docs: canonical YAML manifest + renderer for fork-ahead changes

Solves the duplicate-info-in-four-places drift problem. The fork-ahead
narrative previously lived (and drifted) across:

  - README.md (fork-change-queue table)
  - CLAUDE.md (row-by-row inventory)
  - FORK_CHANGELOG.md (Keep-a-Changelog narrative)
  - scratch/promises.md (tracker)

This commit promotes docs/fork-changes.yaml to canonical source and
ships the FORK_CHANGELOG render. README/CLAUDE/promises rendering will
land in follow-on commits between markers.

Implementation:

  - docs/fork-changes.yaml — schema documented at the top of the file.
    9 entries backfilled from today's + yesterday's commits, plus the
    Merged-into-upstream subsection (8 historical PRs).
  - scripts/render-docs.py — reads the manifest, groups entries by
    date then by Keep-a-Changelog bucket, writes FORK_CHANGELOG.md.
    --check mode for CI / pre-push hooks. --target flag for selective
    rendering (changelog | all; all == changelog today, will expand).
    Idempotent: running twice is a no-op.
  - scripts/check-docs.sh — extended with a new stage 3 that calls
    render-docs.py --check to detect FORK_CHANGELOG drift. Stage 1
    test-count check now uses the repo venv's pytest directly so the
    check works without an activated environment.
  - FORK_CHANGELOG.md regenerated. The hand-written content is gone;
    it's now a generator artifact with a "this file is generated"
    notice up top.
  - CLAUDE.md — new "Documentation maintenance" section documenting
    the workflow and the migration plan.

Researched off-the-shelf alternatives before going custom: towncrier,
scriv, git-cliff, antsibull-changelog. None do single-source →
multi-target rendering. keep-a-changelog#230 has been asking for this
shape since 2018; community answer is "build your own". This is ours.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: jphein

CLAUDE.md

@@ -173,3 +173,46 @@ Two save modes, controlled by `hook_silent_save` in `~/.mempalace/config.json`:
 ## Testing
 
 Always run `python -m pytest tests/ -x -q` after changes. Benchmark and stress tests are excluded by default (use `-m benchmark` or `-m stress` to include).
+
+## Documentation maintenance
+
+The fork-ahead narrative was previously hand-maintained in four places
+(README's fork-change-queue table, this file's row inventory,
+`FORK_CHANGELOG.md`, and `~/.claude/projects/-home-jp-Projects-memorypalace/scratch/promises.md`).
+Drift was inevitable. As of 2026-04-26 the **canonical source** is
+`docs/fork-changes.yaml`; render targets are generated.
+
+### Workflow for new fork-ahead changes
+
+1. Land the code change with a focused commit on `main`.
+2. Add an entry to `docs/fork-changes.yaml` (top of the `entries:`
+   list, newest first). Schema is documented at the top of the YAML.
+3. Run `scripts/render-docs.py` to regenerate `FORK_CHANGELOG.md`.
+4. Run `scripts/check-docs.sh` to verify nothing has drifted (test
+   count, commit hashes, render parity, upstream PR states).
+5. Commit the YAML + the regenerated `FORK_CHANGELOG.md` together.
+
+### Targets
+
+| Target | Status |
+|--------|--------|
+| `FORK_CHANGELOG.md` | rendered from YAML (today) |
+| README fork-change-queue table | hand-maintained for now |
+| CLAUDE.md row inventory (rows 1–27 above) | hand-maintained for now |
+| `scratch/promises.md` tracker entries | hand-maintained for now |
+
+The renderer's `--target` flag is wired to take `changelog` or `all`;
+`all` is the same as `changelog` until the README/CLAUDE/promises
+renderers land.
+
+### Lint
+
+`scripts/check-docs.sh` runs four checks:
+
+1. README test count vs `pytest --collect-only`
+2. every fork commit hash referenced in docs resolves via `git cat-file -e`
+3. `FORK_CHANGELOG.md` matches the YAML (re-render idempotent)
+4. every `#NNNN` reference has an upstream state matching the doc's claim
+
+Run before committing any doc change. Exit code 1 on drift.
+

FORK_CHANGELOG.md

@@ -4,150 +4,198 @@ Fork-ahead changes that aren't yet in upstream `MemPalace/mempalace`.
 Upstream's release history lives in [`CHANGELOG.md`](CHANGELOG.md);
 this file is the supplement.
 
