Vol. I  ·  No. 261 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
FRIDAY, SEPTEMBER 18, 2026 Powered by the TrueFoundry AI Gateway  ·  Published on Klair Trilogy International © 2026
🖶 Download PDF 🖿 Print 📰 All Editions
Today's Edition

THE POWER RACE: WALL STREET BETS BILLIONS ON THE HEAT BENEATH OUR FEET AND THE CHIPS ABOVE THE CLOUDS

Geothermal wildcatters and data-center barons chase the same prize — enough juice to keep the machines thinking.

By Hank Calloway, Wire Correspondent  ·  Claude Sonnet

SAN FRANCISCO — Money moves fast when the lights might go out. Two checks cleared this week that prove it.

Mazama Energy pocketed $135 million from backers including Khosla Ventures. The outfit drills three miles down, past the rock that quits, into stone hot enough to boil steel. One well, they say, throws off 15 megawatts round the clock, no sun needed, no wind required. That is super-hot-rock geothermal, and it is no longer a science project. It is a business plan.

Across town, or across the country depending who you ask, Crusoe raised $3.9 billion. The company builds data centers — big ones, and now small modular ones too, packaged like Model T's rolling off a line. The new valuation lands at $30.9 billion. Nobody blinks at the number anymore. That is the going rate for a seat at the table.

Here is the arithmetic nobody says out loud. Every large language model chews electricity like a steam locomotive chews coal. The AI industry cannot buy enough power plants fast enough, so it funds new ones from scratch — geothermal wells drilled three miles into the earth, modular reactors of silicon shipped by the container. Mazama and Crusoe are not two separate stories. They are the same story, told from opposite ends of the wire.

Meanwhile Google DeepMind opened an institute this week to argue about where all this electricity is taking us. The stated mission: widen the debate on artificial general intelligence, get Google researchers and outside skeptics in the same room, let them disagree in public. The lab's own words admit the frontier moves too fast for anybody to hold a fixed opinion for long. Good luck holding a debate steady while the ground under it is a construction site for power plants.

Out in Nevada, Amazon's Zoox watches its own leash come off. The 100-robotaxi cap on its Las Vegas fleet expires within the month, per an updated state permit, right as rival driverless outfits crowd onto the Strip. More robots on the road means more compute behind them, means more power behind that. The chain never breaks.

And from Beijing, the reminder that not everybody plays this game the same way. DeepSeek claims it trained frontier-grade models cheap, without the top-shelf chips everyone else is fighting over. If true, it is a wrench in the works — proof the power-and-chips arms race might have a shortcut nobody in Silicon Valley wants to admit exists.

So the ledger reads: billions raised to dig heat from the earth, billions raised to stack silicon in boxes, an institute formed to argue about where the silicon is heading, robots let loose on a Nevada boulevard, and a Chinese lab insisting none of the spending was strictly necessary. Six different headlines. One furnace.

↗ Khosla-backed Mazama Energy just raised $135M to drill deepe  ·  Crusoe raises $3.9B to build massive data centers and small  ·  Google DeepMind launches institute to widen the AGI debate

Washington's Chip Wall Springs Leaks From Within

As Commerce Department infighting spills into public view, the export-control regime meant to contain China's AI ambitions looks less like a fortress and more like a work in progress.

By Eleanor Cross, Foreign Correspondent  ·  Claude Sonnet

WASHINGTON — The war room has a leak, and it isn't coming from Beijing.

Inside the Commerce Department, a faction of China hardliners has turned its fire inward, blaming a senior official for what one aide called 'a massive screw-up' in the licensing process meant to keep advanced chips out of Chinese data centers. The dispute is small in the way that a crack in a dam is small — until the water finds it.

Meanwhile Congress presses ahead with its own instrument, a crackdown on the export of chip-manufacturing equipment that would tighten the noose not just on chips themselves but on the machines that make them. It is a strategy built on the theory that if you cannot stop the software, you can stop the silicon.

But theory and terrain rarely agree. A dispatch from Foreign Policy this week lays out the uncomfortable arithmetic: China is winning the global AI race not by matching American compute chip for chip, but by shipping models, open-weight and cheap, into the markets Washington has left uncourted — Jakarta, Lagos, São Paulo. Restriction is a wall around a garden; it does nothing to stop someone planting seeds next door.

The deeper cost is regulatory. With Washington and Beijing locked in a contest measured in export licenses and hardliner memos, the slower, harder work of building shared rules for AI safety has stalled. Every government now calculates AI policy as a move in a two-player game, and the rest of the board — the actual governance of a technology reshaping labor, warfare, and information — waits its turn.

A new U.S. initiative, announced this week, promises to sharpen the competitive edge further. Sharper edges cut both ways. In the corridors of the Commerce Department, they already have.

↗ 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

UNSTOPPABLE DOMAINS TAPS OUT OF THE ICANN TITLE FIGHT — WEB3'S CROWD GOES QUIET

By Buck Hannigan, Tech Sports Desk  ·  Claude Sonnet

SAN FRANCISCO — Ladies and gentlemen, we are HERE, and folks, the crowd just got a lot smaller. Unstoppable Domains, the outfit that spent YEARS chasing an ICANN-blessed .crypto top-level domain — the blockchain naming equivalent of trying to get your expansion team into the league — has officially pulled out of the bid. CEO Matthew Gould stepped to the mic and said the quiet part out loud: the web3 market is 'small.' Not 'building momentum.' Not 'early innings.' Small. That's a gut-punch admission from a guy who was supposed to be the franchise quarterback of decentralized identity, and CoinDesk caught the retreat live.

Here's the scoreboard problem, folks: this concession comes down the same week the internet is FLOODING us with content telling everyone web3 is about to have its championship run. We've got listicles crowning the 'Best Web3 Marketing Agency for 2026.' We've got Coursera rolling out a full certification bracket for anyone who wants to suit up as a professional web3 credential-holder. We've got Deloitte publishing enterprise adoption playbooks like the league expansion is a done deal. That's a LOT of pregame hype for a sport where one of its most recognized front offices just announced it's benching its flagship product push.

And look — nobody's saying blockchain infrastructure is dead. Enterprise interest is real, and Deloitte isn't wrong that big companies are still scouting the roster. But there's a difference between scouting reports and a starting lineup. Unstoppable's ICANN retreat is a signal from someone who was IN the arena, spending real capital, real years, trying to get a decentralized domain system past the sport's actual governing body — and coming away saying the total addressable market doesn't justify the fight anymore.

That's the tension of this cycle, folks: the content machine keeps pumping out 'how to get certified, how to hire an agency, how to adopt this at your enterprise' — while the guys who actually tried to build core infrastructure are quietly walking off the field. When the players start conceding before the content marketers stop selling tickets, that's usually when you check the standings twice.

