From 4bbad24221ff519d1f02942a4f538f2b3e1d1645 Mon Sep 17 00:00:00 2001
From: Garry Tan <garrytan@gmail.com>
Date: Mon, 2 Nov 2026 04:58:56 -0800
Subject: [PATCH 1/6] v1.92.77.0 wave(github): #3724 #3771 #3833 (#3846)

* test-free-shards: --ci-run --jobs k sub-shards inside one CI job (k=1 default), per-file JUnit counts (GSTA-203 PR 2)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(ci): ship ffmpeg in the eval image so the marathon design-motion case can run (#3687)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* shard-jobs: serial sub-shard gets what is left of the shard wall; register shard-jobs.ts in ship-measure touchfiles

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* shard-jobs: validate every sub-shard summary, bounded JUnit read, stricter --ci-verify records, unique count paths, no serial launch without wall

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* shard-jobs tests: non-hex plan id fixture; harness-version: shard-jobs FREE_ONLY entry is non-behavioral

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(test:free): exact root-bound JUnit path matching; serial timeout text names the wall the parallel phase left

Adversarial pass 2 (GSTA-203 PR 2): a plain relative path now matches a planned
file exactly, and an absolute or ../-prefixed path matches only when its prefix is
the child's root, so an unplanned browse/test/x.test.ts is never credited to
test/x.test.ts. The serial sub-shard's FREE_SUBSHARD_TIMEOUT text says how much of
the shard wall the parallel phase left it.

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* docs(testing): real-home tripwire covers --jobs k>1 parallel sub-shards (docs audit docs-203-1)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* ci(free): per-shard sub-shard k in the CI plan (--shard-jobs), free_jobs and windows_jobs dispatch inputs (k=1 defaults)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* ci(free-tests): 10 x ubicloud-standard-8 at 4 sub-shards per shard (GSTA-203 PR 3 A/B winner); Windows stays at k=1

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix: pre-landing review fixes (changelog names main's 20 x standard-4 baseline and the unchanged 80 vCPUs)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* docs(testing): sub-shard default is the plan's per-shard k (docs audit docs-pr3-2)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* ci(free): run the #3680/#3679 BSD-userland regression on macos-named-regressions

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* changelog: macos-named-regressions runs the BSD-userland ubi-runner regression (#3680)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* docs: Windows k=1 applies to the nightly suite since #3806 (docs re-audit after the main merge)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

---------

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---
 .github/docker/Dockerfile.ci                  | 15 +++++
 .github/workflows/free-tests.yml              | 35 ++++++++----
 .github/workflows/windows-free-tests.yml      | 21 ++++++-
 CHANGELOG.md                                  | 12 ++++
 VERSION                                       |  2 +-
 agents-digest/gstack-AGENTS.md                |  2 +-
 docs/TESTING_INTERNALS.md                     | 55 +++++++++++++------
 package.json                                  |  2 +-
 scripts/lib/shard-jobs.ts                     | 31 +++++++++--
 scripts/test-free-shards.ts                   | 50 ++++++++++++-----
 test/ci-image-cli-pin.test.ts                 | 19 +++++++
 test/free-tests-workflow-wiring.test.ts       | 14 +++--
 ...on-issue3680-ubicloud-bsd-userland.test.ts | 10 ++++
 test/shard-jobs.test.ts                       | 55 +++++++++++++++++--
 test/windows-native-workflows.test.ts         | 36 +++++++++++-
 15 files changed, 297 insertions(+), 62 deletions(-)

diff --git a/.github/docker/Dockerfile.ci b/.github/docker/Dockerfile.ci
index 841291cca..1ed8c693d 100644
--- a/.github/docker/Dockerfile.ci
+++ b/.github/docker/Dockerfile.ci
@@ -193,6 +193,21 @@ RUN for i in 1 2 3; do \
     && fc-cache -f \
     && rm -rf /var/lib/apt/lists/*
 
