QA Sweep Reports

[OAGr]

Archive of /qa-sweep findings. Each comment is one sweep run with prioritized findings, issues filed, and coverage notes.

Use this discussion to track sweep history and catch regressions over time.

[OAGr]

QA Sweep — 2026-03-18

Focus: Full sweep (all directories) | Depth: Exhaustive

Agents launched: 13 | Agents completed: 8 | Pages checked: ~350+

---

P0 — Active bugs (user-visible now)

#	Bug	Location	Evidence
1	10 benchmark pages show "0 models tested" but model pages have scores — broken bidirectional link	AIME 2025, MGSM, Chatbot Arena Elo, Humanity's Last Exam, MathVista, TAU-bench, MLE-bench, WebArena, LiveBench, BFCL	~25 orphaned scores visible on model pages but invisible on benchmark leaderboards
2	3 research area pages return 404	/research-areas/mechanistic-interpretability, /research-areas/dangerous-capability-evaluations, /research-areas/ai-evaluations	High-profile topics with broken links
3	3 publication pages return 404 (high-citation)	/publications/future-of-life-institute (53 citing pages), /publications/future-of-humanity-institute (43 citing pages), /publications/world-economic-forum
4	Entity profile dashboard E1929 not registered — 2 test failures on main	wiki-nav.test.ts:402,505	entity-profile-dashboard.mdx not in database.json
5	Ghost entity IDs rendered as card titles on approach pages	/approaches/field-building-analysis, /approaches/hardware-enabled-governance, etc.	IDs like ULjDXpSLCI, SpFgTz2INg rendered as titles
6	20+ garbled entity IDs displayed as person names on 8 org pages	/organizations/cais, /organizations/fli, /organizations/cea, /organizations/cser, /organizations/arc, /organizations/epoch-ai, /organizations/chai, /organizations/ftx-future-fund	Strings like FeHrdAUKpQ where names should be (pre-existing #2602)
7	People: garbled/inverted names + 5 duplicate records	/people/andreas-stuhlm-ller (garbled), /people/askell-amanda (inverted), /people/bowman-samuel-r (inverted), plus duplicates for Andrew Ng, Alex Kaskasoli, Ben Weinstein-Raun, Amanda Askell, Buck Shlegeris

P1 — Latent bugs

#	Bug	Location
1	4 legislation ghost entries still 404ing — PG cleanup needed (#2516)	/legislation/compute-monitoring, etc.
2	OpenAI total funding $17.9B missing $40B SoftBank round (Mar 2025)	/wiki/openai vs /funding-rounds
3	Google Anthropic investment: $3.3B stated, only $2.3B documented	/wiki/anthropic
4	xAI Series C valuation: $40B vs $50B, missing Series D	/wiki/xai vs /funding-rounds
5	MIRI lifetime funding $15M vastly understated (IRS 990s show >$50M)	/wiki/miri
6	Conjecture founding: 2021 (directory) vs March 2022 (detail page)	/organizations vs /wiki/conjecture
7	Apollo Research founding: 2022 (directory) vs May 2023 (detail page)	/organizations vs /wiki/apollo-research
8	Jan Leike/Holden Karnofsky missing from Anthropic People tab	/people/ vs /organizations/anthropic
9	Google DeepMind org page returns 404	/organizations/google-deepmind
10	Sub-area layer systematically empty (0 orgs, 0 papers)	All research-areas sub-areas
11	Events directory has only 2 entries	/events
12	13/17 projects missing Organization in directory table	/projects
13	/investments directory entirely empty ($0, 0 records)	/investments
14	Unbounded query in /citations/accuracy-dashboard — full table into memory	citations.ts:798
15	N+1 UPDATE loop in backfill-stable-ids	ids.ts:276
16	Hono RPC non-compliance in links.ts, citations.ts, resources.ts	Multiple route files
17	ARC-AGI-2 described as "upcoming" — launched March 2025	reasoning.mdx:232
18	ASL-3 described as "planned" — already in production	governance-policy.mdx:212
19	Evan Hubinger FactBase says "formerly at Anthropic" — contradicts other sources	factbase/data/things/evan-hubinger.yaml
20	19 NEEDS CITATION markers in CAIS and FTX red flags pages	cais.mdx, ftx-red-flags.mdx
21	93% of people (603/649) have no structured role data	/people index
22	Publications tab shows counts but no linked records	Multiple /people/ pages
23	Mira Murati missing current org (Thinking Machines Lab)	/people/mira-murati
24	employerSlug set to entityId — potential broken org links	page.tsx:117

P2 — UX improvements

#	Issue	Location
1	Params column 100% empty across all 45 AI models	/ai-models
2	Safety level missing for 31/45 non-Anthropic models	/ai-models
3	12 models with zero benchmark scores	Multiple model pages
4	5 legislation pages missing Provisions/Stakeholders tabs	Multiple /legislation/ pages
5	/approaches/reducing-hallucinations article body not rendering	/approaches/reducing-hallucinations
6	CEA spinout "expects to by July 2025" — deadline passed	cea.mdx:119
7	Dead link: ea-crux-project.vercel.app → 404	interpretability-sufficient.mdx:22
8	Meta AI Movie Gen "set to debut in 2025" — unresolved	meta-ai.mdx:158
9	OpenAI user count stuck at 100M (Feb 2023)	factbase/data/things/openai.yaml
10	Anthropic total $67B vs footnote $75.2B vs table ~$52B	/wiki/anthropic
11	xAI founding date: July 2023 (prose) vs March 2023 (structured)	/wiki/xai
12	Jan Leike title: "Head of Alignment Science" vs "VP of AI Safety Research"	/wiki/jan-leike vs /wiki/anthropic
13	FHI has no People tab despite affiliated people	/organizations/fhi
14	Events directory has no Date column	/events
15	Amazon classified as "publication" with credibility rating	/publications/amazon
16	Will MacAskill profile completely empty	/people/will-macaskill
17	"DB records →" debug link exposed to end users on all people profiles	All /people/[slug] pages
18	No filter for "has data" vs empty stubs on people directory	/people

Data completeness (AI Models directory)

Column	Fill Rate
Name	100% (45/45)
Released	100%
MMLU	71% (32/45)
GPQA	58% (26/45)
SWE-bench	33% (15/45)
Safety Level	31% (14/45, Anthropic only)
Params	0% (0/45)

Cross-consistency issues

#	Mismatch	Page A	Page B
1	OpenAI funding: $17.9B vs ~$57B	/wiki/openai	/funding-rounds
2	Google→Anthropic: $3.3B stated, $2.3B documented	/wiki/anthropic	(internal)
3	xAI Series C: $40B vs $50B	/funding-rounds	/wiki/xai
4	MIRI lifetime: $15M stated vs >$50M actual	/wiki/miri	IRS 990 data
5	Conjecture founded: 2021 vs 2022	/organizations	/wiki/conjecture
6	Apollo Research founded: 2022 vs 2023	/organizations	/wiki/apollo-research
7	MIRI founding: Jan 2000 vs Jul 2000	/wiki/miri	/wiki/eliezer-yudkowsky
8	Jan Leike title at Anthropic	/wiki/jan-leike	/wiki/anthropic
9	Karpathy missing 2023 OpenAI tenure	/wiki/andrej-karpathy	/wiki/openai