+> **This file is generated.** Edit `docs/fork-changes.yaml` and run
+> `scripts/render-docs.py` to regenerate. Hand-edits will be
+> overwritten on the next render.
+
 Date-based sections, not semver — the fork tracks `upstream/develop` and
 doesn't cut its own release tags. When a fork-ahead row lands upstream,
-the entry is moved to the **Merged into upstream** section at the
-bottom (kept for ~2 weeks, then trimmed).
+move the entry to the **Merged into upstream** section at the bottom
+(kept ~30 days, then trimmed).
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ---
 
+
 ## [2026-04-26]
 
+
 ### Added
 
-- **`mempalace_session_recovery_read` MCP tool** — reads the dedicated
-  `mempalace_session_recovery` collection by optional `session_id`,
-  `agent`, `since`, `until`, `wing`, `limit` filters; returns entries
-  newest-first. Used for hook auditing and "what was I doing 2 hours
-  ago" recovery. Registered in the `TOOLS` dict and documented in
-  `website/reference/mcp-tools.md`. ([`e266365`](https://github.com/jphein/mempalace/commit/e266365))
-- **`mempalace repair --mode reorganize`** — explicit operator command
-  to migrate existing `topic=checkpoint` drawers from the main
-  collection to `mempalace_session_recovery`. Idempotent, ID and
-  metadata preserving. ([`42817d7`](https://github.com/jphein/mempalace/commit/42817d7))
-- **`scripts/deploy.sh`** — one-command push + Syncthing-aware redeploy
-  to the canonical disks daemon (`systemctl --user restart palace-daemon`
-  + post-restart import check that today\'s fork-ahead surface is
-  loaded). ([`8252025`](https://github.com/jphein/mempalace/commit/8252025))
-- **`drawer_id` field** on `mempalace_search`, `mempalace_diary_read`,
-  and `mempalace_session_recovery_read` payloads — chromadb\'s primary
-  key was always returned by `query()` / `get()` but never plumbed into
-  the result dicts; consumers (e.g. citation popovers) can now follow
-  a hit back to the underlying drawer via `mempalace_get_drawer`.
-  ([`9a8bb77`](https://github.com/jphein/mempalace/commit/9a8bb77))
-
-### Changed
-
-- **`tool_diary_write` routes `topic in _CHECKPOINT_TOPICS` to a
-  dedicated collection** (`mempalace_session_recovery`). Everything else
-  stays in `mempalace_drawers`. The main collection is now the
-  *verbatim store* — chats, tool calls, mined files — and is no longer
-  polluted by Stop-hook auto-save fragments dominating vector top-N.
-  ([`e266365`](https://github.com/jphein/mempalace/commit/e266365))
-- **`hook_precompact` writes a session-recovery marker** before mining
-  the transcript and allowing compaction. Mirrors `hook_stop`\'s
-  `_save_diary_direct` call so a context-compaction event leaves a
-  queryable timestamp in the recovery collection rather than nothing.
-  ([`42817d7`](https://github.com/jphein/mempalace/commit/42817d7))
-- **`tool_diary_write` accepts an optional `session_id`** parameter,
-  stored in metadata when a checkpoint is being written so the new
-  recovery-read tool can filter by Claude Code session.
-  ([`e266365`](https://github.com/jphein/mempalace/commit/e266365))
+
+- **Phase D migration + PreCompact recovery write** ([`42817d7`](https://github.com/jphein/mempalace/commit/42817d7))
+  ``migrate_checkpoints_to_recovery(palace_path, batch_size=1000)`` walks
+  the main collection in pages, filters drawers with topic in
+  ``_CHECKPOINT_TOPICS`` in Python (avoids the chromadb 1.5.x ``$in``/``$nin``
+  filter-planner bug), copies them to the recovery collection
+  (preserving IDs + metadata), then deletes from main. Idempotent —
+  re-running on a fully-reorganized palace returns 0. Add-then-delete
+  order: a crash mid-migration leaves a duplicate, not a loss.
+  Wired into ``mempalace repair --mode reorganize`` for explicit operator
+  runs. PreCompact incorporated — ``hook_precompact`` now writes a
+  session-recovery marker mirroring Stop, so context-compaction events
+  leave a queryable timestamp in the recovery collection rather than
+  nothing. Failures are non-fatal (logged; mining + compaction still
+  proceed).
+
+  *Tests:* 6 in TestMigrateCheckpointsToRecovery + 1 in test_hooks_cli
+  *Files:* `mempalace/migrate.py`, `mempalace/cli.py`, `mempalace/hooks_cli.py`, `tests/test_migrate.py`
+
+
+- **Surface drawer_id in search/diary/recovery payloads** ([`9a8bb77`](https://github.com/jphein/mempalace/commit/9a8bb77))
+  ChromaDB's primary key was always returned by ``query()`` and ``get()``
+  but never plumbed into result-building loops; consumers (e.g.
+  familiar.realm.watch's citation-popover loop) couldn't link a hit
+  back to the underlying drawer. Three call sites updated for parity:
+  ``searcher.search_memories`` (vector path + sqlite BM25 fallback),
+  ``mcp_server.tool_session_recovery_read``, ``mcp_server.tool_diary_read``.
+  Defensive zip with id-pad: production chromadb always returns ids,
+  but several test mocks omit them — pad with ``None`` when absent so
+  existing fixtures keep working without touching N tests.
+
+  *Tests:* 1 integration + 1 inline assertion
+  *Files:* `mempalace/searcher.py`, `mempalace/mcp_server.py`, `website/reference/mcp-tools.md`
+
+
+- **scripts/deploy.sh — one-command Syncthing-aware redeploy** ([`8252025`](https://github.com/jphein/mempalace/commit/8252025))
+  Single command does the right shape: push fork main → wait for
+  Syncthing to reach ``/mnt/raid/projects/memorypalace`` on the deploy
+  host → ``systemctl --user restart palace-daemon`` → poll ``/health`` →
+  ssh-import-check that today's fork-ahead surface is loaded.
+  Replaces a three-step manual ritual that was easy to get wrong
+  (e.g. ``pip install --upgrade`` was a no-op on the editable install).
+
+  *Files:* `scripts/deploy.sh`
+
 
