Fill the RFC-007 Phase 3 ledger row from git provenance (fixes #618)

[lantisprime]

Closes #618.

PR #617 shipped RFC-007 Phase 3 without an Implementation ledger row. That was deliberate and stated in the PR body: the row's SHA must be the merge commit, which does not exist until the PR lands. Writing a branch-tip SHA there is the defect #582 was filed for and #605 had to repair.

Provenance

Derived from git against the files the row claims to have changed, not from an adjacent row and not from a workplan note:

$ git log --follow --oneline -1 scripts/em-graph.mjs
dafc3c8 RFC-007 Phase 3: wiki-link + composes-with edges ... (#617)

$ git log --follow --oneline -1 tests/test-em-graph-links.mjs
dafc3c8 RFC-007 Phase 3: wiki-link + composes-with edges ... (#617)

Both files resolve to dafc3c8, which is #617's merge commit (gh pr view 617 → mergeCommit.oid = dafc3c89da4fb740ffd957074b434df1162219cd).

Also verified

• CI anchor tests.yml:407-408 checked against the merged file, not assumed from the pre-merge branch.
• node tests/test-em-graph-links.mjs on merged main → 31 passed, 0 failed.

One line, docs only.

🤖 Generated with Claude Code

[lantiscooperdev]

Ledger provenance — one-line docs change, ACCEPT

Closes the deferred obligation from #617. Single table row, no code.

Why this was deferred rather than shipped in #617

The row's SHA must be the merge commit, which does not exist until the PR lands. Writing the feature-branch tip there is the defect #582 was filed for and #605 had to repair. #617's body stated the deferral explicitly and #618 tracked it with all 5 DEFER fields, so this is a closed loop, not a rediscovery.

Provenance, derived not copied

$ git log --follow --oneline -1 scripts/em-graph.mjs
dafc3c8 RFC-007 Phase 3: wiki-link + composes-with edges ... (#617)

$ git log --follow --oneline -1 tests/test-em-graph-links.mjs
dafc3c8 RFC-007 Phase 3: wiki-link + composes-with edges ... (#617)

Both files the row claims resolve to dafc3c8, confirmed against gh pr view 617 --json mergeCommit → dafc3c89da4fb740ffd957074b434df1162219cd. Not taken from the adjacent Phase 2 row, not from a workplan note.

Verified before writing the row

• CI anchor tests.yml:407-408 read from the merged file, not assumed from the pre-merge branch.
• node tests/test-em-graph-links.mjs on merged main → 31 passed, 0 failed.

Post-merge deploy of #617 — all three legs, done and verified at dafc3c8

Leg	Verification	Observed
1. Global	deploy-audit.mjs unfiltered + deployed-copy grep	MISSING=0 DIFFER=0 EXTRA=0 UNDEPLOYED=0 (one cosmetic timestamp); install stamped dafc3c89da4f; deployed em-graph.mjs greps 13 for Phase 3 symbols
2. Consumers	disk-scan shasum sweep, not the installer report	25 of 26 at source 017b1bdd; the 26th is episodic-memory-pi's own source file, the standing deliberate collision
3. Marketplace	git log -1 + grep	clone 4a6fd3a → dafc3c8, Phase 3 symbols present

Leg 1 needed a second pass: the first --install-hooks-force left second-opinion-providers.json stale (DIFFER=1) and only --install-second-opinion cleared it. The audit named the fix itself.

Leg 2 is the one worth flagging: --update-consumers reported files: [] for every consumer, which is exactly the report that concealed three stale v0.2.0 copies on 2026-07-25 (#601). The hash sweep is what proves the fleet current, not the installer's self-report.

Live smoke test of the deployed global copy

$ node ~/.episodic-memory/scripts/em-graph.mjs --nodes rule --edges wiki-link --hubs --top 3
verify-by-artifact-not-narrative-and-run-negative-scenarios   degree 18
surface-tool-errors-offer-to-look-together-before-routing-around  degree 9
use-claude-subagent-not-codex-for-second-opinion-reviews      degree 9

Phase 3 is live against the real rule corpus, not just green in a fixture.

Follow-ups still open from the #617 arc

• #619 — em-console.mjs:114 whitelist omits --nodes/--edges.
• #620 — unclosed code fence does not suppress extraction (affects shipped cites too).
• #621 — recency ordering is minute-granular, so a fork-closure stub and its terminal are unorderable. Found while consolidating a two-session fork; already bit the documented handoff-recall command once.

What user approval is checking

Per Rule 17, your approval in the UI is the review-required gate. This is a one-line docs change with derived provenance and CI 8/8 green. Nothing architectural.

🤖 Posted by lantiscooperdev (bot account) — Claude Opus 5 (1M context).