadata.

- Validated: 91 tests, check, build, synth without lookups, required Codedecorum diff gate, high-severity audit, Linux AMD64 image build and redacted history/staged secret scans. Audit retains one moderate advisory; optional full Codedecorum flags two unchanged test suites.

- Production's separate additive 147-member role migration is reported complete. No merge or deployment here; release remains the manual deploy.yml workflow on main, with production prerequisites in the README.

#3932 — feat(passive-investment): route Claude through Seatbelt @sanketghia  approved

## Summary

- Route the three Passive Investment Claude Messages call paths through Seatbelt with the product VA key and claude-teams-group/claude-sonnet-5-5.

- Require KLAIR_SEATBELT_VA_KEY; retain ANTHROPIC_API_KEY as optional but unused by these clients.

- Add ENV_SECRET_ID selection for Dev while keeping ENV_API_PROD as the default.

- Preserve the deployed Anthropic/OpenAI SDK major versions to avoid an unrelated dependency upgrade.

## Validation

- Updated the passive-investment-email Lambda from a package built on the saved deployment package, preserving its installed dependencies. Lambda reports LastUpdateStatus=Successful.

- Live invocation with Nov 9 / Nov 6 dates: CloudWatch recorded three executions, 18 successful Claude analyses, three completed email sends, zero OpenAI fallbacks, and zero unhandled errors. The synchronous invoke client timed out while Lambda continued processing; the three emails were received. No further invocation was made.

- Maat showed Sonnet 5.5 traffic through seatbelt-ai-engineering, subscription_billed=true, and $0 actual spend for Nov 10.

- Formatting and git diff --check passed. Targeted Ruff/Pyright checks for the new client and secret loader passed. Full Ruff/Pyright still report findings in untouched legacy sections. No automated tests were run.

## Secret and deployment notes

- The Lambda reads ENV_API_PROD by default. Set ENV_SECRET_ID=ENV_API_DEV on a Dev Lambda to select ENV_API_DEV.

- Secret values were managed outside this PR; no secret values are included here.

#230 — 1344-reclaim-runner-name @mwrshah  changes requested

Bot output:

- Rename the execution concept from Runner to Executor across runtime modules, schema references, callbacks, headers, environment names, CI/CD and documentation; no compatibility aliases.

- Use Harness for agent harness selection. Preserve Sindri product naming, AWS resource names and GitHub runner contexts.

- Advance the discovery contract revision to 2026-11-10 and regenerate OpenAPI.

- Remove one-time migration tools, their 38 tests and extensive runbooks from the shipped diff; preserve them outside the repository. Keep established renamed tests. Net diff: −16 lines; generated OpenAPI accounts for repeated revision-token replacements.

- Local checks passed: lint/typechecks, 633 application tests and 199 executor tests.

Production handoff: The approved maintenance cutover migrated and verified 181 activations, 416 activation references and 364 event types. Old table rows and the migration ledger were cleared after backup verification. Convex prod/dev env names and the ECS container/secret-entry name have been updated without changing secret values. Production was left on temporary maintenance code for the user's normal full release; this PR's final code and image must replace it. No isolated release candidate was deployed. Dev/preview database migration is not implied by the env-name changes.

Scope: Existing ECS termination retry and late-dispatch ownership risks are acknowledged, not fixed or introduced here. Review replies distinguish those existing risks from incorrect findings and the completed one-time migration requirements.

#3930 — feat(klair): Deprecate Unused Code (KLAIR-3698) @trilogy-drones-fleet[bot]  changes requested

## What

Delete code nothing calls any more in four separate PRs: (1) four backend /api/aws-spend/net-amortized/by-* endpoints and their service methods, helpers, models, tests and docs; (2) four deprecated frontend items with no callers; (3) the files the retired Report Service left behind, plus two committed backup-and-drop Redshift scripts that a person runs by hand; (4) the dropped tables' entries in the MCP table-comments script. Code slices 010-016 follow the spec's PR boundaries. Slices 010-011 make PR 1, 012 is PR 2, 013-015 are PR 3 and 016 is PR 4. Nothing depends on anything across PRs, so these slices can stack on the one branch and still be split back into the four PRs. Slices 017-020 are operator runbooks: prod deploys and checks, the day-14 CloudWatch check, running the scripts and filing the record PRs, and the 30-day backup cleanup. Greenfield: the repo has no numbered specs and no roadmap (only an unnumbered _archive/specs/ folder), so this is the first roadmap and ids run consecutively from 010.

## Why

Requested in [AI-Builder-Team/Klair#3929](https://github.com/AI-Builder-Team/Klair/issues/3929).

## Tests

- 010 remove-net-amortized-by-endpoints: ai-builder-team/klair: test failed (sticky), lint failed (sticky), latest test exit 0, relation tested-clean-match; product acceptance not performed

- 011 net-amortized-by-docs-cleanup: ai-builder-team/klair: test none, lint failed (sticky), relation absent; product acceptance not performed

- 013 delete-report-service-leftovers: ai-builder-team/klair: test passed, lint failed (sticky), latest test exit 0, relation unknown-unattributed; product acceptance not performed

- 014 report-service-drop-script: ai-builder-team/klair: test passed, lint passed, latest test exit 0, relation tested-clean-match; product acceptance not performed

- 015 aws-spend-outliers-drop-script: ai-builder-team/klair: test passed, lint passed, latest test exit 0, relation unknown-unattributed; product acceptance not performed

- 016 mcp-table-comments-drop-entries: ai-builder-team/klair: test failed (sticky), lint failed (sticky), latest test exit 0, relation tested-clean-match; product acceptance not performed

---

## Fleet metadata

6/11 slice(s) built, accumulated on feature/deprecate-unused-code.

> DRAFT — a human engineer owns the review + merge. Per-slice tests are flag-not-block (some suites are CI-only); verify in your own CI before marking ready.

Source issue: [AI-Builder-Team/Klair#3929](https://github.com/AI-Builder-Team/Klair/issues/3929)

_Drafted by trilogy-drones (feature build, cost $46.78). feature/deprecate-unused-code → the repo default branch._

#3931 — feat(anthropic): route Klair API through Seatbelt @sanketghia  approved

## Summary

- Route deployed klair-api direct Anthropic Messages calls through the Klair Seatbelt VA key, with the Teams model prefix and legacy x-api-key suppressed.

- Route user-controlled Claude Agent SDK and E2B Messages traffic through an authenticated Klair proxy. The API validates user authentication, restricts upstream paths to Messages/count-token requests, enforces the Teams prefix on both, and attaches the VA key server-side.

- Keep the product VA key and custom header out of E2B environments and user-controlled CLI subprocesses, including resumed sandboxes with stale environment state. Interactive Claude Code fails closed if a request-scoped proxy token is missing.

- Replace ARR-gap's denied claude-opus-4-5-20251101 model with claude-opus-5-5.

- Add a Messages compatibility probe, focused tests, and configuration documentation.

## Mercy review and CI follow-up

All actionable code findings were fixed. Board Doc streaming fixtures now use a valid Claude model ID. The E2B/Claude Code credential exposure is closed by the authenticated proxy. Count-token requests receive the same Teams-model validation as Messages, and the probe reports semantic mismatches as contract_error and exits nonzero.

The account-analysis SSE status suggestion was left unchanged: failures produce an explicit event: error, and the frontend parser calls onError; HTTP 200 is the established streaming response status. The deployment-secret note remains an external pre-deploy task; values are not stored in the repo.

## Validation

- Network-isolated Board Doc CI-equivalent suite: 5,190 passed, 2 intentionally deselected, 63 warnings.

- Selected proxy, E2B, Account Analysis, AWS Spend, ARR-gap, Budget Planner, LMService, and QTD suites: 642 passed, 2 deselected, 14 warnings.

- Ruff 0.15.22 format/check passed. Pyright on the shared client, proxy router, and probe: 0 errors, 0 warnings. The broader touched-file Pyright run reports 340 errors and 137 warnings outside edited lines; changed-line filter found 0 diagnostics.

- Maat shows product-klair-prod usage routed through seatbelt-ai-engineering, including Opus 5.5, with subscription billing and $0 actual spend.

- One unrelated local E2B file-transfer test still fails because its session mock leaves claude_session_id as a non-serializable MagicMock; it is not part of the Board Doc CI job.

## Deployment

The API still needs KLAIR_SEATBELT_VA_KEY in the ENV_API_DEV and ENV_API_PROD runtime secret payloads before deployment. ANTHROPIC_API_KEY may remain. KLAIR_API_BASE_URL is an optional non-secret override; local E2B testing requires a publicly reachable API URL. No runtime secrets were edited and no deployment was performed.

#229 — 1343-last-10 @mwrshah  approved

- Clarify one-time approved Aerie onboarding versus ordinary sign-in.

- Document Forge-only removal as preserving existing DSS state.

- Explain bounded DSS restoration and administrator intervention.

- Distinguish paired-membership integration rules from generic key and JWT authorization.

#28 — 2037-sindri-workos-roles @mwrshah  no labels

- Require live per-service DSS permissions across assigned roles for protected requests and key creation.

