earnings-deep-dive — Codex skill: post-earnings analysis

What earnings-deep-dive does

The skill works on numbers that are already published. Official filings and releases come first, the call transcript backs the narrative and any guidance that only came up on the call. Every figure and quote carries a source tag; anything derived by the analyst needs its formula, assumptions and a confidence note. GAAP versus non-GAAP, reported versus constant currency, company guidance versus analyst estimates — all of it stays labelled, so the two do not blur together.

There are several routes. The default is a full deep dive: quarterly key metrics and growth trajectory built on the issuer's own drivers, a Q&A treatment and a debate map. Ask for something short and you get a one-page tear sheet. If you explicitly want a dashboard, the skill keeps ownership of the analysis and hands a public_equity_investing_dashboard.v1 payload to dashboard-builder. A deterministic file mode exists too, for when you supply plan.json, normalized CSVs or model-update inputs.

Before gathering evidence the skill reads the shared runtime contract in ../public-equity-investing/SKILL.md and resolves only the source categories the task needs: filings and IR, transcripts and presentations, internal research, portfolio models and trackers, market data and estimates. When no transcript exists, a full deep dive does not render an empty Q&A table — it shows a visible limitation labelled transcript not provided or transcript source not found and names the missing artifact.