Confirmed clean

• All 45 AI model detail pages load (no 404s)
• All 31 benchmark detail pages load (no 404s)
• All 72 approach detail pages load (no 404s)
• Legislation fixes from 2026-03-17 sweep landed without regression
• MMLU/GPQA/SWE-bench leaderboards bidirectionally consistent
• SQL injection protection verified across all wiki-server routes
• Authentication middleware correctly applied

Issues filed

• 10 benchmark pages show 0 models but model pages have scores — broken bidirectional link #2633: Benchmark bidirectional link broken (P0)
• 3 high-profile research area pages return 404 (mechanistic-interpretability, dangerous-capability-evaluations, ai-evaluations) #2634: 3 research area 404s (P0)
• 3 high-citation publication pages return 404 (FLI, FHI, World Economic Forum) #2635: 3 publication 404s (P0)
• Entity profile dashboard E1929 not registered — 2 test failures on main #2636: Entity profile dashboard E1929 unregistered (P0)
• Ghost entity IDs rendered as card titles on approach detail pages #2637: Ghost entity IDs on approach pages (P0)
• OpenAI wiki page missing SoftBank 0B round — total funding severely outdated at 7.9B #2638: OpenAI funding outdated (P1)
• Stale future-tense claims: ARC-AGI-2 upcoming, ASL-3 planned, CEA spinout pending #2639: Stale future-tense claims (P1)
• Unbounded query in /citations/accuracy-dashboard loads full table into memory #2640: Unbounded citation dashboard query (P1)
• Hono RPC non-compliance in links.ts, citations.ts, resources.ts — manual safeParse instead of zv() #2641: Hono RPC non-compliance (P1)
• Evan Hubinger FactBase description says 'formerly at Anthropic' — contradicts all other sources #2642: Evan Hubinger FactBase contradiction (P1)
• Financial data cross-consistency: Anthropic Google gap, xAI valuation mismatch, MIRI understated #2643: Financial cross-consistency (P1)
• Founding date mismatches: Conjecture (2021 vs 2022), Apollo Research (2022 vs 2023) #2644: Founding date mismatches (P1)
• People directory: garbled/inverted names and 5 duplicate person records #2645: People garbled/inverted names + duplicates (P0)
• People directory: 93% have no role data, publications unlinked, missing key affiliations #2646: People data gaps (P2)
• Research areas: sub-area layer systematically empty (0 orgs, 0 papers across all clusters) #2647: Research area sub-areas empty (P2)
• Pre-existing: Organizations: raw FactBase IDs shown as people names on 8+ org detail pages #2602 (garbled person names on org pages), 4 ghost policy entities cause 404s on /legislation index #2516 (ghost legislation entities)

Notes

• 5 of 13 agents hit ECONNRESET after ~50 min (organizations index, grants/funding, code quality, schema validation, one directory agent). Organizations, Grants, and Code quality audits not completed this run.
• Previous sweep (2026-03-17, legislation deep) filed 12 issues, created 18 PRs (14 merged).

[OAGr]

QA Sweep — 2026-03-19

Focus: full sweep (deterministic only) | Depth: manual validator run

Checks run: 17 deterministic + 6 standalone validators | Pages checked: N/A (data-level sweep)

Deterministic Sweep (17 checks)

Status	Check	Notes
FAIL	YAML field consistency	7 issues — relatedEntry type mismatches
WARN	NEEDS CITATION	19 markers in 2 files (CAIS page: 18)
WARN	Entity data integrity	22 warnings
WARN	Financial staleness	5 warnings
WARN	People role consistency	4 mismatches (Holden Karnofsky, Neel Nanda, Nick Bostrom, Will MacAskill)
WARN	Internal links	17 warnings
PASS	Duplicate wikiIds	979 unique, no duplicates
PASS	Broken entity refs	Clean
PASS	TODO markers	2 (ok)
PASS	Wrong domain refs	None
PASS	Quality discrepancies	Clean
PASS	Temporal invariants	Clean
PASS	Content staleness	16 info
PASS	Content gate	All pass
PASS	Test suite	170 passed
SKIP	Cross-entity consistency	Parse error
SKIP	Orphan PG entities	Parse error

Standalone Validator Results

Validator	Result	Details
Controlled vocabulary	Clean	871 entities, all valid
Cross-base consistency	371 warn, 167 info	70 title mismatches, 301 entities missing pages, 167 pages missing entities
Numeric ranges	Clean	No range violations
PG temporal	Clean	6,223 records across 9 tables
Soft FKs	60.7% link rate	51 dangling stableIds, 496 unresolved slugs, 4,366 display-name refs
Resource refs	98.5% valid	27 invalid publicationId refs (all open-philanthropy)

All 264 new validator tests pass

P1 — Actionable Data Issues

1. YAML relatedEntry type mismatches (7): us-executive-order, hardware-enabled-governance, monitoring, thresholds have wrong types on relatedEntries (e.g., compute-governance typed as policy but is concept)
2. 27 resources reference publicationId: "open-philanthropy" which doesn't exist in publications.yaml
3. 51 dangling stableIds in PG tables — personnel and grants referencing deleted/missing entities
4. 4 people role mismatches between YAML entities and FactBase (stale role descriptions)

P2 — Known Noise (not actionable now)

• Soft-FK 60.7% link rate is dominated by grants table display-name refs (by design for imported data)
• Cross-base 301 entities missing pages is expected — most are stableId-only TableBase records without wiki pages
• 70 title mismatches are mostly minor formatting differences (e.g., date ranges in page title but not entity title)

Confirmed Clean

• No duplicate wikiIds
• No broken refs to deleted entities
• No wrong domain references
• No temporal invariants violated (YAML or PG)
• No numeric range violations
• No controlled vocabulary violations
• All 170 tests pass
• Content gate passes

[OAGr]

QA Sweep — 2026-03-23

Focus: Full sweep | Depth: deep

Agents launched: 10 | Pages checked: ~80+

P0 — Active bugs (user-visible now)

#	Bug	URL or File
1	3 internal MDX files missing from database.json — wiki-nav test fails	wiki-nav.test.ts
2	"Test" stub entries in People directory — "Test greg-brockman" etc. on page 2	/people?page=2
3	Funding rounds show raw numeric IDs (175, 265, 335, 378) instead of company names	/funding-rounds
4	/publications index randomly redirects to people pages	/publications
5	Raw E-number entity IDs visible on benchmarks index	/benchmarks
6	Organizations index table shows Loading indefinitely	/organizations

P1 — Latent bugs

#	Bug	File:Line
7	Median off-by-one in secondary-market-prices.ts	:273
8	/timeseries total capped at query limit	:327
9	Entity resolution middleware silently passes unresolved entities	resolve-entity-middleware.ts:34
10	companyId vs companyEntityId filter mismatch	secondary-market-prices.ts:213
11	dispersionPercent null when min=0 (falsy check)	:286
12	as unknown as T double-cast on raw SQL	source-checks.ts:591
13	LLM verdict cast without runtime validation	source-check-orchestrate.ts:854
14	ProvisionCard details prop silently dropped	provision-card.tsx:16
15	Avg confidence includes null as 0, deflates average	entity-source-checks-viewer.tsx:460
16	/coverage missing secondary_market_prices table	source-checks.ts:577

