alice-opens-the-record-explorer-and-reads-the-whole-developm

Summary

spc-38 delivers itd-136's record explorer: one record.json export of the record graph, and the pages rendered from it — the dashboard, a page per record with its full body, contributors, a foundations page, and a references page rendered by a small in-repo CSL formatter. On the itd-140 boundary every page here is generic-side: input is the record format (.abcd/development/** frontmatter and lifecycle directories, plus the opted-in issue ledger), git history, and CHANGELOG.md dated headings — nothing else — and a missing optional source omits the dependent page and its navigation entry while the build succeeds. No prose anywhere describes the generator as repo-agnostic (itd-140 rule 3).

Settled constraints

• record.json is a pure build artifact, never committed (interview ruling): determinism is asserted by a double-build diff in CI, and production cannot drift from the tree because it is rendered from the tag by the released binary (adr-48).
• Full bodies at /record/<type>/<id>/ (interview ruling); the issue ledger is published because this repo's .abcd/site.json opts in (adr-32 tiering; itd-140 rule 4).
• Verbatim record rendering at /record/** is exempt from the banned-token gate (adr-47 decision 3); the contributors page prints model names under the declared attribution escape.
• The baseline is a ratchet seeded with the eight dangling references the in-repo re-count finds (adr-22→adr-14/adr-15/adr-17, adr-25→adr-8, adr-27→adr-16, adr-28→adr-18, adr-35→adr-4, itd-3→spc-1) — fixing one shrinks it, growing it fails the build.
• One canonical primitive: the graph comes from the record-lint engine's existing typed-reference scan, exported — never a third parser.

Mechanism

record.json (in abcd site build)

• A new exported loader on the lint engine's record scan (the record_schema walker already parses every store, block-sequence spellings included) yields nodes and typed references; internal/core/site consumes it. Landed as a separate behaviour-preserving refactor commit before the feature commit.
• Nodes: id, type, lifecycle from directory, title, path, dates; edges: typed links with mirrored pairs collapsed (intent spec_id ↔ spec implements; related recorded in both files) so each distinct link renders once; body mentions deduplicated against typed edges.
• Dates from one git log --reverse --name-status pass through the repo's isolated-git helpers: created, entered-current-directory, last touched, for every record file; frontmatter date wins where a record carries one.
• Releases from CHANGELOG.md dated headings via the changelog package's own matcher; contributors from git shortlog folded through .mailmap; assistance tallies from Assisted-by: trailers; counts by type and lifecycle; the two precomputed arrangements (spc-39's layouts) ride in the same file; health carries the unresolved references measured against the baseline.
• schema_version: 1; sorted inputs, fixed seeds, no timestamps beyond the build stamp; a golden-file test over an in-process fixture repo pins the export, and CI builds twice and diffs.

Pages

• Dashboard /record/: stat tiles (releases, decisions, intents, specs, issues, principles), lifecycle bars, release cadence, latest decisions, record health — counts, dates, ids and titles only. A visual whose labels and numbers already render as text (a bar with its legend, a chart with labelled ticks) carries no table twin; a twin accompanies only a visual that is the sole carrier of its numbers. The cadence chart labels every tick with version and date, and below 700px a plain list stands in for the strip — exactly one of the two renders at any width.
• Per-record pages: frontmatter, the body rendered verbatim through the Markdown renderer, inbound and outbound typed links, an open-on-GitHub link. Retired ADR ids render as dashed baseline stubs — no tombstone files (interview ruling).
• Contributors /contributors/: authors of record from git shortlog through .mailmap; a separate labelled bots-and-tools row driven by an author-pattern rule (bot-suffixed authors and the mailmap-canonicalised pre-policy commit); the Assisted-by: share and per-model tallies presented as disclosure with CONTRIBUTING.md linked.
• Foundations: cards for principles/ and intents/disciplines/ entries that list and link, never explain; if neither directory exists the page and its navigation entry are omitted.
• References /references/: rendered from .abcd/development/research/references.csl.json by a small in-repo CSL formatter (stdlib-only — the renderer that satisfies the no-Node, no-committed-HTML gate), numbered identically to ACKNOWLEDGEMENTS.md (a check asserts the numbering agrees), DOIs linked, the attribution line the style requires; absent a compatible renderer the page and nav entry are omitted — never a broken page.
• Graceful absence throughout: a missing bibliography, Identity block or CHANGELOG omits the dependent page and nav entry and the build succeeds.

Checks (in abcd site check)

• Unresolved typed cross-references outside .abcd/site-baseline.json fail the build naming the reference; the baseline only shrinks.
• The 390 px static checks run over every explorer route.

Acceptance-criteria mapping

• AC 1 (counts derived at build into an uncommitted record.json; double-build diff; visuals carry their numbers as text) → export + determinism gate + dashboard.
• AC 2 (every record a page with frontmatter, body, typed links, GitHub link) → per-record pages.
• AC 3 (baseline ratchet) → baseline check, seeded with the eight.
• AC 4 (mirrored references collapse) → edge deduplication.
• AC 5 (contributors: mailmap authors, bots row, disclosure framing, model names confined under the escape) → contributors page.
• AC 6 (foundations lists-and-links or is omitted) → foundations page.
• AC 7 (references numbered like ACKNOWLEDGEMENTS or omitted) → CSL formatter + numbering check.
• AC 8 (390 px, no horizontal scroll) → static checks + CI screenshots.

Out of scope

• The relationship chart and genealogy behaviours (spc-39; their layouts are computed here only as record.json fields).
• Any second-instance fixture demonstration (itd-139, drafts).
• Search across the record (plan phase 4).