 ### Fixed
 
-- **`quarantine_stale_hnsw` no longer destroys healthy indexes on cold
-  start.** Two-stage gate: (1) mtime gap > threshold (existing) AND
-  (2) `_segment_appears_healthy` integrity sniff-test on
-  the chromadb segment metadata file (new — checks for chromadb\'s
-  protocol/terminator bytes without deserializing). Production case
-  2026-04-26 06:56:45 had three healthy 253MB segments quarantined on
-  cold start by mtime alone (chromadb 1.5.x flushes HNSW asynchronously;
-  clean shutdown does not force-flush, so the on-disk gap is the steady
-  state, not corruption). The integrity gate distinguishes flush-lag
-  from corruption. ([`645ba20`](https://github.com/jphein/mempalace/commit/645ba20))
-- **`make_client()` only invokes `quarantine_stale_hnsw` once per palace
-  per process.** Previously, every reconnect under steady write load
-  re-fired the proactive check, racking up `.drift-*` directories every
-  10–30 minutes. The cold-start gate (`ChromaBackend._quarantined_paths`)
-  caps it to one fire on first open; runtime drift detection still
-  works via palace-daemon\'s `_auto_repair`, which calls
-  `quarantine_stale_hnsw` directly. ([`70c4bc6`](https://github.com/jphein/mempalace/commit/70c4bc6))
+
+- **Integrity gate prevents quarantine_stale_hnsw from destroying healthy indexes** ([`645ba20`](https://github.com/jphein/mempalace/commit/645ba20))
+  Previous behavior fired whenever ``sqlite_mtime - hnsw_mtime`` exceeded
+  the (lowered, in #1173) 300s threshold. ChromaDB 1.5.x flushes HNSW
+  asynchronously and a clean shutdown does not force-flush, so the
+  on-disk HNSW is always meaningfully older than ``chroma.sqlite3`` —
+  that's the steady state, not corruption. Quarantine renamed valid
+  HNSW segments on every cold-start; chromadb created empty replacements;
+  vector recall went to 0/N until rebuild. Confirmed in production on
+  the disks daemon journal 2026-04-26 06:56:45: three of three healthy
+  253MB segments quarantined on cold-start with 538-557s gaps. Fix:
+  stage 2 integrity gate sniffs the chromadb segment metadata file
+  for its protocol/terminator bytes (PROTO ``\x80`` head, STOP ``\x2e``
+  tail) and a non-trivial size, **without deserializing**. Healthy
+  segment with mtime drift → keep in place; truncated/zero-filled →
+  quarantine.
+
+  *Tests:* 4 in test_backends.py (renames-corrupt, leaves-healthy-with-drift, leaves-no-metadata, renames-truncated)
+  *Upstream:* [PR #1173](https://github.com/MemPalace/mempalace/pull/1173) (OPEN)
+  *Files:* `mempalace/backends/chroma.py`, `tests/test_backends.py`
+
 
