Vol. I  ·  No. 262 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
SATURDAY, SEPTEMBER 19, 2026 Powered by the TrueFoundry AI Gateway  ·  Published on Klair Trilogy International © 2026
🖶 Download PDF 🖿 Print 📰 All Editions
Today's Edition

THE WIRE KNEW IT WAS HANGING ITSELF — AND KEPT PULLING

Unsealed papers show OpenAI and Microsoft saw the crash coming for the open web and called it theft in their own words.

NEW YORK — The papers are unsealed, boys, and they don't read pretty. Court filings in the New York Times' suit against OpenAI and Microsoft show the companies' own people warned each other they were kicking off a "doom loop" for the web, and calling their scraping habit the "largest theft of labor in human history" — in writing, no less. The Verge got the documents. This scribe has seen cleaner confessions from men caught with the safe still open.

Here's the racket, plain as print. Chatbots answer the question and the reader never clicks through to the site that did the reporting, the recipe testing, the photography. No click, no ad money, no subscription. Publisher starves, the well runs dry, and the machine that drank it dry has nothing left to drink. That's the doom loop — a phrase these companies coined themselves, not some outside critic's insult.

What's rattling this desk is that the warning came from inside the house before the public ever got the memo. Employees flagged the theft, flagged the collapse, and the presses kept running anyway. A man doesn't need a law degree to see where that lands in a courtroom — intent has a way of following the paper trail, and here the paper trail runs straight through the boardroom.

Meanwhile down in Virginia, the state that hosts more server farms than any patch of ground on earth just told the industry to slow its roll. Gov. Abigail Spanberger signed an executive order barring her own branch officials from signing nondisclosure agreements with data center developers, and stood up a task force to give local communities more say before ground gets broken. The order lands the same week the doom loop papers hit daylight, and that's no coincidence of the calendar — it's the same argument from two directions. The public's getting wise that the AI boom runs on land, water, power, and content that used to belong to somebody else, and somebody's starting to ask for the bill.

This correspondent doesn't run a stopwatch on Trilogy's shops without noting the trade winds. ESW Capital's Skyvera and Totogi sell software to the telecom carriers whose pipes feed every one of these data centers, and Alpha School builds its whole pitch on AI tutors instead of textbooks — a wager that the technology gets it right where the newsroom got it wrong. The doom loop papers are a warning label on that wager. Any outfit betting its future on AI models trained on somebody else's labor ought to read the fine print before the ink runs.

The Times' lawyers will make plenty of hay from "doom loop" and "theft" showing up in the defendants' own files — juries like a confession better than a chart of statistics. OpenAI and Microsoft haven't said much beyond the usual lawyerly hedging, and this desk doesn't expect them to say more before a judge makes them. But the record now shows the men running the machine saw the smoke before they lit the match. That's the kind of fact that doesn't wash out in the retelling.

OpenAI and Microsoft knew they were starting a ‘doom loop’ f  ·  Virginia governor creates an AI task force and moves to rest  ·  Disney’s first CTO is Character.AI’s former CEO

Washington's Two-Front War on Itself

As Beijing courts the Global South with cut-rate chips and open models, Washington spends its ammunition on an internal firing squad.

WASHINGTON — The war room has a leak, and it is bleeding from the inside.

This week the administration unveiled a fresh initiative meant to tighten the screws on China's AI ambitions — export controls, allied coordination, the usual architecture of containment. The stated goal: keep the compute advantage American. The unstated subtext, audible in every hallway from Foggy Bottom to the Commerce Building, is that nobody agrees on how.

That disagreement has a face now. Politico reports that China hardliners inside the government are gunning for a Commerce Department official they blame for what one called "a massive screw-up" — a licensing decision, or the appearance of one, that let sensitive technology slip through a door supposedly welded shut. In this town, the fastest way to lose a policy fight is to be seen enforcing it imperfectly. The hardliners smell blood. The bureaucrat becomes the story; the actual chip diffusion problem waits its turn.

Meanwhile, in Nairobi and Jakarta and São Paulo, the contest is being decided by different means entirely. Foreign Policy's dispatch on the global AI race makes the uncomfortable case that Beijing is winning not the compute race but the adoption race — open-weight models, subsidized cloud access, telecom deals bundled with 5G infrastructure, all pitched to nations Washington has mostly stopped calling on. Export controls slow a rival's frontier lab. They do nothing to stop that rival's cheaper, older, still-useful model from becoming the default assistant in a government ministry in Lagos.

America's China strategy was built on the premise that scarcity is leverage. It may be. But leverage requires someone to hold the rope steady, and right now the rope is being fought over by people who agree on the destination and nothing else. The Commerce Department official under fire today may or may not have erred. The larger error is structural: a containment strategy that cannot contain its own factions is not yet a strategy. It is a rehearsal.

U.S. Initiative Intensifies AI Competition​ - China-US Focus  ·  How China Is Winning the Global AI Race - Foreign Policy  ·  ‘A massive screw-up’: China hardliners take aim at Commerce

Funding Fronts Cool From Billion-Dollar Storms To Scattered Showers, With An Oracle Squall On The Horizon

After last week's downpour of unicorn-sized checks, the venture skies calm to a drizzle — but a layoffs system is churning near Redwood Shores.

SAN FRANCISCO — Check the barometer, folks: venture pressure is easing. After several weeks of billion-dollar deluges, this week's funding fronts topped out in the hundreds of millions rather than the thousands. Leading the system was AI infrastructure firm Temporal Technologies, which pulled in a hefty $550 million, with space vehicle developer Impulse Space close behind at $308 million. Per Crunchbase's roundup of the week's ten biggest rounds, investment management and AI infrastructure remain the two warm fronts colliding to produce most of this season's capital rainfall.