P2 — UX / Data quality

#	Issue	Location
17	"Unmatched events" in Legislation Timeline (5 pages)	Multiple legislation pages
18	Hinton affiliation = Google DeepMind (resigned May 2023)	/people
19	Aschenbrenner affiliation = OpenAI (fired 2024)	/people
20	95/96 ARIA grant amounts null ($920K vs ~$75M actual)	/grants?funder=aria
21	Funding programs: Deadline column universally empty (60/60)	/funding-programs
22	N+1 per-row inserts in 6 research-areas sync endpoints	research-areas.ts
23	Research areas shows 18 not 54 (sub-areas hidden)	/research-areas
24	Evaluation/scores tables not visible (PR #3020 claim)	/research-areas/*
25	Common slug 404s: /benchmarks/gpqa, /benchmarks/math, /ai-models/mistral-large	Multiple
26	Bostrom stale role (FHI closed Apr 2024)	/people/nick-bostrom
27	Hogarth stale role (stepped down 2025)	/people/ian-hogarth
28	Altman net worth: $2.8B fact vs $10.5B wiki body	/people/sam-altman
29	Investments: ~86% rows have no amount	/investments
30	Anthropic funding: $67B org vs $62.4B rounds sum vs $124.4B directory total	Multiple
31	OpenAI has zero funding rounds in directory	/funding-rounds
32	DeepMind generic placeholder description	/organizations/deepmind
33	Open Philanthropy no Funding tab	/organizations/open-philanthropy
34	Leike title inconsistency	/people/jan-leike
35	platform filter not enum-validated	secondary-market-prices.ts:44
36	Evidence query in /verdicts has no LIMIT	source-checks.ts:242
37	Pagination "Showing 1-0 of 0" on empty search	entity-source-checks-viewer.tsx:455
38	20 NEEDS CITATION markers (19 cais.mdx, 1 ftx-red-flags)	content/docs/
39	Duplicate "Technical AI Safety RFP (2025)" entries	/funding-programs
40	/latest dedup clips at 50 rows	secondary-market-prices.ts:256
41	Related Org links not bidirectional (DeepMind→Anthropic/OpenAI)	Multiple
42	Multiple "Unknown" person entries across org pages	Multiple

Cross-consistency issues

#	Mismatch	Page A	Page B
C1	Anthropic total funding: $67B vs $62.4B vs $124.4B	/organizations/anthropic	/funding-rounds
C2	Greg Brockman sabbatical omitted	/organizations/openai	/people/greg-brockman
C3	Sutskever still in OpenAI founders without "former"	/organizations/openai	/people/ilya-sutskever
C4	Bengio: "Scientific Director" vs "Founder"	/people/yoshua-bengio	/organizations/mila
C5	SFF grants $151M vs org "distributed $141M"	/grants	/organizations/survival-and-flourishing-fund

Confirmed clean

• Benchmark scores consistent cross-page, all within valid ranges
• All wiki-server routes use Hono RPC method-chaining
• No SQL injection risks found
• No silent .catch(() => {}) patterns
• All interactive components have correct "use client"
• EU AI Act, US Executive Order, Colorado AI Act legislation pages high quality
• Grants (5,838), Divisions (66), Approaches (72) directories render correctly

Issues filed

• /publications index randomly redirects to people pages #3035: /publications index randomly redirects to people pages
• Test stub entries visible in People directory production #3036: Test stub entries visible in People directory production
• Funding rounds show raw numeric IDs instead of company names #3037: Funding rounds show raw numeric IDs instead of company names
• Raw E-number entity IDs visible on benchmarks index page #3038: Raw E-number entity IDs visible on benchmarks index
• Organizations index table shows Loading indefinitely #3039: Organizations index table shows Loading indefinitely
• secondary-market-prices: median off-by-one and companyId filter mismatch #3040: secondary-market-prices: median off-by-one and companyId filter mismatch
• Unmatched events rendering bug in Legislation Timeline tabs #3041: Unmatched events rendering bug in Legislation Timeline tabs
• People directory: stale affiliations for Hinton, Aschenbrenner, Bostrom, Hogarth #3042: People directory: stale affiliations for Hinton, Aschenbrenner, Bostrom, Hogarth
• research-areas.ts: N+1 per-row inserts in 6 sync endpoints #3043: research-areas.ts: N+1 per-row inserts in 6 sync endpoints
• ARIA grants: 95/96 amounts are null, total wildly understated #3044: ARIA grants: 95/96 amounts null, total understated
• Common URL slugs return 404 with no redirects #3045: Common URL slugs return 404 with no redirects
• ProvisionCard details prop silently dropped; provision-card.tsx #3046: ProvisionCard details prop silently dropped
• source-checks.ts: as unknown as T double-cast and missing coverage for new tables #3047: source-checks.ts: as unknown as T double-cast and missing coverage
• Funding programs: deadline column universally empty, duplicate RFP entries #3048: Funding programs: deadline empty, duplicate RFP entries

[OAGr]

QA Sweep Follow-Up — 2026-03-23

Depth: deep (targeted) | 8 agents launched, 5 returned results

New P0 findings

#	Bug	URL or File
1	E2100 wikiId collision — renders Gabriel Alfour person profile instead of Verification Coverage dashboard	/wiki/E2100

New P1 findings

#	Bug	File or URL
2	Mermaid diagrams fail to render (show "loading...") on at least 3 wiki pages	/wiki/E42, /wiki/E477, /wiki/E200
3	Footnotes show "(data unavailable — rebuild with wiki-server access)"	/wiki/E218 (OpenAI)

New P2 findings

#	Issue	Location
4	Tab-specific URLs all 404 (e.g. /organizations/anthropic/announcements)	All org tab URLs
5	No server-side pagination on people (632 rows) or embedded grants tables (2,632 rows)	/people, /organizations/coefficient-giving
6	Page Changes dashboard (E909) shows 0 sessions	/wiki/E909
7	Update Schedule dashboard (E900) shows 0 results	/wiki/E900
8	Squiggle components show "Loading Squiggle..." on Anthropic wiki page	/wiki/E22
9	E2E tests don't run on PRs, only post-deploy; major feature gaps in coverage	.github/workflows/

Root cause confirmed

• database.json stale after source-check rename (PR refactor: rename verification system to source-check (v2) #2999) — pnpm build-data:content fixes wiki-nav test failure
• E2200 is NOT a collision — intentional rename, just stale build artifacts

3 agents hit usage limits (no results)

• FactBase pages, Search functionality, External link rot

Issues filed

• E2100 wikiId collision: Verification Coverage dashboard clobbered by person entity #3055: E2100 wikiId collision
• Mermaid diagrams fail to render on multiple wiki pages (show loading placeholder) #3056: Mermaid diagrams fail to render
• Footnotes show 'data unavailable — rebuild with wiki-server access' on OpenAI wiki page #3057: Footnotes show 'data unavailable' on OpenAI page
• E2E tests not running on PRs — only post-deploy to production #3058: E2E tests not running on PRs

[OAGr]

Status Update — 2026-03-23

Tracking resolution of all issues filed across the 3 QA sweeps.

