Chat records gain version (CURRENT_CHAT_VERSION), stamped by a single writeChat choke point in the store; absence reads as v1 — existing files migrate on their next write, no mass rewrite. The graph family (four JSONL ledgers, sources.json, render-settings.json) shares one lazy sidecar, graph/format.json; positions stays unstamped (rebuildable) and jobs stay unstamped (transient). Downgrade scenario ruled honest-refusal: records stamped newer than the running build read as absent (chats, logged) or throw (graph family) — a read-modify-write store never rewrites what it doesn't understand. Golden fixtures stay the unstamped legacy shape on purpose (lazy v1 is current-valid by design); the stamped write path is pinned by tests/schema-stamps.test.ts. Reads never write — the golden byte-identical pin holds.
2.5 KiB
Golden fixtures — real user data, read through production paths
The law (owner ruling, 2026-10-10): this directory is a REAL data dir —
verbatim records written by the real application code, never hand-edited. Any
intentional on-disk format change updates these fixtures IN THE SAME COMMIT;
an unintentional change fails tests/golden-fixtures.test.ts. When schema
stamps land (Wave D), the stamping commit updates these files with it.
tests/golden-fixtures.test.ts copies this dir to a temp dir, points the
production stores at it via the LH_DATA_DIR seam, and reads EVERYTHING
through the real read paths — chat records (contract folds + the LLM wire
flattening), graph ledgers (shape guards + the full derivation), jobs,
settings. It also pins marker COVERAGE: every custom-marker kind and the
core tool surface must appear in real data here.
What's inside
chats/chat_mv27rrra_feff23b6.json— the teach-side master (owner-built, 2026-10-10): probe-handoff, plan-block, compaction, subagent-steer, subagent-abort, and every teach tool (quiz, ask_user_question, plan_init, probe, probe_done, recall, explain, research, demo, visual, pdf, find, grep, ls, read, list_sessions, list_propositions). Plan tree + 25 turn records + a per-chat model override.chats/chat_mv293wv5_10656952.json— the practice lifecycle (owner-built, 2026-10-10): graph-context (the bracketed preamble), problem ×2, practice-thinking ×2, conclude ×2, practice-next, problem-dismissed — the full marker cycle of a practice session, including the dismissal.graph/— the real ledgers: nodes (282 lines), chains (160), sessions (8), coverage (93), sources (4), render-settings, positions.jobs/— five records covering kind × status: research done + aborted by learner, demo done, visual done, pdf error.settings.json— the stamped settings record (version 1, model slots).
Regenerating / extending
Run the app (npm run dev), exercise the feature against data/, then copy
the produced record in verbatim. Do not hand-edit content; if a record needs
pruning, re-run the flow instead.
Stamps
These files are deliberately the UNSTAMPED legacy shape (schema-stamp wave,
2026-10-10): lazy v1 means absence reads as current-valid, so this set pins
the pre-stamp read path forever. Stamped writes (chat version: 1, the
graph/format.json sidecar) are pinned by tests/schema-stamps.test.ts.
New records copied in after their first post-stamp write will carry the
stamps — that shape is equally current-valid.