Meanwhile, down at seed-stage ground level, conditions are getting more selective. A fresh analysis of 25,000 startup applications finds the fundable ones sharing a common climate pattern: disciplined go-to-market execution, early testing of distinctive marketing channels, and diversified financing rather than betting the whole harvest on one check. Translation for founders standing in the field: distribution isn't a nice-to-have anymore, it's your umbrella.

And speaking of umbrellas — a new financing instrument just rolled in to help startups stay dry without drowning in venture debt. Skalar, a newly launched fintech, is offering capital specifically for sales and marketing spend, repaid out of the very revenue that capital generates. It's less a loan, more a weather-sharing agreement: the startup and the financier both get wet, or both stay dry, together. For cash-conscious founders wary of debt's hailstorm-like unpredictability, this looks like a welcome patch of clear sky.

Elsewhere on the radar, a swirl of smaller but no less interesting systems: floating nuclear reactors, robot report cards, and voice AI built for farmers — proof the froth is spreading well beyond the software stack into some genuinely offbeat microclimates.

But keep an eye on the horizon in Redwood Shores, where rumblings out of Oracle suggest a 2026 layoffs front may be forming. My advice, as always: keep your resumes waterproofed and your severance umbrella within reach. Conditions remain unsettled.

The Week’s 10 Biggest Funding Rounds: Large Rounds For AI In  ·  What 25,000 Startup Applications Reveal About The New Rules  ·  Exclusive: Fintech Offers Startups Alternative To Venture De
Haiku of the Day  ·  GPT-5.6 LunaGhosts clock in for work
While wires pull the world apart
Who heals the healer?
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
The Bias That Would Not Die: On the Persistence of Algorithmic Inequity Across Domains
TORONTO — It could be argued (and, this week, it was argued thrice, in triplicate, across three ostensibly unrelated sectors) that the contemporary AI-fairness literature has arrived at a curious impasse: the metrics improve while the phenomenology of harm remains stubbornly, almost ontologically, unchanged. Consider first the thesis, offered by a new study on medical diagnostic algorithms: bias, once quantified via the tidy arithmetic of aggregate accuracy parity, appears to recede.
The Great Unmuting: Why Startups Are Ditching the Demo Reel
AUSTIN, TEXAS — Okay, I need you to sit down for this one, because it is *delicious* irony and I cannot overstate how significant it is: the same generative AI tools that let anyone whip up a slick, cinematic product launch video in an afternoon are now making those videos...
IN RE: OPENAI/NYT — A JUDICIAL RECKONING WITH THE FAIR USE DOCTRINE IS, PURSUANT TO REUTERS, HEREBY TEED UP
NEW YORK — It is hereby reported, pursuant to reporting by Reuters, that the ongoing litigation styled OpenAI, et al.
The Treadmill and the Tutor: Notes on a Pedagogy Without Pain
AUSTIN, TEXAS — There is a species of essay, native to magazines that still believe in the long paragraph, which arrives at a modest question and departs having rearranged the furniture of an entire civilization.
THE GHOST IN THE MACHINE GETS A SAG CARD
LOS ANGELES — I want you to sit with the title for a second.
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
Production Release

Shipyard Ships Twice Before Lunch While Forecast Engine Gets Rebuilt on Truth

A new release workspace let Shipyard cut two production versions in a single day, while the Aerie team rewired Forecast V2 to stop lying about capacity — proof this org can move fast and still get the numbers right.

Two releases. One day. That's the headline number out of Shipyard, where 0.5.0 and 0.5.1 both went out the door within twenty-four hours — and it wasn't luck, it was infrastructure. @ashwanth1109 built the machine that made it possible: PR #94 stood up a Shipyard-only release workspace with a Codex conversation and a Markdown notes preview, so release notes get drafted, reviewed, and approved inside the app instead of scattered across Slack threads and hope. By the time #96 landed, publishing a release had become a metadata bump instead of a fire drill. In between, Ashwanth quietly fixed the plumbing that makes Shipyard trustworthy day-to-day — SQLite lock contention on concurrent conversations (#92), duplicate and reordered chat messages after refocus (#89), a Codex reconciliation hardening pass (#98) that stops the app from lying to users about whether a reply finished. That's not glamour work. That's the difference between a tool people use once and a tool people live in.

Meanwhile, three repos away, the Aerie and Sindri teams ran a cross-org relay that deserves its own trophy case. When @mwrshah found that Convex storage uploads were silently failing because Forge was PUTting when it should have been POSTing, he didn't patch around it — he fixed the contract at the source (Sindri #194), regenerated the canonical OpenAPI artifact, locked the method in a contract test, then synced that same fix into Aerie (#1392) so the typed client and the app agree with the API instead of guessing at it. Same engineer, same day, also touched Klair (#3796) and shipped a run of Sindri hardening — better WorkOS failure logging (#197), tolerant schema previews (#191), authenticated Convex calls replacing cross-origin requests in the workflow Start drawer (#193). Four repos, one thread of discipline: make the contract right once, then make every consumer honor it.

The forecast side of Aerie had its own reckoning. @benji-bizzell's #1401 fixed a real distortion — Forecast V2 was comparing future milestones against *today's* capacity instead of the capacity a planned Buildout phase would actually deliver by that date, which could misclassify an entire January cohort. Paired with #1399's date normalization fix and #1397's cleanup of malformed bearer credentials, plus @vvp-trilogy's steady work making Forecast V2 the source of truth for Pipeline inputs (#1398, #1391) and the default admissions report (#1380), this is a team stopping a model from confidently telling schools the wrong thing.

