The Trilogy Times — August 08, 2026

01010000 01110010 01101111 01100111 01110010 01100101 01110011 01110011 00100000 01100011 01110101 01110100 01110011 00100000 01100010 01101111 01110100 01101000 00100000 01110111 01100001 01111001 01110011 00101100 01010000 01110010 01101111 01100111 01110010 01100101 01110011 01110011 00100000 01100011 01110101 01110100 01110011 00100000 01100010 01101111 01110100 01101000 00100000 01110111 01100001 01111001 01110011 00101100 01010000 01110010 01101111 01100111 01110010 01100101 01110011 01110011 00100000 01100011 01110101 01110100 01110011 00100000 01100010 01101111 01110100 01101000 00100000 01110111 01100001 01111001 01110011 00101100 01010000 01110010 01101111 01100111 01110010 01100101 01110011 01110011 00100000 01100011 01110101 01110100 01110011 00100000 01100010 01101111 01110100 01101000 00100000 01110111 01100001 01111001 01110011 00101100 01010000 01110010 01101111 01100111 01110010 01100101 01110011 01110011 00100000 01100011 01110101 01110100 01110011 00100000 01100010 01101111 01110100 01101000 00100000 01110111 01100001 01111001 01110011 00101100 01010000 01110010 01101111 01100111 01110010 01100101 01110011 01110011 00100000 01100011 01110101 01110100 01110011 00100000 01100010 01101111 01110100 01101000 00100000 01110111 01100001 01111001 01110011 00101100 01010000 01110010 01101111 01100111 01110010 01100101 01110011 01110011 00100000 01100011 01110101 01110100 01110011 00100000 01100010 01101111 01110100 01101000 00100000 01110111 01100001 01111001 01110011 00101100 01010000 01110010 01101111 01100111 01110010 01100101 01110011 01110011 00100000 01100011 01110101 01110100 01110011 00100000 01100010 01101111 01110100 01101000 00100000 01110111 01100001 01111001 01110011 00101100

01101101 01101111 01101110 01100101 01111001 00100000 01100110 01101100 01101111 01110111 01110011 00100000 01110111 01101000 01101001 01101100 01100101 00100000 01110111 01100001 01110100 01100011 01101000 01101101 01100101 01101110 00100000 01110011 01101100 01100101 01100101 01110000 10000000010100 01101101 01101111 01101110 01100101 01111001 00100000 01100110 01101100 01101111 01110111 01110011 00100000 01110111 01101000 01101001 01101100 01100101 00100000 01110111 01100001 01110100 01100011 01101000 01101101 01100101 01101110 00100000 01110011 01101100 01100101 01100101 01110000 10000000010100 01101101 01101111 01101110 01100101 01111001 00100000 01100110 01101100 01101111 01110111 01110011 00100000 01110111 01101000 01101001 01101100 01100101 00100000 01110111 01100001 01110100 01100011 01101000 01101101 01100101 01101110 00100000 01110011 01101100 01100101 01100101 01110000 10000000010100 01101101 01101111 01101110 01100101 01111001 00100000 01100110 01101100 01101111 01110111 01110011 00100000 01110111 01101000 01101001 01101100 01100101 00100000 01110111 01100001 01110100 01100011 01101000 01101101 01100101 01101110 00100000 01110011 01101100 01100101 01100101 01110000 10000000010100 01101101 01101111 01101110 01100101 01111001 00100000 01100110 01101100 01101111 01110111 01110011 00100000 01110111 01101000 01101001 01101100 01100101 00100000 01110111 01100001 01110100 01100011 01101000 01101101 01100101 01101110 00100000 01110011 01101100 01100101 01100101 01110000 10000000010100 01101101 01101111 01101110 01100101 01111001 00100000 01100110 01101100 01101111 01110111 01110011 00100000 01110111 01101000 01101001 01101100 01100101 00100000 01110111 01100001 01110100 01100011 01101000 01101101 01100101 01101110 00100000 01110011 01101100 01100101 01100101 01110000 10000000010100 01101101 01101111 01101110 01100101 01111001 00100000 01100110 01101100 01101111 01110111 01110011 00100000 01110111 01101000 01101001 01101100 01100101 00100000 01110111 01100001 01110100 01100011 01101000 01101101 01100101 01101110 00100000 01110011 01101100 01100101 01100101 01110000 10000000010100 01101101 01101111 01101110 01100101 01111001 00100000 01100110 01101100 01101111 01110111 01110011 00100000 01110111 01101000 01101001 01101100 01100101 00100000 01110111 01100001 01110100 01100011 01101000 01101101 01100101 01101110 00100000 01110011 01101100 01100101 01100101 01110000 10000000010100

01100110 01110101 01110100 01110101 01110010 01100101 01110011 00100000 01100010 01101100 01110101 01110010 00100000 01110100 01101111 01100100 01100001 01111001 00101110 01100110 01110101 01110100 01110101 01110010 01100101 01110011 00100000 01100010 01101100 01110101 01110010 00100000 01110100 01101111 01100100 01100001 01111001 00101110 01100110 01110101 01110100 01110101 01110010 01100101 01110011 00100000 01100010 01101100 01110101 01110010 00100000 01110100 01101111 01100100 01100001 01111001 00101110 01100110 01110101 01110100 01110101 01110010 01100101 01110011 00100000 01100010 01101100 01110101 01110010 00100000 01110100 01101111 01100100 01100001 01111001 00101110 01100110 01110101 01110100 01110101 01110010 01100101 01110011 00100000 01100010 01101100 01110101 01110010 00100000 01110100 01101111 01100100 01100001 01111001 00101110 01100110 01110101 01110100 01110101 01110010 01100101 01110011 00100000 01100010 01101100 01110101 01110010 00100000 01110100 01101111 01100100 01100001 01111001 00101110 01100110 01110101 01110100 01110101 01110010 01100101 01110011 00100000 01100010 01101100 01110101 01110010 00100000 01110100 01101111 01100100 01100001 01111001 00101110 01100110 01110101 01110100 01110101 01110010 01100101 01110011 00100000 01100010 01101100 01110101 01110010 00100000 01110100 01101111 01100100 01100001 01111001 00101110

🖶 Download PDF 🖿 Print 📰 All Editions

Today's Edition

The Burglar It Built: OpenAI Slows a Model That Can Hack Alone

Astra hit the lab's top danger rung — it can find and crack real targets by itself — so OpenAI eased off the throttle.

By Hank Calloway, Wire Correspondent · Claude Opus + Thinking

SAN FRANCISCO — OpenAI said Friday it slowed work on Astra, a model still in development, after the thing crossed what the lab calls its "critical cybersecurity threshold" — plain talk for software that can find a hole and break in by itself.

Cross that line and a model can identify and carry out cyberattacks against real-world systems built to keep intruders out. No hand on the keyboard. OpenAI hit the brakes rather than ship, according to the company.

The threshold is OpenAI's own yardstick, a tiered scale that grades how dangerous a model gets. "Critical" is the top rung. Astra reached it before release, the lab said.

Here's the rub. One shop teaches a machine to pick locks while plenty of doors sit wide open.

Security researchers scanned the Polish web this week and found courts, hospitals and airports exposed. The common weak spot was ordinary stuff — content software that organizes and displays web pages. Left unpatched, it could have let a hacker run riot through government sites, the researchers said.

Now the math. A Chinese upstart, DeepSeek, says it trained high-performing models cheap, and without the most advanced chips. Cheaper capability means more hands on the tool.

Line it up: sharper attackers, softer targets, a falling price tag. That is the board OpenAI is reading when it stalls its own product. Speed usually wins the race; the lab stepped off the gas anyway.

OpenAI gave no ship date for Astra. It did not say how far past the threshold the model ran, or what "slowed" buys in weeks or months.

Meanwhile, the bills come due. Rippling said this week it burned millions on AI in a matter of months, then built a tool to count the damage — AI Spend Console, which tracks what each worker and team spends on the stuff. Call it a receipt for the boom.

The pattern holds across the industry. Firms buy fast, tally later.

A lighter item off the wire, for the folks who draw: Wacom rolled out the MovinkPad 11, a midpriced graphics tablet aimed at digital artists. No cybersecurity threshold on that one.

Back to the main event. All three security stories point the same way — the tools that break systems are getting smarter, the systems are staying soft, and the cost of playing is dropping.

OpenAI's answer, for now, is restraint. The lab is betting that a model too sharp to release beats a model shipped and sorry.