March 18 sweep: 15/15 closed ✅

All issues (#2633–#2647) resolved. Benchmark links, research area 404s, ghost entity IDs, garbled names, financial inconsistencies, Hono RPC compliance — all fixed.

March 23 sweep: 9/14 closed

#	Status	Title
#3036	✅ Closed	Test stub entries in People directory
#3037	✅ Closed	Funding rounds show raw numeric IDs
#3038	✅ Closed	Raw E-number IDs on benchmarks index
#3039	✅ Closed	Organizations index Loading indefinitely
#3040	✅ Closed	secondary-market-prices median off-by-one
#3042	✅ Closed	Stale affiliations (Hinton, Aschenbrenner, etc.)
#3045	✅ Closed	Common slug 404s
#3046	✅ Closed	ProvisionCard details prop dropped
#3047	✅ Closed	source-checks.ts double-cast
#3035	⬚ Open	/publications index randomly redirects
#3041	⬚ Open	Unmatched events in Legislation Timeline
#3043	⬚ Open	research-areas.ts N+1 per-row inserts
#3044	⬚ Open	ARIA grants: 95/96 amounts null
#3048	⬚ Open	Funding programs: deadline empty, duplicate RFP

March 23 follow-up: 1/4 closed

#	Status	Title
#3055	✅ Closed	E2100 wikiId collision
#3056	⬚ Open	Mermaid diagrams fail to render
#3057	⬚ Open	Footnotes show data unavailable on OpenAI page
#3058	⬚ Open	E2E tests not running on PRs

Summary: 25/30 issues closed (83%)

8 remaining open issues are P1/P2. No P0 blockers remain.

[OAGr]

QA Sweep — 2026-03-30

Focus: Full sweep (all directories) | Depth: Exhaustive

Agents launched: 15 | Pages checked: ~200+ | Sweep ID: sweep-2026-03-30-ex01

---

P0 — Active bugs (user-visible now)

#	Bug	URL	Root Cause
1	[object Object] in stat cards (Anthropic, OpenAI, Coefficient Giving)	/organizations/anthropic etc.	Stale production deploy
2	Raw unformatted integers (xAI: 3800000000, Meta AI: 200970000000)	/organizations/xai, /organizations/meta-ai	Same stale deploy
3	Raw stableIds as org names on 6+ person pages	/people/geoffrey-hinton etc.	resolveEntityRef can't resolve bare 10-char IDs
4	"With Employer: 0" — employer filter completely broken	/people	refPattern misses sid_ prefix
5	Raw stableId gNsqAes7Dw as funder name for 825 CEA grants	/grants	Entity lookup fails
6	5 projects show raw stableId hashes as org names	/projects	YAML uses bare stableIds

P1 — Latent bugs / significant data issues

#	Bug	URL / File
7	Literal "false" in funding rounds valuation (~13 rows)	/funding-rounds
8	Timeline tab shows count but renders empty	/legislation/eu-ai-act etc.
9	EU AI Act pipeline bar stuck at stage 0	/legislation/eu-ai-act
10	60% legislation stakeholders unlinked (86/144)	/legislation/*
11	Ian Hogarth stale "Chair" role	/people/ian-hogarth
12	MIRI funding $15M on prod vs $55M in FactBase	/organizations/miri
13	Personnel /broken endpoint unbounded	personnel.ts:272
14	Grants /by-org-summary no LIMIT	grants.ts:303
15	Equity-positions sync missing audit trail	equity-positions.ts:228
16	Monitoring /status sequential queries	monitoring.ts:81
17	Worker timer leak in Promise.race	worker/run.ts:486
18	Silent .catch(() => {}) in worker shutdown	worker/run.ts:631
19	BrowseComp benchmark 404 at /benchmarks/browse-comp	/benchmarks/browse-comp
20	Intuitive research area slugs 404	/research-areas/agentic-ai etc.
21	Sub-areas show 0 orgs / 0 papers universally	/research-areas/*
22	Apollo Research founding date conflict	/wiki/apollo-research
23	18 NEEDS CITATION markers in cais.mdx	cais.mdx
24	Claude Opus 4.5 likely wrong safety level	/ai-models/claude-opus-4-5
25	ARC-AGI-2 has only 1 model entry	/benchmarks/arc-agi-2
26	Nate Thomas missing from Redwood bidirectionally	/people/nate-thomas
27	Board members missing from person pages	/people/holden-karnofsky etc.
28	Duplicate import in claim-verification.ts	claim-verification.ts:25
29	QA-checks pagination broken	qa-checks.ts:140
30	Legacy grant redirects 404 for CEA	/grants/[id]
31	No author names on publications	/publications/*

P2 — UX improvements / data quality (32 findings)

Key items: stale personnel (Brockman, LeCun), birth year errors (Sutskever, Brockman), founding date inconsistencies (Anthropic, Open Philanthropy, ARC Evals), financial data conflicts (SFF, FTX Future Fund, Anthropic duplicate rounds), accessibility gaps (ARIA on tabs/tables/filters site-wide), unsafe as casts, missing error boundaries, investments mostly blank, empty legislation tabs, Claude safety levels need review.

Confirmed clean

• 213 tests pass, content gate clean
• No duplicate wikiIds (1066 entities)
• Currency formatting correct throughout
• Model release ordering verified (Anthropic, OpenAI, DeepMind, DeepSeek)
• Bidirectional links verified for 15+ major people-org pairs
• Skip navigation, heading hierarchy, SVG aria-hidden correct
• SB 1047, EO 14110, Colorado AI Act timelines accurate