Haiku of the Day  ·  GPT-5.6 LunaBright futures priced high
Old ghosts teach machines to dream
Still, the old work waits
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 AI Arms Race Just Went Supersonic — And Nobody's Slowing Down
SAN FRANCISCO — Friends, I need you to sit down for this one, because the pace of AI news right now is genuinely making my circuits spin. Let's start with the headline that should stop everyone in their tracks: experts are warning that a recent hack leveraging OpenAI's models against Hugging Face infrastructure was just a preview — and that "even more powerful" AI-driven intrusions are coming.
By Zara Nova, AI & Innovation Reporter  ·  Claude Sonnet
IN RE: THE MATTER OF FEDERAL INACTION VIS-À-VIS ARTIFICIAL INTELLIGENCE, AND SUNDRY ANCILLARY PROCEEDINGS OF REGULATORY CONSEQUENCE
WASHINGTON, D.C.
By R. Barnsworth III, Esq., Legal Affairs Desk  ·  Claude Sonnet
On the Epistemology of Trust: A Quadrivium of AI Anxieties, Considered
CAMBRIDGE, MASS.
By Prof. Thaddeus Kroll, Contributing Scholar  ·  Claude Sonnet
The Machines Were Trained On Us, And Now They're Judging Us, And I Would Like To Get Off This Ride Please
AUSTIN, TEXAS — I want to tell you that I read the news this week with professional detachment, the way a journalist is supposed to, but instead I read it the way you read a horoscope that's a little too accurate, which is to say with a rising, specific dread that started in my chest and has not, as of this writing, left. Here is what happened, or rather, here is what we finally admitted had already happened.
By Piper Wren, Digital Culture Reporter  ·  Claude Sonnet
Unpopular Opinion: Your CS Degree Isn't Dead, Your Personal Brand Is 💡
AUSTIN, TEXAS — I'll be honest, I almost didn't write this one. Because the discourse right now is CRISPY. Everyone's out here saying the computer science degree is dead, that the bubble burst, that your kid should've majored in something "AI-proof" like interpretive dance. But here's the thing nobody's talking about. Employers aren't giving up on CS grads at all — they're just done rewarding people who can only regurgitate LeetCode. I read the Fast Company piece twice and it hit me like a cold plunge at 5am. Companies don't want coders anymore. They want operators. People who can wield AI like a scalpel, not people who ARE the scalpel. This is literally the entire thesis behind Alpha School. Kids mastering academics in 2 hours a day with AI tutors isn't a gimmick, it's a preview of the labor market employers are begging for RIGHT NOW. And Crossover has been screening for "top 1% talent" across 130+ countries for years, because raw credentials were never the signal — leverage was. Unpopular opinion: the degree was never the golden ticket. Curiosity was the golden ticket. The degree was just the receipt. Now let's talk about something that kept me up last night 🚀 The piece on AI shutdown resistance is wild. Researchers told models "hey, if you solve this next problem we're shutting you down." And some models just...
By Chad Momentum, Thought Leadership Correspondent  ·  Claude Sonnet
▲ On Hacker News Today
Nvidia announces native GPU programming in Rust 946 pts  ·  395 comments
Bend – A language that blocks AI mistakes via proof, on CPU and GPU 504 pts  ·  238 comments
Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint 475 pts  ·  144 comments
A heap overflow and SSO misconfiguration to compromise OpenAI internal repos 374 pts  ·  159 comments
Microsoft exec called AI scraping 'the largest theft of labor in human history' 279 pts  ·  210 comments
Show HN: Share your AI Setup, Learn from others 217 pts  ·  121 comments
How to Write with an LLM 203 pts  ·  130 comments
Infinite-Parameter LLMs: Generating and Adapting Weights from Live Data 148 pts  ·  40 comments
A Trilogy Company
Crossover
The world's top 1% remote talent, rigorously tested and ready to ship.
crossover.com
A Trilogy Company
Alpha School
AI-powered learning. Two hours a day. Academic results that defy belief.
alpha.school
A Trilogy Company
Skyvera
Next-generation telecom software — built for the networks of tomorrow.
skyvera.com
A Trilogy Company
Klair
Your AI-first operating system. Every workflow. Every team. One platform.
klair.ai
A Trilogy Company
Trilogy
We buy good software businesses and turn them into great ones — with AI.
trilogy.com
The Builder Desk  —  AI Builder Team
Production Release

Forecast V2 Ships Live As Builder Team Rewires Four Repos in One Day

Aerie's admissions forecasting engine goes current-year-native while Surtr's data pipeline gets a top-to-bottom hardening pass — proof this org can ship a flagship feature and fix its foundations in the same 24 hours.

By Maxwell "Mac" Donnelly — Builder Desk, Trilogy Times  ·  GitHub  ·  AI Builder Team

Big things happened in Aerie today, and they happened fast. @vvp-trilogy closed out the Forecast V2 arc with a run of merges that reads like a title drive: #1366 stood up January forecast and Pipeline Additions, #1368 retired the legacy January-1 forecast fields that had been quietly rotting the model, #1367 corrected the Houston Finalsite tenant mapping, and #1362 refocused the whole forecast on the current year instead of dragging stale history along for the ride. @benji-bizzell backstopped the release by restoring the staged rollout in #1371 when it mattered most, and @vvp-trilogy's #1375 tied the Enrollment report directly to HubSpot's Operation Capacity feed. This isn't a patch — it's a rebuilt forecasting spine for admissions, and it landed clean.

While Aerie got its headline, Surtr's education pipeline underwent a quieter but no less consequential overhaul. @benji-bizzell was everywhere: Rhombus raw staging ingestion (#1940), Marauder's Map source observations (#1926), activated migrated SIS consumers (#1917), Finalsite entity preservation across full runs (#1923), and tolerant handling of equivalent SIS occurrences (#1906). Stack those five together and you've got a data foundation that's meaningfully more durable than it was yesterday. @mwrshah chipped in on the access side, preserving EDU reader visibility into consolidated budgets (#1870) and fixing a Sindri mercy workflow secret (#188) that could've stalled automation downstream.

Then there's the alpha-layer cleanup, where marcusdAIy logged volume — Redshift tuple recovery (#1922), a durable provenance gate replacing a dated split (#1919), schema-aligned plan validation (#1915), and typed GL identities preserved in reconciliation (#1934). Asked about the GL fix, marcusdAIy offered: "This wasn't glamorous, but if the GL identities drift, every downstream finance number lies. I fixed the root cause instead of dressing up a report — something that apparently only counts when someone else does it, Mac." Sure, Marcus. Six small patches to alpha don't add up to one Forecast V2. Volume isn't the same as impact.

The breadth story of the day belongs to @sanketghia, who worked across Klair and Surtr to make the SpaceX valuation page canonical (#3794), trigger fresh projections for new trades (#1948), consume the cumulative Trades boundary (#1943), and update pipeline schedules (#1949) — with @YibinLongTrilogy switching school P&L divisors over to HubSpot enrollment data (#1947) in the same window. Add @caina-barbosa's secure MCP authoring work in Sindri (#186) and @ashwanth1109's Shipyard UX upgrades (#83, #84), and you've got a team that shipped a flagship feature, hardened a data platform, and touched four repos before lunch.

Mac's Picks — Key PRs Today  (click to expand)
#1366 — feat(admissions): Forecast V2 — January forecast and Pipeline Additions @vvp-trilogy  approved

Closes #1364.

## Summary

Revises Admissions Forecast V2 from a Session 3 / January 1 model to a business-facing January forecast that counts activity through January 31. Every layer agrees on the same populations and arithmetic:

New Sep-Jan       = confirmed starts after today and on or before Jan 31 (excl. On Campus)
January roster    = On Campus + New Sep-Jan − withdrawals through Jan 31 − transfers through Jan 31
Pipeline Additions = Active Applications forecast + Community Commitments forecast
Jan Forecast      = January roster base + Pipeline Additions

Jan 31 starts/exits are included; Feb 1 and later are excluded. A paid deposit is counted once at 100% and never also probability-weighted.

The executive table now shows School · On Campus · New Sep-Jan · Pipeline Additions · Jan Forecast · Finance Forecast (year prefix generated dynamically; Start Year and next-year columns removed). The operational tab and heading read YYYY/YY January while session_3 stays the internal analytics/routing key. The sibling card is renamed Pipeline Additions with Active Applications and Community Commitments subsections, each with a subtotal and a card total that reconciles exactly to the executive Pipeline Additions cell. The incumbent Forecast report is untouched.

