Earnings review artifacts

One file per reporting period: memory/earnings/<TICKER>/<period-label>.json. The file is the canonical record — which first-party documents were read, the comparable history, the promise ledger, and the provenance manifest behind every published number. Markdown write-ups are renderings of it, never the source.

python3 scripts/data/earnings_review.py validate        memory/earnings/<TICKER>/<period>.json
python3 scripts/data/earnings_review.py review          memory/earnings/<TICKER>/<period>.json
python3 scripts/data/earnings_review.py thesis-evidence  memory/earnings/<TICKER>/<period>.json
python3 scripts/data/earnings_review.py promises \
    memory/earnings/<TICKER>/<previous>.json memory/earnings/<TICKER>/<current>.json

Shape: config/earnings_review.schema.json. Workflow and source policy: skills/earnings-review/SKILL.md. Worked US and HK examples live in tests/fixtures/earnings/.

Rules this directory relies on:

• the newest period’s file holds the live promise ledger; earlier files are the history, so there is no second commitment store to drift out of sync;
• a promise may never be dropped, and met / partial / missed are terminal;
• basis, currency and unit are constant across a comparable history;
• footnote claims require a primary issuer document covering the period;
• earnings evidence flows into memory/theses/*.json through the registry’s own drift evaluator. Nothing here changes thesis state on its own.