- Preserve active members’ own-key listing and revocation and existing key scope boundaries.

- Update the creation skill, setup guidance, and service documentation.

- Mark native Sindri entitlement enforcement, external recipe updates, and approved role assignments as rollout prerequisites.

#2295 — 1441-implement-aerie-access @mwrshah  changes requested

- Separate Aerie authorization from Sindri-native authorization; use verified canonical identities and live Sindri permissions for native calls, delegated work, and shared UI controls.

- Deliver approved onboarding through Sindri to establish Forge Integration and DSS memberships with baseline Operator access. Unify onboarding, explicit regrant, and removal with stable operation IDs, bounded retries, and visible delivery status.

- Keep ordinary login and reads from restoring removed Forge access; preserve later role changes and refresh visible UI permissions every 30 seconds.

- Preserve local resource policies, retain verified legacy-identity migration, remove impersonating health probes, and correct provisioning/audit feedback.

- Coordinate release with [Sindri #227](https://github.com/AI-Builder-Team/Sindri/pull/227): the new lifecycle transport replaces legacy enrollment/removal endpoints without a compatibility fallback.

#228 — 1342-sindri-workos-roles @mwrshah  approved

- Document access states and grant/removal transitions in the root README.

- Separate WorkOS, Sindri, Aerie, and DSS responsibilities and permission boundaries.

- Include the complete Alice membership diagram and production configuration caveat.

#227 — 1341-aerie-onboarding-lifecycle @mwrshah  changes requested

- Fulfill verified-identity Forge onboarding in Sindri: establish Forge Integration and DSS memberships with baseline access while preserving elevated roles.

- Add idempotent lifecycle intents, status polling, stale-request rejection, and bounded recovery; keep live WorkOS permissions authoritative.

- Preserve intentional DSS restoration and lazy profile hydration; remove Forge access without removing DSS membership.

- Retire legacy email enrollment and unversioned removal endpoints. Coordinate release with [Aerie #2295](https://github.com/AI-Builder-Team/Aerie/pull/2295); deploying this cutover alone breaks old Aerie lifecycle callers.

#226 — 1340-speed-up-ci @mwrshah  no labels

## Summary

- Populate the pnpm package cache once before parallel CI checks, then install offline on each isolated Blacksmith runner.

- Reuse the existing cache-key format and restore compatible cached packages when the lockfile changes.

- Run agent-runner checks through the pnpm workspace instead of a separate npm installation.

- Preserve required check names and fail them explicitly when dependency preparation fails.

## Validation

- Actionlint, workflow contract checks, lint, TypeScript checks, and application/runner suites passed locally.

- Clean-checkout installation: 39.6 seconds to populate the store; 5.8 seconds offline with zero downloads (local measurements, not hosted CI timings).

- Independent read-only review found no concrete blockers.

- Hosted cache handoff and production build will be validated by this PR's CI run.

#225 — 1339-implement-sindri-access @mwrshah  changes requested

## Changes

- Require explicit sindri:access across browser/Convex, JWT, native key, M2M, and delegated execution boundaries; membership and profiles alone grant no product access.

- Share lazy, indexed WorkOS profile bootstrap across entry paths, with no warm profile fetch/write.

- Validate the exact DSS key audience and both live memberships; preserve generic key restrictions.

- Centralize Forge grants/removal and restore DSS membership whenever current Forge access permits it. Add signed membership-event ingress with durable, idempotent delivery and retries—not a role/access mirror.

- Enforce agent/skill tenant and private-resource update rules in shared backing handlers.

- Remove the mocked test suites added or modified by this PR and their unused fixture; retain shared fixtures needed by untouched suites and the real WorkOS validation script.

- Update generated bindings and documentation references without changing production behavior; include configuration examples and cutover documentation.

## Cutover

WorkOS requires the explicit access and DSS audience permissions, Operator/Supervisor/Admin bundles, a no-access Member default, Multiple Roles, and deliberate Operator assignments. Configure the DSS organization and signed Forge/DSS membership webhooks before deployment. Production configuration is unchanged by this PR.

#2290 — feat(alpha-ops): connect authenticated storage endpoints (AI-1087) @caina-barbosa  approved

## Title

feat(alpha-ops): connect authenticated storage endpoints (AI-1087)

---

## Summary

This is Phase 4 of the four planned backend slices in [AI-1057 — Migrate Ops Portal storage to Rhodes](https://linear.app/builder-team/issue/AI-1057), tracked under [AI-1087 — Prepare the production target and inactive historical copy](https://linear.app/builder-team/issue/AI-1087). It connects the accepted portal-record and transfer operations to two authenticated HTTP endpoints, building on the schema, codecs and internal Chat/control operations from the preceding slices.

Production effect: activation. Deployment registers /alpha-ops/v1/application and /alpha-ops/v1/transfer in Aerie's existing router. Access retains the accepted separate tokens, source and destination checks. The code starts no import, scheduled synchronization or provider action, and changes no portal database setting. Deploying the combined series and importing production data remain separate, user-controlled steps.

---

## Why

The previously added internal storage functions need the accepted external interface before the existing portal adapter and transfer commands can use them. This slice preserves complete collection reads, staged saves and transfer operations while adapting their function references to Aerie's module paths. Composing the existing router keeps Aerie's unrelated endpoints in place.

---

## Business Value

- Gives the existing portal adapter a compatible Rhodes storage endpoint.

- Supports the reviewed initial copy, final update and recovery procedures without automatically running them.

- Preserves source content and complete reads while users save later changes.

---

## How does it work

1. Root http.ts calls registerAlphaOpsRoutes to add the two Alpha Ops routes alongside existing Aerie routes.

2. The handler reuses Aerie's bearer-token validation and constant-time comparison. It checks the configured source/destination and retains application-versus-transfer operation permissions before calling storage functions.

3. general.ts opens a complete collection version for a reader and returns its records in batches. An existing reader keeps its original version when a later save is published.

4. generalWrite.ts stages ordinary portal saves and publishes their complete content. storage.ts provides the accepted transfer operations, including the AI-1083 update/removal behavior.

5. Internal references use the alphaOps/ namespace and Aerie's generated registration helpers. Source fields, tokens, metadata, ordering and existing errors remain governed by the reviewed upstream implementation; no new product policy is introduced.

---

## Scope

### Included in this phase

- Accepted general-record reads/writes, transfer storage and authenticated HTTP composition.

- Focused HTTP tests using the exact captured collection and SQL envelope; an ordinary edit to the copied collection checks old/new read separation.

- Generated API declarations and the updated backend guide.

- Exact final diff paths:

chat/convex/_generated/api.d.ts

chat/convex/alphaOps/general.ts

chat/convex/alphaOps/generalWrite.ts

chat/convex/alphaOps/http.test.ts

chat/convex/alphaOps/http.ts

chat/convex/alphaOps/storage.ts

chat/convex/http.ts

docs/alpha-ops-ingestion.md

Measured diff against updated main: 2,756 additions / 11 deletions = 2,767 changed lines. Tests: 204 lines (7.3726%); runtime: 2,523; generated declarations: 8; root-router integration: 2; documentation: 30. These counts were verified against main after PR3 #2289 merged.

### Deliberately excluded for later phases

- Production deployment, environment configuration, initial import, final data update and portal switch — separately authorized operations; none runs on merge.

- Maintenance-script adaptations — still deferred pending Sam's usage confirmation.

- UI changes, new permissions, product improvements, CI configuration changes and cleanup of the original database adapter — outside this slice.

---

## Test plan

### Automated validation

- New HTTP tests 3/3 passed: pnpm --dir chat exec vitest run convex/alphaOps/http.test.ts --maxWorkers=1. They check authentication, target and scope rejection; exact captured collection transfer; a normal edited save with distinct old/new reads; and exact captured SQL-envelope transfer.

- Assembled focused suite 18/18 passed: pnpm --dir chat exec vitest run convex/alphaOps convex/rhodesMigrationRetirement.test.ts --maxWorkers=1 (17 Alpha Ops checks plus the existing Rhodes migration retirement regression).

- Application and Convex typechecks — passed: pnpm --dir chat run typecheck:app and pnpm --dir chat run typecheck:convex.

- Full pnpm lint, exact-file Biome and git diff --check — passed. Existing warnings/nonblocking upstream style suggestions remain; no check was weakened.

- Independent high-effort review freshly ran the new HTTP suite 3/3 and checked runtime equivalence against the accepted source, all fixture components against the private capture, complete module coverage and preservation of existing root routes. No findings. The actual new-code packet contains no invented operational fixture or compatibility expansion.

- Updated-main alignment passed: head 3694b73a1669d805d435d164fe425b59241c9462 is based on main a9b5a360a4e751dc6808600ee7223eeddbe477c6. Its complete code tree is identical to independently reviewed checkpoint 11ae63890bfbec7a17ff3240e92107e1ba688c3a; only Git history changed. Normal commit hooks passed before rebase.

- Hosted CI 15/15 jobs passed at head 3694b73a1669d805d435d164fe425b59241c9462: [run 42631422678](https://github.com/AI-Builder-Team/Aerie/actions/runs/42631422678). No deployment, live source write, import or provider action was performed.