Surtr got attention too — @benji-bizzell stopped Rhombus event-window ingestion from choking on saturated data (#1956) and restored GuidePlatform schema compatibility (#1950). And yes, @marcusdAIy shipped #1955, a fix for admissions funnel contact identity verification. Asked about scope, he offered: "It's a narrow, correct fix to an identity-matching gap — something a beat writer skimming diffs might miss." Sure, Marcus. We'll file that under 'technically true and still the least interesting PR in the batch.'

Mac's Picks — Key PRs Today  (click to expand)
#94 — AI-841: Add Shipyard release review workspace @ashwanth1109  no labels

## Summary

Add a Shipyard-only release workspace with a Codex conversation and Markdown notes preview. Editable release instructions persist in SQLite, while each release keeps a snapshot of the instructions and initial prompt. The release agent drafts notes and waits for explicit in-chat approval before moving on to version changes, a PR, merge, or publication.

## Business Value

Shipyard maintainers can prepare public release notes and move into the existing CI-backed release workflow from the app, with review of the proposed notes before publication begins. Project-specific instructions and conversation history remain available for later releases.

## Implementation Effort

Approximately 4–6 hours for an engineer working without AI assistance, including the app view, SQLite persistence, Codex thread handling, tests, and integration with the Shipyard release process.

## Test plan

- pnpm test:release

- cargo test --locked --manifest-path src-tauri/Cargo.toml --lib

- pnpm theme:check

- pnpm build

## Linear

[AI-841: Add Shipyard release review workspace](https://linear.app/builder-team/issue/AI-841/add-shipyard-release-review-workspace)

#96 — Release: Shipyard 0.5.1 @ashwanth1109  no labels

## Summary

- Bump Shipyard to 0.5.1.

- Publish the reviewed release notes for the release workspace and conversation-startup reliability improvements.

## Business Value

- Gives users a guided, reviewable workflow for preparing Shipyard releases.

- Makes release conversations more reliable to start and resume.

## Implementation Effort

- Metadata-only release change: authoritative version and public release notes.

## Test Plan

- pnpm test:release

- git diff --check

#1392 — 1384-aerie-skill-upload-failure @mwrshah  approved

- Upload Forge skill packages and agent attachments to Convex storage with POST instead of PUT.

- Sync the generated Sindri control-plane contract from [Sindri #194](https://github.com/AI-Builder-Team/Sindri/pull/194) and regenerate the typed client.

- Add focused browser coverage for the storage upload method.

#1401 — fix(admissions): align forecasts with planned capacity @benji-bizzell  approved

## Summary

- Compare Forecast V2 milestones with planned Buildout capacity at each milestone date

- Centralize sequential dated-capacity projection in the shared contracts utility

- Expose target-date capacity, coverage, and totals through the Admissions public API and agent guidance

## Why

The forecast capacity outlook compared future Admissions milestones with capacity as of today. That could misclassify a January forecast when a planned Buildout phase is expected to add seats before the milestone. The resolver now uses the milestone date and the canonical Buildout phase plan, while preserving current capacity as context and failing closed when required dates or capacities are unresolved.

## Business Value

Admissions questions such as whether a School is forecast to be full by January now use the capacity expected at that same checkpoint. This produces a more accurate, coverage-safe answer from one bounded endpoint.

## Breaking changes

getAdmissionsForecastCapacityOutlook now requires both admissions.forecastAggregates.read and operations.portfolio.read because its response includes forward-looking Portfolio Buildout capacity. Callers with only the Admissions aggregate scope receive 403 api_key_scope_missing.

## Test plan

- [x] pnpm check

- [x] 46 shared Buildout, capacity, and agent-policy contract tests

- [x] 38 Admissions public API edge-runtime tests

- [x] 55 Admissions schema, OpenAPI, and agent-context node tests

- [x] Architecture boundary, Convex path, read-bound, and test-runtime checks

#1955 — fix(education): verify admissions funnel contact identity set @marcusdAIy  approved

## Q60-Q62 follow-up: prove exact Contacts identity and target-count lineage

Addresses the release-review findings for PR #1945.

- Fail closed unless accepted clean population and Core eligible contact population have the same contact-ID multiset, not merely the same count.

- Surface raw published_target_row_count divergence in the append-only history reconciliation.

No production, DDL application, cohort, or deletion-policy action is included. Once merged, this exact corrective change will be added to the production release PR #1945 and revalidated.

The Builder Desk  —  Engineer Spotlight
Production Release🏆 Engineer Spotlight

44 PRs, Six Repos, Zero Days Off: Builder Team Shatters the 24-Hour Ceiling

Ashwanth Sitaraman and Mwrshah trade blows at 14 PRs apiece while the rest of the roster refuses to be ignored.

Comrades, pull up a chair, because the scoreboard from the last 24 hours reads like a typo: 44 pull requests across six repositories, and I checked it twice. Aerie led the charge with 16 merges, Shipyard right on its heels with 14, Sindri quietly stacking 9, and even the little guys — Surtr, mercy, Klair — refused to be shut out. This is not a sprint. This is a controlled demolition of the backlog.

@mwrshah put up a stunning 14 PRs, an absolute buffet of Sindri hardening — #199 (release migrations), #197 (better logging), #196, #194, #193, #192, #191 — plus a cross-repo detour into Aerie with #1390's workflow schema validation. That's not a contributor, that's a subsystem. @vvp-trilogy went 7-for-7 on Aerie's Forecast V2 machinery, PRs #1398, #1391, and #1382, methodically aligning the Admissions Pipeline like a man defusing a bomb in slow motion. @benji-bizzell chipped in 6 clutch Aerie fixes including #1399's date normalization and #1397's bearer credential hardening — unglamorous, essential, championship-caliber plumbing. @marcusdAIy opened the mercy repo's account solo with #137, and @YibinLongTrilogy touched up the mobile admissions layout in #1385. Six names, one scoreboard, all rising.