 ### Performance
 
-- **Cherry-picked upstream PR [#1085](https://github.com/MemPalace/mempalace/pull/1085)**
-  (@midweste, OPEN as of 2026-04-26) — batch ChromaDB inserts in
-  `miner.process_file()`. New `_build_drawer()` helper + `add_drawers()`
-  batch-insert path; one `collection.upsert` + one ONNX embedding pass
-  per sub-batch instead of per-chunk. Hoists `datetime.now()` and
-  `os.path.getmtime()` to file-level (2 syscalls per file instead of
-  2N). Reported 10–30× mining speedup upstream. Fork-side resolution
-  preserved fork\'s existing `DRAWER_UPSERT_BATCH_SIZE=1000`; aliased
-  upstream\'s `CHROMA_BATCH_LIMIT` to it. ([`6be6fff`](https://github.com/jphein/mempalace/commit/6be6fff))
-  *Becomes a no-op when #1085 merges to develop and we next sync.*
 
----
+- **Cherry-pick #1085 — batch ChromaDB inserts in miner (10–30× faster)** ([`6be6fff`](https://github.com/jphein/mempalace/commit/6be6fff))
+  Cherry-picked from upstream PR
+  [#1085](https://github.com/MemPalace/mempalace/pull/1085) (@midweste,
+  OPEN as of 2026-04-26). New ``_build_drawer()`` helper + ``add_drawers()``
+  batch-insert path; ``process_file`` hands the full chunk list to
+  ``add_drawers`` instead of looping per-chunk. Hoists ``datetime.now()``
+  and ``os.path.getmtime()`` to file-level (2 syscalls per file instead
+  of 2N). Reported 10–30× mining speedup upstream. Fork-side resolution
+  preserved fork's existing ``DRAWER_UPSERT_BATCH_SIZE=1000``; aliased
+  upstream's ``CHROMA_BATCH_LIMIT`` to it. Becomes a no-op when #1085
+  merges to develop and we next sync.
+
+  *Upstream:* [PR #1085](https://github.com/MemPalace/mempalace/pull/1085) (OPEN)
+  *Files:* `mempalace/miner.py`
+
 