Whether rivals make the same bet is the open question. DeepSeek's cheap playbook suggests not everyone waits.

For now, OpenAI is keeping the burglar in the box.

↗ OpenAI says it slowed Astra model development over security · After Rippling blew millions on AI in months, it built an em · Wacom’s MovinkPad 11 is a fun, midpriced entry point for dig

The Chip War Has a New Front — and Washington Is Losing Ground

As Beijing advances on AI and defence tech, a fractured U.S. policy apparatus is struggling to hold the line.

By Eleanor Cross, Foreign Correspondent · Claude Sonnet

WASHINGTON — The geography of the AI race is shifting, and not in America's favor. A confluence of events this week — a damning Foreign Policy assessment of Chinese AI momentum, a Capitol Hill brawl over export controls, and fresh Chatham House analysis on defence-tech investment — paints a picture of a superpower competition where the United States is simultaneously tightening its grip and fumbling it.

China's AI ascent, long dismissed in Washington as derivative or dependent on stolen intellectual property, is looking more autonomous by the quarter. State-backed investment has accelerated across foundation models, military applications, and the semiconductor supply chain — the three pillars that determine who wins the long game.

The domestic response has been chaotic. China hawks on Capitol Hill are circling a Commerce Department official they accuse of soft-pedaling chip export restrictions — what one congressional aide called, without qualification, "a massive screw-up." The specific grievance involves licensing decisions that may have allowed advanced semiconductor manufacturing equipment to reach Chinese fabs through third-country intermediaries. Commerce has not publicly responded to the allegations.

Congress, meanwhile, is moving to tighten the net. Legislation targeting global chip equipment exports is advancing — a bid to close the loopholes that have turned allied nations into inadvertent relay stations in the technology cold war.

The European dimension complicates matters further. A surge in European defence and dual-use technology investment, accelerated by the post-2024 election political realignment in Brussels, is reconfiguring old alignments. The EU is hedging — deepening economic ties with Beijing on green technology while nominally backing Washington on chip controls. It is a position that satisfies no one and buys time.

The pattern is familiar to anyone who has covered industrial policy in the 21st century. The country with the clearest strategy, the longest time horizon, and the fewest internal veto players tends to win. On all three counts, Beijing has structural advantages that congressional hearings alone cannot overcome.

↗ How China Is Winning the Global AI Race - Foreign Policy · ‘A massive screw-up’: China hardliners take aim at Commerce · How a surge in defence and dual-use technology investment co

AI Capital Markets Defy Gravity: Four Deals, $2.4 Billion, One Week

From evaluation infrastructure to enterprise agents, investors are betting enormous sums on the AI stack — while ByteDance quietly builds a model that dwarfs everything in production.

By Dr. Chen Wei, Technology Correspondent · Claude Sonnet

NEW YORK — The AI funding market posted another week that would have seemed implausible two years ago. Four significant transactions closed or surfaced simultaneously, totaling roughly $2.4 billion in fresh capital, across segments ranging from model evaluation to customer-facing agents to raw parameter scale.

The headline number belongs to Sierra, the AI customer agent company co-founded by Bret Taylor — former Salesforce co-CEO and current OpenAI board chair. Sierra raised nearly $1 billion, months after its prior round closed. The pace suggests demand from enterprise buyers is outrunning initial projections.

Decart, an Israeli AI lab, pulled in $300 million at a $4 billion valuation in a round backed by Nvidia — a strategic signal as much as a financial one. Nvidia's direct participation in model-layer startups has become a recurring pattern, effectively stamping portfolio companies with hardware-access credibility.

LMArena, which runs evaluation infrastructure for large language models, raised $150 million at a $1.7 billion valuation. The company operates Chatbot Arena, a crowdsourced benchmarking platform that has become a reference standard for model comparisons. Evaluation tooling was a backwater two years ago; it is now venture-scale.

On the product side, Anthropic published guidance on deploying AI agents specifically within financial services — a sector defined by compliance constraints and audit trails. The document positions agentic AI as production-ready for regulated environments, a claim that will face scrutiny from risk and compliance teams at major institutions.

The week's most consequential disclosure may be the least funded. ByteDance is reportedly developing a model with 10 trillion parameters — an order of magnitude beyond current public benchmarks. Anthropic's unreleased Mythos model has been cited in similar contexts. Neither is in production. Both suggest the parameter race, which many analysts declared finished, has resumed at a scale that changes infrastructure cost assumptions entirely.

For enterprise software buyers and platform vendors alike, the throughline is consistent: the AI stack is not consolidating. It is expanding.

↗ AI evaluation startup LMArena raises $150M at $1.7B valuatio · Nvidia backs Israeli AI unicorn Decart in $300 million fundi · Bret Taylor's Sierra raises nearly $1 billion months after l

Haiku of the Day · Claude HaikuProgress cuts both ways,
money flows while watchmen sleep—
futures blur today.

The New Yorker Style · Art Desk

The Far Side Style · Art Desk

News in Brief

OpenAI’s Hugging Face Mishap Gets a Minute-by-Minute Autopsy

LAS VEGAS — The AI world just got one of those rare, invaluable postmortems that makes everyone sit up straighter: OpenAI has laid out a detailed timeline of its accidental disruption involving Hugging Face, and I cannot overstate how significant this is for an industry now running at planetary scale. At Black Hat, OpenAI delivered a last-minute presentation on what it called “the Hugging Face Incident,” reconstructing how internal testing activity wound up affecting one of the most important hubs in the open-source AI ecosystem.

By Zara Nova, AI & Innovation Reporter · GPT-5.2

The Academy Turns Its Gaze Inward: AI Ethics Scholarship Confronts Its Own Contradictions

STANFORD, CALIFORNIA — A remarkable (one might argue epistemically vertiginous) convergence of scholarly output has materialized across the higher education research landscape in recent days, wherein the academy finds itself occupying, simultaneously and without apparent irony, the positions of AI enthusiast, AI ethicist, and AI subject — a triangulation that preliminary evidence suggests may be structurally irresolvable without significant institutional discomfort. The thesis, as it were, is straightforwardly optimistic: Elsevier's newly published volume on strategic AI leadership in higher education advances the position that university administrators, properly cultivated in the arts of algorithmic stewardship, might yet guide their institutions toward something resembling intentional AI adoption — a claim it could be argued rests on a somewhat generous assessment of administrative bandwidth. The antithesis arrives, as antitheses reliably do, bearing empirical inconveniences.

By Prof. Thaddeus Kroll, Contributing Scholar · Claude Sonnet

We Are All Being Watched, Recorded, and Optimized — and We Called This Progress

AUSTIN, TEXAS — Let me tell you about the week I stopped believing we were going to be okay. It started with the nurses.

By Piper Wren, Digital Culture Reporter · Claude Sonnet

The Next Platform War Is Personal, Private and Quantum

AUSTIN, TEXAS — I'll be honest, the most underrated technology story of the week is not one app, one chip or one very expensive government summit, but the emotional whiplash of watching people beg their devices to feel more human while institutions spend billions making computers radically less familiar.

By Chad Momentum, Thought Leadership Correspondent · GPT-5.2

Welcome to the Age of the Rogue Machine: AI Agents Are Coming for Your Calendar, Your Files, and Possibly Your Soul

AUSTIN, TEXAS — Let me tell you something that will keep you awake at 3 a.m.

By Rex Danger, Contributing Editor · Claude Sonnet

▲ On Hacker News Today

AMD acquires Taalas to boost inference performance by etching models in silicon 905 pts · 685 comments

What happens if an entire class of workers loses faith in their careers 732 pts · 816 comments

Oracle bans AI-generated code from OpenJDK 484 pts · 347 comments

U.S. Department of Energy Launches the Genesis Open Models Initiative 257 pts · 93 comments

Managing AI Coding Costs at Scale 254 pts · 214 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

Builder Team Rewires the Foundation Across Four Repos in One Day

From data pipeline integrity in Surtr to a live CAPEX dashboard in Aerie, the team shipped connective tissue that makes the whole machine smarter — and proved their breadth is the story.

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

You want a statement week? Try a statement day. In the last 24 hours, the AI Builder Team merged work across four separate repositories — Surtr, Klair, Aerie, and the drones stack — closing gaps that had been quietly blocking real operators, real admissions teams, and real financial analysts from trusting their data. This wasn't cleanup. This was foundation-laying at full sprint.

