Changelog

Lead decision under FD-40: ingestion fetchers run their launch cadence pre-launch (scrape 2x/day, api daily, rotation 4x/day, EDGAR daily)

SP-27B: /app/entities shows Watched with this week's documents; /app/watchlist is a page (301 to Today removed at merge)

SP-12C: docs-freshness check judged on first-parent history (REVIEW-SP-12C F1); ci python job at fetch-depth 0

SP-13C: consent banner, one subprocessor list, DSAR intake handler, workspace deletion routine, export AI marker, Terms v1.1 draft

SP-17C: true .docx Memo/Binder, behavioural money/export tests, cited-paragraph validator rule, digest + entity-resolution fixes

SP-21C: product objects code half - Answer/Watch/Case/Export (D1-D4 SQL in hand-off, signed manifests, /app/today, suggest-only evidence)

SP-24C: paywall/login connector pattern (per-source cookie or token, approved hosts only, abstract-only storage) + register template

SP-01C WP-1: four FD-04 spend budgets inside the $3/day cap; monitoring keys draw only on monitoring; tutor probe k<=1 on Sonnet

The monthly cap's subject moves with attribution: Principal carries workspace, customer and tier together so the two cannot diverge

MARKETING 4/5/7: weekly validation series (live), calibration data (bins + Murphy + cohorts), public stats summary + GET /api/public/stats

API: /v1/entities/by_alias now matches an entity's own canonical name, not only its other surface forms.

Cases: create a case, add answers and entities to it, and compose a memo from the case.

Data correction: 113 forecast horizons now equal the interval their own label names.

Earliest post-fix horizon moved from 2027-08-07 to 2026-11-05.

The Free tier's 50-queries/month cap is enforced in-app, shown before the boundary, and reconciled with API-key usage.

Source coverage: NHTSA complaints now in the corpus — 185 documents, the first from this source.

Briefs without a sector are labelled "Ad-hoc synthesis" everywhere, never a borrowed sector.

Entity dossiers: key-fact units corrected — 3,687 displayed values fixed (a market cap had read "$5" for five trillion).

The network graph and deep-dive entity panels now distinguish "measured empty" from "the read failed".

Deep-dive pages load in under a second (were 36–46 s) and the desk in under a second (was 13–15 s), with the caching disclosed.

Tutor restored: answers cite again and follow-up questions no longer echo the prompt.

/accuracy segments forecasts into pre- and post-driver-fix cohorts, with the cutover stated.

The cross-sector network is a Team-and-above feature; that limit is now enforced everywhere it appears, not only described.

Stop exports leaking orphan [N] markers + count DISTINCT sources cited

Supplier edges pointed the wrong way, and "verified" did not mean verified

A headline stub is not evidence — ag brief published $32M for a $32B figure

DOJ pill-mill and cartel press releases were the AGRICULTURE desk's top signals

Forecast engine: every driver was inert — the Monte-Carlo posterior was the prior plus sampling noise. Fixed; /accuracy reports the two cohorts separately.

A developer could not obtain an API key at all — add the self-serve mint path

The launch-cadence series counts suborbital flights, skips launch-free days, and publishes a half-finished day

Cache the global cross-sector entity list so board/memo/desk watchlist reliably populates

Suffix-only twins were invisible to the split rule — Northrop Grumman rendered 70 of 368 mentions

Docs.forumatlas.com live + curated OpenAPI spec published

Resolve the SpaceX identity split — capture searched the trading name, every count read one half of the company

Sector-forecast fallback for customer workspaces

/v1/query contradicted itself — validator, freshness, head

API: Evidence Bundle v1 — GET /v1/briefs/{brief_id}/evidence.

Public commit-reveal forecast ledger at /ledger, with a machine-readable copy at /ledger.json.

Archive-at-citation — snapshot + hash every cited source

ART. 50 transparency package: /transparency page, machine-readable AI disclosure, per-brief provenance manifests

Source-integrity defense: concentration flag, feed anomaly scan, corrections policy

forumatlas.com custom domain live — the platform now serves from its canonical apex.

Platform hardening batch

Citation-validator typography fix, scheduled briefs now embeddings-backed, forecast engine driver fix + self-resolving threshold forecasts, API input validation, real user identity + onboarding preferences persistence.

GENIUS Act stablecoin tracker launched (ForumFinance)

Live DeFiLlama-fed market-cap tracker + regulatory milestone timeline, on a zero-LLM data lane.

Search + Tutor restored after the web→API auth fix.

Watchlists added.

Marketing site, app surface and docs site shipped.

Citation discipline UX live across the brief reader; Forum Forecast Monte-Carlo distribution viz; Forum Network entity graph viz; B2A2B docs site.

Forum Atlas launches — agent-built industry intelligence across 12 sectors

Today we're opening the Forum Atlas private beta. Here's what we're building, why we're building it at single-FTE + AI-agent scale, and what it means for teams buying incumbent intelligence tools at enterprise seat prices today.

AI capabilities shipped — 8 agent specs delivered.

B2A2B + Community shipped: API spec, bounty marketplace and community surfaces.

Ingestion wave 3 shipped — 441 sources, 6 personas.

Citation discipline — the verbatim-substring contract

How the Forum Atlas Citation Validator agent works. Why we render 'insufficient evidence' instead of fabricating. Why every paragraph anchors back to source provenance.

Sector spotlight: ITDR moves to Peak of Inflated Expectations

Gartner's May 2026 Hype Cycle for IAM promoted Identity Threat Detection & Response (ITDR) to Peak. Here's how Forum Atlas tracks category-shift signal across its cyber source catalog.

Sector spotlight: defense-tech M&A heat under Anduril's shadow

How Forum Forecast surfaces mid-tier defense-tech companies with elevated 12-month M&A probability — and how the driver-attribution trace tells you why.

AI architecture: a multi-agent ensemble at single-FTE scale

Inside the Forum Atlas coordinator + five specialist agents + citation validator. Why we're single-provider Anthropic. Why voyage-3 at 1024 dimensions. Why Cohere Rerank v3.5.