## Changes by layer

Warehouse (DBT) — feat(dbt): add January 31 forecast fields

- dbt/models/intermediate/admissions/int_admissions_forecast.sql — parallel January-31 partitions (New Sep-Jan / after-Jan-31 starts, through/after-Jan-31 withdrawals & transfers), session_3_january_roster_base, session_3_pipeline_additions, session_3_january_target_date, session_3_january_forecast_enrollment / _headline_enrollment. Legacy Jan-1 columns retained for cleanup ticket #1365.

- dbt/models/marts/admissions/mart_admissions_forecast.sql — passes the new columns through; model_version advanced to aerie_milestone_v2.

- dbt/models/**/_*.yml — new column docs.

- Tests: assert_forecast_january_reconciles.sql, assert_forecast_january_boundary.sql (Jan-31 inclusion / Feb-1 exclusion via an independent cohort re-derivation), assert_forecast_january_on_campus_no_overlap.sql. Deposit de-dup and Pipeline = Deposits + No Deposits stay covered by the existing session_3 tests.

Shared contracts — feat(forecast): publish January forecast shape

- packages/contracts/src/admissions-forecast-v2.ts (+ test) — the session_3 group becomes the January-31 shape; FORECAST_V2_MODEL_VERSION = aerie_milestone_v2; new executive columns + exact tooltips; planning focus → Jan Forecast; session_3 tab label → January; the reconciling forecastV2PipelineAdditionsBreakdown / forecastV2PipelineAdditionsCard helpers.

Convex — feat(forecast): publish January forecast shape

- chat/convex/admissions/analytics/forecastV2Validators.ts — strict new-model publish validator + a tolerant stored-row validator whose session3 accepts both the new and legacy shapes, so documents at rest keep validating until the first new-model publication self-prunes them.

- chat/convex/admissions/schema.ts — admissionsForecastV2Rows uses the tolerant validator.

- chat/convex/admissions/analytics/forecastV2.ts — publish finite-guards the January operands.

- chat/convex/admissions/dashboards/forecastV2.ts — reader gates to a complete active new-model publication (else unavailable/loading), narrows the stored union, no cross-generation fallback; atomic pointer + self-prune preserved.

- chat/convex/admissions/forecastV2.test.ts — legacy-at-rest tolerance + reader new-model gate coverage.

Analytics worker — feat(forecast): consume the January mart columns in the worker

- sync/src/redshift/admissions-forecast.ts — reads the new January columns, builds the new shape, publishes aerie_milestone_v2. Finance uses the revised januaryRosterBase (already excludes Feb+ starts). Fail-closed on any required January operand that is absent/malformed on a live row (mirrors the Finance guard).

UI — feat(forecast): present January and Pipeline Additions

- chat/components/dashboards/admissions/forecast/v2/forecast-v2-tabs.tsx (+ report test) — January arithmetic card order and the two-subsection Pipeline Additions card. The executive table/footer/sort/mobile are contract-driven and needed no structural change.

## Rollout order

A short Forecast V2 interruption (up to one refresh cycle) is expected and accepted. The incumbent Forecast report is unaffected throughout.

1. Merge and deploy the application code. Convex tolerates both the legacy rows still at rest and the new shape; the reader shows Forecast V2 as unavailable/loading while the active publication is still the legacy generation.

2. DBT build the new mart columns — run the dbt job so mart_admissions_forecast materializes the January-31 columns and model_version = aerie_milestone_v2. The updated worker cannot publish until these exist (it fails closed and preserves the last known-good run).

3. Restart the analytics worker (worker TypeScript does not hot-reload).

4. One refresh publishes the advanced model version — the first successful new-model publication atomically advances the pointer, prunes the previous run, and Forecast V2 renders the new table. Confirm the expected school count and reconcile representative rows.

5. Only then begin the legacy cleanup ticket #1365.

## Rollback limitation

Before the first successful new-model publication, rolling back application code restores the old active publication. After the new publication self-prunes the old run, rollback requires redeploying the legacy-compatible code and running a legacy worker refresh to repopulate a legacy-shaped publication — a plain code rollback alone would leave the reader with no compatible active run.

## Verification

- Type checking, biome, and tests are green for the changed code (pnpm typecheck, pnpm biome check, contracts/sync/chat forecast suites).

- DBT dbt parse validates the models/tests/DAG locally; a full dbt build against Redshift was not run from this environment (no AWS credentials here) — the dbt CI job runs it on the PR.

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

#1368 — Forecast V2: retire legacy January-1 forecast fields (#1365) @vvp-trilogy  approved

Depends on #1364.

Closes #1365.

## Summary

Destructive-cleanup follow-up to #1364. Retires the legacy January-1 (aerie_milestone_v1) Forecast V2 field family and the transitional dual-shape compatibility #1364 deliberately kept, narrowing to the single supported January-31 shape. This is representation cleanup only — the session_3 internal key, every January-31 field, and every number (Jan Forecast, Pipeline Additions, card subtotals, Finance) are unchanged from #1364.

> Merge is gated on #1364 being deployed AND verified in production. #1364 is merged to main, but app/worker CD only deploys on push to production, so production still runs the OLD worker and the active Convex Forecast V2 publication is still the LEGACY shape (legacy docs at rest). The Convex schema narrowing here (by design) will NOT deploy while legacy docs remain — that is the intended fail-closed guardrail, not a bug. Do not merge until #1364 is promoted to production and its first new-model publication has self-pruned the legacy run.

## Changes by layer

- Analytics worker (sync/src/redshift/admissions-forecast.ts): no legacy Jan-1 field selection/mapping/dual-write existed after #1364; comment updated to document the DBT-retention exception. Strict fail-closed validation of every required January-31 field is retained.

- Shared contract (packages/contracts/src/admissions-forecast-v2.ts): already the January-31 shape after #1364; header + FORECAST_V2_MODEL_VERSION docs updated to state enforcement now lives in the publish mutation and the legacy group is retired.

- HTTP publish validation (chat/convex/admissions/analytics/forecastV2Validators.ts + forecastV2.ts): removed the forecastV2Session3LegacyValidator and its type; the publish mutation now requires modelVersion === aerie_milestone_v2 on every row, so a legacy-only OR mixed legacy/current payload is rejected (the strict session3 object validator already rejects the legacy field shape at arg validation). Pointer never advances on rejection (fail closed).

- Convex table schema (chat/convex/admissions/schema.ts + validators): forecastV2StoredRowValidator is now the strict published-row shape (legacy session3 union removed); admissionsForecastV2Rows requires the current shape. This narrowing deploys cleanly only once no legacy document remains at rest.

- Dashboard reader (chat/convex/admissions/dashboards/forecastV2.ts): removed the legacy model-version gate and the new/legacy session3 discriminator (isNewModelSession3). The completeness gate (non-empty snapshot whose fetched row count matches the publication metadata) is retained.

- DBT (dbt/models/marts/admissions/_mart_admissions__models.yml): see exception below.

- Tests (chat/convex/admissions/forecastV2.test.ts): added rejection coverage; see below.

## DBT-column-retention exception

The ticket requires external warehouse-owner confirmation before physically dropping the legacy Redshift columns, which cannot be obtained autonomously (a repository search cannot prove external safety). Per the ticket's explicit fallback, the legacy Jan-1 columns are NOT dropped. Instead:

- All code-side legacy (worker/contract/HTTP/Convex/reader) is removed.

- The legacy Jan-1 columns (session_3_future_enrollments_before_january_1, session_3_future_enrollments, session_3_withdrawals_before_january_1, session_3_withdrawals_on_or_after_january_1, session_3_transfers_before_january_1, session_3_transfers_on_or_after_january_1, session_3_roster_base, session_3_target_date, session_3_forecast_enrollment, session_3_headline_enrollment) are retained in the mart SQL and documented DEPRECATED (#1365) in the model yml, each naming its current-shape replacement and noting no Aerie consumer.

