Changelog 2026-06-10 (second batch) — EDGAR-only structured-data pipeline

User-approved Tier 2 change. Implements hard-change recommendation #3 ("structured data only — no free-text figures") from the 2026-06-10 kit review, on an EDGAR-only basis (user decision: no paid connectors; Daloopa free tier and bigdata.com declined; SEC EDGAR is free and sufficient).

What shipped

1. _Tools/edgar/ toolkit (new):
   • edgar_normalize.py — normalizer: raw XBRL companyconcept JSON → uniform alphasteve-edgar-fundamentals-v1 format (JSON + citable markdown). Hard entity gate (CIK mismatch = no output), point-in-time dedupe with restatement flags, figure-level citations (form + accession + filed date + tag), DERIVED labels on computed metrics.
   • concept-map.json — canonical concept → prioritized us-gaap tag lists (19 concepts + 4 derived).
   • cik-registry.json — resolver-generated entity cache (seeded: MP, CAG).
2. edgar-fundamentals-standard (new, 08-Frameworks) — the three-step protocol: atom resolver → companyconcept fetch → normalize. Includes what XBRL does not cover (non-GAAP, segments, guidance → read the filing) and refresh discipline.
3. data-integrity-gate amended — Rule 2 rerouted from "Daloopa/bigdata connectors" to the EDGAR pipeline; Rule 3 changed from hand-typed entity confirmation to resolver-generated; checklist updated.
4. Scheduled-task prompts updated — thesis-builder, AM scan, PM scan, intraday monitor now route fundamentals through the pipeline.

Validation performed at adoption

• Full pipeline run on MP: entity gate passed (CIK 0001801368 = MP Materials Corp. / DE), 16 concepts resolved, 6 annual periods + 8 discrete quarters, output written to 09-Theses/MP/_data/. FY2025: revenue $224.4M, operating income −$149.4M, D&A $89.3M, GAAP-derived EBITDA −$60.1M — a standing cross-check against the adjusted-EBITDA confusion in audit #013.
• Resolver run on CAG: CONAGRA BRANDS INC. ↔ CIK 0000023217, FYE 05-31. In the process, found and fixed a wrong hand-typed CIK inside Rule 3's own example (it read 0001045309) — direct evidence for the change.
• Constraint discovered and documented: the shell sandbox cannot reach sec.gov; web fetch can. Bulk company_tickers.json truncates; per-ticker atom resolution is the supported method.

Expected impact

• Closes the #001/#012/#013 failure classes mechanically rather than by checklist.
• Shrinks the daily primary-source verification surface: pipeline-sourced figures carry their own citations and need no re-verification; the verification step now covers only non-pipeline claims (prices, non-GAAP, qualitative).

Counter-test (per methodology-calibration discipline)

If a fundamentals-class audit entry (#012/#013 type) recurs on a name with a generated fundamentals file, the pipeline failed its purpose — log it and debrief. If the pipeline is bypassed "for speed" in any thesis, that is a gate violation, not a shortcut.

Open follow-ups

• Retrofit PLTR and remaining thesis names (BLDR, FCN, GIS, LULU, PLAB, RMD, ROCK) into the registry + fundamentals files at next touch (Tier 1, mechanical).
• MP thesis EPV recompute (open P2, audit #013) should now cite MP-fundamentals.md.
• IFRS filers (ADRs) use ifrs-full taxonomy — extend concept-map when the first non-US filer enters the pipeline.