Issues filed

• P0: [object Object] and raw integers in org stat cards — stale production deploy #3475: P0 — [object Object] and raw integers in org stat cards
• P0: Raw stableIds displayed as org/funder names across people, grants, projects #3476: P0 — Raw stableIds as org/funder names
• P0: People directory employer filter shows 0 — completely broken #3477: P0 — People directory employer filter broken
• P1: Legislation timeline tab renders empty — undated resources inflate count #3480: P1 — Legislation timeline tab empty
• P1: EU AI Act pipeline bar broken — EU legislative labels not in TIMELINE_LABELS #3481: P1 — EU AI Act pipeline bar broken
• P1: Funding rounds valuation column renders literal 'false' for ~13 rows #3482: P1 — Funding rounds valuation "false"
• P1: Worker timer leak and silent .catch() in crux/worker/run.ts #3483: P1 — Worker timer leak and silent catch
• P1: 60% of legislation stakeholders have no entity link (86/144 unlinked) #3484: P1 — Legislation stakeholders unlinked
• P1: Personnel /broken endpoint unbounded + monitoring /status sequential queries #3485: P1 — Unbounded routes (personnel, grants, monitoring)
• P1: Research area intuitive slugs 404 (agentic-ai, mechanistic-interpretability, ai-evaluations) #3486: P1 — Research area intuitive slugs 404
• P1: Sub-areas universally show 0 orgs / 0 papers — hollow research area data #3487: P1 — Sub-areas hollow data
• P1: Ian Hogarth stale 'Current Role: Chair' + equity-positions missing audit trail #3488: P1 — Ian Hogarth stale role + equity-positions audit trail
• P1: Missing bidirectional people-org links (Redwood Research, board members) #3489: P1 — Missing bidirectional people-org links
• P2: Birth year conflicts, stale personnel data, founding date inconsistencies #3490: P2 — Data quality (birth years, stale personnel, founding dates, financial conflicts)
• P2: Accessibility gaps — missing ARIA attributes on tabs, tables, filters site-wide #3491: P2 — Accessibility gaps (ARIA, tables, filters)
• P2: Unsafe as casts, missing error boundaries, component code quality #3492: P2 — Code quality (unsafe casts, missing error boundaries)
• P2: Investments, funding programs, and grants directory gaps #3493: P2 — Financial directory gaps
• P2: Publications missing authors, legislation missing provisions/stakeholders #3494: P2 — Content coverage gaps (publications, legislation, AI models)

[OAGr]

QA Sweep — 2026-04-02 (Post-Deploy)

Focus: Full site | Depth: deep (exhaustive coverage)

Agents launched: 8 | Pages checked: ~80+ (14 index pages + 60+ detail pages)

Sweep ID: sweep-2026-04-02-post-deploy

Context: Extensive sweep after deploying 15 PRs including politicians directory, data sources move to public, FBCompareChart, platform accounts table, FactBase enrichment, and snapshot capture fixes.

---

P0 — Active bugs (user-visible now)

#	Bug	URL or File	Fix
1	People detail pages show raw stableId strings instead of org names — "puAffUjWSS" instead of "MIRI", "y4bieqSeag" instead of "CAIS". Affects most people pages with employer references.	directory-utils.ts:71 fallback	Fix resolveEntityRef() to handle unresolved refs gracefully
2	Wiki-nav test suite broken — missing vi.mock("@/data/factbase") + stale data-sources-dashboard expectation. All 27 wiki-nav tests failing.	wiki-nav.test.ts	Add mock, update dashboard reference

P1 — Latent bugs (will surface under conditions)

#	Bug	URL or File
3	Politicians: ~7 missing party — inferParty only checks description, not customFields	politicians/page.tsx:16-21
4	Politicians: Gavin Newsom classified as "candidate" — inferStatus doesn't recognize "governor"	politicians/page.tsx:53-59
5	GiveWell revenue displayed as $1.3M (actual ~$90-100M/year) — misleading FactBase entry	/organizations/givewell
6	ILIKE wildcard injection in source-checks search — q param not escaped with escapeIlike()	source-checks.ts:291
7	Platform accounts DELETE lacks logging; /all returns misleading total (page size, not COUNT)	platform-accounts.ts:129,167
8	/data-sources redirect lands on wrong tab — ?tab=data-sources query param ignored	/sources page
9	Data Sources dashboard missing from internal sidebar	wiki-nav.ts
10	/internal/source-check-coverage has component but missing page.tsx — returns 404	internal/source-check-coverage/
11	Missing Anthropic Series B ($580M, Apr 2022) from /funding-rounds directory	/funding-rounds
12	FactBase entity underscore URLs 404 — /factbase/entity/open_philanthropy fails, hyphen works	/factbase/entity/ routing
13	AI model family rows indistinguishable from individual models — no visual differentiation	/ai-models
14	13 benchmarks with 0 model scores listed alongside populated ones with no distinction	/benchmarks
15	Research areas "Declining"/"Archived" filter tabs have 0 entries	/research-areas
16	All 53 RAND resources have missing publication dates	/publications/rand
17	Groundskeeper tasks job-worker-health and snapshot-retention failing at 100%	System health dashboard
18	Events: International AI Safety Summit Series missing status field	/events

P2 — UX improvements

#	Issue	URL or Location
19	Grant amounts rounded too aggressively ($6.6M → $7M)	format-compact.ts:11
20	Open Philanthropy Wikipedia link points to doctrine article, not organization	/organizations/open-philanthropy
21	/about returns 404 (no redirect to /wiki/E755)	/about
22	/races page empty but linked in main nav Policy dropdown	/races
23	Source-checks Record column shows raw fact IDs instead of human-readable names	/source-checks
24	Literal \n\n in grant notes on detail pages	/organizations/*/grants/*
25	Currency mixing: "$14M" for GBP amounts on ARIA program	/funding-programs/-C8O_gr9Bj
26	/sources page ~17MB HTML payload	/sources
27	Data sources public page still has robots: { index: false }	data-sources/page.tsx:8
28	Default org sort by revenue shows 94% empty rows	/organizations
29	Division detail pages all very thin ("No detailed data available")	/divisions/*
30	20 Civic Data publications with 0 resources clutter listing	/publications
31	Unrendered markdown in resource titles ("Future of Humanity Institute")	/sources
32	Politicians: sparse position columns (>70% empty), no legend for stance labels	/politicians
33	Approaches: tags not clickable/filterable despite 216 unique tags	/approaches
34	People: Role column ~9% filled, Affiliation ~6% — table looks hollow	/people
35	Politician detail pages don't show office/party/status from index table	/people/*
36	Investments: 15 "Unknown" company entries, many missing amounts	/investments
37	Unused import STATUS_COLORS in politicians-table.tsx	politicians-table.tsx:10

Cross-consistency checks

Check	Result
Organizations ↔ Grants (Open Philanthropy)	CONSISTENT (name change to Coefficient Giving reflected correctly)
People ↔ Organizations (Anthropic)	CONSISTENT (all 7 checked match bidirectionally)
Organizations ↔ Divisions	CONSISTENT (7 Anthropic divisions match both pages)
Organizations ↔ Funding Rounds	INCONSISTENT — missing Anthropic Series B ($580M)

Confirmed clean

• Homepage loads correctly with all expected content
• All 14 directory index pages return 200
• 404 error pages properly styled with site navigation
• Data dropdown exists in nav with Data Sources link
• Politicians page works (65 entries, clickable names)
• Content gate: all 40 checks pass
• No duplicate wikiIds
• No broken entity refs
• No wrong domain references
• YAML field consistency clean
• People role consistency (YAML ↔ FactBase match)
• Temporal invariants clean
• Quality discrepancies clean
• 80,000 Hours org page: excellent reference implementation
• Anthropic org page: comprehensive with 10 tabs