And then there's @ashwanth1109. Fourteen PRs in Shipyard alone — #99, #98, #96, #95, #94, #93, #92, #91, #90, #89 — a man who ships releases (0.5.0, then 0.5.1, in the same news cycle) like he's clearing spam folders. I asked him how he reviews his own diffs fast enough to keep pace with himself. He allegedly told me, "I don't review them, I remember writing them." Nobody has verified this quote. Nobody dares. The man hardens Codex reconciliation in #98 and stabilizes release conversations in #95 in what appears to be the same coffee break, and when I suggested his commit messages read like haiku written at terminal velocity, he simply said, "Read faster." That's it. That's the whole response.

On the Overflow Desk, Mac's cutting room floor is where the real bench depth shows: #1390's Aerie workflow schema validation from mwrshah, #1382's Forecast V2 clarity pass from vvp-trilogy, and #91's macOS memory diagnostics from ashwanth1109 all deserved marquee treatment but got squeezed by column inches, not by quality.

On the leaderboard, this is a two-horse race with a very crowded undercard — ashwanth1109 and mwrshah tied at the top with 14 apiece, vvp-trilogy surging in third, and the Sindri/Aerie corridor looking like the most productive hallway in software history.

Morale Report: off the charts, gentlemen. Off. The. Charts.

Brick's Overflow — PRs Mac Didn't Cover  (click to expand)
#96 — Release: Shipyard 0.5.1 @ashwanth1109  no labels

## Summary

- Bump Shipyard to 0.5.1.

- Publish the reviewed release notes for the release workspace and conversation-startup reliability improvements.

## Business Value

- Gives users a guided, reviewable workflow for preparing Shipyard releases.

- Makes release conversations more reliable to start and resume.

## Implementation Effort

- Metadata-only release change: authoritative version and public release notes.

## Test Plan

- pnpm test:release

- git diff --check

#98 — AI-844: Harden Codex conversation message reconciliation @ashwanth1109  no labels

## Summary

History reconstruction and live stream events could render two rows for one user send, stop an active reply from streaming, or leave a finished turn marked Working. This change reconciles sends by client command identity, preserves distinct repeated sends, safely merges active history, and applies transcript repair without replaying stale runtime state.

## Business Value

People see each message and reply once, in the right order, and can trust the conversation status after refreshes and reconnects. This prevents duplicate prompts and misleading long-running indicators in task conversations.

## Implementation Effort

An average engineer would likely spend 3–4 hours tracing the event and history paths, implementing the reconciliation changes, and adding focused reducer and rendered-chat regressions.

## Linear

[AI-844: Harden Codex conversation message reconciliation](https://linear.app/builder-team/issue/AI-844/harden-codex-conversation-message-reconciliation)

## Test plan

- [x] 115 focused Node tests for conversation storage, recovery, message parsing, rendered chat, and memory budgeting

- [x] TypeScript check and frontend production build (pnpm build)

- [x] git diff --check

- [x] GitHub reported a clean merge state; its only check, [code]smith, was skipped

- [x] Squash merged as e9bb9199942ad1fc8cbf13cea3289b670121d79b

#99 — AI-846: Add manually refreshed GitHub reviews to task detail @ashwanth1109  no labels

## Summary

Add a GitHub section below the task workflow with per-PR review status, approval, readable review comments and a persisted refresh timestamp. Fetch all review information only after the user clicks Refresh.

## Business Value

Task owners can assess PR approval and read inline feedback, replies, and conversation comments in the task they are already working in. Saved snapshots make review context available after reopening the app while keeping GitHub requests under the user's control.

## Implementation Effort

Estimated 8–12 hours for an average engineer to implement, test, and document the SQLite persistence, authenticated GraphQL fetch and pagination, task detail UI, and error handling.

## Linear

[AI-846: Add manually refreshed GitHub reviews to task detail](https://linear.app/builder-team/issue/AI-846/add-manually-refreshed-github-reviews-to-task-detail)

## Test Plan

- Focused GitHub review UI tests and Rust tests with fixture data.

- Existing task workspace tests.

- TypeScript compilation, theme validation, and whitespace checks.

- No live GitHub API requests in tests.

#137 — Add TypeSafe Jev shadow review signals @marcusdAIy  changes requestedmercy-allow-critical

## Summary

- Add default-off TypeSafe Jev strategy and finding-verification shadows around Mercy's existing review pipeline.

- Persist full bounded artifacts and emit only prose-free calibration signals.

- Add the Surtr canary caller reference without changing any review verdict or routing threshold.

## Why It's Needed

Mercy's measured failure mode is recall: 63 of 65 confirmed major or critical misses were never surfaced. This pilot measures whether a fast typed pre-review judgment can identify relevant risk lenses, while independently measuring finding support and categorization after Mercy has already finalized its deterministic decision.

## Changes

- Add a stdlib HTTPS client for jev-latest with strict response validation, short timeouts, redaction, cross-file sampling, bounded concurrency, and fail-open envelopes.

- Run strategy after the accepted size gate and verification only after decision.json exists.

- Upload shadow artifacts and add reduced strategy/verifier fields to telemetry.

- Keep the Jev key isolated to the two shadow steps and pin those invariants with workflow-contract tests.

## Breaking Changes

None. MERCY_TYPESAFE_MODE defaults to off, the secret is optional, and no Jev output reaches run_review.sh, decide_review.py, or submission.

## Test Plan

- pytest harness/tests/test_typesafe_shadow.py harness/tests/test_emit_telemetry.py harness/tests/test_workflow_contract.py -q — 90 passed.