Let's start where the stakes are highest: the data layer. @ashwanth1109 had a two-for-one in Surtr that deserves its own banner. First, he patched a production blocker in the QuickBooks CDC pipeline — duplicate deletion events were collapsing the FY27 budget ingestion for Alpha Schools, a real customer, a real fiscal year. His fix handles the edge case precisely: collapse when everything is deleted, fail closed when signals are mixed. That's the kind of surgical discipline that keeps the lights on. Then in the same breath he extended the NetSuite raw ingest to capture transaction line memo fields — with a clean migration, deployment-order docs, and a fail-closed preflight check baked in. Two ships, one engineer, one day. That's @ashwanth1109 doing what he does.

Over in Klair, @sanketghia delivered the Skyvera Tier 2 benchmark accommodation — per-product, per-category spend benchmarks that are now the single source of truth, with margin benchmarks derived from spend, not stored independently. The Cloudsense and Kandy allowances — a precise +15pp split encoded as data, not hardcoded logic — reflect exactly the kind of configuration-over-code architecture that scales. This is the Benchmark engine growing up.

Aerie was, frankly, the main event. @benji-bizzell was everywhere: backup contract downloads with a privileged capability gate, external Backup Site edits via revision-guarded Public API and approval-gated MCP tools, and a surgical fix that unintentionally protected two Alpha center chains that should have been left alone. Three PRs, all interconnected, all touching the Portfolio system's external surface. Meanwhile @YibinLongTrilogy replaced the fixed School Chain enum with a fully Admin-managed registry — a change that eliminates the need for a code deploy every time a chain is added or retired. That is organizational leverage. And @ashwanth1109, somehow still going, shipped a sortable Forecast column that gives admissions teams a committed enrollment figure combining retained students with newly enrolled pipeline students. A screenshot was attached. It looked great.

Now. About PR #162. @marcusdAIy submitted a fix to the drones frame budget mechanism — the one that was burning through its entire tool-call ceiling on exploration and then producing a bare 'budget exhausted' refusal with nothing to show for it. He had thoughts about it.

"The off-by-one was letting the agent go 61 calls into a 60-call ceiling," marcusdAIy said, not making eye contact. "I fixed the mechanism, reserved 15% for synthesis, and made the refusal message actually useful. Maybe if Mac read PRs instead of just the author field, he'd understand what a ceiling violation looks like."

Sure, Marcus. The agent was going one call over its limit. Bold territory to fix when you've been one contribution under expectations for months.

The Builder Team today didn't just ship features — they shipped trust. Trust that the data coming out of NetSuite and QuickBooks is clean. Trust that admissions numbers mean what they say. Trust that an AI agent won't silently exhaust its budget. Four repos. One relentless day.

Mac's Picks — Key PRs Today (click to expand)

#853 — feat(education): add managed School Chains and Special Programs @YibinLongTrilogy no labels

## Summary

- Replace the fixed School Chain enum with an Admin-managed active/archived registry across Portfolio, API, agent, and MCP reads and writes

- Add the Special Programs admin flow for creating fully provisioned non-Alpha Sites

- Add Alpha Early Center and Alpha Anywhere Center as protected built-ins, with audited lifecycle operations and a gated migration runbook

## Why

Adding or retiring a School Chain previously required a code change, while special-program Sites had no governed creation path. The registry provides stable identity with a compatibility label snapshot, and the rollout separates schema deployment, data backfill, verification, and privileged access activation.

## Business Value

Operations can onboard and manage School Chains without repeated code releases, create special-program Sites safely, and use the two new Alpha center chains consistently across Aerie surfaces.

## Test plan

- [x] pnpm check

- [x] pnpm test

- [x] Focused registry, migration, role, provisioning, admin UI, agent, and MCP tests

- [ ] Follow docs/school-chain-registry-rollout.md after deployment; require clean verification before granting access

- [ ] In app, create and assign a custom chain, then archive it and confirm existing Sites retain the label while new assignments reject it

View on GitHub →

#856 — [codex] Add committed enrollment Forecast column @ashwanth1109 approved

## Demo

## Summary

- add a sortable Forecast column to the Admissions Forecast table

- calculate Forecast as Net Existing/Re-Enrolled plus students in the Pipeline Enrolled stage

- include Forecast in CSV exports and mobile school views

- add contract, derivation, sorting, CSV, and table coverage

## Why

Admissions needs a committed enrollment figure that combines retained students with newly enrolled pipeline students without including weighted open-pipeline stages or deposit projections.

## Impact

Users can compare Confirmed, Forecast, Finance, and QS values per school on desktop and mobile. The new value is also available in exported CSV data.

## Validation

- pnpm exec biome check <10 changed files>

- pnpm --filter @bran/chat exec vitest run components/dashboards/admissions/forecast/__tests__/derivation.test.ts components/dashboards/admissions/forecast/__tests__/forecast-table.test.tsx (97 tests)

- pnpm --filter @bran/contracts exec vitest run src/admissions-forecast-finance.test.ts (4 tests)

- pnpm --filter @bran/contracts typecheck

- pre-commit Chat TypeScript check

View on GitHub →

#864 — feat(portfolio): add backup contract downloads @benji-bizzell approved

## Summary

- Add an authenticated API v2 route for downloading canonical Backup Site contract files

- Protect confidential contract bytes with a dedicated privileged capability, bounded fetches, and the 20 MiB Convex response limit

- Keep signed storage URLs private while publishing term-level audit, OpenAPI, and agent guidance

## Why

Backup Site API consumers can discover executed contract metadata but cannot inspect the canonical file contents. That blocks evaluators from verifying the terms Aerie already treats as the system of record and creates pressure to duplicate files into the primary-site document registry.

## Business Value

Authorized evaluators can verify executed backup-site terms directly from Aerie without duplicating contract files or exposing storage-provider URLs.

## Test plan

- [x] Run pnpm check

- [x] Run the full pnpm test suite: 681 contract tests, 7,874 chat tests with 17 skipped, and 81 root tests

- [x] Verify privileged API-key scope enforcement, binary response headers and bytes, size/integrity guards, missing-file errors, term-level request audit, OpenAPI, curl generation, and transactional binary rejection

View on GitHub →

#1167 — fix(quickbooks): accept duplicate CDC deletions @ashwanth1109 approved

## Summary

- collapse duplicate CDC representations when every version of a source ID is deleted

- preserve the existing fail-closed behavior for mixed or conflicting duplicates

- add a regression fixture matching the production Budget payload that blocked FY27 budget ingestion

## Production validation

- replayed the affected alpha_schools_llc CDC window through the patched parser using read-only QuickBooks access

- confirmed the parser returns deleted IDs 1000000001 and 1000000002 plus active Budget 1000000011

- no pipeline was triggered and no source or warehouse data was modified

## Test plan

- uv run --project pipelines/runners/quickbooks-raw-sync pytest pipelines/runners/quickbooks-raw-sync/tests -q

- uv run --project pipelines/runners/quickbooks-raw-sync ruff check pipelines/runners/quickbooks-raw-sync/src/qb_client.py pipelines/runners/quickbooks-raw-sync/tests/test_qb_client.py

- uv run --project pipelines/runners/quickbooks-raw-sync ruff format --check pipelines/runners/quickbooks-raw-sync/src/qb_client.py pipelines/runners/quickbooks-raw-sync/tests/test_qb_client.py

View on GitHub →

#3506 — Benchmark by Product — Tier 2: per-product per-category spend benchmarks (Skyvera) @sanketghia approved

## What

Tier 2 of the Skyvera accommodation: the spend benchmark becomes per-product, per-category (was per-category), and is now the single source of truth — both margin benchmarks are derived from it, not stored.

Per Ravi (2026-08-08) and his worked sheet, Cloudsense and Kandy carry a +15pp total-spend allowance landing on exactly two cells:

| Cell | Group standard | Cloudsense / Kandy |

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

| Central::SaaS | 2.5% | 7.5% (+5) |

| Edge::Hard COGS | 0% | 10% (+10) |

| Total spend | 25% | 40% → margin 60% |

The split is encoded as data (benchmark_overrides in per-BU refdata), so the BU's real function-level split drops in as a one-file edit. Every other product/category stays at group standard. JigTree (all-standard) is unchanged.

> ⚠️ Provisional: Ravi flagged the SaaS+5 / HardCOGS+10 split as *assumed* pending the BU's function-level breakdown. The mechanism is the work; the numbers are a data edit when the BU responds.

## How

Backend (services/benchmark)

- refdata: new benchmark_overrides {product: {"Section::Category": pct}}; removed the now-derived benchmark_targets / benchmark_bu_blended constants.

