Derived pack: Fama–French ratios on free-float mcap, geo revenue segmentation, capex/opex

[febuz]

Owner requirement (2026-07-06): extend finfacts.derive with: Fama & French-style ratios computed on FREE-FLOAT market capitalization (owner confirmed free-float as the basis), geographic revenue segmentation facts, capex and opex concepts. All exact math (scaled ints / Decimal under pinned localcontext), derived_from CID chains to the source facts.

[febuz]

Delivered in c74ebd5: capex/opex TTM (finfield:capex_ttm, finfield:opex_ttm) and Fama–French ratios on free-float mcap (finfield:book_to_float_mcap, finfield:earnings_to_float_mcap, finfield:capex_to_float_mcap — on dei:EntityPublicFloat, with staleness ≤400d, USD-unit and denominator>0 guards). Validated against the live feed: 87,313/87,313 records decoded, 4,182 B/M ratios (median 0.538, 74.6% in 0.1–10), 3 entities hand-recomputed exact, deterministic across passes.

Still open on this issue: geographic revenue segmentation — blocked by source: the SEC companyfacts API carries no XBRL dimensional data (segments live in the full filings). Needs a full-filing/frames adapter; TTM-on-feed additionally waits on FinField/feed#1 (≥4 quarters in the core pack).

🤖 Generated with Claude Code

[febuz]

Progress via #3 (merged): momentum_12_1 (equity+crypto, live BTC proof), asset_growth_yoy (F&F investment), operating_profitability (EUR/EUR pairs from ESEF derive now), IFRS aliases across the concept tuples. Float-mcap ratios + capex/opex TTM were already in.

Still open on this issue: geographic revenue segmentation — blocked on a dimensions field in the FinFact model (segment facts are deliberately skipped by finscrapers.esef until the model carries them). Equity momentum additionally waits on scrapers#7 (stooq bot-wall).

🤖 Generated with Claude Code

[febuz]

Dimensions chain complete: facts#4 (model field, CID-compat proven) → scrapers#8 (ESEF geo-segment capture) → knit#1 (weave round-trip) — all merged, 115 tests green across the three repos.

Live coverage caveat (probed on ASML + Nokia 2025): ESEF mandatory detail-tagging stops at the primary statements — geographic tables in the notes are block-tagged text, so numeric geo yield from ESEF is currently ~zero. The reliable geo-revenue source is US 10-K segment footnotes via a full-filing XBRL scraper (companyfacts excludes segments) — that's the remaining follow-up on this issue.

🤖 Generated with Claude Code