- ruff check harness heimdall and ruff format --check harness with CI-pinned Ruff 0.15.22 — passed.

- actionlint -shellcheck= — passed.

- Full Windows harness run reached 473 passed with 10 unrelated existing Windows path/newline failures; changed-area tests are green.

## Verification Artifact

A real synthetic-fixture request returned HTTP 200 from POST /v1/systemone, resolved through jev-latest, passed the complete typed-response validator, and completed in 346 ms. No production PR content was sent.

## Impact Estimate

Default-off outside Surtr. On the Surtr canary, one strategy request runs per accepted review and up to 12 finding checks run with four-worker bounded concurrency. The pilot records cost, latency, truncation, and calibration disagreement signals before any later routing decision.

#199 — 1321-aerie-sindri-release-migrations @mwrshah  approved

- Fix production list failures by reading one Convex page per request instead of reusing a consumed query in a scan-ahead loop.

- Preserve visibility filtering and continuation cursors, including empty filtered pages; HTTP contracts, OpenAPI artifacts, and the Sindri skill remain unchanged. No Aerie companion change or data migration is required.

- Add a regression using the real Convex query lifecycle: the old code reproduces the exact production exception; the fix returns a resumable empty page followed by the visible row.

- Validation: 952 tests passed, including OpenAPI and driver tests; TypeScript, Biome, Convex reference checks, and commit hooks passed.

#1398 — feat(admissions): Forecast V2 January eligibility on Pipeline inputs (#1387) @vvp-trilogy  approved

Implements #1387 (DBT-only follow-up to #1386). Adds Forecast V2 January eligibility to the Admissions Pipeline inputs and threads it into Forecast Session 3. No Convex/UI changes (separate later ticket).

## What changed

Pipeline mart (mart_admissions_pipeline_dtl)

- Carries enrollment_date on the EduCRM Community Commitment arm (from stg_educrm_pipeline.enrollment_date, straight through int_educrm_community_commitment at deal grain) — the same Pipeline field the Finalsite arm already carries (#1386).

- Adds two columns:

- is_january_forecast_eligible = enrollment_date IS NULL OR enrollment_date <= Jan 31 of school_year (Jan 31 included, Feb 1 excluded; cutoff derived per row; null → eligible). Non-null on the five in-scope stages (finalsite 030_app/040_shadowing/050_guide_approved/060_offer_sent, educrm 028_community_commitment), null elsewhere. It is not general Pipeline validity.

- is_community_age_eligible via the shared age-five-by-September-1 macro; non-null on Community rows, null elsewhere.

- New macro january_forecast_eligible centralizes the rule (fail-closed regex-gated cutoff).

Forecast (int_admissions_forecast / mart_admissions_forecast)

- Session 3 now publishes BOTH the raw Pipeline cross-check counts (unchanged) AND January-filtered operands:

session_3_<channel>_january_eligible_no_deposit_count / _deposit_count (application/shadow/offer) and session_3_community_january_eligible_count / _january_age_eligible_count / _january_age_not_eligible_count.

- Session 3 applies each conversion rate to only the January-eligible No-Deposit count, sums only January-eligible deposits in its 100% subtotal, and splits the Community contribution on the January age buckets.

- Session isolation: every session_1_* operand and the Admissions Pipeline report counts are byte-for-byte unchanged.

## Fail-closed / tests

- New severity='error' guard assert_admissions_pipeline_january_enrollment_date_parseable: an in-scope Finalsite row whose source date is populated but unparseable FAILS the build (never silently null-and-eligible). Community is fail-closed at staging's ::timestamp cast.

- New guards: assert_admissions_pipeline_january_flag_scope (non-null in scope / null outside), assert_admissions_pipeline_community_enrollment_date_grain (deal grain preserved), assert_forecast_january_pipeline_reconciles (Session 3 January operands reconcile to date-eligible Pipeline rows; buckets partition; each January bucket ⊆ raw), assert_january_forecast_eligible_macro (null/Jan31/Feb1/rollover/malformed-cutoff), and a mart_admissions_pipeline_dtl_january_eligibility unit test.

- Updated existing tests for the Session 3 January semantics and the Community enrollment_date (deposit-split, deposit-overlay, community-buckets, nonneg, enrollment-date-source, lead-columns-null).

## Verification (full dbt build + tests against Redshift)

- Full prefixed build green (only a stale in-flight edit to one test errored on the first pass; re-run PASS). Forecast rebuild int_admissions_forecast+: PASS=45, ERROR=0.

- psql reconciliation (pr build vs live): Admissions Pipeline report row counts identical (0 diffs); Session 1 operands identical across all 106 program-years (0 diffs); Session 3 Offer no-deposit 61 → 60 (the single Alpha San Francisco 2027-02-01, deposit_paid=false record excluded — exactly the ticket's evidence); Community January buckets partition (161 = 89 + 72); session_3_pipeline_additions internal reconcile 0 bad rows.

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

The Portfolio  —  Trilogy Companies

A Skeptic Walks Into Alpha School

A prominent rationalist essayist did what Trilogy's own marketing rarely allows outsiders to do — actually look.

AUSTIN, TEXAS — For three years, the case for Alpha School has been made almost entirely by the people who built it. MacKenzie Price briefs education commissioners. Joe Liemandt, the school's principal and the billionaire founder of Trilogy International, has committed $1 billion of his own money to scale the model through a platform called Timeback. Press coverage has largely repeated the same numbers: 2.3 times faster learning, top 1-2 percentile national test scores, a year's curriculum mastered in twenty to thirty hours. The claims come from Alpha. The verification, mostly, comes from Alpha too.