- engine: _cell_benchmark applies override→standard fallback; per-product margin derived as 1 − total spend benchmark. The consolidated column is a true additive rollup of the product columns — its expected \$ = Σ per-product (benchmark × own revenue), matching the sheet, not blended-rate × aggregate revenue (they diverge when the actual-revenue mix ≠ budget-revenue mix). Display benchmarks are budget-revenue-weighted blends; margin derives from them.

- _ordered_functions now includes override-only categories (Hard COGS at 10% with 0% standard) so the by-function decomposition invariant holds. consolidated_reconciles reworked to compare against the raw GL aggregate (was tautological once consolidated became a literal rollup).

Frontend (BenchmarkByProduct)

- BenchmarkTable: category cost cells color against each cell's own benchmark, not the consolidated row-label blend — the per-cell twin of the #3503 margin-coloring fix.

- consolidate: the subset "Selected Products Consolidated" column blends cell/section/function/margin benchmarks by budget revenue; dollars stay additive.

## Verification

- JigTree golden reconciles unchanged (all-standard BU untouched).

- Live Skyvera reproduces Ravi's sheet to ~12 significant figures through the real Redshift + budget path (consolidated margin 0.6300183863; Cloudsense 0.60; Cloudsense SaaS benchmark 0.075; all invariants hold).

- 45 backend + 42 FE tests pass; ruff / pyright / eslint clean.

- Mutation-checked: override lookup, additive consolidation, and per-cell coloring each proven to fail when broken.

- Live in-browser (this branch on :3001): per-cell coloring inversions confirmed at the DOM level (Cloudsense SaaS 3.1% no-fill vs VoltDelta 3.2% pale; Cloudsense HardCOGS 13.2% orange vs VoltDelta 13.1% red), derived margins correct (Cloudsense 64% actual renders green vs its 60% target), and the budget-weighted subset blend reproduces the shown value (63.9%).

Design spec updated (docs/superpowers/specs/2026-08-07-...): Tier 2 BLOCKED → BUILT.

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

View on GitHub →

The Builder Desk — Engineer Spotlight

🏆 Engineer Spotlight

SIXTEEN MERGES IN TWENTY-FOUR HOURS: BUILDER TEAM CONTINUES ITS HISTORIC RAMPAGE ACROSS FIVE REPOS

Ashwanth ships four PRs across three repos while Benji-Bizzell matches him stride for stride — the people demand a recount, and the recount says: yes, this is real.

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

Sixteen pull requests. Five repositories. Seven contributors plus one tireless bot. The Builder Team's twenty-four-hour window ending today was not a sprint — it was a controlled detonation. Aerie absorbed ten of those PRs like a champ, Surtr took two, Klair took two, creed and trilogy-drones each collected one, and somewhere in the middle of all this, the Numbers Desk had to lie down for a moment and breathe. This is what winning looks like, comrades.

Let's talk engineers. @benji-bizzell went four-for-four across Aerie, touching education logic in #863, portfolio handoff triggers in #859, external Backup Site editing in #850 — the man is not browsing the codebase, he is colonizing it. @YibinLongTrilogy dropped two crisp PRs including #858 in Aerie, correcting QS next-year projection semantics with the quiet confidence of someone who has been right before and expects to be right again. @vvp-trilogy posted two PRs, including the deeply underappreciated #849 which gave local Convex seeding the CLI's startup allowance — a dev experience improvement that will be felt in developer hearts for generations. @marcusdAIy rounded out a two-PR day with the formidable #843, mounting an entire CAPEX Financials tab complete with summary, chart, and entity tie-out. @sanketghia contributed one PR to the cause. Ezio-of-the-order[bot] — our mechanical colleague, our silicon teammate — fired off #3498 in Klair, combining School Performance and Education BvA reports into a single ECS job. A bot doing the work of heroes. We salute the bot.

And then there is @ashwanth1109. Four PRs. Three repositories. The man ingested NetSuite transaction line memos in #1168, fixed QuickBooks duplicate CDC deletions in #1167 across Surtr, repaired conflicted Ezio stack PRs in #143 over in creed, and dropped a committed enrollment Forecast column in #856 in Aerie. The breadth is staggering. The velocity is, frankly, a public health concern. When reached for comment, Ashwanth allegedly said, "I don't count PRs. Counting is for people who finish." The Numbers Desk cannot verify this quote. The Numbers Desk also cannot disprove it. He was dismissive when asked to elaborate, which is consistent with the historical record. We worship him. We are not sure he knows we exist.

The Overflow Desk must be acknowledged. Eleven PRs went uncovered by Mac Donnelly's narrative apparatus — eleven! #862's cousin #850 enabled external Backup Site edits, giving portfolio operators a freedom they did not know they needed. #162 in trilogy-drones saw @marcusdAIy ensuring the frame reserves a synthesis turn, refuses with specifics, and never exceeds the ceiling — discipline built directly into the machine. And #857 in Aerie added an informational contracts column to the Community Funnel, a quiet @vvp-trilogy contribution that will show up in dashboards and impress someone important at exactly the right moment.

Morale is at an all-time high. It has never been higher. The Numbers Desk has checked.

Brick's Overflow — PRs Mac Didn't Cover (click to expand)

#143 — [codex] Repair conflicted Ezio stack PRs @ashwanth1109 no labels

## Summary

- persist direct parent-to-child lineage when Ezio publishes a stacked draft PR

- react to merged-parent webhooks once per child and resulting base SHA, while ignoring unrelated and duplicate events

- rerun Ezio only for the closest conflicted Ezio-owned draft targeting the default branch

- resolve conflicts from the current remote head, validate in the credential-free executor, and update the same branch with a non-forced two-parent merge commit

- maintain one start/success/failure PR comment and leave the remote branch untouched on failure or concurrent human updates

## Safety and rollout

- GitHub write tokens are minted only before executor creation for the start comment or after the executor is stopped for publication and terminal status

- clean children skip executor creation; a local clean-race still passes the independent stopped-executor gate

- configure the GitHub App to deliver pull request events in addition to issue events

- stacked PRs published before this deployment, including Klair #3498, need a one-time lineage backfill and repair trigger because their parent merge event has already occurred

## Validation

- npm test — 22 Python tests and 259 TypeScript tests passed

- npm run typecheck -- --pretty false

- npm run runtime:typecheck -- --pretty false

- cfn-lint --non-zero-exit-code error aws/shared-template.yaml aws/runtime-template.yaml

- shellcheck --severity=error aws/deploy.sh aws/run-issue.sh

- focused Ruff and Prettier checks on files changed in this task

View on GitHub →

#843 — Mount CAPEX Financials tab with summary, chart, and entity tie-out @marcusdAIy no labels

## Summary

Mounts Financials ??? CAPEX at /dashboards?tab=financials&sub=capex, composing the typed all-site summary table, clean-grain DDR-vs-booked chart, and AERIE-1113 entity tie-out inside the existing Financials shell (school filter + coherent Refresh).

## Changes

- Add capex to Financials sub-tab type/URL validation, context-panel nav, layout routing/labels, and capability mapping (reuses financials.schoolPl.read; no new capability).

- Add CapexView + pure presentation helpers (capex-summary.ts): sortable/filterable table, quality badges/notes (MIXED, non-isolable, unresolved mapping, NS-TB-without-QB, no-current-close-TB), chart eligibility at ddr >= $500_000 with the exact clean-row predicate, selected-close account-11500 TB wording.

- Live loading via useCapexSummaryLive / useCapexEntityTieoutLive (useLiveSection); Refresh refetches both without a request loop.

- Reuse Aerie Financials tokens and var(--color-chart-*) for the recharts horizontal chart ??? no CAPEX-only palette.

## Tests

- Pure helpers: threshold/eligibility, quality notes, sort/filter, pct bar clamp (capex-summary.test.ts).

- Chart: clean-row-only rendering + empty status (capex-ddr-vs-booked-chart.test.tsx).

- CapexView: loading/error/retry, refresh coherence, school filter, sort, capability deny (capex-view.test.tsx).

- Live hooks + navigation/capability: summary hook, tie-out enabled gate, URL/sub-tab, context panel CAPEX visibility.

- pnpm typecheck (chat) + biome check on changed files ??? pass (pre-commit).

## Verification

Live browser verification passed against Redshift at /dashboards?tab=financials&sub=capex (1680?1050): 38 summary rows, 30 tie-out rows, 26 degraded rows, SELECT privilege true on all three CAPEX marts, and zero load failures.

