Parent PRD
#12
What to build
Prove the hybrid generation workflow end-to-end on real content before scaling.
Extraction script — a small CLI tool in this repo that reads a commit range from the sibling auditmos/ogsfrompoly checkout and emits a structured markdown bundle (commit messages with SHAs, file-level diffs scoped to the range, a top-level "context for the article" preamble) ready to paste into Claude. Small interface, hidden implementation. Boundary snapshot tests are deliberately out of scope per PRD #12 — correctness is verified by manual inspection of the bundle output.
Build Log Episode 1 — "Replicating sign-randomization on real Polymarket data" — drafted via the hybrid path: extract a commit range from poly-track's earliest sign-randomization work, paste into Claude, curate for voice and opinion, run disclosure-policy self-review against docs/disclosure-policy.md, commit. Episode 1 replaces the placeholder article from #20. The article embeds code/diff excerpts inline (upstream repo is private, so no clickable commit links).
See plans/slice-11-build-log.md Phase 3 for full details.
Acceptance criteria
Blocked by
User stories addressed
Reference by number from PRD #12:
- User story 4 (institutional reader — Build Log audit trail)
- User story 7 (institutional reader — embedded code excerpts)
- User story 9 (technical peer — why decisions were made)
- User story 10 (technical peer — what didn't work)
- User story 11 (technical peer — source papers cited inline)
- User story 17 (author — extraction script enables 30-min drafting)
- User story 21 (author — milestone-tied cadence)
Parent PRD
#12
What to build
Prove the hybrid generation workflow end-to-end on real content before scaling.
Extraction script — a small CLI tool in this repo that reads a commit range from the sibling
auditmos/ogsfrompolycheckout and emits a structured markdown bundle (commit messages with SHAs, file-level diffs scoped to the range, a top-level "context for the article" preamble) ready to paste into Claude. Small interface, hidden implementation. Boundary snapshot tests are deliberately out of scope per PRD #12 — correctness is verified by manual inspection of the bundle output.Build Log Episode 1 — "Replicating sign-randomization on real Polymarket data" — drafted via the hybrid path: extract a commit range from poly-track's earliest sign-randomization work, paste into Claude, curate for voice and opinion, run disclosure-policy self-review against
docs/disclosure-policy.md, commit. Episode 1 replaces the placeholder article from #20. The article embeds code/diff excerpts inline (upstream repo is private, so no clickable commit links).See
plans/slice-11-build-log.mdPhase 3 for full details.Acceptance criteria
/articles/<slug>withepisode_number: 1,series: "build-log", ~800–1200 words, narrative-ledcommit_rangefrontmatter captures the SHAs the article narrates (internal metadata only, not rendered in HTML or.mdbody)docs/disclosure-policy.mdpnpm types && pnpm test && pnpm lint && pnpm knipall pass in CIBlocked by
User stories addressed
Reference by number from PRD #12: