DR-129: ast-dataflow public docs are authored fresh; private ledger artefacts do not migrate

Context

DR-128 moved the tool’s backlog to tool-repo GitHub issues but said nothing about the documentation set: the id-50 spec family (PRODUCT/TECH/ROADMAP), the id-375/id-377 research and trials, and the sidecar contract (DR-102). The S533 owner board ruled on the full inventory (31 docs: 3 migrate-class, 4 split, 24 stay). Three facts shaped the ruling: two recent sessions leaked client identity into public history; the docs-site ROADMAP was stale twice over and carried a denylist token; and the S533 invariant audit found PRODUCT.md no longer describes the shipped binary (8 invariants broken, 5 reinterpreted without amendment, 3 queries with no invariant — tool-repo issue #7).

Decision

Public-facing ast-dataflow documentation is authored fresh in the tool repo from measured shipped behaviour — never migrated as files from the docs-site. Shipped under this ruling: the tool-repo ROADMAP, docs/gap-register.md, and docs/SIDECAR.md (the DR-102 contract restated on its own authority). The docs-site artefacts stay private: the id-50 spec family (ROADMAP retired in place), trials Part 2, the id-375 research set including the market scan, and the 25-hard-case oracle. Anything crossing the private→public boundary gets a ledger-identifier + denylist-token scrub and lands as a PR the owner reviews — the scrub is best-effort by construction (agents cannot hold the private denylist), so owner review is the gate, not a courtesy. PRODUCT.md (id-50) is no longer the spec of record; a future public spec is authored fresh against the shipped envelope (issue #7). The tool repo is licensed Apache-2.0 (S533 owner pick, tool-repo PR #6).

Alternatives Considered

• Migrate the files as-is — rejected: stale content (ROADMAP wrong on OPS-T1; PRODUCT.md fails its own invariants) and denylist tokens in the migration path.
• Publish nothing beyond the README — rejected: the README’s “Known caveats” already forward-referenced gap numbers the public could not read, and the sidecar extension point had no public contract.

Consequences

• Remaining origin-repo identifier residue in the public repo is tracked as tool-repo issue #10, including the owner-held UNDECIDABLE on whether bid_projects/bids/feed_articles are the same leak class.
• The docs-site id-50 ROADMAP carries a retirement banner (654483e1); its client token and R-WP12’s six were redacted per the id-115 map.
• research/market.md and the 25-case oracle remain private; revisit only if third-party sidecar demand materialises (S533 board D4/D5).