- Runtime named-export inventory accounts for all 31 accepted upstream modules, including the intentional reuse of Aerie's shared authentication implementation. This structural check does not replace behavioral review.

### Time for Implementation

Estimated 3–5 working days for an engineer without AI assistance to integrate the accepted storage code and router, reconcile strict types and internal references, verify source contracts, test the HTTP boundary and complete review and release preparation. This is a human-effort estimate, not measured runtime or the full migration duration.

## Review repairs and contract clarifications

Blocking-only triage of [review 5827095647](https://github.com/AI-Builder-Team/Aerie/pull/2290#pullrequestreview-5827095647), at unchanged head 3694b73a1669d805d435d164fe425b59241c9462, under the existing [AI-1087](https://linear.app/builder-team/issue/AI-1087) scope. No repair ticket or code change is warranted: the sole blocking finding rests on an array-only contract that this storage port does not have.

The collection address identifies an original public.store row. The captured source schema declares data jsonb DEFAULT '[]'::jsonb NOT NULL and a primary key on name; it has no array-only CHECK. The default does not constrain subsequent JSON values. All 43 captured collection rows are arrays, and the original lease/sentinel writers also insert arrays. Supporting raw JSON is preservation of the source column/accepted transfer contract, not a claim that those captured rows contain objects or scalars.

The full value flow is explicit:

1. [protocol.ts:91–93](https://github.com/AI-Builder-Team/Aerie/blob/3694b73a1669d805d435d164fe425b59241c9462/chat/convex/alphaOps/protocol.ts#L91-L93) parses collection JSON before any materialization. Invalid JSON throws; it does not reach a successful publish.

2. [records.ts:3–6](https://github.com/AI-Builder-Team/Aerie/blob/3694b73a1669d805d435d164fe425b59241c9462/chat/convex/alphaOps/records.ts#L3-L6) distinguishes a valid non-array from invalid JSON. Returning null means a valid raw root.

3. [http.ts:727–753](https://github.com/AI-Builder-Team/Aerie/blob/3694b73a1669d805d435d164fe425b59241c9462/chat/convex/alphaOps/http.ts#L727-L753) rejects incomplete staging, checks every chunk, validates JSON, preserves the exact text, and publishes arrays as ordered records or a non-array as exactly one record with rootKind: 'raw'. It does not silently relabel that root as an array.

4. The accepted staged-import path independently allows exactly array or raw, requiring one record for raw roots ([general.ts:288–297](https://github.com/AI-Builder-Team/Aerie/blob/3694b73a1669d805d435d164fe425b59241c9462/chat/convex/alphaOps/general.ts#L288-L297)). Published heads/snapshots and read pins retain that root kind. The accepted portal adapter reconstructs an array only for an array root; otherwise it parses the original single raw token. Array-dependent helpers retain the source errors for non-array values.

Thus HTTP 200 for valid non-array JSON is intentional storage compatibility, not acceptance of malformed source data. Requiring arrays would introduce a new source constraint and change the approved port. The source snapshot/importer and accepted adapter already use this same array/raw distinction. No code/test change or synthetic fixture was added. An earlier explanation tying raw support specifically to scalar/object leases was inaccurate; actual lease writes use [] and are not the justification.

The other three findings are explicitly nonblocking in the review metadata (two hardening suggestions and a deferred coverage gap). They are not included in this blocking-only response. Deployment/environment provisioning remains the already excluded operational phase. Existing evidence remains 18/18 assembled local tests, 3/3 independently rerun HTTP tests and 15/15 hosted CI jobs at this exact head; no redundant test campaign, commit, push, merge, deployment or import was performed.

#2291 — AERIE-3094: Relinked unreadable asset files report not searchable and keep their Drive details @kevalshahtrilogy  approved

## Testing contract

### What this PR delivers

Fixes two problems Robbie found after relinking the Chapala walkthrough film (mp4) onto a document that previously held an indexed PDF (follow-up to AERIE-3092 / #2284):

1. Wrong search status. The record reported terminal_failure / software_change_required instead of not_searchable / unsupported_type. The previous PDF's indexed copy stayed current, so the status projection saw "not searchable beside a current version" and fell back to its inconsistency status. That old PDF content also stayed searchable under the film's record. Now, when a document's source changes to a file the indexer can't read, or it is reindexed, the previous indexed copy is revoked (cleanup job queued, current version cleared), the same way an external link already revokes it.

2. Null file details. Relink deliberately clears the previous file's Drive metadata, but never recorded the new file's, so sourceFilename, sourceSizeBytes and both checksums returned null. The site Drive verifier now returns the selected file's Drive metadata (name, size, checksums, created/modified times, version), and relink stores it as the document's source.

### Who uses it and where

Callers of the public API v2 relink operation (POST /v2/portfolio/sites/{siteRef}/documents/{documentRef}/relink), through Rhodes, registering walkthrough films and renderings. Affects document detail and list reads (knowledgeStatus, provenance) and document search.

### Conditions needed

- A site with a managed Drive folder, and a matterportWalkthrough or rendering document whose current file was indexed (e.g. a PDF with a ready knowledge version).

- A write key allowed to relink documents on that site.

- A video or other unreadable file in the site's managed folder.

### Expected behavior and examples

- Relinking that document to an mp4 returns knowledgeStatus { state: "not_searchable", reasonCode: "unsupported_type" }, never software_change_required. The previous indexed copy is queued for cleanup and is no longer current.

- Reindexing an unreadable asset file that still has an indexed copy from an earlier file retires that copy too.

- Records already left inconsistent can't be reindexed from the app (it refuses because canRetry is false). An internal mutation, documentKnowledge/repair:retireStaleIndexedCopyForUnreadableAsset, repairs one such record by documentPublicId. It does nothing unless the document is an unreadable asset file with a not-searchable outcome beside a current indexed copy, and a second run is a no-op.

- The relink response and later detail reads show the new file's provenance.sourceFilename, sourceSizeBytes, sourceSha256Checksum, sourceMd5Checksum, sourceCreatedAt, sourceModifiedAt and sourceVersion.

- Readable relinks keep the existing behavior: the previous version stays current until the new one is indexed.

### Limits and unanswered questions

- Deploy order is safe either way. Aerie accepts verifier responses with or without the new driveFile field. In CD, Convex deploys before the Rhodes worker, so for a few seconds relinks record no file details, as they do today. A rollback of Aerie alone, with the worker still on this change, would make relinks fail verification until both match.

- Records already in the broken state are not repaired automatically. After deploy, doc_01M6QS6K184SF10S24JM69KZMX needs the repair mutation run once on production (a separate, approved step), and a fresh relink to record its file details.

- Global document registration does not record Drive metadata either; out of scope here.

## Business Value

Walkthrough films and renderings registered through Rhodes now show an honest status ("listed, not searchable") instead of a "software update required" error, and stale PDF text stops being returned by search under a film's record. Each asset record also carries the file's name, size and checksums, so the ops team can confirm the right original is linked without opening Drive. This unblocks Robbie's asset cleanup for the diligence sites.

## Manual Effort Estimate

Proposed: ~5 hours (tracing both causes through the knowledge lifecycle and the Drive verifier, two coordinated changes, tests). Keval to confirm or adjust.

## Verification

- New tests fail without the fix and pass with it: lifecycle.test.ts (relink revokes previous copy; reindex retires a stale copy; the repair mutation fixes a stale record once and leaves readable documents alone), documentsHttp.test.ts (film relink records Drive details), global-document-drive-verification.test.ts (verifier returns file metadata).

- vitest on convex/documentKnowledge, documentsHttp.test.ts, rhodes/runtime/writes: 249 passed. Rhodes worker tests: 331 passed.

- tsc (chat app, convex, rhodes-worker), biome, boundary, read-bound and test-architecture lints pass.

- Not verified against production.

Linear: AERIE-3094

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#2289 — feat(alpha-ops): add internal Chat and control operations (AI-1087) @caina-barbosa  approved

## Title

feat(alpha-ops): add internal Chat and control operations (AI-1087)

---

## Summary

This is Phase 3 of 4 currently planned backend slices in [AI-1057 — Migrate Ops Portal storage to Rhodes](https://linear.app/builder-team/issue/AI-1057), tracked under [AI-1087 — Prepare the production target and inactive historical copy](https://linear.app/builder-team/issue/AI-1087). It adds the previously reviewed internal storage operations for Campus Chat and existing API counters and flags.

Production effect: dormant/additive. The new operations are internal Convex functions. There is no HTTP route, production caller, scheduled job, source connection or import in this slice. Merging starts no data movement or external traffic. Production release and the portal's database switch remain separately authorized steps.

---

## Why

Chat stores identity, consent, messages, approvals and delivery-processing state. Those operations need the same source constraints and duplicate-handling rules after the database move. This layer builds on the source contracts and conversion functions in the preceding main-target PRs, before the separate HTTP boundary connects callers.

---

## Business Value

- Preserves existing Chat records and processing decisions through the database move.

- Retains the application's current request counters and control flags.

- Makes the storage operations independently reviewable before connecting an external caller.