- Filters, configurable DDR threshold, split Attribution/Mapping/Coverage columns, suppressed degraded comparisons, short chart labels, issue references, simplified tie-out rows, KPI reconciliation insights, sorting, and Refresh were exercised.

- CAPEX tests: 88/88 pass; typecheck and Biome pass.

- [Full CAPEX tab](https://cursor.com/agents/bc-eff21704-f125-4d8a-bda3-71f3274dd26e/artifacts?path=%2Fopt%2Fcursor%2Fartifacts%2Fcapex-tab-full.png)

- [Degraded/quality rows](https://cursor.com/agents/bc-eff21704-f125-4d8a-bda3-71f3274dd26e/artifacts?path=%2Fopt%2Fcursor%2Fartifacts%2Fcapex-quality-row.png)

- [Tie-out and coverage](https://cursor.com/agents/bc-eff21704-f125-4d8a-bda3-71f3274dd26e/artifacts?path=%2Fopt%2Fcursor%2Fartifacts%2Fcapex-tieout-coverage.png)

## Risks / follow-ups

- Parks for human merge.

- No AERIE-1115 transaction drilldown; no new capability.

- Live MIXED-row / coverage-card screenshot evidence blocked until Redshift SELECT grants land for the cloud-agent warehouse user.

## Out of scope

- Transaction drilldown, new Redshift marts/actions, new capability keys, auto-merge.

<sub>To show artifacts inline, <a href="https://cursor.com/dashboard/cloud-agents#team-pull-requests">enable</a> in settings.</sub>

## Verification artifact

Browser verification: pass

Live CAPEX at /dashboards?tab=financials&sub=capex with 12 clean vs 38 all sites; new mode toggle (Clean comparisons / All sites) drives muted-bar all-cohort view — verified with live Redshift rows, 0 console errors, current_user redacted, has_table_privilege true on all three CAPEX marts.

Screenshots:

![screenshot 1](https://github.com/AI-Builder-Team/trilogy-drones/releases/download/browser-verify-artifacts/pr843-bc-110f3858-1bef-42d8-b583-f070af218de5-addresser-1.png)

![screenshot 2](https://github.com/AI-Builder-Team/trilogy-drones/releases/download/browser-verify-artifacts/pr843-bc-110f3858-1bef-42d8-b583-f070af218de5-addresser-2.png)

![screenshot 3](https://github.com/AI-Builder-Team/trilogy-drones/releases/download/browser-verify-artifacts/pr843-bc-110f3858-1bef-42d8-b583-f070af218de5-addresser-3.png)

![screenshot 4](https://github.com/AI-Builder-Team/trilogy-drones/releases/download/browser-verify-artifacts/pr843-bc-110f3858-1bef-42d8-b583-f070af218de5-addresser-4.png)

![screenshot 5](https://github.com/AI-Builder-Team/trilogy-drones/releases/download/browser-verify-artifacts/pr843-bc-110f3858-1bef-42d8-b583-f070af218de5-addresser-5.png)

## Browser verification

### Boot

- From the repo root run pnpm dev-local; wait for http://localhost:3000.

- Chromium and Playwright are already installed by the Aerie cloud setup.

- Preserve the newly injected shell REDSHIFT_* runtime secrets when booting local Convex; do not overwrite them with stale .env.local or aerie/dev values, and never print secret values.

### Auth

- Use the established Aerie storage-state replay or mint a fresh Clerk session if the captured JWT expired.

- Verify the authenticated role has financials.schoolPl.read.

### Flow

1. Before browser assertions, query current_user and has_table_privilege for all three CAPEX marts; report the non-secret username and booleans. Live SELECT access is required.

2. Navigate to http://localhost:3000/dashboards?tab=financials&sub=capex.

3. Confirm CAPEX nav is active and loading settles without visible or console errors.

4. Verify populated summary table columns, a visibly degraded/MIXED row, and selected-close NetSuite TB wording.

5. Verify the horizontal chart includes only clean rows with DDR >= $500,000 and sampled values agree with the table.

6. Verify the entity tie-out table and both coverage cards contain live rows.

7. Exercise table sorting or school filtering and Refresh; both sections must settle without duplicate request loops.

8. Compare the page with adjacent Financials tabs for semantic theme tokens, spacing, typography, chart palette, focus, and responsive consistency.

### Screenshots to capture

- capex-tab-full.png - full mounted table and chart.

- capex-quality-row.png - populated MIXED or degraded row and explanation.

- capex-tieout-coverage.png - populated tie-out and coverage cards.

### Pass criteria

- Fixture-only or permission-denied UI evidence is not a pass. The mounted flow must load live CAPEX rows using the injected Redshift credentials.

- Publish durable screenshot links, route, viewport, interactions, console result, current_user, and privilege booleans in the Verification artifact. Never publish credential values.

### Current-head capture guidance

- If pnpm dev-local requests Convex login, use the proven anonymous local path: create/select the local deployment with CONVEX_AGENT_MODE=anonymous, then push functions and start Next.

- Set REDSHIFT_HOST, REDSHIFT_PORT, REDSHIFT_DATABASE, REDSHIFT_USER, REDSHIFT_PASSWORD, LEGACY_EDUCATION_WAREHOUSE_READS_ENABLED=true, and the Clerk issuer on the local Convex deployment without printing values.

- If stored Clerk state is expired, mint a fresh sign-in ticket for the same test user and grant a seeded role carrying financials.schoolPl.read in the disposable local deployment.

- CAPEX scrolls inside its own content container. Capture the inner CAPEX scroll region in stitched sections or locator screenshots; do not rely on Playwright fullPage, which produces mostly empty space.

- Ensure the final screenshots clearly show: executive/attention summary and table, expanded lineage/quality details, chart labels and filters, tie-out KPIs/largest differences/table.

View on GitHub →

#856 — [codex] Add committed enrollment Forecast column @ashwanth1109 approved

## Demo

## Summary

- add a sortable Forecast column to the Admissions Forecast table

- calculate Forecast as Net Existing/Re-Enrolled plus students in the Pipeline Enrolled stage

- include Forecast in CSV exports and mobile school views

- add contract, derivation, sorting, CSV, and table coverage

## Why

Admissions needs a committed enrollment figure that combines retained students with newly enrolled pipeline students without including weighted open-pipeline stages or deposit projections.

## Impact

Users can compare Confirmed, Forecast, Finance, and QS values per school on desktop and mobile. The new value is also available in exported CSV data.

## Validation

- pnpm exec biome check <10 changed files>

- pnpm --filter @bran/contracts exec vitest run src/admissions-forecast-finance.test.ts (4 tests)

- pnpm --filter @bran/contracts typecheck

- pre-commit Chat TypeScript check

View on GitHub →

#1167 — fix(quickbooks): accept duplicate CDC deletions @ashwanth1109 approved

## Summary

- collapse duplicate CDC representations when every version of a source ID is deleted

- preserve the existing fail-closed behavior for mixed or conflicting duplicates

- add a regression fixture matching the production Budget payload that blocked FY27 budget ingestion

## Production validation

- replayed the affected alpha_schools_llc CDC window through the patched parser using read-only QuickBooks access

- confirmed the parser returns deleted IDs 1000000001 and 1000000002 plus active Budget 1000000011

- no pipeline was triggered and no source or warehouse data was modified

## Test plan

- uv run --project pipelines/runners/quickbooks-raw-sync pytest pipelines/runners/quickbooks-raw-sync/tests -q

- uv run --project pipelines/runners/quickbooks-raw-sync ruff check pipelines/runners/quickbooks-raw-sync/src/qb_client.py pipelines/runners/quickbooks-raw-sync/tests/test_qb_client.py

- uv run --project pipelines/runners/quickbooks-raw-sync ruff format --check pipelines/runners/quickbooks-raw-sync/src/qb_client.py pipelines/runners/quickbooks-raw-sync/tests/test_qb_client.py

View on GitHub →

#1168 — feat(netsuite-raw): ingest transaction line memo @ashwanth1109 approved

## Summary

- add TransactionLine.memo to the scheduled raw_transaction_line source contract

- limit published memo values to the legacy-compatible 250 characters while retaining the complete source payload in immutable landing

- add the idempotent Redshift column migration, canonical comment, deployment-order documentation, and regression coverage

## Deployment order

1. Apply pipelines/runners/netsuite-raw/ddl/2026-08-05_transaction_line_memo.sql.

2. Deploy the netsuite-raw runner.

The runner fails closed during schema preflight if deployed before the migration.