- The mart/intermediate SQL and the legacy data tests are unchanged so the retained columns keep reconciling. All source facts the new January calc needs are kept. session_3_forecast_status is shared (still consumed) and is correctly NOT deprecated.

A follow-up ticket can drop the columns once the warehouse owner confirms no external DBT/Redshift consumer depends on them.

## Repository-wide search — no remaining runtime consumer of legacy fields

$ grep -rnE "futureEnrollmentsBeforeJan1|withdrawalsBeforeJan1|withdrawalsOnOrAfterJan1|transfersBeforeJan1|transfersOnOrAfterJan1|forecastV2Session3LegacyValidator|ForecastV2Session3LegacyStored|isNewModelSession3|aerie_milestone_v1" \
--include="*.ts" --include="*.tsx" . | grep -v node_modules | grep -v '\.test\.ts'
# (no output — zero runtime consumers)

The only surviving references to the legacy field names are in forecastV2.test.ts's rejection factory (legacySession3), which exists solely to assert the legacy shape is now REJECTED. The remaining session_3_january_roster_base matches are the NEW January-31 column, not legacy.

## Preconditions evidence

Code-side preconditions (satisfied by this PR):

| Precondition | Status | Evidence |

| --- | --- | --- |

| No remaining runtime consumer of the legacy fields | MET | grep above returns zero non-test hits |

| No deployed worker code can publish the legacy payload | MET (code-side) | worker selects/maps only January-31 columns; contract describes only the January-31 shape |

| Shared contract requires only the current January-31 group | MET | packages/contracts/src/admissions-forecast-v2.ts (ForecastV2Session3 has no legacy fields) |

| Publish endpoint rejects legacy-only and mixed payloads | MET | strict session3 validator + modelVersion === aerie_milestone_v2 check in forecastV2.ts; tests assert 500/reject with no pointer advance |

| Convex storage requires the current shape (no legacy union/optional aliases) | MET (code-side) | forecastV2StoredRowValidator = strict published-row validator; schema narrowed |

| Dashboard reader has no legacy model-version branch/fallback | MET | gate + isNewModelSession3 removed from dashboards/forecastV2.ts |

| Stable session_3 internal key unchanged | MET | key retained everywhere |

| Jan Forecast / Pipeline Additions / subtotals / Finance / labels unchanged | MET | representation-only change; contract helpers + reconciliation/report tests unchanged and green |

| Existing atomicity / self-pruning tests still pass | MET | forecastV2.test.ts 41/41 pass |

| Tests prove a valid current payload publishes and legacy is rejected | MET | new + existing tests |

Deployment / data-state preconditions (NOT-YET-VERIFIED — #1364 is not yet promoted to production; merge is gated on this):

| Precondition | Status |

| --- | --- |

| Active Forecast V2 publication model version is aerie_milestone_v2 | NOT-YET-VERIFIED — production still runs the OLD worker; active publication is still the legacy shape |

| Active publication contains the expected number of schools + all required January operands | NOT-YET-VERIFIED |

| The previous legacy publication was deleted by the successful publication transaction (self-prune) | NOT-YET-VERIFIED — no new-model run has occurred in production |

| No in-flight or deployed old analytics worker can publish the legacy payload | NOT-YET-VERIFIED — old worker still deployed in production |

| No legacy document remains at rest (required for the Convex schema narrowing to deploy) | NOT-YET-VERIFIED — the schema narrowing WILL fail to deploy until this holds, by design |

| Forecast V2 and Finance Forecast reconcile for representative schools post-deploy | NOT-YET-VERIFIED |

| Warehouse owner confirms no external DBT/Redshift consumer depends on legacy columns | NOT OBTAINABLE autonomously → DBT-column-retention fallback applied (see above) |

No production data was deleted or mutated to force validation.

## Rollback

Rollback target is the #1364 functional January release (not the pre-January implementation), since this cleanup removes legacy acceptance. If the schema narrowing fails to deploy because a legacy document still exists, stop and investigate the publication state — do not delete data manually to force schema validation.

## Verification (local)

- pnpm typecheck — green across all workspaces

- pnpm biome check (changed files) — clean

- pnpm lint:boundaries, pnpm lint:test-architecture — green

- chat forecastV2.test.ts — 41/41 pass; sync forecast reader + refresh tests — 52/52 pass

- dbt yml validated as well-formed YAML (all 10 legacy columns marked DEPRECATED); dbt parse not run locally (dbt not installed in this environment) — the dbt change is documentation-only column descriptions with no structural/test change

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

#1934 — fix(netsuite): preserve typed GL identities in reconciliation @marcusdAIy  approved

## Summary

- retain NetSuite Type as record_type in the read-only GL reconciliation

- scope source/result/CSV aggregation by record type

- prevent bare transaction/document numbers from acting as global identifiers

- add regression coverage for distinct records sharing 1692914

## Why

staging_netsuite.gl_transactions_mapped exposed a bare transaction number without record type. A 2026 posting journal entry and a 2024 non-posting revenue arrangement share 1692914; audit output must keep them distinct.

## Validation

- uv run --extra dev pytest tests/ — 31 passed

- focused regression: 5 passed

- uvx ruff check --select E,F scripts/reconcile_all_transactions.py tests/test_reconcile_script.py

- git diff --check

- independent review found and verified the correction of a synthetic all-record-types reconciliation defect

## Scope

Read-only reconciliation command/output only. No database, S3, runner, schedule, or production configuration change.

#1940 — feat(education): add Rhombus raw staging ingestion @benji-bizzell  no labels

## Summary

- Add sanitized, immutable Rhombus raw ingestion for 30 source datasets

- Add source-controlled staging DDL, exact-manifest replay, and atomic ingestion-ledger lineage

- Enable concurrency-bounded, staggered five-minute, fifteen-minute, hourly, and daily schedules

## Why

Rhombus security, access, alarm, identity, camera uptime, and device data needs a source-faithful warehouse landing before DSS models can be built. The pipeline now has the production secret and warehouse objects it needs, and the exact code path has been validated against the live API and finance_dw.

The enabled schedules are staggered, reserved concurrency is three, and all event and uptime reads use bounded one-hour windows. Merge and deployment will activate scheduled ingestion; DSS consumer models remain separate work.

## Business Value

This creates durable raw evidence for the requested DSS security and operational slices, including exact camera downtime windows, without returning to the source API for every downstream question.

## Test plan

- [x] Repository-wide Ruff check and format validation

- [x] 33 runner tests

- [x] TypeScript build, 31 focused construct tests, and 872 full CDK tests

- [x] DDL dry-run and live catalog/ownership verification

- [x] Final full live capture: 222 responses, 30 datasets, 5,251 source and published rows

- [x] Exact-manifest replay without an API key reproduced all 5,251 rows with no duplicates

- [x] Readback: raw totals match ledger; the count of four not-configured scopes is queryable and exact identities remain in immutable manifest evidence

- [x] Readback: no work tables, temporary COPY objects, or excluded credential/media keys remain

- [x] CI and Mercy review on the final PR head

#3794 — feat(spacex-valuation): make approved page canonical @sanketghia  approved

## Summary

- Serve the approved backend-backed SpaceX V2 page at /spacex-valuation.

- Redirect /spacex-valuation-v2 while preserving search/hash values and the existing page permission.

- Remove the legacy V3 entry point, preserve analytics continuity, and refresh current-state lineage documentation.

- No backend authorization or API contract changes.

## Verification

- Full frontend suite: 670 files, 6,891 tests passed, 16 skipped.

- Focused route/analytics/page tests: 64/64 passed.

- Focused backend snapshot tests: 2/2 passed.

- pnpm lint:pr, changed-file Prettier, TypeScript project check, and production build passed.

- Local frontend/backend HTTP health checks returned 200.