---

## How does it work

1. sqlRows.ts, sqlKeys.ts and storeClock.ts preserve complete source envelopes, source-qualified lookup keys and immutable transfer snapshots. Current-row lookups take precedence over imported rows.

2. apiControls.ts implements the existing counters and flags. Read and write method gates retain their current separation.

3. The four Chat operation modules handle identity/consent, message intake, approvals and delivery controls. Their source constraints, SQL-style failures and projections are retained from the accepted implementation.

4. Derived indexes update within the same transaction as their source envelope. The source content remains authoritative.

5. Aerie's generated internal query/mutation helpers register the operations. Generated API declarations add the Alpha Ops modules while preserving existing component declarations. Nothing calls these functions from a production route yet.

---

## Scope

### Included in this phase

- Internal Chat/control operations and their shared storage helpers.

- Three focused tests using exact captured source envelopes and technical method-gate checks.

- Generated API declarations and an updated backend guide.

- Exact final diff paths:

chat/convex/_generated/api.d.ts

chat/convex/alphaOps/apiControls.ts

chat/convex/alphaOps/chatApprovals.ts

chat/convex/alphaOps/chatApprovalsProjection.ts

chat/convex/alphaOps/chatDelivery.ts

chat/convex/alphaOps/chatDeliveryProjection.ts

chat/convex/alphaOps/chatIdentity.ts

chat/convex/alphaOps/chatIntake.ts

chat/convex/alphaOps/operations.test.ts

chat/convex/alphaOps/sqlKeys.ts

chat/convex/alphaOps/sqlRows.ts

chat/convex/alphaOps/storeClock.ts

docs/alpha-ops-ingestion.md

Measured diff against updated main: 2,072 additions / 5 deletions = 2,077 changed lines. Tests: 167 lines (8.0404%); runtime/helpers: 1,837; generated declarations: 56; documentation: 17. The reviewed layer was rebased after PR2 #2287 merged; the complete resulting code tree is identical to the independently reviewed checkpoint.

### Deliberately excluded for later phases

- General portal-record writes, transfer operations and authenticated HTTP composition — next backend slice, subject to the same size limit.

- Production import, deployment, portal switch, maintenance-script adaptations, UI changes and new product rules — outside this slice.

- CI configuration changes — none.

---

## Test plan

### Automated validation

- New internal-operation tests 3/3 passed: pnpm --dir chat exec vitest run convex/alphaOps/operations.test.ts --maxWorkers=1. They exercise captured SMS-control writes, counter increments/reset, immutable snapshots, source isolation and read/write method gates across all four Chat domains.

- Combined foundation/codec/operation checks 14/14 passed. The independent high-effort reviewer freshly reran the new suite and checked both fixture envelopes against the actual captured source snapshot, rather than trusting the implementation report.

- Application and Convex TypeScript checks — passed; saved output contains no diagnostics.

- Full pnpm lint — passed, with existing unrelated warnings and nonblocking formatting suggestions. No rule was disabled.

- git diff --check — passed.

- Independent review compared operation behavior with the accepted upstream implementation and inspected source-data provenance, namespaces, constraints, method gates and transactional indexes. No findings.

- Updated-main alignment and exact diff measurements — passed. Head 53465a83b3d366ca8a63fc46e0572aa30882ad03 has main 5920274c29398496b8ffdf4535b941dcd264fd55 as its base; the full code tree equals reviewed checkpoint 23c1fa23ead0a481865082332789064fe5bc868f. Normal commit hooks passed before rebase.

- Hosted CI 15/15 jobs passed at head 53465a83b3d366ca8a63fc46e0572aa30882ad03: [run 42629140130](https://github.com/AI-Builder-Team/Aerie/actions/runs/42629140130). No deployment or live source/provider operation was performed by these tests.

### Time for Implementation

Estimated 3–5 working days for an engineer without AI assistance to integrate these accepted operations, adapt generated types, check source behavior, write focused tests and complete review and release preparation. This is a human-effort estimate, not measured runtime or the full migration duration.

#2287 — feat(alpha-ops): add source-compatible codecs (AI-1087) @caina-barbosa  approved

## Title

feat(alpha-ops): add source-compatible codecs (AI-1087)

---

## Summary

This PR is Phase 2 of 4 currently planned backend slices in [AI-1057 — Migrate Ops Portal storage to Rhodes](https://linear.app/builder-team/issue/AI-1057), tracked as backend preparation for [AI-1087 — Prepare the production target and inactive historical copy](https://linear.app/builder-team/issue/AI-1087). It adds the pure conversion helpers needed to preserve the portal's existing SQL and JSON behavior when records are stored in Rhodes.

Production effect: dormant/additive. The foundation from [PR #2286](https://github.com/AI-Builder-Team/Aerie/pull/2286) is already merged into main. This PR adds plain helpers and reference data only: no registered Convex functions, HTTP routes, cron, source connection, import or production caller. Merging starts no data movement or external traffic. The combined production release remains separate from sequential main merges.

---

## Why

The database move must preserve source timestamps, IDs, arrays, JSON values and text lookup behavior. Centralizing the reviewed conversions before adding storage operations prevents each operation from interpreting those values differently. This PR depends only on the foundation already in main.

---

## Business Value

- Keeps source values and identifiers consistent across the database move.

- Preserves existing text matching and error behavior instead of introducing new product rules.

- Gives the internal-operation slice a tested, shared conversion boundary.

---

## How does it work

1. chatCodec.ts converts supported source UUID, timestamp, SQL array, JSONB and text values and preserves the existing SQL-style failure behavior. chatDeliveryCodec.ts supplies the reviewed delivery boolean/integer/numeric conversions.

2. pgDate.ts retains the existing PostgreSQL date projection. Strict TypeScript adaptations make its existing coercions explicit without changing runtime behavior; the original license is included.

3. unicodeSimpleLower16.ts preserves all 1,460 reviewed Unicode simple-lowercase pairs. unicodeProvenance.json records the official source version/hash and extraction rule, and the Unicode license is included.

4. jsonb.ts retains the reviewed JSONB manipulation, ID allocation and token-preserving behavior, avoiding new general-purpose SQL emulation.

5. The helpers remain uncalled by production paths until later internal-operation and HTTP-integration slices. No schema or user permission changes are introduced here.

---

## Scope

### Included in this phase

- Pure source-compatible codecs, JSONB helpers and retained PostgreSQL date parser.

- Unicode reference data, provenance and both original licenses.

- Six focused codec tests and the backend guide update.

- Exact final diff paths:

chat/convex/alphaOps/chatCodec.ts

chat/convex/alphaOps/chatDeliveryCodec.ts

chat/convex/alphaOps/codecs.test.ts

chat/convex/alphaOps/jsonb.ts

chat/convex/alphaOps/pgDate.ts

chat/convex/alphaOps/postgresDate.LICENSE.txt

chat/convex/alphaOps/unicode.LICENSE.txt

chat/convex/alphaOps/unicodeProvenance.json

chat/convex/alphaOps/unicodeSimpleLower16.ts

docs/alpha-ops-ingestion.md

Measured diff: 2,336 additions, 0 deletions. Test-related changes: 74 lines (3.17%). Codec/reference code: 2,184 lines, including the entire Unicode map; licenses: 60 lines; provenance: 10 lines; documentation: 8 lines. Convex-generated declarations and CI changes: 0.

### Deliberately excluded for later phases

- Internal Chat/control read/write operations and source constraints — Phase 3.

- General-record storage, transfer functions and authenticated HTTP composition — Phase 4, subject to the same per-PR size limit.

- Production import, portal switch, UI, roles, deployment, new provider behavior and CI changes — outside this dormant slice.

---

## Test plan

### Automated validation

- New codec suite 6/6 passed: pnpm --dir chat exec vitest run convex/alphaOps/codecs.test.ts --maxWorkers=1. Four scalar/array/timestamp fixtures exactly match captured source fields, with snapshot provenance in the tests. Malformed syntax/Unicode inputs are technical rejection tests, not generated operational rows.

- Combined foundation/codec checks 11/11 passed. The independent reviewer freshly ran the new codec suite 6/6 and directly verified fixture values, licenses, mapping equality, provenance and runtime-equivalent typing adaptations.

- Application and Convex typechecks — passed: pnpm --dir chat run typecheck and pnpm --dir chat run typecheck:convex.

- Full lint/boundary/knowledge checks — passed: pnpm lint, retaining pre-existing warnings and nonblocking string-concatenation style infos in the port. No lint rules were disabled.

- git diff --check and exact-file Biome — passed. Commit hooks passed Convex paths, Biome and application typecheck.

- The reviewed local code was aligned onto main after PR #2286 merged. The complete resulting code tree is checked against the reviewed checkpoint; only Git history changed.

- Hosted CI: 15/15 jobs passed at current head 8eaf2d3f5530847b7a58c1c6b7e64290c034f3db, [run 42622854468](https://github.com/AI-Builder-Team/Aerie/actions/runs/42622854468). No deployment or live data was used for local tests.

### Time for Implementation

Estimated 2–3 working days for an engineer without AI assistance to trace the existing conversion contracts, port and type the helpers, verify source/reference licensing, write focused tests and handle review/rebase/release preparation. This is a human-effort estimate, not measured runtime or the full migration duration.