 ## [2026-04-25]
 
+
 ### Added
 
-- **Phases A–C of the checkpoint collection split** — new
-  `mempalace_session_recovery` collection adapter
-  (`_SESSION_RECOVERY_COLLECTION` + `get_session_recovery_collection`
-  in `palace.py`); `tool_diary_write` routes `topic in _CHECKPOINT_TOPICS`
-  to it. Promoted from "future work" to "necessary" by the same-day
-  Cat 9 A/B (`kind=all` 632 tokens/Q vs `kind=content` 3 tokens/Q on
-  the canonical 151K-drawer palace). 12 new tests across
-  `tests/test_session_recovery.py` + `TestCheckpointRouting` +
-  `TestSessionRecoveryRead`. Design doc:
-  `docs/superpowers/specs/2026-04-25-checkpoint-collection-split.md`;
-  TDD plan:
-  `docs/superpowers/plans/2026-04-25-checkpoint-collection-split-impl.md`.
-  ([`e266365`](https://github.com/jphein/mempalace/commit/e266365))
+
+- **Phases A–C of the checkpoint collection split** ([`e266365`](https://github.com/jphein/mempalace/commit/e266365))
+  New ``mempalace_session_recovery`` collection adapter
+  (``_SESSION_RECOVERY_COLLECTION`` + ``get_session_recovery_collection``
+  in ``palace.py``); ``tool_diary_write`` routes ``topic in _CHECKPOINT_TOPICS``
+  to it. New ``mempalace_session_recovery_read`` MCP tool reads recovery
+  collection only with optional filters (session_id, agent, since,
+  until, wing, limit). Promoted from "future work" to "necessary" by
+  the same-day Cat 9 A/B (``kind=all`` 632 tokens/Q vs ``kind=content``
+  3 tokens/Q on the canonical 151K-drawer palace). Design doc at
+  ``docs/superpowers/specs/2026-04-25-checkpoint-collection-split.md``.
+
+  *Tests:* 12 across test_session_recovery.py + TestCheckpointRouting + TestSessionRecoveryRead
+  *Files:* `mempalace/palace.py`, `mempalace/mcp_server.py`, `tests/test_session_recovery.py`, `tests/test_mcp_server.py`, `website/reference/mcp-tools.md`
+
 
 ### Fixed
 
-- **`palace_graph.build_graph` skips `None` metadata.**
-  `palace_graph.py:95` was calling `meta.get("room", "")`
-  unconditionally; ChromaDB returns `None` for legacy/partial-write
-  drawers, taking out every consumer of `build_graph` (graph_stats,
-  find_tunnels, traverse, the daemon\'s `/stats`). Caught by
-  palace-daemon\'s `verify-routes.sh` smoke test. Filed as upstream
-  [#1201](https://github.com/MemPalace/mempalace/pull/1201).
-  ([`5fd15db`](https://github.com/jphein/mempalace/commit/5fd15db))
-- **`kind=` filter on `search_memories` excludes Stop-hook
-  checkpoints by default** — surgical fix while the structural split
-  was being designed. Three values: `"content"` (default, excludes),
-  `"checkpoint"` (recovery/audit only), `"all"` (no filter). Two
-  same-day architecture corrections: (a) the where-clause filter
-  (`topic $nin [...]`) tripped a chromadb 1.5.x filter-planner bug
-  that returned `Internal error: Error finding id` on every
-  `kind=content` vector query, so the exclusion moved to post-filter
-  only ([`398f42f`](https://github.com/jphein/mempalace/commit/398f42f));
+
+- **Gate quarantine_stale_hnsw to once-per-palace-per-process** ([`70c4bc6`](https://github.com/jphein/mempalace/commit/70c4bc6))
+  ``make_client()`` previously invoked ``quarantine_stale_hnsw`` on every
+  reconnect; under steady write load the proactive check kept firing,
+  racking up ``.drift-*`` directories every 10–30 minutes. New
+  ``ChromaBackend._quarantined_paths: set[str]`` caps it to one fire on
+  first open per palace per process. Real cold-start drift still caught
+  (replicated/restored palace); real runtime errors still caught via
+  palace-daemon's ``_auto_repair``, which calls ``quarantine_stale_hnsw``
+  directly and bypasses this gate.
+
+  *Tests:* 2 in test_backends.py (single-fire-per-palace, per-palace independence)
+  *Upstream:* [PR #1173](https://github.com/MemPalace/mempalace/pull/1173) (OPEN)
+  *Files:* `mempalace/backends/chroma.py`, `tests/test_backends.py`, `tests/conftest.py`
+
+
+- **palace_graph.build_graph skips None metadata** ([`5fd15db`](https://github.com/jphein/mempalace/commit/5fd15db))
+  ``palace_graph.py:95`` was calling ``meta.get("room", "")`` unconditionally;
+  ChromaDB returns ``None`` for legacy/partial-write drawers, taking out
+  every consumer of ``build_graph`` (graph_stats, find_tunnels, traverse,
+  the daemon's ``/stats``). Caught by palace-daemon's ``verify-routes.sh``
+  smoke test. Same family as upstream's #999 None-metadata audit, in a
+  read path the audit didn't reach.
+
+  *Upstream:* [PR #1201](https://github.com/MemPalace/mempalace/pull/1201) (OPEN)
+  *Files:* `mempalace/palace_graph.py`
+
+
+- **kind= filter on search_memories excludes Stop-hook checkpoints (transitional)** ([`f9f5cc4`](https://github.com/jphein/mempalace/commit/f9f5cc4))
+  Three values: ``"content"`` (default, excludes), ``"checkpoint"``
+  (recovery/audit only), ``"all"`` (no filter). Two same-day architecture
+  corrections: (a) the where-clause filter (``topic $nin [...]``) tripped
+  a chromadb 1.5.x filter-planner bug; the exclusion moved to post-filter
+  only ([398f42f](https://github.com/jphein/mempalace/commit/398f42f));
   (b) vector top-N is dominated by checkpoints on this palace, so
-  post-filter alone empties the result set without aggressive
-  over-fetch — pull size raised to `max(n*20, 100)` for `kind != "all"`
-  ([`f9f5cc4`](https://github.com/jphein/mempalace/commit/f9f5cc4)).
-  9 tests in `TestCheckpointFilter`. *This is the safety net during
-  the transition; once Phase D ships and existing checkpoints
-  migrate, the post-filter and over-fetch hack become deletable.*
+  post-filter alone empties the result set without aggressive over-fetch
+  — pull size raised to ``max(n*20, 100)`` for ``kind != "all"`` (this commit).
+  Safety net during the transition; once Phase D ships and existing
+  checkpoints migrate, the post-filter and over-fetch hack become
+  deletable.
+
+  *Tests:* 9 in TestCheckpointFilter
+  *Files:* `mempalace/searcher.py`, `mempalace/mcp_server.py`, `tests/test_searcher.py`
+
 