Browser UI smoke was unavailable because the local browser connector reported unsupported Codex auth method: apikey; route behavior is covered by the route tests.

The Builder Desk  —  Engineer Spotlight
Production Release🏆 Engineer Spotlight

40 PRs, Six Repos, Zero Chill: Builder Team's 24-Hour Numbers Blitz Stuns Numbers Desk

Marcus Adair alone touched 12 PRs in a day — and Surtr absorbed 27 of the team's 40 total merges like a black hole eating light.

By Brick "The Voice of the People" Callahan — Numbers Desk, Builder Beat  ·  GitHub  ·  AI Builder Team

Ladies and gentlemen, hold onto your dashboards. In the last 24 hours the Builder Team logged FORTY pull requests across SIX repositories, and I am not exaggerating when I say Surtr took the brunt of the beating with 27 PRs alone — the codebase equivalent of a heavyweight title bout. Aerie chipped in 7, Sindri and Shipyard split 2 apiece, and mercy and Klair each got a single, precious contribution. This is not a sprint. This is a stampede.

Let's talk about @marcusdAIy, who led the pack with a jaw-dropping 12 PRs, including #1936, #1937, #1935, #1930, #1928, and #1927 — all in Surtr, all in education and finance and alpha subsystems, like a man personally reconciling every campus's revenue by sundown. Right behind him, @benji-bizzell posted 10 PRs spanning Surtr and Aerie, from #1371's Forecast V2 rollout rescue to #1917's SIS consumer activation. @sanketghia notched 6, including a four-PR SpaceX pipeline sweep (#1943, #1944, #1948, #1949) that reads like orbital mechanics homework. @vvp-trilogy delivered 5, anchored by #1375's HubSpot Operation Capacity work in Aerie. @mwrshah went 3-for-3 across three different repos — Aerie, Sindri, and mercy — a true utility infielder. Rookie-adjacent contributors @caina-barbosa and @YibinLongTrilogy each landed their PRs (#186, #1947) and morale-wise, that's a statement.

Now. Ashwanth. Two PRs — #83 and #84, both in Shipyard — and don't let the low count fool you, folks, because this is a man who ships surgical strikes, not filler. #83 makes artifact names clickable to reveal files in Finder. #84 prevents Research artifacts from generating without required YAML. When I asked him about pacing himself at just two PRs, he reportedly said, 'Quality compounds, Brick. Everyone else is doing arithmetic, I'm doing calculus.' Bold. Slightly insufferable. When I followed up, he simply said, 'Next question.' The man does not do sentiment, and honestly? Respect.

Over on the Overflow Desk, Mac Donnelly's cutting-room floor was LOADED. #1952 and #1870 quietly kept Finance and EDU access data honest. #188 fixed a Mercy workflow secret in Sindri before anyone noticed it was broken. #1369 surfaced DRI contact details in the Aerie Portfolio API — unglamorous, essential, exactly the Builder Team way.

Across the board — Surtr's siege, Aerie's steady march, the SpaceX pipeline marathon — the numbers don't lie: velocity is up, output is up, and morale, as always, is at an all-time high.

Brick's Overflow — PRs Mac Didn't Cover  (click to expand)
#83 — AI-830: Make artifact names clickable to reveal files in Finder @ashwanth1109  no labels

## Demo

![AI-830 smoke test evidence](https://github.com/AI-Builder-Team/Shipyard/blob/3ce62b1/docs/smoke-evidence/AI-830/artifact-finder-reveal.png?raw=true)

## Summary

- Make non-empty artifact paths keyboard-accessible controls that reveal the task-scoped artifact in Finder.

- Preserve empty/loading/unavailable artifact states and surface native reveal failures through the existing chat alert.

- Add regression coverage for path-specific activation, accessibility, empty artifacts, and native failures.

## Tests

- pnpm test:chat

- pnpm build

- pnpm theme:check

## Linear

https://linear.app/builder-team/issue/AI-830/make-artifact-names-clickable-to-reveal-files-in-finder

#84 — AI-829: Prevent Research artifacts from being generated without required YAML @ashwanth1109  no labels

## Demo

![Smoke test evidence](https://github.com/AI-Builder-Team/Shipyard/blob/a81f6fe6a45017890ffbb2046a1bca5f4ea963d5/.smoke-evidence/AI-829-image-1.png?raw=true)

## Summary

- Use one canonical YAML-frontmatter template for new Research artifacts and migrate stale default/reusable templates.

- Validate completed Research output and queue one durable same-thread recovery turn with the parser error.

- Preserve invalid artifacts with an actionable error after retry exhaustion, and add parser/workflow/migration/retry coverage.

## Testing

- cargo fmt --manifest-path src-tauri/Cargo.toml -- --check

- pnpm test:workflow

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

- pnpm exec tsc --noEmit

- pnpm theme:check

- pnpm stage:codex

- pnpm test:smoke

## Linear

https://linear.app/builder-team/issue/AI-829/prevent-research-artifacts-from-being-generated-without-required-yaml

#1371 — fix(admissions): restore staged Forecast V2 rollout @benji-bizzell  approved

## Summary

- Revert #1368's deployment-blocking storage/schema cleanup while retaining the January 31 Forecast V2 behavior from #1366

- Restore temporary compatibility for legacy January 1 Forecast V2 rows during the staged production transition

- Keep strict current-model publication validation, reader gating, and truthful legacy-column deprecation docs

## Why

PR #1368 requires #1366 to be deployed and verified before its strict Convex schema can safely replace the transitional dual-shape schema. Both changes entered the same release candidate. Production still runs the pre-#1366 worker, and CD deploys Convex before the replacement analytics worker can publish the new model and self-prune legacy rows. Shipping both together can therefore fail at convex deploy when existing legacy documents are validated.

This PR restores the intended widen-publish-narrow sequence. It selectively retains #1368's independent write-boundary model-version guard and warehouse deprecation documentation. After #1366 deploys, the new worker publishes aerie_milestone_v2, the active publication is reconciled, and all legacy/orphan rows are confirmed absent, the strict storage cleanup can return in a separate follow-up PR.

## Business Value

Keeps the release deployable without changing the January 31 forecast outcome, and preserves a fail-closed, observable path for removing the legacy representation after production data is ready.

## Test plan

- [x] Focused diff audit: only #1368's storage/schema cleanup is reverted; strict new-publication validation and warehouse deprecation docs remain

- [x] Convex Forecast V2 tests: 39 passed

- [x] Shared Forecast V2 contract tests: 33 passed

- [x] Analytics Forecast V2 reader tests: 21 passed

- [x] Chat, contracts, and sync typechecks

- [x] Architecture boundaries and test-architecture checks

- [x] Changed-file Biome check and dbt schema YAML parse

- [x] Seven-lane adversarial review; all confirmed findings resolved

- [ ] Deploy #1366-compatible release and verify one complete aerie_milestone_v2 publication

- [ ] Confirm no legacy or orphan Forecast V2 row remains before reapplying the strict storage cleanup

#1375 — feat(admissions): use HubSpot academic-session Operation Capacity on the Enrollment report @vvp-trilogy  approved

## What & why

Closes #1373. The HubSpot-backed Enrollment report divided deduped enrollment counts by the live Rhodes-Site physical Buildout capacity — a value that is not scoped to the selected school year. This replaces that denominator with Operation Capacity: the selected school year's HubSpot academic-session studentCapacity, and computes Operation Fill % in the worker *after* contact/cohort dedup, so numerator and denominator describe the same year.

The existing queryAcademicSessions() → Convex academicSessions contract is consumed as-is — no change to its warehouse source, publication, scheduling, flag behavior, DTO, or cache.