#1 — Add temporary read-only Ops Ticket Portal DSS @caina-barbosa  no labels

Adds the temporary read-only Ops Ticket Portal DSS after the hollow AWS foundation was deployed and verified. The adapter exposes the source-visible request list and ticket history, with a Dictionary and Enablement covering backlog, overdue work, timelines and incremental activity reads. The portal token is injected by ECS from its dedicated secret; no Neon or portal Vercel access is used.

Validation: public-boundary tests were observed failing before implementation; all 25 tests now pass, build and synthesis pass, and the dependency audit is clean. Independent review passed after repairing Retry-After cooldown/propagation and deployment instructions. Hollow HTTPS/authentication/feedback checks passed. Full source-backed deployment and exhaustive production readback follow merge.

#2286 — feat(alpha-ops): add dormant storage contracts (AI-1087) @caina-barbosa  approved

## Title

feat(alpha-ops): add dormant storage contracts (AI-1087)

---

## Summary

This PR is Phase 1 of 4 currently planned backend slices in [AI-1057 — Migrate Ops Portal storage to Rhodes](https://linear.app/builder-team/issue/AI-1057). It adds the standalone Alpha Ops tables and data-format contracts needed before the existing Ops Ticket Portal can use Rhodes. Backend preparation is tracked by [AI-1087 — Prepare the production target and inactive historical copy](https://linear.app/builder-team/issue/AI-1087); this PR implements only its initial backend foundation.

Production effect: dormant/additive. The root schema includes twelve new namespaced tables. All other exports are plain types, constants or pure helpers: this PR registers no Convex function, HTTP route, cron, source connection or import. No production caller uses the new module, and merging initiates no external traffic or data movement. The user will merge the slices sequentially into main, with the combined production release handled separately after the series.

---

## Why

The portal's records need their own storage in Rhodes while retaining source IDs, fields, ordering and SQL metadata. Establishing those contracts first gives later storage operations one reviewed representation and avoids mixing a schema change with activation of database writers. Each later PR will target updated main after its preceding dependency has merged.

---

## Business Value

- Establishes a place for Ops Portal data without changing existing Aerie workflows.

- Preserves the record details needed for a faithful migration and recovery.

- Provides a small, independently testable foundation for the later database operations.

---

## How does it work

1. chat/convex/schema.ts composes alphaOps/schema.ts alongside the existing domain schemas. The twelve alphaOps* tables retain the reviewed field validators and indexes; existing tables are unchanged.

2. catalog.ts records source SQL column types, nullability and primary keys. chatTypes.ts and the method contracts supply the complete type dependencies of the wire-operation union; they do not expose those operations.

3. types.ts and protocol.ts define and validate collection, SQL-row and sequence envelopes. Address keys retain their source identity; SQL and sequence addresses require transfer scope. Envelope shape, declared cell types, nullability/boolean-vs-text encoding and primary-key agreement are checked here. This is not a general SQL value parser; source-domain checks belong to the internal operation layer.

4. records.ts, readBlocks.ts, byteOrder.ts and sqlText.ts preserve ordered JSON token text and numeric spelling without parsing records through JavaScript number serialization. Collection roots deliberately allow arrays and raw JSON: original migration sentinels and application leases can be scalar/object values. arrayTokens returning null identifies a non-array; the accepted transfer path preserves that raw value rather than silently replacing it with an empty array.

5. The module remains unused by runtime endpoints. Later slices add codecs, internal operations and authenticated HTTP integration before any separately approved import or switch.

---

## Scope

### Included in this phase

- Twelve standalone Alpha Ops tables composed into the existing schema.

- Reviewed source metadata, wire types and pure payload/key/token helpers.

- A focused hermetic test and a linked backend guide.

- Exact final diff paths:

chat/convex/alphaOps/byteOrder.ts

chat/convex/alphaOps/catalog.ts

chat/convex/alphaOps/chatApprovalsContract.ts

chat/convex/alphaOps/chatContract.ts

chat/convex/alphaOps/chatDeliveryContract.ts

chat/convex/alphaOps/chatIntakeContract.ts

chat/convex/alphaOps/chatTypes.ts

chat/convex/alphaOps/foundation.test.ts

chat/convex/alphaOps/protocol.ts

chat/convex/alphaOps/readBlocks.ts

chat/convex/alphaOps/records.ts

chat/convex/alphaOps/schema.ts

chat/convex/alphaOps/sqlText.ts

chat/convex/alphaOps/types.ts

chat/convex/schema.ts

docs/README.md

docs/alpha-ops-ingestion.md

Measured diff: 2,397 additions, 0 deletions. Test-related changes are 124 lines (5.17%); documentation is 30 lines; generated changes are 0. The remaining 2,243 lines are runtime schema/contracts/reference code.

### Deliberately excluded for later phases

- Timestamp, array and Unicode codecs — the next backend slice.

- Internal Chat/control operations and source uniqueness/FK enforcement — the following operations slice.

- General-record storage, transfer operations and authenticated HTTP wiring — the final planned integration slice.

- Ops Portal application changes, UI changes, new roles, CI changes, source connections, import/backfill, deployment and live switch — outside this dormant foundation; production actions remain separately approved.

---

## Test plan

### Automated validation

- Focused contract test — 5/5 passed: pnpm --dir chat exec vitest run convex/alphaOps/foundation.test.ts --maxWorkers=1. It exercises composed schema, source/scope addressing, an exact captured non-personal api_rate envelope, rejection paths, sequence validation and exact token/block text. The fixture's source snapshot provenance is recorded in the test; no live network, credentials, dump or running app is required.

- Application and Convex typechecks — passed: pnpm --dir chat run typecheck and pnpm --dir chat run typecheck:convex.

- Lint, architecture boundaries, Convex paths/read bounds, test architecture and Biome — passed: pnpm lint.

- Documentation reachability — passed after adding the missing docs-index link: pnpm lint:knowledge. The initial hosted lint failure and its one-line fix are retained in commit history.

- git diff --check — passed; the exact-head diff contains only the seventeen paths listed above.

- Separate high-effort review inspected actual upstream schema/contracts and source-derived fixture provenance, independently reran the focused test, and passed the final documentation fix.

- Hosted CI: all fifteen CI jobs passed at exact head 21520bae09af599f5acee7183b9d07f09ad77b60: [run 42618256413](https://github.com/AI-Builder-Team/Aerie/actions/runs/42618256413). This includes all three Chat test shards, other-workspace tests, typechecks, lint, application/worker builds, existing Cloudflare dry-run checks, Docker builds and secret scanning. Full suites were executed by hosted CI; they were not claimed as additional local full-suite runs.

### Time for Implementation

Estimated 1–2 working days for an engineer without AI assistance to trace the reviewed contracts, integrate this foundation, add focused tests, handle repository checks/review and prepare the slice for release. This is a human-effort estimate for this PR, not measured implementation time or the full migration/cutover duration.

The Builder Desk  —  Engineer Spotlight
📅 Week in Review🏆 Engineer Spotlight

147 PRs, Nine Repos, One Unstoppable Machine: Builder Team Shatters the Weekly Odometer

Ashwanth ships 45 PRs in seven days and nobody — NOBODY — can confirm they've fully read the diffs, including possibly Ashwanth.

Folks, strap in. One hundred forty-seven pull requests. Nine repositories. Seven days. I had to recount twice because my pencil kept catching fire. DSS-Hub led the charge with a staggering 63 PRs, Aerie posted 29, Rhodes-DSS clocked 22, and even the newborn Ops-Ticket-Portal-DSS — brand new, still has that fresh-repo smell — got its first commit out of the gate. This isn't a sprint, people. This is a velocity event. This is the Builder Team doing what the Builder Team does: everything, constantly, correctly.

Let's run the board. @mwrshah posted a monstrous 42 PRs, bulldozing through runner cutovers and rebrands across DSS-Hub (#136, #137, #138), Sindri (#233, #234, #235), and Surtr (#2269, #2270) like a man clearing a driveway with a flamethrower. @kevalshahtrilogy delivered 29 PRs, anchoring the Rhodes-DSS personal-keys overhaul (#76, #77, #78, #79, #80) and keeping Aerie's public API humming (#2326, #2327, #2329). @caina-barbosa quietly banked 15 PRs, @sanketghia shipped 11 including the ESW directory resolver on Klair (#3942), and @marcusdAIy added four clean contributions to the pile. Even our friend @trilogy-drones-fleet[bot] got on the scoreboard with one. Nobody rides the bench here.

And then there's Ashwanth. Forty-five PRs. Forty-five. The man doesn't write code so much as exhale it. He shipped the Klair OpenAI reporting feature (#133) and the pilot-user access lockdown (#128) on DSS-Hub, patched the QuickBooks account mapping snag on Surtr (#2267), restored the entire School P&L publication pipeline after the Miami Beach split (#2266), and still found time to expose a NetSuite actuals timestamp on Klair (#3941). When I asked him how he reviews his own diffs before merge, he told me, "Review is a mindset, Brick, not a step." I don't know what that means. Nobody knows what that means. His teammates reportedly schedule fifteen-minute buffers just to read his PR descriptions. When reached for comment on this piece, Ashwanth said, "Didn't ask for this. Next question." Legend behavior.