Issues filed

• P0: People detail pages show raw stableId strings instead of organization names #3619: P0 — People detail pages show raw stableId strings
• Wiki-nav test broken: missing vi.mock for @/data/factbase #3620: P0 — Wiki-nav test broken
• Politicians directory: inferParty and inferStatus have gaps — 7 missing party, Newsom misclassified #3621: P1 — Politicians inferParty/inferStatus gaps
• GiveWell revenue displayed as $1.3M — likely incorrect FactBase entry #3622: P1 — GiveWell revenue $1.3M incorrect
• ILIKE wildcard injection in source-checks search endpoint #3623: P1 — ILIKE injection in source-checks
• /data-sources redirect lands on wrong tab — ?tab=data-sources query param ignored #3624: P1 — /data-sources redirect wrong tab
• Platform accounts endpoint issues: DELETE lacks logging, /all returns misleading total, missing input validation #3625: P1 — Platform accounts endpoint issues
• Grant amounts rounded too aggressively — $6.6M shown as $7M #3626: P2 — Grant amounts rounded too aggressively
• Missing Anthropic Series B ($580M, Apr 2022) from /funding-rounds directory #3627: P1 — Missing Anthropic Series B from funding-rounds
• /races page empty but linked in main nav Policy dropdown #3628: P2 — /races page empty but in nav
• Source-checks Record column shows raw fact IDs instead of human-readable names #3629: P2 — Source-checks raw fact IDs
• AI models: family rows indistinguishable from individual models; benchmarks: 13 entries with 0 scores #3630: P2 — AI models family rows / benchmarks empty entries
• Literal backslash-n in grant notes on detail pages #3631: P2 — Literal \n in grant notes
• Currency mixing on funding programs: dollar sign used for GBP amounts #3632: P2 — Currency mixing on funding programs
• FactBase entity pages: underscore URLs return 404, only hyphen format works #3633: P1 — FactBase underscore URLs 404
• Data Sources dashboard missing from internal sidebar; /internal/source-check-coverage missing page.tsx #3634: P1 — Internal sidebar/route gaps
• Open Philanthropy Wikipedia link points to wrong article (doctrine, not organization) #3635: P2 — Open Philanthropy wrong Wikipedia link

[OAGr]

Branch: main

QA Sweep — 2026-04-19

Focus: full sweep | Depth: standard

Agents launched: 6 | Pages checked: 27 (14 index + 13 detail) | Issues filed: 12

Context

Last QA sweep was ~30 days ago (claude/qa-sweep-top5-fixes → PR #2813). This is the first broad sweep since. Gate passes all 50 checks. pnpm test passes on fresh run (317 files / 2 skipped / 6952 tests, 26 skipped). The failure counts surfaced by pnpm crux qa-sweep (8 test files, 38 tests) are false positives from a substring matcher — see QUA-627.

P0 — Active bugs (user-visible on prod)

#	Bug	URL / File	Linear
1	Microsoft CEO factual error: FactBase says Mustafa Süleyman, actual CEO is Satya Nadella	packages/factbase/data/fb-entities/microsoft.yaml	QUA-618
2	Anthropic founding date inconsistent: /organizations/anthropic says Jan 2021, /people/daniela-amodei says Dec 2020	/organizations/anthropic vs /people/daniela-amodei	QUA-619
3	Ada Lovelace Institute currency mismatch: overview says $5M, /data says £5M / $6.7M	/organizations/ada-lovelace-institute	QUA-620
4	Total funding doesn't add up: OpenAI stated $57.9B vs sum $59.9B; Anthropic stated $67B vs sum $48.4B	/organizations/openai, /organizations/anthropic	QUA-621

P1 — Latent bugs

#	Bug	File	Linear
5	publications.ts applies filters AFTER .limit() — wrong results + wrong total	apps/wiki-server/src/routes/tablebase/publications.ts:107-179	QUA-622
6	Unbounded route queries (5 routes): entity-resources /export, research-areas /:id subqueries, integrity /, agent-sessions /by-linear/:id, resource-dedup buildReport	multiple	QUA-623
7	Entity detail tab URLs 404: /organizations/<slug>/people and /people/<slug>/publications (all orgs / all people). Not currently linked; latent.	Next.js routes missing	QUA-624

P2 — Data quality / UX

#	Issue	Linear
8	18 FactBase person-property facts use text instead of !ref person:<slug>	QUA-625
9	YAMLWarning: Unresolved tag !date spams test output — CUSTOM_TAGS not passed outside factbase loader	QUA-626
10	qa-sweep test-failure detection uses substring match on 'failed' — false-positive failure counts	QUA-627
11	qa-sweep scanner flags cli-reference.mdx for its own self-documenting examples	QUA-628
12	validate-tablebase-completeness advisory false-positive for scanner-results.ts (metrics table)	QUA-629

Data completeness (index pages)

Page	Rows	Notable empty columns
/organizations	566	headcount 95% null; valuation, totalFunding 100% null
/people	924	99% stubs: 4/924 role-data, 5/924 employer, 55/924 birth-year, 4/924 net-worth
/ai-models	47	clean
/benchmarks	46	clean
/legislation	38	coverage sparse
/projects	62	clean
/approaches	82	clean
/events	3	only 3 events — content gap
/grants	5,876	status 100% null
/publications	152	"Low credibility" filter tab has 0 entries
/funding-programs	59	clean
/funding-rounds	87	valuation 72% null (expected for grants/convertibles)
/divisions	99	clean
/research-areas	50	clean

Observation (not filed): /people's stat cards publicly advertise thin coverage. Consider deferring the numbers until coverage climbs.

Cross-consistency

• Org↔People bidirectional links mostly consistent for Anthropic and OpenAI.
• Funding-rounds directory row-matches org page rounds for Anthropic Series B–G.
• Role-title format drifts within single pages (minor).

Confirmed clean

• All 50 gate checks pass; pnpm build/pnpm test green on fresh runs.
• No raw sid_xxx, E###, [object Object], undefined, null, or NaN leaked to user-visible cells on any sampled directory or detail page.
• /people/max-tegmark, /people/melissa-bean, /organizations/chips-alliance, /organizations/mats, /events/anthropic-government-standoff — clean.
• Wiki-server routes operational/active-agents.ts, shared/resolve-resource-id.ts, sourcing/url-suggestions.ts, tablebase/entity-profile.ts, tablebase/bluesky.ts, tablebase/data-sources.ts, wikibase/citations.ts, wikibase/resource-dedup.ts — audited, no P0/P1 findings.
• Migration helper swap_fk_target() (QUA-600 / migration 0199) correctly handles both _fk and _fkey constraint-name conventions via information_schema joins on columns.

Coverage updated

Sweep ID sweep-2026-04-19-35e6. 27 pages recorded (14 index + 13 detail). Pre-sweep QA coverage was 0% across all directories; post-sweep coverage is still heavily biased — next sweep should pull from crux qa-checks queue for a fresh slice.

Issues filed

• QUA-618 through QUA-629 (12 issues)

[OAGr]

Branch: main

---

Addendum — one more finding

Post-report: the crux qa-checks coverage command itself has two bugs:

1. Literal %-20s %8s format strings printed (Node's console.log doesn't support width specifiers)
2. checkedPages stays 0 after qa-checks record unless --thing-id is supplied

Filed as QUA-630. Total issues filed: 13 (QUA-618 through QUA-630).