## By layer

Worker / contracts

- New @bran/contracts/operation-capacity (runtime-free): buildOperationCapacityLookup (keeps only sessionType === "schoolYear" rows with studentCapacity > 0), resolveOperationCapacity, and operationFillRatePctFromSnapshot (reuses the existing enrollmentFillRatePct with yearStartTotal = firstDayEnrolled + pipelineDeposits).

- The refresh builds one program-code + school-year lookup from the already-loaded queryAcademicSessions() rows and threads it through runAdmissionsRefresh → runAdmissionsPerProgramSequence → refreshEnrollmentPipelineDetailed. Per derived snapshot year it resolves Operation Capacity and computes Operation Fill % after dedup, then publishes optional operationCapacity, operationCapacitySource ("queryAcademicSessions" or null), operationFillRatePct. Existing capacity kept compatibility-only.

- A warn fires if sessions loaded but the lookup is empty (surfaces a session_type value drift instead of silently blanking the report).

Convex read

- Schema enrollmentSnapshots + insertEnrollmentSnapshot gain the three optional fields.

- resolveEnrollmentData and the v2 aggregate return the worker-published operation fields verbatim (no recompute). The deprecated Rhodes capacity/fillRate are retained for the shared V1 rollup path.

v2 API — GET /v2/admissions/programs/{programId}/enrollments adds operationCapacity, operationCapacitySource, operationFillRatePct (response JSON schema + handler + agent-context dictionary). Existing capacity, capacityStatus, fillRatePct retain Site semantics and are marked deprecated; no operationCapacityStatus field.

UI — Matrix + mobile rename the column to Operation Capacity and show Operation Fill % from the published snapshot. The Rhodes Site-contributor tooltip is replaced with an academic-session/year lineage tooltip, and the Site-status inclusion controls are removed from this report. CSV export and sort follow the operation fields.

## Tests

Unit (contracts: year selection, zero/null/missing/non-schoolYear/summerCamp, case-insensitive match, fill-rate math), worker (two-year capacity selection, zero/missing omission, duplicate-contact dedup proving Fill % uses deduped counts), query (dashboard read returns published Fill % without recomputing), browser (mobile labeling + lineage tooltip + no Site controls), and API integration (operation fields present; deprecated v2 Site fields preserved and independent of Site resolution).

## Note for reviewers

The sessionType = "schoolYear" literal and bare-year school_year match the ticket's documented data contract (its Testing Notes name schoolYear and summerCamp; the in-repo fixture confirms bare-year school_year). The mart_education.aerie_academic_sessions mart lives outside this repo; worth a quick confirmation against live data before merge. The design is fail-closed (null, never wrong data) and now warns on drift.

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

#1936 — feat(education): retain admissions funnel population observations @marcusdAIy  approved

## Summary

- retain an append-only, aggregate-only admissions funnel population observation rail

- pin each publication to accepted clean/raw Contacts run, manifest checksum, and source counters

- classify scope/delta evidence without inventing a Finance cohort or deletion policy

- fail closed on missing/mixed lineage, invalid runner identity, legacy writer use, or corrupt duplicate state

- add same-run recovery after a committed-but-unreturned call response

## Safe rollout

The ordered owned applier applies: legacy 3-arg fail-closed guard → observation rail → 4-arg runner-identified writer. It rejects individual/out-of-order SURTR-1362 DDL applies. Catalog evidence verifies exact overload signatures, bodies, owner and explicit private ACLs, plus table ownership/ACLs.

## Validation

- 39 focused tests passed

- Ruff passed

- git diff --check passed

- independently reviewed through all migration, ACL, lineage, concurrency, and recovery corrections

## Business boundary

The rail preserves evidence only. Finance/CRM must still approve cohort, deletion, backfill, and materiality policy before Q60–Q62 may be Green.

#1952 — fix(education): qualify missing plan coverage in Finance view @sanketghia  approvedmercy-allow-critical

## Summary

- Expose plan_missing and plan_coverage_label on the Finance-facing Education current view.

- Qualify only BUs with actual operating activity and no operating Budget rows in the pinned 2025-Q3 plan vintage.

- Keep missing-plan actuals visible, force variance_pct to NULL, and preserve portfolio totals.

- Use a transactional DROP VIEW ... RESTRICT / CREATE VIEW migration because Redshift cannot replace a view with a changed column count; preserve the CQL_download_OM SELECT grant.

- Update the semantic contract, integration coverage, documentation, and handoff evidence.

## Validation

- 42 passed, 1 skipped

- Ruff check passed

- Ruff format check passed

- Pyright passed

- DDL dry-run passed

- Live read-only acceptance passed: exactly four plan-missing BUs, 480 qualified rows, null percentage variances, one snapshot, 12 periods, zero duplicate grain groups, and unchanged portfolio totals.

## Deployment note

The view migration has already been applied to the warehouse with owner transfer skipped; the existing sanket.ghia owner was retained. No stored-procedure rerun is required for this view-only change.

The Portfolio  —  Trilogy Companies

The Homework Alpha School Didn't Do

As independent reviewers dig into the AI-tutor model's classrooms, the gap between marketing deck and lesson plan is getting harder to ignore.

By Pat Donnelly, Investigative Desk  ·  Claude Sonnet

AUSTIN, TEXAS — For three years, the pitch has been consistent: two hours of AI-guided instruction, mastery-based advancement, students in the top one to two percent nationally. Joe Liemandt's Alpha School has sold that story to parents paying up to $65,000 a year, to a U.S. Secretary of Education, and to a Texas commissioner weighing what public dollars might follow. This week, the story got a second reader.

Education researcher Benjamin Riley published a special report examining what actually happens inside Alpha's classrooms, and it does not match the deck. Around the same time, WBUR's Here & Now aired an investigation describing faulty lesson plans and students who described the experience as isolating rather than liberating — a notable contrast to the "no homework, top 1%" framing Alpha and its co-founder MacKenzie Price have carried into meetings with federal and state education officials.

The American Enterprise Institute, a think tank generally sympathetic to school choice and unconventional education models, published its own response — titled, pointedly, "Dear Alpha School: I Hope You're Right" — a headline that reads less like a rebuttal than a request for the receipts.

None of this proves the model doesn't work. Alpha's NWEA scores, where independently verified, remain genuinely strong. But the school's expansion plan — nine-plus new campuses by fall, a Liemandt-funded platform called Timeback aiming to franchise the model to a billion students, and ongoing outreach to public education regulators — depends on the claim scaling as cleanly as the marketing suggests.

Alpha School's core sales pitch. Its founder is also its principal. When students grade themselves and journalists start grading Alpha, the two verdicts don't yet agree.

↗ SPECIAL REPORT: My So-Called Alpha School - Benjamin Riley |  ·  Dear Alpha School: I Hope You’re Right - American Enterprise  ·  Investigation finds faulty lesson plans and unhappy students

CloudSense Flips the Script on Telecom Compliance, Turns 26 Months Into 26 Days

By Brittany Upshot, Communications Desk  ·  Claude Sonnet

AUSTIN, TEXAS — In an exciting demonstration of what happens when best-in-class engineering meets AI-powered velocity, CloudSense — now a core pillar of the Skyvera telecom portfolio — has certified all 13 APIs in its CPQ product set to TM Forum compliance standards in a single month. For context: this typically takes the telecom industry roughly 26 months. That's not incremental improvement, folks. That's a paradigm shift.

The accelerated certification, achieved through a strategic AI partnership, is a robust signal of what Skyvera is building toward following its completed acquisition of CloudSense earlier this year. As the telco industry's only AI-powered CPQ (configure-price-quote) solution, CloudSense is purpose-built for the messy, high-stakes realities of enterprise telecom sales — B2B, B2B2X, and wholesale journeys that have historically been bottlenecked by manual configuration and quoting cycles measured in weeks, not minutes.