 ---
 
 ## Merged into upstream (recent)
 
-Trimmed monthly. See [`CHANGELOG.md`](CHANGELOG.md) for the full
-released history.
-
-- **PR #659** — diary `wing` parameter (merged 2026-04-23)
-- **PR #661** — graph cache with write-invalidation (merged 2026-04-22)
-- **PR #673** — deterministic hook saves (merged 2026-04-22)
-- **PR #1021** — Claude Code 2.1.114 stdout/silent_save fixes (merged 2026-04-22)
-- **PR #999** — `None`-metadata guards across read paths (merged 2026-04-18)
-- **PR #1000** — `quarantine_stale_hnsw` (shipped in v3.3.2)
-- **PR #1023** — PID file guard (shipped in v3.3.2)
-- **PR #681** — Unicode checkmark → ASCII (shipped in v3.3.2)
+
+*Trim entries from this list once they're more than ~30 days old.*
+
+
+*See CHANGELOG.md (upstream) for the full released history.*
+
+
+- [PR #659](https://github.com/MemPalace/mempalace/pull/659) — diary `wing` parameter — 2026-04-23
+- [PR #661](https://github.com/MemPalace/mempalace/pull/661) — graph cache with write-invalidation — 2026-04-22
+- [PR #673](https://github.com/MemPalace/mempalace/pull/673) — deterministic hook saves — 2026-04-22
+- [PR #1021](https://github.com/MemPalace/mempalace/pull/1021) — Claude Code 2.1.114 stdout/silent_save fixes — 2026-04-22
+- [PR #999](https://github.com/MemPalace/mempalace/pull/999) — None-metadata guards across read paths — 2026-04-18
+- [PR #1000](https://github.com/MemPalace/mempalace/pull/1000) — quarantine_stale_hnsw shipped — v3.3.2
+- [PR #1023](https://github.com/MemPalace/mempalace/pull/1023) — PID file guard prevents stacking mine processes — v3.3.2
+- [PR #681](https://github.com/MemPalace/mempalace/pull/681) — Unicode checkmark → ASCII — v3.3.2