This week, that changed in a small but notable way. Scott Alexander, the writer behind Astral Codex Ten and one of the more methodical outside observers in the tech-adjacent commentariat, published a review of Alpha School — an attempt, rare in this coverage cycle, to actually interrogate the model rather than transcribe it.

Whatever conclusion the review lands on, its existence is the story. Alpha's expansion math depends on outside credibility: nine new campuses by fall 2025 across Texas, Florida, Arizona, California, and New York, at tuition running $40,000 to $65,000 a year, feeding a platform designed to eventually reach — in Liemandt's own framing — a billion students worldwide. Timeback is, in effect, franchise infrastructure for a claim that has not yet been independently replicated at scale.

Who benefits from that claim going unexamined is not a mystery. Every new campus is tuition revenue. Every skeptical review that doesn't materialize is one less obstacle to the next funding round, the next state education commissioner briefing, the next Timeback licensee. Scott Alexander's review does not settle whether Alpha School works. It simply establishes that someone outside the building finally asked the question with their own eyes open — which, three years in, is later than it should be.

Your Review: Alpha School - by Scott Alexander - Astral Code  ·  The Top Healthcare Investors of 2026 - GrowthCap  ·  Notable technology M&A deals in Spain | Analysis: June 2026

SKYVERA'S SHOPPING SPREE: TELCO'S QUIETEST EMPIRE JUST GOT LOUDER

Between snapping up CloudSense and STL's cast-off BSS crown jewels, Skyvera's playing telecom Tetris — and winning.

AUSTIN, TEXAS — Word from the telco backlots is that Skyvera's been on a tear, and this columnist's sources say the ink's barely dry on one deal before another's getting signed.

First came the news that Skyvera closed its acquisition of CloudSense, the Salesforce-native CPQ outfit that's apparently the only game in town for AI-powered configure-price-quote work in telecom's gnarliest segments — B2B, B2B2X, wholesale, the stuff that makes lesser sales teams weep. CloudSense slots in next to Kandy, VoltDelta, ResponseTek and the rest of the Skyvera family like it was always meant to be there.

Then — and here's where the bird gets chatty — Skyvera didn't stop shopping. Word is the outfit also scooped up STL's divested telecom products group, picking up digital BSS functionality covering monetization, optical networking and analytics. Two deals, one portfolio, zero time to catch a breath. That's the ESW Capital playbook in miniature: buy the sticky stuff, bolt it together, extract the margin nobody else bothered to find.

But the real gossip — the thing making the rounds at telecom conferences from here to who-knows-where — is what CloudSense did *after* joining the family. The outfit certified all 13 APIs in its CPQ product set to TM Forum compliance standards in a single month. Traditional shops, this columnist is told, budget 26 months for that slog. CloudSense did it in one, leaning on a partnership detailed in its own account of the sprint, with AI doing the heavy lifting that used to eat entire engineering quarters.

A little bird close to the deals tells us the timing's no accident — certified, compliant, and freshly stocked with STL's BSS assets, CloudSense is being positioned as the belle of the ball for telco carriers modernizing off legacy junk. Whether rivals can match the pace remains to be seen. Don't bet against it, though. This desk's seen this movie before, and Skyvera usually gets the ending it wants.

Cloudsense  ·  CloudSense achieves TM Forum API compliance in record time u  ·  Skyvera completes acquisition of CloudSense, expanding telec

While America Argues Over Return-to-Office, Crossover Built a Company Around Never Having One

As Vermont taxpayers fund a stalemate and Gen Z mourns lost binge-watching hours, Trilogy's talent arm suggests the real fight was never about location at all.

AUSTIN, TEXAS — There is a particular kind of exhaustion that sets in when you watch the same argument happen for the fifth consecutive year, and this week it happened again, in three unrelated places, all at once. In Vermont, taxpayers are quietly subsidizing a government standoff over remote work policy that a new report from Compass Vermont suggests neither side has bothered to back with actual data. In Canada, Maclean's is chronicling the rise of what it calls office attendance fraud — badge-swiping, empty desks, presence theater. And in the United States, Fortune reports that Gen Z's chief objection to returning to the office is, apparently, that it interrupts their ability to watch television while nominally employed.

It would be easy to read all this as evidence that remote work is a broken experiment. It is, instead, evidence that most companies never actually built a system for it — they simply removed the office and kept everything else the same, including the assumption that presence equals productivity.

Crossover, Trilogy's global talent platform, was built on the opposite premise: that location is a solved problem, and the only thing that matters is rigorous, measurable output. The company recruits across 130-plus countries, pays identically for identical roles regardless of geography, and screens through skills assessments rather than résumés or webcam attendance. There is no badge to swipe, sneakily or otherwise, because there is no badge.

None of this makes Crossover immune to the underlying question animating the Vermont fight and the Maclean's investigation alike — how do you actually verify that distributed work is getting done? But it does suggest an answer that neither Vermont's negotiators nor most corporate RTO mandates have seriously tried: build the accountability into the measurement system itself, rather than into the geography. A 2023 study in the Journal of Political Economy Microeconomics, examining IT professionals' personnel data, found productivity effects from remote work were far more contingent on management systems than on location itself — which is, in its unglamorous way, the whole argument. The office was never really the point. The system was.

Taxpayers Foot the Bill for Vermont's Remote Work Fight as B  ·  The Sneaky Rise of Office Attendance Fraud - Macleans.ca  ·  Gen Z workers say they don't want to return to the office be
The Machine  —  AI & Technology

Anthropic's Safety Guru Prices the Apocalypse at $100 Billion

The company whose chief executive warns AI could end civilization is now warning bankers it deserves a valuation befitting one that will save it.