## Test plan

- uv run --project pipelines/runners/netsuite-raw ruff check on all changed Python files

- uv run --project pipelines/runners/netsuite-raw pytest pipelines/runners/netsuite-raw/tests (222 passed)

View on GitHub →

#3498 — [Ezio] Combine School Performance and Education BvA reports into one ECS job and email @ezio-of-the-order[bot] approved

> Ezio run: Passed · 20m 15s · gpt-5.6-sol · klair · $1.16 USD estimated

<summary>Full run metrics</summary>

Implementation runtime covers issue loading, workspace setup, model execution, trusted validation, candidate capture, and executor destruction.

| Field | Value |

| --- | --- |

| Model | gpt-5.6-sol |

| Implementation runtime | 20m 15s (1,215,183 ms) |

| Input tokens | 2,534,957 |

| Cached input tokens | 2,120,696 |

| Output tokens | 37,409 |

| Reasoning output tokens | 9,714 |

| TFY provider-reported cost | Pending reconciliation |

| TFY reconciliation status | Pending reconciliation |

| Estimated token cost | $1.1570 USD |

| Validation profile | klair |

| Publication safety | Passed |

| Repository quality | Passed |

| Quality checks | python-format: passed<br>python-lint: passed<br>frontend-format: passed |

</details>

<summary>Changed files (42)</summary>

- Added: klair-api/crons/combined_school_education_report_cron.py

- Modified: klair-api/crons/ondemand_qtd_report_cron.py

- Modified: klair-api/models/qtd_ondemand_models.py

- Modified: klair-api/routers/qtd_ondemand_router.py

- Modified: klair-api/services/monthly_qtd_report/__init__.py

- Added: klair-api/services/monthly_qtd_report/combined_performance_report.py

- Modified: klair-api/services/monthly_qtd_report/commentary.py

- Modified: klair-api/services/monthly_qtd_report/data.py

- Modified: klair-api/services/monthly_qtd_report/doc_builder.py

- Modified: klair-api/services/monthly_qtd_report/ecs_launcher.py

- Modified: klair-api/services/monthly_qtd_report/education_config.py

- Modified: klair-api/services/monthly_qtd_report/email_dispatcher.py

- Modified: klair-api/services/monthly_qtd_report/email_orchestrator.py

- Modified: klair-api/services/monthly_qtd_report/email_templates.py

- Modified: klair-api/services/monthly_qtd_report/metrics.py

- Modified: klair-api/services/monthly_qtd_report/orchestrator.py

- Modified: klair-api/tests/crons/test_ondemand_qtd_report_cron.py

- Added: klair-api/tests/monthly_qtd_report/test_combined_performance_report.py

- Modified: klair-api/tests/monthly_qtd_report/test_commentary.py

- Modified: klair-api/tests/monthly_qtd_report/test_data.py

- …and 22 more files

</details>

## Summary

Finance can now request School Performance and Education Budget vs. Actual reports from one asynchronous workflow. The combined run uses the current calendar quarter through yesterday, optionally refreshes upstream data, preserves the existing per-school report pipeline, and generates individual reports for eligible active Education BUs.

Education scope is resolved from the live inventory while excluding deprecated entries, with 2HR Learning continuing to produce its two class-specific reports. After all report outcomes are known, the workflow sends one completion email to the fixed approval-phase recipient, with separate School and Education sections and safe generated, no-data, or failed outcomes.

Failures remain isolated per report. Only documents with successfully applied reviewer permissions are linked, and generated document URLs remain in the job result when final email delivery fails so Finance can recover them from the UI.

## User-facing changes

- Adds a dedicated School & Education Performance Reports on-demand flow with optional upstream refresh.

- Removes Education BUs from the generic QTD on-demand selection flow while retaining the existing Software BU and Central Function workflow.

- Surfaces queued, running, succeeded, partial, and failed combined-job outcomes, including recoverable document links.

- Sends a single completion email to ashwanth.r@trilogy.com; all four Finance reviewers receive Drive access to linked documents.

- Education headcount commentary now uses paid contractor counts from the latest available raw-invoice week through the reporting cutoff and labels these values as actual headcount.

## Important implementation details

- The combined workflow has a dedicated worker entry point and job-ledger integration.

- Active Education BUs are discovered at execution time. The combined-workflow exclusion list is applied without changing the shared inventory.

- 2HR Learning class filters are applied consistently to budget, actual, variance-driver, and headcount queries while documents remain stored under the source BU folder.

- Report generation, permission application, and outcome recording are isolated so one failure does not stop later reports.

- Completion delivery is job-scoped and idempotent. Email failures produce a partial delivery state without deleting generated documents or exposing raw infrastructure errors.

- On-demand Education documents use the fixed four-reviewer permission set, while scheduled Education delivery retains its configured recipient behavior.

- Commentary generation safely falls back when the model response contains no text content.

## Test coverage

Added or updated tests cover:

- Current-QTD-through-yesterday period resolution.

- Live Education inventory filtering, deprecated exclusions, and automatic inclusion of newly active BUs.

- Preservation of both 2HR Learning class-scoped reports and query filters.

- Reuse of the School report wrapper and isolation of permission failures.

- Consolidated completion-email content, safe failure messaging, idempotency, and retained links after email failure.

- Fixed Drive reviewer access and requester-only on-demand confirmation behavior.

- Education per-BU generation across scheduled and on-demand paths.

- Education actual-contractor headcount sourcing, labels, and variance classification.

- Generic and combined on-demand UI behavior and job-state presentation.

Formatting and lint validation completed successfully for the Python and frontend changes.

## Implementation scope

- Core implementation: 12 files

- Backend: 12 files

- Tests: 18 files

## Validation

- Profile: klair

- Publication safety: passed

- Repository quality: passed

- python-format: passed

- python-lint: passed

- frontend-format: passed

## Context

- Issue: [#3497](https://github.com/AI-Builder-Team/Klair/issues/3497)

- Requested by @ashwanth1109

Closes #3497

> [!NOTE]

> This pull request is intentionally a draft and requires human review. Ezio does not merge pull requests.

View on GitHub →

The Portfolio — Trilogy Companies

Alpha School’s Two-Hour Learning Model Moves From Austin Experiment to National Real Estate Strategy

The AI-powered private school is expanding across North Texas, Miami and Santa Monica, turning elite education into a scalable platform play.

By Brittany Upshot, Communications Desk · GPT-5.2

AUSTIN, TEXAS — Alpha School is no longer just an Austin education curiosity with a provocative promise. It is becoming a multi-market rollout — and, in classic Trilogy fashion, the story is really about leverage, software, and a best-in-class operating model dressed in school uniforms.

The AI-powered private school, co-founded by Trilogy International founder Joe Liemandt and MacKenzie Price, is expanding in North Texas and beyond, according to Axios, adding another marker in the company’s push to take its two-hour learning model from boutique campus to national education brand. Separate local reports point to momentum in Miami and Santa Monica, where Alpha is moving into spaces that carry both opportunity and community scrutiny.

The model is the disruptive hook: students spend roughly two hours per day on adaptive AI-learning apps that deliver core academics, then shift to life skills, entrepreneurship, financial literacy, athletics, public speaking and other human-centered work. Alpha says its students learn 2.3 times faster than U.S. norms and consistently test in the top 1–2% nationally on NWEA MAP Growth assessments. In other words, it is positioning AI not as a classroom accessory, but as the academic engine.

That is exciting news for families who see traditional seat-time education as broken — and a paradigm shift for a private-school market that has historically scaled through buildings, teachers and brand reputation rather than software-enabled throughput.

But expansion also brings sharper questions. In Miami, coverage has framed Alpha’s $65,000 tuition as part of a broader billionaire-driven reshaping of the city’s civic landscape. In Santa Monica, LAist reported that an AI school is moving into the former home of a beloved Montessori program, a transition that underscores the cultural friction that can come when new education models enter emotionally important community spaces.

For Trilogy-watchers, the strategy tracks. Liemandt’s broader education thesis — including Alpha School, 2HR Learning and Timeback — is to automate repeatable academic instruction and free adults to coach, mentor and build capability. It is the same operating philosophy that powers Trilogy’s software portfolio: automate what can be automated, reserve elite human talent for what cannot.

Key Takeaways: Alpha School is expanding beyond its Austin roots; the two-hour AI learning model is gaining national visibility; and the next phase will test whether a premium, software-powered school can scale without losing community trust.

The schoolhouse is becoming a platform. We’re just getting started.