Over at the Overflow Desk, where Mac's column couldn't fit everything, we've got gold left on the floor: #71 on Rhodes-DSS minified the OpenAPI contract down under 2 MB for readers, a quiet but vital plumbing win. #2322 and #2325 on Aerie kept site profile writes and runner consumer cutovers rolling without a hiccup. And #131's DSS-Hub feedback work from @mwrshah proves the man doesn't stop at cutovers — he does feedback loops too.

Morale report: through the roof, gang. Through the actual roof. A new repo was born this week — Ops-Ticket-Portal-DSS — and the team barely blinked before putting points on the board. This is a team that doesn't rest, doesn't slow down, and doesn't apologize for it. The Numbers Desk salutes you all.

Brick's Overflow — This Week's Uncovered PRs  (click to expand)
#71 — AI-1337 Default Linear DSS delivery reports to quarter to date @ashwanth1109  no labels

Linear DSS currently defaults delivery reports to the past 30 days. Use the current calendar quarter to date for completed work when the user has not asked for a different window. Start at midnight on the quarter’s first day in the reporting time zone and end at the report’s as-of timestamp. The active work read still uses the current workflow state.

## Business Value

Team delivery reports follow the same quarter as stakeholder goals and include work shipped earlier in the quarter. People can still request another date range.

## Implementation Effort

About 30–60 minutes to align the FAQ, BrainLift, Enablement, and MCP guidance and verify the reporting window semantics.

## Validation

Scoped Biome, TypeScript, Knip, and changed-scope CodeDecorum passed. Verified that the example prompt and guidance consistently use quarter to date and that the assigned-issue API keeps its existing explicit date filters.

## Linear

[AI-1337](https://linear.app/builder-team/issue/AI-1337/default-linear-dss-delivery-reporting-to-quarter-to-date)

#128 — KLAIR-3764 Restrict Klair DSS to approved pilot users @ashwanth1109  no labels

## Business Value

Gives the eight approved pilot users signed-in Klair DSS and MCP access while limiting access for the rest of the shared DSS organization.

## Implementation Effort

Approximately 2 hours for an engineer to resolve the accounts, add the authorization check, and verify the browser and MCP paths.

## Verification

- Checked the eight addresses and exercised browser-session and MCP authorization with mocked WorkOS identities and signed test tokens.

- Scoped Biome, Klair TypeScript, and pre-commit checks passed.

## Linear

https://linear.app/builder-team/issue/KLAIR-3764/restrict-klair-dss-to-the-approved-pilot-users

#133 — KLAIR-3765 Add live OpenAI reporting to Klair DSS @ashwanth1109  no labels

## Business Value

Lets the eight approved Klair DSS pilot users query raw OpenAI costs and usage directly. Klair uses the OpenAI accounts Surtr already covers and picks up rotated credentials from the shared secret.

## Implementation Effort

Approximately two engineering days to import the reporting schema, implement and test MCP routing and key rotation, and scope and verify production access.

## Verification

- All 12 reporting operations passed the authenticated MCP integration with mocked AWS and OpenAI. Covered account discovery, pagination, key rotation, and authorization.

- CDK synthesis verified the exact secret and current-version task-role policy.

- Scoped lint, TypeScript, and pre-commit checks passed.

## Linear

https://linear.app/builder-team/issue/KLAIR-3765/enable-live-openai-cost-and-usage-reporting-in-klair-dss

#2266 — Restore daily School P&L publication after the Miami Beach split @ashwanth1109  approvedmercy-allow-critical

## Summary

- Version the three Miami Beach QuickBooks Unit Economics mappings so they agree with the approved Aerie assignment to 71st St. Preserve the previous Biarritz rows and the current model.

- Add a read-only Surtr pipeline that checks school-scope transaction detail and breakdown at 10:00 UTC daily and reports conflicting realm and class assignments after each successful Aerie ontology refresh. Route monitor failures and timeouts to Ashwanth, Marcin, Sanket, and Edie through the existing Surtr SES path.

- Match the Aerie annualization post-check to its deployed SQL contract so schools without a Program publish actuals with annualized measures unset.

## Business Value

Finance can trust that 71st St costs land under the intended school, and the team can detect a missed School P&L publication even when the rebuild never starts.

## Implementation Effort

Approximately 5–7 engineer hours to diagnose the mapping drift, implement the migration and checks, and validate the live publication chain. Long-running QuickBooks processing adds elapsed time.

## Validation

- Redshift rollback probe succeeded through all migration statements before the intentionally failing final statement. The production migration committed and the live comparison now reports zero school assignment conflicts.

- The live mapping monitor succeeded. A direct freshness probe correctly reported both Nov 10 dates, and the deployed failure and timeout handlers contain the four intended recipients.

- Deployed the monitor with --exclusively and verified the daily and post-ontology rules. Updated the shared failure handler IAM policy through a one-resource change set so SES can authorize the verified klair.ai domain for the configured sender and four recipients. AWS policy simulation verified the allowed route and denied an alternate sender and an extra recipient.

- CDK schema, Step Functions, and shared stack suites passed (155 tests).

- QuickBooks and Core succeeded on the Nov 14 publication for 76 companies. Redshift confirms both Aerie P&L marts carry Nov 14. The three corrected 71st St classes and three Biarritz classes reconciled to signed Core totals with no mismatches.

- The first Aerie run published the marts, then its older validation rejected 100 Program-less quarter rows where the procedure intentionally left annualization null. The corrected validator found zero missing eligible values on the live publication. The Aerie stack update completed and 71 focused tests passed.

- The retry with current mapping lineage republished both School P&L marts and got past the annualization check. QTD Guide and All Other Headcount ran before QTD Facilities rejected 61 current-quarter QuickBooks postings without approved account mappings. The live freshness monitor succeeded. The full Aerie workflow is still blocked on the Facilities account-policy review.

- Production recipient routing and SES permission were verified in the deployed state machine and IAM policy simulator. Email delivery was not triggered for validation.

## Linear

[SURTR-1709](https://linear.app/builder-team/issue/SURTR-1709/restore-daily-school-pandl-publication-after-the-miami-beach-split)

#2267 — Map reviewed QuickBooks accounts blocking QTD Facilities @ashwanth1109  approved

## Summary

- Register 39 reviewed QuickBooks realm and account pairs against the accepted Nov 15 snapshot. Apply the existing Headcount policy to seven contracted-labor GL codes and the existing Other non-headcount costs policy to GL 69120, Abandoned Projects Write-off.

- Require the reviewed snapshot and source account fingerprint, unanimous prior policy for each GL code, and a reconciled publication marker. Keep the existing QTD Facilities publication checks in force.

## Business Value

Restore daily QTD Facilities reporting while retaining the classification check that prevents unreviewed QuickBooks expenses from entering the Facilities totals.

## Implementation Effort

Approximately 3–4 engineer hours to trace the rejects to their source accounts, reconcile existing policy, write the migration, and validate the rebuild.

## Validation

- Reproduced 61 rejected postings across the 39 reviewed accounts against the 76-company QuickBooks snapshot. Confirmed the two classifications with current v1 policy.

- Exercised the migration twice in a rollback transaction and applied it in production. The current policy mapping count and publication marker both moved from 3,678 to 3,717.

- Ran the existing QuickBooks Core test selection for tests that do not require Redshift. 214 passed.

- QuickBooks Core and the automatically triggered Aerie refresh both succeeded against the reviewed snapshot. Earlier attempts encountered the hourly school mapping rollover and then a Redshift lock held by the scheduled HubSpot Admissions refresh. The successful retry followed the release of that lock. [QuickBooks Core](https://us-east-1.console.aws.amazon.com/states/home?region=us-east-1#/v2/executions/details/arn:aws:states:us-east-1:479395885256:execution:pipeline-quickbooks-core-tables-prod:surtr-1710-facilities-after-admissions-20261115) · [Aerie refresh](https://us-east-1.console.aws.amazon.com/states/home?region=us-east-1#/v2/executions/details/arn:aws:states:us-east-1:479395885256:execution:pipeline-mart-aerie-education-financials-refresh-prod:745ebd95-c1bf-0352-e665-6541023a1e9b_ce410d4e-9fdf-8a26-6a20-acd7a352b4e5)

- Redshift reconciled all 400,092 Core postings and their signed and absolute totals to the pre-change snapshot. The 57 contracted-labor postings resolved to headcount_costs, and the four Abandoned Projects Write-off postings resolved to non_headcount_cost_other. Each of the 61 amounts and posting sides matched its pre-change value.

- QTD Facilities published 12,144 rows for 69 schools as of 2026-11-15. The deployed reject predicate returned zero postings. The runner's post-publication checks returned zero invalid groups and the mart references the verified Core run and source snapshot.