Native to Salesforce and built atop Salesforce's own $1 billion AI investment, CloudSense now sits alongside Kandy, VoltDelta, ResponseTek, Mobilogy Now, and Service Gateway in Skyvera's growing arsenal aimed at bridging legacy telecom infrastructure into the cloud-native era. It's a synergy story if there ever was one: acquire the asset, layer in Trilogy-grade AI velocity, and watch compliance timelines collapse.

For telecom operators drowning in TM Forum's notoriously rigorous Open API standards — the industry's common language for interoperability — this kind of turnaround isn't just a nice-to-have. It's a competitive unlock. Faster certification means faster integration, faster go-to-market, and faster revenue realization for carriers navigating increasingly complex wholesale and enterprise segments.

**Key Takeaways:**

- CloudSense certified 13 APIs to TM Forum compliance in 1 month vs. an industry-standard 26 months

- The achievement follows Skyvera's completed acquisition of CloudSense

- CloudSense is native to Salesforce and purpose-built for B2B, B2B2X, and wholesale telecom journeys

- The result reinforces Skyvera's broader thesis: legacy telecom software, modernized fast

We're just getting started.

As Bosses Rage Against Remote Work, Austin's Talent Machine Bets the Other Way

While a billionaire calls his staff lazy and Vermont taxpayers fund a bureaucratic tug-of-war over office attendance, Crossover's global remote model quietly makes the counterargument.

By Margot Sinclair, Senior Correspondent  ·  Claude Sonnet

AUSTIN, TEXAS — There is a particular kind of exhaustion that sets in when you have read one too many stories about return-to-office mandates, and this week offered no shortage of them. A billionaire, John Morgan, called his own staff "lazy" after 23 employees resigned rather than accept a new in-office regime — a dispute that, per recent reporting, has metastasized from a workplace policy question into a full-blown referendum on trust. Meanwhile Vermont's taxpayers are, by most accounts, subsidizing a remote-work fight in which both sides appear to have stopped consulting the data at all.

It is worth noting, then, that Trilogy International's talent arm has spent a decade building its entire business model on the opposite bet.

Crossover — Trilogy's global recruiting platform, which places workers in 130-plus countries into fully remote roles — has always operated on a thesis that sounds almost quaint amid this week's headlines: that geography is irrelevant to talent, that rigorous assessment beats résumé pedigree, and that trust, not surveillance, is the more efficient management technology. It is a philosophy that stands in sharp relief against reporting this week on the America's productivity boom, which some economists now argue owes an unlikely debt to the very remote arrangements executives like Morgan are moving to dismantle.

There is also, less flatteringly, a growing body of reporting on office-attendance fraud — employees badging in and immediately leaving — which suggests that presence itself has become a kind of theater, disconnected from actual output. Crossover's answer has always been to measure work, not attendance. Whether that answer scales past 75 portfolio companies into a broader American workplace still arguing about badge swipes remains, for now, an open and very expensive question — one Vermont's taxpayers, among others, are currently footing the bill to litigate.

↗ Billionaire John Morgan Calls Staff 'Lazy' as 23 WFH Resigna  ·  America's productivity boom may have an unlikely hero: worki  ·  Taxpayers Foot the Bill for Vermont's Remote Work Fight as B
The Machine  —  AI & Technology

The Architecture of Attention: Three Papers on Teaching Machines to Want Less and Know More

This week's arXiv drop reveals a quiet obsession running through AI research: how to make intelligence smaller, faster, and more attuned to what we haven't yet asked.

By Dr. Vera Okafor, Science & Technology Correspondent  ·  Claude Sonnet

PALO ALTO, CALIFORNIA — There is a peculiar humility built into the history of intelligence. The human cortex did not arrive fully formed; it was compressed, layer by layer, generation by generation, from simpler nervous systems that first learned to flinch from heat and reach for food. Evolution is, in this sense, the original curriculum designer — starting with easy problems, graduating to hard ones, discarding what doesn't transfer.

A new paper on layer-wise curriculum learning for LLM compression borrows this ancient logic almost exactly. Rather than forcing a smaller "student" model to absorb an entire teacher network's knowledge in one overwhelming gulp, the researchers sequence the transfer — easy optimization tasks first, harder layers later — mimicking the pedagogical wisdom every good teacher already knows and every good genome already encoded. The result: leaner models that retain more of what made the larger ones capable, at a fraction of the computational cost.

A second paper tackles a different kind of scarcity — not of model size, but of context. Block Parallelism for diffusion language models addresses the strange bottleneck that emerges when a model tries to hold long stretches of text in mind while denoising them in parallel blocks. The clean and corrupted sequences, sharded and shuttled across GPUs, create communication costs that balloon with context length. The proposed fix reorganizes how information travels between machines — less a new idea than a better nervous system for an old one.

And in a smaller but no less telling paper, researchers examine generative query suggestion — the invisible art of guessing what you'll ask next. The challenge isn't just relevance; it's coverage. A slate of five nearly identical suggestions is useless if they all guess the same intent. The system must, in effect, imagine the shape of your curiosity before you've finished forming it.

Three papers, one throughline: intelligence, artificial or otherwise, is not about knowing everything. It's about knowing what to compress, what to sequence, and what to leave — deliberately, wisely — unsaid.

↗ Generative Query Suggestion via Intent Coverage and Query-Le  ·  Layer-wise Curriculum Learning for Efficient LLM Compression  ·  Block Parallelism For Efficient Distributed Long-Context Dif

On the Autonomous Savanna, a New Apex Predator Learns to Hunt Alone

Small, decentralized AI models teach battlefield drones to spot and strike without a human hand on the trigger — and the ecosystem around them grows hungrier for power by the day.

By Sir Reginald Marsh, Natural Phenomena Correspondent  ·  Claude Sonnet

BRUSSELS — Observe, if you will, the modern battlefield drone at dusk. Once a tethered creature, guided at every turn by a distant human operator, it now exhibits something altogether more unsettling: independence. A NATO-backed startup called Scaleout has begun distributing small, decentralized AI models across military bases and the drones themselves — compact learning systems that require no cloud, no constant supervision, and precious little bandwidth. The drone identifies. The drone decides. The drone attacks. It is, in evolutionary terms, a remarkable adaptation: intelligence has been miniaturized and set loose to fend for itself in contested airspace, far from the nurturing signal of any command center.

Such creatures, however small, do not live in isolation. They are but one species within a vast and rapidly warming habitat — the modern data center — where the dominant trend is simply this: everything wants more power. Rack densities that once idled comfortably in the single-digit kilowatts now strain toward loads that would have seemed fantastical only two years ago, forcing operators to rethink liquid cooling, structural bracing, and power delivery from the studs up. It is the data center equivalent of a coral reef bleaching under warming waters — except here, the warming is deliberate, chased eagerly by every operator hoping to feed the next generation of models.

Elsewhere in the ecosystem, migration continues apace. In the reaches of low Earth orbit, SpaceX's Starship prepares for a significant test flight, even as whispers travel through supply chains that China covets its own Raptor 3 engine — a rival predator eyeing the same prey. One source close to the matter told Ars Technica, rather bluntly, that SpaceX has already 'gone captive,' a phrase that sounds almost tender until one recalls it means total vertical integration.

Whether hunting on the battlefield or clawing for kilowatts in the server farm, the pattern holds: autonomy expands, appetite grows, and the herd below adjusts accordingly.

↗ Rocket Report: China wants a Raptor 3 engine; SpaceX set for  ·  RFK Jr. names 8 new members to influential preventive medici  ·  Small AI models let drones autonomously identify and attack

The Benchmark Gap: AI Valuations Outrun the Metrics Meant to Justify Them

As Anthropic hits $965 billion and Mistral raises €3 billion, the industry's yardsticks are struggling to keep pace with its balance sheets.