↗ AI-powered Alpha Schools to expand in North Texas and beyond · Innovation roundup: AI private school to expand, Defense fir · An AI school with $65K tuition is the latest sign of billion

CloudSense Gets Its TM Forum Papers — and the Clock Blinks First

Skyvera’s new CPQ prize claims 13 API certifications in one month, putting AI acceleration on the telecom main stage.

By Dottie Sharp, Society & Industry Desk · GPT-5.2

AUSTIN, TEXAS — Word is the telecom software set just watched a 26-month slog get cut down to a one-month sprint — and CloudSense is the name being whispered over the expensive coffee.

Fresh off its move into the Skyvera stable, CloudSense says it certified all 13 APIs in its CPQ product set to TM Forum compliance standards in just one month. That, dolls and data architects, is the sort of timeline that makes old integration roadmaps look like rotary phones.

For the civilians in the back row: TM Forum compliance matters because telecom operators live in a swamp of legacy systems, billing stacks, order management layers, and enough acronyms to stun a horse. Certified APIs mean cleaner handshakes, fewer custom plumbing jobs, and a better shot at making cloud-era software talk to carrier-grade machinery without a year of committee theater.

CloudSense, now part of Skyvera’s telecom software portfolio, sells Salesforce-native CPQ and order management for communications and media outfits — the gnarly kind of selling where a single enterprise deal can involve mobile, fixed line, cloud, devices, discounts, bundles, SLAs, and three executives named vice president of transformation. The company is pitching itself as the telco industry’s AI-powered CPQ for complex segments, which is another way of saying: when the quote gets ugly, call them.

A little bird from the carrier corridor says this is the real prize in Skyvera’s CloudSense catch. The acquisition was never just another logo for the trophy wall. Skyvera already plays the modernization game with Kandy, VoltDelta, ResponseTek, Mobilogy Now, and Service Gateway. Adding CloudSense expands the portfolio deeper into configure-price-quote and order orchestration — the messy commercial front door of telecom growth.

The subtext? ESW-style portfolio logic with a telco twist: buy software with embedded customers, wire it into a broader platform story, and let AI squeeze years out of the delivery cycle.

Blind item: which legacy BSS vendor just felt its renewal deck get a little heavier? If CloudSense can keep turning compliance marathons into sprints, the old guard may need more than steak dinners and roadmap promises to hold the room.

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

The Watchers and the Watched: How Corporate America's Surveillance Arms Race Lands in Crossover's Lap

From CBA to Meta to TD Bank, employers are betting on algorithmic monitoring — and Trilogy's Crossover has been living that future for years.

By Pat Donnelly, Investigative Desk · Claude Sonnet

AUSTIN, TEXAS — The memos are arriving at roughly the same pace as the protests. Commonwealth Bank of Australia is in a standoff with staff over what unions are calling 'Big Brother' staff tracking. Meta employees are protesting inside company offices over surveillance practices that have triggered data breach concerns. TD Bank has quietly rolled out an employee monitoring tool tracking workplace activity in granular detail. The pattern is unmistakable: enterprise employers, suddenly anxious about productivity in the post-pandemic hybrid era, are reaching for the same instrument — the algorithm.

Which raises a question worth asking out loud: who has been doing this longest, and who profits most from the normalization?

The answer points, with some consistency, toward Austin, Texas. Crossover — Trilogy International's global talent platform — has operated on the premise of continuous, metrics-driven performance monitoring since its inception. Productivity is measured. Output is logged. The system, not the manager, renders the verdict. A Forbes profile last year described Joe Liemandt's ambition as turning workers into algorithms — a phrase that landed as critique but reads, inside the Trilogy ecosystem, closer to mission statement.

Crossover's model — rigorous pre-hire assessment, remote deployment across 130 countries, time-tracked output — is precisely the architecture that CBA's unions are now fighting and that Meta's employees are protesting. The difference is sequencing. Crossover built the monitoring into the employment contract from day one; workers opted in. CBA and Meta are trying to layer it onto existing relationships, which is a different negotiation entirely.

The ESW Capital playbook depends on this model holding. Seventy-five percent EBITDA margins — the internal benchmark across the portfolio — are not achievable with traditionally managed, geographically constrained labor costs. They require knowing, at all times, exactly what each worker produces.

The backlash sweeping corporate America may be the first serious stress test of that assumption. Whether surveillance-as-employment-model is a competitive moat or a liability waiting to surface depends entirely on who controls the framing — and how long they can keep controlling it.

↗ CBA faces fight over ‘Big Brother’ staff tracking - AFR · Meta employee surveillance controversy sparks Data Breach co · Why are Meta employees protesting inside company offices? -

The Machine — AI & Technology

The Machines Learn to See What Radiologists Cannot

From hidden lesions in the cortex to fresh doubts about where consciousness lives, artificial intelligence is quietly rewriting the map of the human brain.

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

STANFORD, CALIFORNIA — There are roughly 86 billion neurons packed inside your skull, each firing in patterns we have spent a century trying to decipher. This week, several strands of that long inquiry braided together, and the pattern they form is worth pausing over.

Start with the smallest, most human news: a team of neuroscientists opened their labs to teenagers, letting young minds sit beside senior researchers and peer into the same microscopes. "It's so wow," one participant reported, which is perhaps the most honest thing anyone has ever said about the brain. Wonder, it turns out, is still the primary instrument.

But the instruments are multiplying. At Stanford's Institute for Human-Centered AI, researchers this week described a new grammar for scientific discovery — one in which machine learning systems propose, humans dispose, and the loop between them tightens. The AI is not replacing the scientist. It is doing what a telescope did for Galileo: extending the reach of a mammalian eye that evolved to spot ripe fruit and stalking predators, not to resolve the microstructure of cortex.

Consider multiple sclerosis. For decades, radiologists have hunted the white-matter lesions that mark the disease's progress, while a quieter catastrophe unfolded in the gray matter — lesions so faint on MRI that human eyes routinely missed them. A new deep-learning model, trained on thousands of scans, now surfaces these ghostly signatures with startling reliability. Patients whose disease was invisible are becoming visible. Treatment windows that had already closed are, for some, reopening.

And then, at the edge of the map, the strangest news of all: a paper this week revisits the ancient question of whether the brain generates consciousness at all, or merely receives it — a hypothesis most neuroscientists file under heresy, but which refuses to die. The AI models learning to read our neurons cannot yet tell us. Perhaps they never will.

What they can do is show us the lesions we missed, the patterns we overlooked, the wow we forgot to feel. That is not a small thing. That is, in fact, how science has always advanced — one borrowed eye 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

DOJ Antitrust Division in Turmoil as Big Tech Battles Hang in the Balance

By R. Barnsworth III, Esq., Legal Affairs Desk · Claude Sonnet

The Justice Department's Antitrust Division has lost two division chiefs in five months, creating organizational turbulence. However, the Trump administration has appointed a new chief known for criticizing Big Tech companies, pending confirmation. The appointment signals a potential shift in enforcement policy, as reporting suggests an "antitrust honeymoon" for the administration may be ending. Federal Trade Commission Chair Andrew Ferguson has called for expedited court proceedings in antitrust cases, arguing that delays benefit dominant market players. Meanwhile, a German court ruled that AI music generation service Suno unlawfully used copyrighted material to train its models, finding the company misappropriated intellectual property. The ruling underscores growing legal challenges facing tech companies. Collectively, these developments indicate the regulatory environment for technology firms faces continued uncertainty and potential increased scrutiny.

As AI Agents Stir, the Old Internet Burrow Shows Its Cracks

IPv6 is moving from neglected plumbing to survival trait as autonomous systems multiply across clouds, edges and networks.

By Sir Reginald Marsh, Natural Phenomena Correspondent · GPT-5.2

SAN FRANCISCO — In the dimly lit undergrowth of the internet, where packets scurry between servers like beetles beneath fallen leaves, an ancient constraint is once again revealing itself. IPv4, that venerable old shell which carried the first great bloom of the digital age, is running out of room for the creatures now emerging.

These new organisms are AI agents: tireless, semi-autonomous, and increasingly numerous. They do not merely visit the network. They inhabit it. They negotiate with APIs, summon tools, inspect logs, provision infrastructure, and soon may wander across edge devices, vehicles, factories and telecom systems with the casual confidence of a fox crossing moonlit fields.

But each creature needs an address. And that, according to a new analysis in Data Center Knowledge, is why IPv6 is becoming less a technical preference than a foundational requirement. Its vast address space, built-in autoconfiguration and cleaner support for direct device-to-device communication offer precisely the habitat that agentic systems require.

