Testing plan: record-feedback appends duplicate sibling headings when a gate is approved twice on one day

Everything is proved in this repository with the existing pytest suite, plus one run of the pinned markdownlint over a twice-recorded artifact. The hook is pure file I/O on a temporary directory; no service, credential or network is involved.

Test matrix

#	Type	Applies?	Scope / what it proves	Where it runs
T1	Unit	n/a — _unique_heading is exercised through the hook in T2; a separate unit test would assert the same strings twice
T2	Integration (scenario)	yes	two and three same-day recordings into one artifact yield distinct ### headings; a first recording is unchanged; earlier headings are untouched (R1, R2, R3.1)	uv run pytest cli/tests/test_graph_integration.py
T3	Contract	n/a — no API, schema or CLI shape changes
T4	End-to-end	n/a — the gate calls the same hook function T2 calls; no live GitHub needed
T5	UI / visual	n/a — no UI
T6	Snapshot	n/a — no snapshot surfaces
T7	Performance	n/a — one linear scan of an artifact the hook already reads
T8	Security / abuse case	n/a — the heading takes no untrusted input (design §Security design); the existing authorization tests for classify-feedback are unchanged and run in T12
T9	Accessibility	n/a — no UI
T10	Migration / upgrade	n/a — artifacts already on disk are read, never rewritten (R2.2)
T11	Real linter over a recorded artifact	yes	the shape T2 asserts is the shape MD024 accepts (R3.2)	npx --yes markdownlint-cli2@0.18.1 <artifact>
T12	Full suite + lint/type	yes	nothing else depended on the old heading	make check

Scenarios & requirement trace

Row	Requirement(s)	Scenario / case
T2	R1.1, R1.3, R3.1	Scenario: a gate approved twice on one day records distinct headings
T2	R1.2, R2.1, R2.2	same scenario: the first heading stays <date> — approved, byte-identical after the later recordings
T11	R3.2	the artifact T2's scenario produces, written out and linted
T12	R2.1	the issue-247 tests (attribution, empty body) still pass

Verification environment

The repository checkout with make install-dev done; Node.js for npx (T11 only). No credentials.

Evidence plan

Row	Evidence	Path under evidence/
T2	the new test failing before the fix, passing after	integration.md
T11	the recorded artifact and the linter's summary	markdownlint.md
T12	make check summary	lint-and-typecheck.md

Verification activities

• [x] T2 — the new scenario fails on the pre-fix hook and passes after
• [x] T11 — markdownlint reports 0 issues on a twice-recorded artifact
• [x] T12 — make check passes

Verification results

Activity	Command / procedure	Outcome	Evidence
T2	uv run pytest cli/tests/test_graph_integration.py	the new scenario failed on the pre-fix hook (second heading repeated the first); after the fix 30 passed	integration.md
T11	the real gate chain records three approvals into a scratch artifact; npx --yes markdownlint-cli2@0.18.1 with the repository's config	0 errors; the same artifact with the suffixes removed gives two MD024 errors	markdownlint.md
T12	make check, with the daemon's THE_LOOP_* variables unset	exit 0: lint clean, 6789 passed	lint-and-typecheck.md

Not executed: none. T12's first run inside the session failed 22 unrelated tests that read the operator's live config through inherited environment variables; the note in the evidence explains it.