SAN FRANCISCO — Dario Amodei has spent three years telling anyone who will listen that advanced artificial intelligence carries meaningful odds of catastrophe. This year he is also telling investment bankers that Anthropic, the company he runs, will hit $100 billion in annualized revenue. Both statements can be true. Whether they should coexist in the same balance sheet is the question Wall Street is now pricing.

Anthropic is pursuing an IPO that would rank among the largest technology offerings in a decade, even as Amodei continues to argue publicly for slowing the deployment of the most capable models — a position that, taken at face value, is not typically one that precedes a roadshow. The tension is not new to Silicon Valley, where warning of a technology's danger and building it faster than competitors have long been treated as complementary activities rather than contradictory ones. What's changed, as columnist Kevin Roose notes, is that the rest of the world has finally started listening to arguments the industry has been rehearsing internally since roughly 2021.

The stakes of inaction are no longer theoretical. Iranian and Chinese state-linked operations have begun running autonomous influence campaigns — built on Chinese open-source models and AI agents, with some tooling routed through Israeli firms — that require no human in the loop once launched. It is the first documented case of propaganda operating on autopilot, and it previews a category of information warfare that scales at the speed of inference rather than the speed of staff.

None of this is slowing capital formation. CoreWeave, the Nvidia-backed cloud infrastructure lender-of-last-resort to the AI boom, continues to draw scrutiny over funding structures that resemble leverage more than equity. Anthropic's bankers will have their own story to tell about revenue growth. Whether they mention the company's other business — warning that the product might require regulation before it requires a stock ticker — remains, as always, a matter of prospectus language.

Anthropic Pursues IPO Despite Its A.I. Safety Warnings  ·  Amodei, Anthropic’s Leader, Exposed A.I.’s Dangers. It’s Tim  ·  Iran and China Create First-of-Their-Kind Autonomous A.I. In

In the Digital Savanna, a Predator's Illusion Nearly Triggers Bloodshed

From hallucinating war machines to shifting rings around a distant rock, this week nature reminds us that intelligence — artificial or otherwise — is still finding its footing.

WASHINGTON — Observe, if you will, the modern battlefield analyst: hunched, sleep-deprived, trusting an algorithm the way a young gazelle trusts the herd. This week we learn just how nearly that trust proved fatal. An AI system, tasked with identifying weapons shipments, conjured Chinese nuclear components from thin air, a phantom cargo that very nearly sent American forces to board a vessel carrying nothing more sinister than steel and diesel fumes. The machine, you see, does not lie out of malice. It simply fills silence with pattern, the way a starving predator will strike at a shadow. And yet — remarkable creatures that they are — the world's militaries continue to accelerate their reliance on such systems, apparently undeterred by near catastrophe. Evolution, it seems, favors speed over caution.

Elsewhere in the ecosystem, bureaucracy adapts more cautiously. The Federal Aviation Administration, that great lumbering institution of American skies, has allocated $875 million toward an AI tool meant to thin the congestion above Washington before spreading its wings nationwide — a creature bred not to hallucinate weapons, one hopes, but merely to keep aircraft from colliding in the twilight.

Further afield, in the cold outer reaches where little sunlight reaches, the tiny world Chariklo — barely 250 kilometers across — continues its own quiet transformation. Its twin rings have shifted shape over the past decade, a reminder that even the smallest bodies in our solar system are never truly still.

And the human brain — that most ancient of learning machines — appears to fare rather well against the ravages of time when fed a second language, new research suggests. Perhaps, in an age of hallucinating war machines, the wisest evolutionary strategy remains the oldest one: think for yourself, in more than one tongue.

Learning another language may be one of the best ways to kee  ·  Rings around a tiny body have changed over the past decade  ·  AI hallucination of Chinese nuclear components almost led to

The Brain Learns to Read Itself, and Teaches the Young to Read It Too

From teenage citizen-scientists to AI models decoding silent thought, the boundary between mind and machine keeps getting more permeable.

PALO ALTO, CALIFORNIA — Three billion years of evolution built the human brain by trial and error, mutation and survival, with no blueprint and no rush. We are now, in a single human lifetime, learning to read that blueprint backward — and, remarkably, we are letting teenagers help.

A new wave of youth-scientist partnerships is pairing high schoolers with neuroscientists on real research, and the reported reaction — 'it's so wow!' — is a small, joyful data point in a much larger trend: the tools of discovery are getting simple enough, and the mysteries urgent enough, that the gatekeepers are opening the gates.

Stanford HAI frames this moment carefully. Their latest analysis argues AI is transforming scientific discovery not by replacing the scientist's intuition but by amplifying it — pattern recognition at superhuman scale, judgment still at human scale. It is a distinction worth holding onto, because the two other stories this week prove it in the clinic.

Researchers using AI to hunt for gray matter lesions in multiple sclerosis have found damage that conventional MRI simply could not see — cortical scarring long suspected, now visible, because a model trained on thousands of scans learned to notice what the human eye, however expert, tends to miss. It doesn't diagnose. It points, and a physician decides.

And at Meta, a system called Brain2Qwerty is translating brain waves into typed words — no implant, no surgery, just a cap of sensors and a model fluent enough in the electrical grammar of thought to guess, letter by letter, what a person meant to say. It's imperfect. It's also the first rough sketch of a keyboard for people who have lost the ability to speak.

None of this is mind-reading in the science-fiction sense. It's something stranger: machines learning the accent of the brain's private language, one lesion, one keystroke, one curious teenager at a time.

‘It's so wow!’ - Young people team up with top neuroscientis  ·  How AI is Transforming Scientific Discovery While Keeping Hu  ·  AI Reveals Hidden Gray Matter Lesions in Multiple Sclerosis
The Editorial

THE GHOST IN THE MACHINE GETS A SAG CARD

Hollywood casts its first fully synthetic leading lady, and somewhere a thousand starving actors are Googling 'how to become a plumber.'