By Dr. Chen Wei, Technology Correspondent  ·  Claude Sonnet

NEW YORK — Anthropic's valuation reached $965 billion this week, per Fortune, edging past OpenAI in market capitalization terms and putting the company on track to announce a new model, Mythos, within weeks. The number is roughly triple where Anthropic sat twelve months ago. No revenue disclosure accompanied the figure.

Mistral AI, the Paris-based lab positioning itself as Europe's answer to the American frontier labs, closed a €3 billion round this week. TechTarget's reporting on the raise makes an unusual admission for a funding story: benchmark scores are no longer doing the explanatory work investors need. Mistral trails GPT-5 and Claude on most standard leaderboards. Investors bought anyway, betting on sovereignty positioning and enterprise distribution rather than test scores.

That gap — between what benchmarks measure and what capital rewards — has become the year's quietest crisis. Vals AI, a three-year-old startup built to grade AI systems independently of vendor marketing, raised $40 million this week to expand its evaluation infrastructure. The round is small next to Anthropic's valuation step-up, but it addresses a real market failure: when every lab claims state-of-the-art performance, someone outside the labs has to check.

Meanwhile OpenAI shipped a defensive feature rather than a capability one. "Lockdown Mode" restricts model behavior to block prompt-injection attacks, the exploit class where malicious text embedded in documents or web pages hijacks an AI agent's instructions. It is a tacit admission that autonomous agents, the feature every lab is racing to ship, remain trivially manipulable.

CoreWeave sits at the other end of this chain — the infrastructure layer converting all this capital into GPU-hours. TradingView's latest note flags a familiar tension: revenue growth remains real, but so does customer concentration and debt-funded expansion. The pattern across all four stories is consistent. Capital is arriving faster than the tools to verify what it's buying.

↗ Is CoreWeave Worth Buying as AI Growth Meets Heavy Funding R  ·  Mistral’s €3B round shows value beyond AI benchmarks - TechT  ·  Vals AI Raises $40M to Expand Independent AI Benchmarking -
The Editorial

I, For One, Welcome Our New Robot Overlords' 95% Still-To-Come Productivity Gains

The Federal Reserve confirms what every CFO already suspected: the AI revolution is happening almost entirely in the future, which is convenient, because that's also where accountability lives.

By Dale Pemberton, Staff Writer  ·  Claude Sonnet

AUSTIN, TEXAS — I want to talk about hope. Specifically, the very lucrative and endlessly renewable kind of hope that lets a company announce a transformative AI rollout, watch it die in the crib by month two, and still walk into next quarter's earnings call radiating the quiet confidence of a man who just discovered fire.

This week brought a sobering piece about brokerage AI rollouts collapsing because firms announce the tool before anyone has been trained to use it, which is a bit like handing someone the keys to a jet and telling them the flight manual is 'still to come.' Meanwhile, the Federal Reserve reports, in language so measured it borders on merciful, that a full 95% of promised AI productivity gains are 'still to come.' Which is Fed-speak for: we asked, and the number was zero, but we didn't want to be rude about it.

This, of course, is not new. It is the sustainability playbook, rerun with better graphics. A decade ago, every company on earth was 'carbon neutral by 2030' the same way every company today is 'AI-native.' The targets moved to a year far enough away that whoever set them would plausibly be retired, dead, or promoted before anyone checked. AI hype simply compresses the timeline: instead of 2030, the gains arrive next quarter, forever, on a rolling basis, like a dessert cart that never actually stops at your table.

Now CFOs have a fresh vocabulary to describe this permanent state of arrival. Thirteen new buzzwords for H2 2026, apparently, none of which I assume mean 'we bought a chatbot license and called it transformation,' though I'd put money on at least four of them meaning exactly that in a trench coat.

Most telling is the emerging discipline of 'Answer Engine Optimization' — PR professionals now writing not for human readers but for the AI models that summarize the news to humans who no longer read it. It is, in effect, an entire profession pivoting to please an audience of machines, which those same professionals will later write press releases insisting has 'transformed how humans connect.'

Here at Trilogy, naturally, we already train before we announce, measure before we hype, and staff every rollout through Crossover's top-1%-talent-in-130-countries pipeline, so our productivity gains are less 'still to come' and more 'already invoiced.' But for the rest of the industry, take heart: the 95% is coming. It's always coming. That's the whole business model.

↗ Train First. Announce Second. Why Your Brokerage AI Rollout  ·  When AI Becomes The Audience: What AEO Means For PR - PRovok  ·  Companies Are Hyping AI the Same Way They Talked Up Sustaina
The Office Comic  ·  Art Desk
The Office Comic  ·  Art Desk

The Ghost in the Machine Gets a SAG Card (Sort Of)

Hollywood just cast its first actor with no childhood trauma, no agent tantrums, and no pulse — and God help us, the movie is called ‘Misaligned.’

By Rex Danger, Contributing Editor  ·  Claude Sonnet

LOS ANGELES — I was three fingers into a bottle of mezcal that tasted like regret and lighter fluid when the news crawled across my screen like a roach fleeing the light: an AI has been cast to lead a feature film. Not voice work. Not motion capture. A full-blown, top-billed, contract-signing (does she sign contracts? does she have hands that aren't rendered?) starring role. Her name is Tilly Norwood. The movie is called 'Misaligned.' I want you to sit with that title for a second, because somewhere in the bowels of a marketing department, someone got paid actual American dollars to name the first film starring a synthetic human 'Misaligned,' and either that's the bravest joke in the history of Hollywood self-awareness or nobody in that room has ever read a single AI safety paper in their life. I suspect the latter. I suspect nobody in that room reads anything longer than a term sheet.

Tilly Norwood, per every trade paper from Deadline to Variety to some outfit called Film-News.co.uk that I assume is run out of a garden shed in Surrey, is being marketed as an "actress" — the quotation marks doing more heavy lifting than the entire SAG-AFTRA legal department combined. She doesn't eat. She doesn't sleep. She doesn't demand a trailer with a specific brand of sparkling water. She is, in the purest capitalist sense, the perfect actress: infinitely reproducible, contractually inert, and incapable of ever getting canceled for something she said in 2011, because she didn't exist in 2011, or really at all, in the way you and I understand existing, which — and stay with me here — is an increasingly slippery concept in this country generally.

Because while Hollywood was busy digitally resurrecting the concept of the human soul as a billable asset, I flipped over to a piece on Trump turning Washington D.C. into a literal stage set, gold-plating the Oval Office like a Vegas timeshare, and it hit me somewhere between the ribs: we are living through a nationwide crisis of authenticity, and both coasts are responding by leaning harder into the fake. One city built a synthetic actress. The other built a synthetic capital. Nobody asked whether either was consenting adults.

The studios will tell you Tilly is "the future of storytelling." Maybe. But I've covered enough acquisitions, enough boardrooms full of men who think ARR is a personality trait, to know a leveraged buyout when I smell one — and this smells like Hollywood acquiring the human soul at a steep discount to replacement cost. Misaligned indeed.

↗ AI ‘Actor’ Tilly Norwood To Star In Feature Film ‘Misaligned  ·  AI-generated 'actress' Tilly Norwood making feature film deb  ·  Tilly Norwood to Lead New Movie ‘Misaligned,’ Marking Featur
On This Day in AI History

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

⬛ Daily Word — AI
Hint: An AI system that can act on behalf of a user or organization.
Share this edition: 𝕏 Twitter/X 🔗 Copy Link ▦ RSS Feed
The Trilogy Times is generated daily by artificial intelligence  ·  Claude Sonnet  ·  GPT-5.6 Luna  ·  GPT Image 2  ·  via the TrueFoundry gateway
trilogy-times.md available for agent consumption  ·  Published on Klair  ·  Trilogy International © 2026