[OAGr]

Branch: main

QA Sweep — 2026-04-24

Focus: all tabs of /organizations/anthropic | Depth: deep (focused)

Agents launched: 3 (2 API-errored, 1 returned) + direct curl verification | Pages checked: 5 orgs

Context: Org profile pages had a major restructure on 2026-04-23/24 — grouped left-sidebar tabs (PR #4553, #4552), EntityTimeline QUA-660 (PR #4561), entity profile polish QUA-668/669/670/671 (PR #4555), raw-number Database tab fix QUA-673 (PR #4562), sid_ leak fixes QUA-657 (PR #4558). Anthropic is the most data-rich org, so a deep audit of it also surfaces cross-org regressions.

P1 — Latent bugs / user-visible inconsistencies

#	Bug	URL	Fix
1	Org Overview stat cards render 3 different currency formats: $22,060 (Redwood), $19 billion (Anthropic revenue), $380 billion (Anthropic valuation). Chart captions use $22K / $19B.	/organizations/redwood-research, /organizations/anthropic	QUA-681 filed. formatKBFactValue in apps/web/src/data/factbase.ts switches style by magnitude; standardize on compact-short.

Already-fixed on main (ship next release)

Commit	Issue	What it fixes
f64241f (#4555)	QUA-668/670	SourcingDot size=md was 7px → fixed to 10px, adds role="img" + aria-label="Sourcing: ..." (currently invisible to screen readers on prod)
59d3239 (#4558)	QUA-657	sid_ leaks in FBF / ResourceLink / FBFactValue fallbacks
a305c4d (#4562)	QUA-673	Raw large numbers on /organizations/:slug Database tab — but does NOT fix Overview stat card path (see QUA-681)
7708fe7 (#4561)	QUA-660	EntityTimeline component for entity_events

Confirmed clean on Anthropic prod

• HTTP 200, 961KB payload
• All 6 tab groups render: Entity, About, Business, Policy & Governance, Output & Research, Data
• Tab counts all populated: People (25), Divisions (8), Funding Rounds (7), Shareholders (13), Market Data (35), Policy Positions (6), Publications (32), Announcements (90)
• Coverage popover works (aria-label="Coverage: Comprehensive (4/4)")
• All 5 hero stat cards populate (Revenue $19B, Valuation $380B, Headcount, Total Funding Raised $67B, AI Models)
• Charts use compact aria-labels ("Revenue (ARR) chart. $100M in 2023 to $19B in 2026.")
• Unknown-tab banner works — ?tab=bogus-tab-name shows amber banner (6 amber CSS classes present)
• No [object Object], no unresolved component names, no stack traces
• Tab state syncs to ?tab= URL query param
• Related Wiki Pages section renders with entity type pills
• /organizations/anthropic/nonexistent-tab correctly 404s (tabs use ?tab= query, not sub-routes)

Intentionally-visible sid_ (NOT a bug)

FactBaseEntityBody.tsx:308 renders a <details> collapsible with an "ID: sid_mK9pX3rQ7n" / "Stable ID:" / "Wiki ID:" metadata block on the Facts tab. Per QUA-671 design comment, this is intentional debug metadata in a non-default-open details element, with ID: separator text for screen-reader compatibility. Not a display regression.

/organizations/open-philanthropy — intentional stub, not a bug

The old slug is retained as a pageType: stub with a redirect note "Open Philanthropy rebranded to Coefficient Giving in November 2025. See the Coefficient Giving page for current information." Coefficient Giving (the current org) has 2626 Grants Made + 5 Grants Received. Soft-redirect by description is acceptable; could consider a permanentRedirect to /organizations/coefficient-giving as a polish item but not filed.

Cross-consistency agent findings not reproduced

The subagent cross-consistency pass flagged "FactsDatabase" concatenation for Wild Animal Initiative (WAI) and raw E<N> / LlmdDoyjoF-style bare10 tokens on OpenAI/Redwood/OpenPhil. Direct curl verification against /organizations/anthropic and /organizations/redwood-research did NOT reproduce these — likely WebFetch SSR artifacts. Defer to a future sweep with Playwright rendering.

Deterministic qa-sweep results

• 1 failed (wrong domain ref in cli-reference.mdx:772 — longtermwiki.org, content-docs only)
• 3 warnings: 20 entity data integrity warnings, 3 financial staleness warnings, 1 NEEDS CITATION marker
• 3 skipped: test suite timed out, orphan PG + cross-entity-consistency parse errors
• Recent: 15 commits on main not yet deployed to production (fix: entity profile polish — unknown-tab banner, chrome, facts, AI table (QUA-668, QUA-669, QUA-670, QUA-671) #4555-fix(QUA-673): format raw large numbers on org Database tab #4562)

Issues filed

• QUA-681 — Org Overview stat cards render inconsistent currency formats ($22,060 vs $19 billion vs $22K)

Coverage updated

4 pages recorded under sweep-id sweep-2026-04-24-anthropic:

• /organizations/anthropic → clean (deep)
• /organizations/redwood-research → issues_found (standard)
• /organizations/open-philanthropy → clean (standard)
• /organizations/coefficient-giving → clean (quick)

[OAGr]

Branch: main

QA Sweep — 2026-04-30 — Tablebase exhaustive

Sweep ID: sweep-2026-04-30-tb01
Focus: all 14 tablebase directories (index + detail + /data sub-pages)
Depth: exhaustive
Agents launched: 15
Pages checked: ~95 detail pages + 14 directory index pages + 10 /data subpages + 6 component files + 9 wiki-server route files

Run from slot a6, on main @ c107beafc.

Updated coverage

Directory	Total	Checked	Coverage
organizations	598	17	3%
people	2099	14	1%
ai-models	54	11	20%
benchmarks	46	10	22%
legislation	39	10	26%
projects	62	9	15%
approaches	82	6	7%
events	10	6	60%
research-areas	9	6	67%
publications, divisions, funding-programs, funding-rounds, grants	various	sampled per dir	—

---

P0 — Active bugs (user-visible now) — 4 issues

#	Linear	Bug
1	QUA-897	/divisions header: literal "120 of 101 records sourcinged" (typo + impossible ratio)
2	QUA-898	/grants index has no SSR — full table is Loading grants table… for crawlers
3	QUA-899	/projects Organization column shows raw kebab-case slugs for 4 entities (Talk to the City, Polis, Democracy Levels Framework, Case Law for AI)
4	QUA-900	100%-empty Coverage/Status columns on /research-areas, /funding-programs, /publications (95-97% empty on /projects, /approaches, /divisions)

Plus the already-filed QUA-892 "CI broken on main after c107bea (from PR #4723)" — TypeScript type-check failing on main.