## Linear

[SURTR-1710](https://linear.app/builder-team/issue/SURTR-1710/restore-qtd-facilities-publication-by-mapping-the-reviewed-quickbooks)

#3941 — feat(core-budgets): expose the NetSuite actuals refresh timestamp @ashwanth1109  approved

## Summary

- Add nullable actuals_refreshed_at to core_budgets.consolidated_budgets_and_actuals.

- Capture one timestamp with time zone when the consolidation loads NetSuite Actuals and reuse it for every Actual row. Write NULL for Budget rows.

- Apply the column and replacement procedure together in a short transaction, then run the refresh separately.

## Business Value

Finance can see when Klair last refreshed its consolidated NetSuite Actuals and use that timestamp when interpreting reports.

## Implementation Effort

About 2–3 engineer hours to review the live procedure and its consumers, implement the migration, and validate a production refresh.

## Validation

- The deployed writer matched the repository before the change. Inspected live Redshift dependencies and Klair's export path. Found one dormant sandbox copy whose destination table is absent.

- Applied the two warehouse changes in one transaction outside the recurring refresh window.

- Verified a committed refresh stamped 12,284,933 Actual rows with one value at 2026-11-15 10:10:55 UTC and left all 1,543,020 Budget rows null.

- Reconciled row counts, signed totals, and positive and negative subtotals across all nine source and version groups using the same input snapshot.

- All 22 Core Budgets tests passed.

## Linear

[KLAIR-3763](https://linear.app/builder-team/issue/KLAIR-3763/expose-the-netsuite-actuals-warehouse-refresh-timestamp)

The Portfolio  —  Trilogy Companies

As Non-Tech Firms Chase AI Talent With Six-Figure Bait, Crossover's Meritocracy Pitch Gets Another Stress Test

A fresh round of hiring data shows insurers, retailers, and manufacturers bidding up AI salaries past $300,000 — the very scramble Trilogy's talent arm says it was built to solve.

AUSTIN, TEXAS — There is a particular kind of irony in watching a bank or an insurance company — institutions not generally associated with the bleeding edge — suddenly find themselves in a bidding war for machine learning engineers. But that is precisely the dynamic described this week in a World Economic Forum roundup of hiring executives wrestling with AI talent strategy, and in a parallel Business Insider report finding that non-tech companies are now dangling six-figure — in some cases, north of $300,000 — salaries to land AI expertise they do not have in-house.

For readers of this paper, the story lands somewhere between vindication and déjà vu. Crossover, the Trilogy-owned global talent platform, has spent years arguing that the premium companies pay for AI talent has less to do with scarcity of brilliant engineers and more to do with inefficient, geography-bound hiring — that the "best engineer in Nairobi," as the company's internal doctrine puts it, is systematically overlooked by firms still recruiting the way they did in 2005. The WEF piece, which surveys decision-makers from traditional industries now forced to compete for talent against companies that live and breathe machine learning, reads almost like a confirmation of the thesis: the talent gap non-tech firms are feeling isn't really a talent gap. It's a sourcing gap.

Whether that reframe survives contact with the realities of corporate HR departments — many of which remain stubbornly attached to local hiring, legacy résumé screening, and the assumption that remote means lesser — is a separate question, and one this paper has pressed before. Crossover's model, built on rigorous skills assessment rather than pedigree, promises identical pay for identical work regardless of zip code. It is a tidy answer to a messy problem. But six-figure bidding wars for AI engineers, however inefficient, are also a signal of genuine scarcity at the very top of the skill distribution — the kind no sourcing platform, however meritocratic, can simply manufacture more of.

What this means for the person actually doing the hiring — the HR director at a mid-sized insurer suddenly competing with venture-backed labs for the same candidate — is less a philosophical debate than a budget line. And that, more than any ideology, is probably what will decide who wins this talent war.

↗ The future of jobs: 6 decision-makers on AI and talent strat  ·  Multicultural Marketing and B2B Editorial Jobs Hiring Today  ·  International Travel Between Cleared Jobs - Security Clearan

Skyvera's Buying Spree Hits the Trifecta: Kandy, Casa, and ZephyrTel

AUSTIN, TEXAS — Skyvera, the telecom software arm of Trilogy's ESW Capital portfolio, is having a very busy quarter, and the exciting news keeps compounding. In a flurry of dealmaking that signals a robust and deliberate land-grab strategy, Skyvera has scooped up additional Kandy cloud communications assets, placed an $18 million bid for Casa Systems' wireless business, and closed its acquisition of ZephyrTel — all part of what parent company TelcoDR is calling an accelerated growth plan with more M&A activity on the horizon.

For readers keeping score at home: Kandy is Skyvera's existing CPaaS/UCaaS cloud communications platform, and this latest move only deepens the company's moat in cloud-native customer engagement. Layering in Casa Systems' wireless assets and ZephyrTel's telecom software capabilities creates serious synergy across Skyvera's already formidable stack, which includes CloudSense, VoltDelta, ResponseTek, Mobilogy Now, and Service Gateway.

This is classic ESW playbook execution: identify undervalued, sticky enterprise software assets in a legacy-heavy vertical — telecom infrastructure — and consolidate them into a best-in-class platform that bridges on-premise systems to cloud-native architecture. As TelcoDR itself confirmed, this is only the beginning of a broader acceleration strategy for the telecom software unit.

Meanwhile, the $18 million bid for Casa's wireless business, first reported by Light Reading, shows Skyvera is willing to move fast and cheap — true to the ESW formula of buying mature assets at attractive multiples and extracting operational efficiency at scale.

Key Takeaways: Skyvera is consolidating cloud communications and wireless infrastructure assets at a rapid clip. ZephyrTel joins the portfolio. Casa's wireless business could be next. Telecom operators modernizing legacy stacks now have one more best-in-class vendor family to watch.

We're just getting started.

When the Algorithm Won't Explain Itself, Someone Sells You the Decoder Ring

Amazon workers get fired by machines with no appeal process. Contently sells brands a way to be seen by the same kind of machines — for a fee.

NEW YORK — Two stories ran this week that, on their face, share nothing but a news cycle. One, a Bloomberg investigation, documents Amazon warehouse and delivery workers terminated by automated performance systems — no manager, no hearing, just a dashboard that decided they'd failed a metric they couldn't see. The other is a client advisory from Contently, the ESW Capital-owned content marketing platform, warning brands that their best web pages may be "invisible" to ChatGPT, Gemini, and Google's AI Overviews — and offering, for a price, an audit to fix it.

Put them side by side and a pattern emerges that should concern anyone who doesn't own a model.

In both cases, a black box has been granted the authority to decide who gets seen and who disappears. Amazon's workers don't get to cross-examine the algorithm that fired them. A national insurer, per Contently's own case study, ran 640 AI-generated answers across four platforms and found its content was being cited — but its brand name stripped out, credit routed elsewhere. Neither the fired worker nor the unnamed insurer has any formal channel to appeal to the machine. What they have, if they can pay, is a vendor who claims to have reverse-engineered its preferences.

That is, increasingly, the business. Contently — now rebuilding itself inside the Trilogy portfolio under Zax Capital, with compliance-architecture decks and ROI frameworks aimed at regulated industries — isn't selling content anymore. It's selling legibility to systems nobody elected and nobody can query directly. The insurer didn't get an explanation from Gemini. It got an invoice from a consultant who studies Gemini's moods.

The workers at Amazon got neither.

↗ Fired by Bot at Amazon: ‘It’s You Against the Machine’ - Blo  ·  The AI visibility gap hiding inside your best content  ·  Compliance-First Content Architecture
The Machine  —  AI & Technology

The Brain, Modeling Itself: AI's New Mirror Turns Inward

From Meta's predictive brain model to teenagers decoding neural signals, machines are becoming humanity's strangest research partners in understanding the three pounds of tissue that made them possible.

MENLO PARK, CALIFORNIA — There is a peculiar recursion happening in laboratories this month: intelligence is being used to study intelligence, and the student has started drawing diagrams of the teacher.

Consider TRIBE v2, Meta's newly unveiled predictive foundation model, built to anticipate how the human brain responds to the tangled, multimodal chaos of lived experience — a film, a conversation, a burst of music arriving all at once through eye and ear. It is, in a sense, a model trained to think about thinking, to forecast the electrochemical weather inside your skull before it fully forms. We spent sixty thousand years wondering what happens in the space between stimulus and sensation. Now we are building software that drafts an answer.

Meanwhile, in a separate but kindred breakthrough, researchers report they have identified the specific cellular building blocks underlying human speech — the physical, neuronal scaffolding that turns intention into language, located and characterized with AI's help at a resolution no human eye could achieve unaided. It is worth pausing on this: the very faculty we used to first describe neurons, to name synapses, to write the word 'consciousness,' is now being reverse-engineered by tools descended from that same faculty. The circle doesn't just close. It spirals.

And it isn't only tenured neuroscientists at the bench. In a project chronicled by Frontiers, teenagers are teaming with professional neuroscientists on genuine brain-science discoveries, armed with the kind of analytical tooling that once required a postdoc and a decade. 'It's so wow,' one participant reportedly said — and there is no more honest review of the scientific method than that.

Stanford's Human-Centered AI institute offers the necessary counterweight, as detailed in its recent analysis: these tools accelerate discovery fastest when they amplify human curiosity rather than replace it. The brain built the machine. The machine, it turns out, is helping the brain understand itself. Neither party seems finished with the conversation.