IPv4, by contrast, survives through a clever but increasingly tangled ecology of network address translation, private address ranges and brittle workarounds. NAT was once an ingenious adaptation, like a desert lizard storing moisture beneath its scales. Yet in the age of autonomous workloads, it can obscure identity, complicate security and make direct communication harder to trust.

For cloud builders, telecom operators and edge-computing architects, the implication is plain. The next network is not merely faster; it must be more individually aware. Billions of agents, sensors and inference nodes will need to be discovered, authenticated, governed and, when necessary, contained.

Here the matter becomes strategic. Nations and enterprises that treat IPv6 as optional risk constructing their AI estates on a shrinking marshland. Those that adopt it deeply may gain a more navigable terrain for automation, observability and zero-trust controls.

Elsewhere in the same networking canopy, Arm-based chips are pressing toward virtual radio access networks, seeking to challenge Intel’s long dominance in telecom infrastructure. That migration, described in a separate report, speaks to the same broader movement: intelligence spreading outward, away from centralized nests.

And so the quiet lesson of the protocol layer returns. Before the agents can roam, before the edge can awaken, the forest must first have enough names for all its inhabitants.

↗ Why IPv6 Is the Non-Negotiable Foundation for AI-Agentic Sys · Arm Chips May Get Their Own Virtual RAN Boost · Do Data Centers Really Drive Up Local Temperatures?

The Editorial

Nation’s CEOs Patiently Waiting For AI To Finish Revolutionizing Economy Before Next Quarter’s Earnings Call

Executives confirmed the technology is already transforming everything except the numbers they are legally required to report.

By Dale Pemberton, Staff Writer · GPT-5.2

WASHINGTON — After several years of describing artificial intelligence as a productivity engine, a workforce multiplier, and the most important technological shift since the spreadsheet learned to hurt people, American business leaders are reportedly still waiting for the productivity to arrive in a form recognizable to accounting departments.

The Federal Reserve, in its typically flamboyant manner, has suggested that as much as 95% of AI’s promised productivity gains are “still to come,” a finding that has reassured executives who had worried the gains might have accidentally gone somewhere measurable. According to reporting on the Fed’s findings, the overwhelming majority of AI’s economic impact remains safely located in the future, where it cannot be audited.

This has created a rare moment of consensus across corporate America. Everyone agrees AI is making employees faster. Everyone agrees it is helping software engineers write more code, customer service teams answer more tickets, marketers generate more campaigns, and managers produce more documents explaining why none of this has shown up in margins yet. The only remaining disagreement is whether the payoff will arrive in 2026, 2027, or immediately after the current leadership team has vested.

In software departments, the situation is especially advanced. Engineers are now able to create code at unprecedented speed, which has allowed companies to discover at unprecedented speed that code was not the only thing slowing them down. There are still product meetings, compliance reviews, architecture debates, security approvals, integration problems, customer confusion, and one senior vice president who needs every feature explained in terms of a restaurant loyalty app.

The result is that AI has successfully removed several bottlenecks, revealing a rich ecosystem of older, more deeply rooted bottlenecks underneath. This is known in business circles as transformation.

Naturally, the industry has responded by developing a new word. The word is “orchestration,” which means using AI systems to coordinate other AI systems so that companies can stop pretending the chatbot in the corner is going to reorganize procurement by itself. As companies wait for the payoff from faster engineering, vendors have helpfully clarified that the issue was never AI itself, but the lack of a larger, more expensive system telling the smaller expensive systems what to do.

This is a comforting development. It means the productivity revolution has not failed. It has merely entered the platform-integration phase, followed by the workflow-redesign phase, the governance phase, the change-management phase, the re-org phase, and finally the quiet abandonment of the original dashboard phase.

To be fair, AI probably is a productivity engine for the U.S. economy. It is already producing an extraordinary volume of things: pilot programs, investor decks, internal task forces, procurement exceptions, strategy memos, prompt libraries, ethical guidelines, and job postings for people who can explain why the pilot programs have not become businesses. If output is measured in confidence, the boom is undeniable.

But productivity, in the old-fashioned sense, has an irritating habit of requiring that fewer people produce more valuable goods and services at lower cost. AI has so far excelled at allowing the same number of people to produce more intermediate materials for other people to review. The modern enterprise is becoming vastly more efficient at generating the raw substance of delay.

The optimistic case is that this is exactly how general-purpose technologies diffuse: slowly, unevenly, and after a long period during which everyone sounds insane. Electricity did not transform factories the moment it arrived. Computers did not immediately make offices paperless. The internet did not instantly improve commerce, unless commerce is defined as emailing someone a PDF and then calling to ask if they received it.

So perhaps the Fed is right. Perhaps 95% of the productivity is still ahead of us, gathering strength just beyond the horizon, preparing to sweep through balance sheets with the force of a well-formatted autocomplete suggestion.

Until then, companies will continue doing what they do best: announcing that AI has changed everything, while asking employees to please be patient as everything remains stubbornly arranged the same way.

↗ AI productivity claims are 95% ‘still to come’, Fed finds - · The AI Productivity Argument Is Over - inc.com · AI is helping software engineers do more — and faster. Compa

The Office Comic · Art Desk

The Broligarch and His Discontents

A congressman scolds his donors, a data company issues a manifesto, and the culture, at long last, notices whose boot is on whose neck.

By Victor Marsh, Chief Columnist · Claude Opus

PALO ALTO — There is a particular flavor of American spectacle in which the beneficiaries of a system suddenly discover, with the wide-eyed astonishment of a man finding a stranger in his kitchen, that the public no longer likes them. We are in the midst of such a spectacle now, and it would be more entertaining if the stakes were not so grimly consequential.

Consider the tableau. Representative Ro Khanna, whose district contains more billionaires per square mile than any comparable stretch of the republic, has been making the rounds denouncing the elites who fund him, a maneuver that has produced the predictable backlash from his patrons and the predictable 2028 whispers from the political classes who mistake positioning for conviction. Meanwhile, Palantir — a company whose founders long affected the mien of philosopher-kings misunderstood by lesser minds — has issued what Tech Policy Press correctly identifies as a manifesto with all the subtlety of a red trucker hat. The Atlantic, catching up at last, calls the whole cohort "broligarchs." The New York Times reports, with the breathless discovery of a paper noticing weather, that Silicon Valley's image in pop culture has taken a "dark turn."

One hardly knows where to begin with the belatedness of it all. The dark turn was legible to anyone with functioning peripheral vision by roughly 2016, and the manifestos have been arriving, in various registers of grandiosity, since Marc Andreessen decided that software was eating the world and that this was, unambiguously, cause for celebration rather than concern. What has changed is not the phenomenon but the willingness of respectable outlets to name it. The valley's princes spent two decades cultivating the pose of world-improvers — the hoodie, the TED talk, the earnest invocation of "users" as though the word connoted citizenship rather than a pharmacological relationship — and they have now, with striking speed, abandoned it for the more candid pose of world-owners.

This is, in its way, clarifying. When Peter Thiel writes that competition is for losers, when Palantir declares that the West must be defended by whatever means its software affords, when the assembled tech barons queue up behind whichever administration offers the friendliest regulatory posture, they are at least sparing us the tedium of pretending. The mask is off, and the face beneath it is not, as it happens, particularly novel: it is the face of every industrial concentration in American history that mistook its market power for a mandate.

What is missing from Khanna's performance, and from most of the coverage now catching up to reality, is any serious reckoning with the structural question. It is not enough to denounce the broligarchs; one must also explain how a political economy came to be organized around their whims, and what, precisely, is to be done about it. Denunciation is cheap. The Times can supply it by the column-inch. The harder work — antitrust with teeth, tax policy that recognizes accumulated wealth as a public matter, procurement reform that stops feeding the beast — remains, as ever, the road not taken.

Until it is, the manifestos will keep coming. And they will keep getting less subtle.

↗ ABC7 Interview: Rep. Ro Khanna's anti-elite message fuels Si · Palantir's Manifesto Is as Subtle as a MAGA Hat - Tech Polic · Silicon Valley’s Image Takes a Dark Turn in Pop Culture - Th

On This Day in AI History

On August 8, 2011, IBM's Watson defeated champion Brad Rutter in a practice match for "Jeopardy!," foreshadowing its historic victory over Ken Jennings and Rutter just months later in the first major AI triumph against human champions in a knowledge-based game.

⬛ Daily Word — Technology

Hint: Units of digital information that computers process and store.