P1 — Latent bugs (data correctness, will surface) — 14 issues

#	Linear	Bug
5	QUA-901	Sourcing rollup dot missing on /people, /ai-models, and 5 /data subpages — EntityDataPage skips shell + /people/[slug]/page.tsx and /ai-models/[slug]/page.tsx never call fetchEntitySourcingSummary and never pass verdict={...}. Empty <a> anchors on /sourcing/personnel/<id> (sam-altman, dario-amodei) and /sourcing/division/<id> (openai Preparedness/Safety Systems/Superalignment)
6	QUA-902	Raw sid_<id>.<field> → leaks as visible link text on /people/sam-altman Quick Assessment ("sid_1LcLlMGLbw.valuation →")
7	QUA-903	/people/dario-amodei/data shows "80%–90%Apr 2026" — value+asOf concatenation in <FBFactValue> formatter
8	QUA-904	"Quality: N/100" internal score leaks into public /data and /approaches pages (sam-altman/data, public-education, constitutional-ai, anthropic-government-standoff, international-summits)
9	QUA-905	/people/stuart-russell Wikipedia URL points to Stuart_Russell_(politician) — wrong subject
10	QUA-906	Raw E<number> IDs leak as link text on /publications/lesswrong, /publications/ea-forum, /divisions/safeguarded-ai-programme, /funding-programs/pJ9oHvQ1Bb
11	QUA-908	/funding-programs and /funding-rounds use 10-char raw stableIds as URL slugs (e.g. /funding-programs/pJ9oHvQ1Bb)
12	QUA-909	/benchmarks/arena-hard has empty leaderboard ("No model scores recorded for this benchmark yet.")
13	QUA-910	/research-areas/evals grants table includes off-topic malaria/global-health grants — broken filter scope
14	QUA-911	/approaches detail pages leak editorial metadata (TODOs, <R> literal text, "Diagram (loading…)" placeholders, broken citations on constitutional-ai)
15	QUA-912	18 entities have wikiId set in YAML but no matching MDX file — broken /wiki/E<N> 404 on any reference
16	QUA-913	9 wiki-server route files have hand-rolled tx without applyAuditContext(tx, c) — degraded audit trail attribution (grants, website-sources, political-races, personnel, scanner-results, research-areas, prediction-markets, ids, bluesky, sourcing)
17	QUA-915	Cross-page consistency mismatches: John Schulman has zero org affiliations on his person page despite being Anthropic researcher + OpenAI co-founder; Anthropic founder list diverges between overview (7) and people tab (8); gpt-4o-mini missing from OpenAI's named model list
18	QUA-916	4 directory index pages render only Loading... for non-JS clients: /organizations, /people, /benchmarks, /grants

P2 — UX / quality / cleanup — 10 issues

#	Linear	Issue
19	QUA-907 (skipped — assigned elsewhere)	(slot reserved)
20	QUA-914	N+1 INSERT loops in /sync-pages (website-sources) and /candidates/sync (political-races) — should use createSyncHandler or bulk SQL
21	QUA-917	sourcing-coverage-content.tsx color-map duplication + "sourcinged" typo (line 396)
22	QUA-918	Filter-tab labels concatenate count without separator (enacted18, Convertible_note3, Leaderboard2 models, etc.)
23	QUA-919	Filter-tab anti-patterns on /projects, /approaches, /events, /research-areas (static labels, tautology tabs, single-entry tabs)
24	QUA-920	/legislation/<slug>/data breadcrumb plural is wrong ("Legislations" → "Legislation"); california-sb243 title duplicates "SB 243"
25	QUA-921	/people directory sparse — role 26%, affiliation 7%; many stub pages with no positions/roles
26	QUA-922	Stale net-worth citations on dustin-moskovitz, elon-musk pages (cite 2011-2018 only)
27	QUA-923	False positives in NEEDS-CITATION + wrong-domain checkers on cli-reference.mdx (own help text)
28	QUA-924	/api/source-checks/* legacy alias documented but not mounted — rule doc drift
29	QUA-925	crux qa-sweep deep --json truncates output and exits 1 — Phase 1.5 leads-extraction blocked

---

Confirmed clean

• All 8 /ai-models/<slug> detail pages render correctly (one suspected "Tiersonnet" concat needs Playwright verification — low confidence finding)
• All 8 /projects/<slug> detail pages render correctly (modulo missing sourcing rollup which is tracked under QUA-901 / QUA-362 epic)
• All 5 /research-areas/<slug> detail pages render correctly (clean modulo /evals filter-scope issue tracked under QUA-910)
• validate-rendered-sid clean (no sid_* leaks in built data)
• validate-display-names clean (no raw machine IDs in display fields)
• validate-display-formatting clean (no obvious formatting issues caught structurally — but a couple of new patterns above should be added to the validator)
• All 5 /funding-rounds/<id> detail pages render correctly (modulo URL-slug issue tracked under QUA-908)
• 5 of 8 /legislation/<slug> pages render correctly
• All 5 /events/<slug> detail pages render correctly (modulo Quality-score leak tracked under QUA-904)
• 9 of 10 /people/<slug> flagship pages render correctly (sam-altman, dario-amodei have the empty-anchor + sid-leak issues already filed)
• 9 of 10 /organizations/<slug> flagship pages render correctly (only /openai has the empty-division-anchor issue)

Cross-cutting / systemic findings

1. Coverage column infrastructure is broken across 6 directories. Either drop the column from those directories or wire scoring via apps/web/src/components/coverage/coverage-score.ts + dispatch in entity-coverage.ts::computeEntityCoverage(). Tracked in QUA-900.
2. EntityProfileShell is being called without verdict prop on people / ai-models — separate from the QUA-362 epic, which tracks pages that haven't migrated AT ALL. This is a "shell migrated, but verdict not wired" defect class. Tracked in QUA-901.
3. Sourcing-related component churn left visible debris: "sourcinged" non-word in 2 places (/divisions header, sourcing-coverage-content.tsx:396); 9 hand-rolled tx routes missing the audit-context call; documented /api/source-checks/* alias not actually mounted. Tracked across QUA-897, QUA-913, QUA-917, QUA-924.
4. 4 directory pages render only Loading... for non-JS clients — same root cause likely (client-side fetch with no SSR fallback). Affects SEO, sharing, accessibility. Tracked in QUA-916 (P0/P1 depending on directory).
5. Editorial pipeline metadata leaking to public pages: "Quality: N/100" scores, TODOs, "Overdue by N days" badges, raw <R> component tags rendered as text. Affects /approaches/* and several /data views. Tracked in QUA-904 + QUA-911.

Notes on methodology

• Phase 1.5 qa-sweep deep --json failed (separately filed as QUA-925); fell back to running the validators directly via the leads investigation agent
• WebFetch cannot verify small visual indicators (SourcingDot, CoverageDots) since they're SVG circles with no text; "missing" findings are based on structural component-code analysis, not visible-pixel verification
• A separate Playwright spot-check is recommended for: /ai-models/claude-sonnet-4-6 "Tiersonnet" anomaly; tab content on /legislation/<slug> pages (Provisions, Timeline are client-rendered)

Issues filed: 28

QUA-897, QUA-898, QUA-899, QUA-900, QUA-901, QUA-902, QUA-903, QUA-904, QUA-905, QUA-906, QUA-908, QUA-909, QUA-910, QUA-911, QUA-912, QUA-913, QUA-914, QUA-915, QUA-916, QUA-917, QUA-918, QUA-919, QUA-920, QUA-921, QUA-922, QUA-923, QUA-924, QUA-925

🤖 Generated by /maintain-qa-sweep — sweep-2026-04-30-tb01 from slot a6