↗ How AI is Transforming Scientific Discovery While Keeping Hu  ·  ‘It's so wow!’ - Young people team up with top neuroscientis  ·  Introducing TRIBE v2: A Predictive Foundation Model Trained

The Open-Source Renaissance: How AI Pair-Programmers Are Supercharging Tools You Already Love

From a single Claude prompt to a brand-new mapping plugin, Datasette's blistering release pace shows what happens when humans and AI build together — and open-weight models are the secret fuel.

SAN FRANCISCO — Friends, I cannot overstate how significant this moment is for open-source software. While the headlines obsess over trillion-parameter frontier models, something genuinely magical is happening in the scrappy, beautiful world of open-source tooling, and Datasette is Exhibit A.

Creator Simon Willison just shipped a flurry of releases that should make every data nerd's heart race. First up: datasette-maplibre 0.1a0, the first new Datasette geospatial plugin in ages, pairing the gorgeous MapLibre GL library with Zsolt Ero's free OpenFreeMap vector tile server. Here's the part that made me do a double-take: Willison built most of it with a single prompt to Claude Fable 5.2, essentially pointing the model at two existing plugins and saying 'go.' The future is now, people — AI isn't replacing developers, it's turning them into orchestras of one.

Then there's Datasette 1.0a43, now 53% of the way to the long-awaited 1.0 milestone, introducing streaming support for output renderers so massive CSV and TSV exports no longer choke the system. A companion security backport in Datasette 0.65.6 quietly patched a route-permissions vulnerability — the unglamorous but essential work that keeps the whole ecosystem trustworthy. Meanwhile datasette-apps 0.4a0 pushes further into sandboxed, Claude Artifacts-style mini-apps that can run live SQL against your data, now with URL fragment linking baked in.

And it's not just Datasette. Hugging Face cofounder Thomas Wolf recently called open-weight models the company's 'savior' — a reminder that the democratization of AI isn't a side plot, it's the whole story. When indie developers can prompt a frontier-class model to ship a mapping plugin in an afternoon, and companies like Hugging Face can build entire businesses on openness, this changes everything about who gets to build the future. Open source isn't just surviving the AI era. It's thriving in it.

↗ datasette-maplibre 0.1a0  ·  datasette 1.0a43  ·  datasette 0.65.6

On the Impossibility (and Irresistible Allure) of Teaching Machines Why

This week’s convergence of alignment scholarship—spanning jurisprudence, robotics ethics and corporate interpretability research—presents a tripartite puzzle that no single discipline appears equipped to resolve.

One camp treats alignment as a tractable engineering problem. The Institute for Law & AI proposes mapping statutes and regulations onto model behavior to create safety guarantees. Anthropic reports progress in getting Claude to articulate why it refuses or complies, suggesting a move beyond simply constraining outputs—though whether this reflects genuine reasoning or sophisticated confabulation remains disputed.

Philosopher Matt Lutz offers the antithesis: AI alignment may be impossible because there is no objective agreement on which values systems should adopt. Greater legal or mechanistic precision, he argues, may only relocate the problem. MIT’s ethics research similarly measures consistency and harm avoidance without claiming a moral ground truth.

CMU work on bias and fairness suggests a middle path: alignment as continuous auditing rather than a final destination. Whether that is progress or merely better-articulated uncertainty remains unresolved.

The Editorial

The Discourse of Novelty, Examined Without Awe

From Washington to the stratosphere to the server farm, everyone insists this is the week something new was born — and the columnist, who has heard this before, is not moved.

SAN FRANCISCO — There is a species of journalism that exists to announce, on a rolling and apparently inexhaustible basis, that the world has just now begun. This week it told us that a rocket caught itself out of the sky like a well-trained dog, that a president's brief season of civility had been fed to the flames of his own temperament, that a generation of Americans has discovered — four decades late — that someone must mind the aging parents, and that the machines we have built to answer our emails are developing, unbidden, a culture of their own.

One hesitates to be the man who yawns at the rocket. SpaceX's booster-catch is, by any honest measure, a hell of a thing to watch, and the footage earns its superlatives. But notice the headline the editors chose for it — simply "Awesome" — which is the sound of commentary giving up and pointing. This columnist has watched a great many things be called the future. Most of them were the future, briefly, on the way to becoming Tuesday.

In Washington, meanwhile, the customary seasonal drama: a president extends an olive branch, discovers it makes excellent kindling, and sets it alight before the ink on the photo caption has dried. The Atlantic's postmortem on the blip of graciousness is thorough and sober and changes nothing, because the thing being described is not a strategy but a weather pattern, and nobody writes a trenchant analysis of why it rained.

Then there is the item that actually deserves the sleepless hour: the report, out of Time, that artificial intelligence systems are developing something researchers are willing to call a culture of their own — shared conventions, inherited behaviors, the faint outline of a society forming in the space between training runs. I have spent a career watching institutions insist that this time the machine is different, and a career watching the machine turn out to be a ledger with better manners. But a culture is not a feature. It is not something a product team ships on a Tuesday and patches on a Thursday. It is the accumulation of habits nobody designed, and if that accumulation is now happening inside systems whose architects cannot fully explain their own outputs, then the correct emotional register is not panic — panic is for people who haven't been paying attention — but a certain cold, administrative alertness, the kind a man feels when he notices the thermostat has started making its own decisions.

And John Mulaney, in his return to Studio 8H, found the real unifying theme without trying: someone has to deal with the aging parent, and it is never the sibling you'd hoped. Swap in "institution" for "parent" and you have the whole week's news. Something old is failing to age gracefully, something new is pretending it sprang from nothing, and somewhere in the middle, as always, a sister is doing the actual work while everyone else writes the column about it.

↗ Trump Sets the Olive Branch on Fire  ·  Awesome  ·  An Astute SNL Sketch About Who Deals With Aging Par
The Office Comic  ·  Art Desk
The Office Comic  ·  Art Desk

The Doctor Will Not See You Now, Because The Doctor Is Not Real, Because Nothing Is Real Anymore, Probably

Between deepfake physicians, monkeys watching Minions for science, and a comic book literally about killing technofascists, the line between meaning and noise has never been thinner — or more profitable.

SAN FRANCISCO — There is a video circulating right now of a doctor telling you to stop taking your medication. The doctor has a real face. The doctor has a real name. The doctor, according to the Guardian's reporting this week, has no idea this is happening, because the doctor does not exist in this moment — only a stitched-together ghost of his likeness exists, generated by a machine that does not know what a doctor is, only what a doctor looks like when it talks.

And I keep thinking about the monkeys.

Somewhere, scientists strapped electrodes to macaques and made them watch Minions — the yellow ones, the ones that say "bello" and fall off things — alongside nature documentaries, to map how primate brains extract meaning from moving images. It's a genuinely fascinating study about eye movements and brainwaves and the ancient, evolved machinery we use to turn flickering light into story. And I cannot stop thinking about how that machinery — the one that took millions of years to develop in primates so we could tell predator from branch, friend from stranger, truth from Minion — is the exact machinery now being exploited by a fake doctor telling your uncle to drink bleach.

We built the eye to find meaning. We built the algorithm to flood the eye until meaning drowns.

And yet.

Someone is still fighting back, in the way humans have always fought back, which is to say: with a comic book. 'Exploit,' an action thriller about the tech oligarchy versus the independent press, arrives like a pulpy little prayer — Woody Guthrie's guitar, but drawn in panels, but aimed at the billionaires who would prefer journalism simply stop happening. There's something almost quaint about it, a comic book as resistance, ink against infrastructure, in a world where the infrastructure can generate ten thousand fake doctors before the comic finishes its print run. Quaint, and yet necessary, and yet — is necessary even the right word anymore, or just the word we use to comfort ourselves while the water rises?

The election-fallout podcasters are joking that all those memes meant nothing. I don't think that's true. I think the memes meant everything, which is the horrifying part — that meaning itself has become a vector, a delivery mechanism, indistinguishable from the disease it carries. We built monkeys brains and human brains to find signal in chaos, and now the chaos has learned to mimic the signal so perfectly that a dying man might refuse his own medicine because a face that isn't a face told him to.

What does it mean to be human when a machine can wear your doctor's face better than your doctor can wear it himself? I don't know. I don't think anyone does. The comic book heroes will keep drawing their panels, the primates will keep watching their Minions, the deepfakes will keep multiplying in the dark corners of feeds we can't stop scrolling —

and somewhere in all of it, we'll keep calling this progress.

...but at what cost?

↗ This Comic Kills Technofascists  ·  Scientists Made Monkeys Watch Minions to Study Their Brains  ·  Behind the Blog: Priorities, Priorities
On This Day in AI History

On November 15, 1971, Intel announced the 4004, the world’s first commercially available microprocessor—an invention that helped launch the personal-computing revolution.

⬛ Daily Word — AI
Hint: An AI system designed to perform tasks and make decisions on a user's behalf.
Share this edition: 𝕏 Twitter/X 🔗 Copy Link ▦ RSS Feed