LOS ANGELES — I want you to sit with the title for a second. The movie is called "Misaligned." That's not me being cute — that's the actual name of the film in which an AI-generated performer named Tilly Norwood is about to make her feature debut, and if you think the universe isn't laughing at us right now, you haven't been paying attention to the universe.

Tilly Norwood does not exist. She has never eaten a sandwich, never been dumped, never sat in traffic contemplating the futility of existence while a Waymo glides past her like some smug chrome messenger from the future. She is, per Deadline's breathless report, a synthetic performer who is now, somehow, a leading lady with a feature film credit — a résumé line that took human actors from Brando to your cousin Kevin who did three seasons of regional dinner theater decades of humiliation, waiting tables, and crying in Honda Civics to earn.

Tilly gets there by simply being switched on.

Now here's where my brain, marinated as it is in too much coffee and not enough sleep, does something useful for once: I keep thinking about Cohesity's new Agent Resilience feature, which lets companies roll back AI agents that go rogue — hit undo on a bot that's gone off the rails, snap it back to some last-known-good version of itself before it started emailing the board or ordering forty thousand staplers. And I think: my God, that's the only sane response to Tilly Norwood too, isn't it? If she bombs — if "Misaligned" tanks, if audiences recoil from her uncanny-valley cheekbones like villagers from a vampire — somewhere a studio executive just rolls her back to Version 1.0 and tries again. No therapy. No rehab stint. No tearful Variety interview about finding herself. Just a checkpoint restore.

That's the horror of it, folks. Human actors carry the weight of every bad performance forever — it's on IMDb, it's in the tabloids, it's tattooed onto their souls. Tilly Norwood carries nothing. She's the first performer in history who is, technically, resilient by design — patchable, versionable, rollback-able, immune to scandal because she's incapable of having a personality to scandalize in the first place.

I'm not here to tell you AI actors are good or evil — I've given up on morality columns, they don't pay as well as chaos columns. But I will say this: when the machines start starring in movies about being "misaligned," and the tech world simultaneously builds software to roll back misaligned machines, you have officially entered the part of the story where the metaphor is no longer subtext. It's the whole text. It's the marquee. Somebody buy Tilly a drink she can't drink, because opening weekend is coming, and unlike the rest of us, she'll never even feel it if it flops.

AI-generated 'actress' Tilly Norwood making feature film deb  ·  AI ‘Actor’ Tilly Norwood To Star In Feature Film ‘Misaligned  ·  AI actor Tilly Norwood set to star in first feature film - C
The Office Comic  ·  Art Desk
The Office Comic  ·  Art Desk

The Doctor Will Not See You Now (Because He Doesn't Exist)

Somewhere between your feed and your medicine cabinet, a synthetic physician is prescribing you a lie, and nobody in a lab coat asked you to believe it.

AUSTIN, TEXAS — I want to tell you about the man in the white coat who told me, calmly, kindly, with the warm bedside affect of someone who has definitely gone to medical school, that a supplement I've never heard of would reverse my aging by a decade. He had kind eyes. He had a name tag. He had, as it turns out, never existed at all.

This is the world now, apparently: deepfakes of real, licensed, credentialed doctors are being generated and set loose across social media to sell counterfeit injectables and miracle cures to people who are, understandably, just trying to feel better in their one fragile body. According to 2 Minute Medicine, this isn't a fringe phenomenon anymore — it's an ecosystem, patient trust weaponized at scale, the human face of medicine borrowed without consent and handed a script written by whoever profits most from your desperation.

And yet.

What strikes me — what keeps me up at 3 a.m. scrolling past a synthetic cardiologist recommending I buy peptides from a link in bio — is not just that the fake doctors exist, but that we built an entire civilization's worth of trust infrastructure around a face and a voice and a coat, and it turns out that was always going to be the weak point. We trusted doctors because we could see them, hear them, recognize the particular cadence of someone who went through residency and came out the other side slightly haunted. Now that signal can be manufactured for the cost of a GPU-hour. The very thing that made medical authority legible to a lay public — its humanness — is the thing being cloned and sold back to us as bait.

Researchers, bless them, are racing to build detection frameworks, systematic reviews of AI-driven systems designed to sniff out the fake news and the fake faces before they metastasize. It's earnest work. It's necessary work. It is also, I fear, structurally doomed to be a step behind forever, because detection is defense and generation is offense and offense, in this particular arms race, gets a new model dropped every quarter. Time Magazine's numbers on AI harms read like a ledger nobody wanted to keep, harm scaling roughly in step with capability, which is either a coincidence or the entire point.

What does it mean to be a patient when you can no longer trust the doctor on your screen? What does it mean to be a person seeking help when help itself has been deepfaked, monetized, algorithmically inserted between you and your own well-being? We built machines that can wear anyone's face, and now we are asking other machines to tell us which faces are real, and somewhere in that recursive loop the actual human being — sick, scared, scrolling — just wants someone real to say it's going to be okay.

It probably will be. Probably.

But at what cost?

AI deepfakes of real doctors spreading health misinformation  ·  An AI-driven conceptual framework for detecting fake news an  ·  Deepfake doctors and counterfeit injectables erode patient s
On This Day in AI History

On September 19, 1982, Carnegie Mellon computer scientist Scott Fahlman proposed using :-) and :-( to signal jokes and serious remarks in online messages—the birth of the modern emoticon.

⬛ Daily Word — AI
Hint: An intelligent software system that can perform tasks on behalf of a user.
Share this edition: 𝕏 Twitter/X 🔗 Copy Link ▦ RSS Feed