+# ffmpeg (with ffprobe): /design-motion's checks and exports shell out to it
+# (lib/motion-encode.ts), and the marathon design-motion-showcase case runs
+# that flow end to end. The image had none, so every trial stopped at
+# FFMPEG_MISSING (#3687). The build fails if the packaged ffmpeg ever lacks a
+# filter or encoder the checks and exports use.
+RUN for i in 1 2 3; do \
+      apt-get update && apt-get install -y --no-install-recommends ffmpeg && break || \
+      (echo "ffmpeg install retry $i/3"; sleep 10); \
+    done \
+    && rm -rf /var/lib/apt/lists/* \
+    && command -v ffprobe \
+    && filters="$(ffmpeg -hide_banner -filters)" && encoders="$(ffmpeg -hide_banner -encoders)" \
+    && for f in psnr gblur palettegen paletteuse signalstats tile; do echo "$filters" | grep -qw "$f" || { echo "ERROR: ffmpeg lacks the $f filter" >&2; exit 1; }; done \
+    && for e in libx264 gif; do echo "$encoders" | grep -qw "$e" || { echo "ERROR: ffmpeg lacks the $e encoder" >&2; exit 1; }; done
+
 # Pre-install dependencies (cached layer — only rebuilds when package.json or
 # bun.lock changes). Copy BOTH so install is deterministic and matches local
 # resolution. Without bun.lock here, bun install resolved transitive deps
diff --git a/.github/workflows/free-tests.yml b/.github/workflows/free-tests.yml
index 966419ad3..0d3cff19e 100644
--- a/.github/workflows/free-tests.yml
+++ b/.github/workflows/free-tests.yml
@@ -41,13 +41,21 @@ on:
         description: 'Runner-sizing experiments only: shard count, 1-40 (empty keeps FREE_SUITE_SHARDS)'
         type: string
         default: ''
+      free_jobs:
+        description: 'Runner-sizing experiments only: sub-shards per CI shard (--shard-jobs), 1-8 (empty keeps FREE_SUITE_JOBS)'
+        type: string
+        default: ''
 
-# The one source of the free-suite runner and shard count. pull_request and
-# push runs never see workflow_dispatch inputs, so free-plan resolves these
-# (or a dispatch override) into outputs that free-suite reads.
+# The one source of the free-suite runner, shard count and sub-shards per
+# shard. pull_request and push runs never see workflow_dispatch inputs, so
+# free-plan resolves these (or a dispatch override) into outputs and the plan
+# that free-suite reads. FREE_SUITE_JOBS is k in `--ci-plan --shard-jobs k`:
+# each shard runs min(k, its parallel files) sandboxed sub-shards side by side
+# (scripts/lib/shard-jobs.ts); the exclusive host-state shard stays one child.
 env:
-  FREE_SUITE_RUNNER: ubicloud-standard-4
-  FREE_SUITE_SHARDS: '20'
+  FREE_SUITE_RUNNER: ubicloud-standard-8
+  FREE_SUITE_SHARDS: '10'
+  FREE_SUITE_JOBS: '4'
 
 # Keyed on the PR number, not head_ref: a bare branch name carries no fork
 # prefix, so same-name branches from two forks would share one group and a
@@ -120,10 +128,12 @@ jobs:
         env:
           RUNNER_OVERRIDE: ${{ inputs.free_runner }}
           SHARDS_OVERRIDE: ${{ inputs.free_shards }}
+          JOBS_OVERRIDE: ${{ inputs.free_jobs }}
         run: |
           set -euo pipefail
           runner="${RUNNER_OVERRIDE:-$FREE_SUITE_RUNNER}"
           shards="${SHARDS_OVERRIDE:-$FREE_SUITE_SHARDS}"
+          jobs="${JOBS_OVERRIDE:-$FREE_SUITE_JOBS}"
           case "$runner" in
             ubicloud-standard-2|ubicloud-standard-4|ubicloud-standard-8) ;;
             *) echo "::error::free_runner '$runner' is not ubicloud-standard-2, -4 or -8"; exit 1 ;;
@@ -132,10 +142,14 @@ jobs:
             ''|*[!0-9]*) echo "::error::free_shards '$shards' is not a whole number"; exit 1 ;;
           esac
           if [ "$shards" -lt 1 ] || [ "$shards" -gt 40 ]; then echo "::error::free_shards $shards is outside 1-40"; exit 1; fi
-          if [ "$runner" != "$FREE_SUITE_RUNNER" ] || [ "$shards" != "$FREE_SUITE_SHARDS" ]; then
-            echo "- Runner-sizing experiment: $shards shards on $runner (defaults $FREE_SUITE_SHARDS on $FREE_SUITE_RUNNER); not a PR or train Free Tests result" >> "$GITHUB_STEP_SUMMARY"
+          case "$jobs" in
+            ''|*[!0-9]*) echo "::error::free_jobs '$jobs' is not a whole number"; exit 1 ;;
+          esac
+          if [ "$jobs" -lt 1 ] || [ "$jobs" -gt 8 ]; then echo "::error::free_jobs $jobs is outside 1-8"; exit 1; fi
+          if [ "$runner" != "$FREE_SUITE_RUNNER" ] || [ "$shards" != "$FREE_SUITE_SHARDS" ] || [ "$jobs" != "$FREE_SUITE_JOBS" ]; then
+            echo "- Runner-sizing experiment: $shards shards on $runner at $jobs sub-shards each (defaults $FREE_SUITE_SHARDS on $FREE_SUITE_RUNNER at $FREE_SUITE_JOBS); not a PR or train Free Tests result" >> "$GITHUB_STEP_SUMMARY"
           fi
-          matrix=$(bun run scripts/test-free-shards.ts --ci-plan "$RUNNER_TEMP/free-plan.json" --shards "$shards")
+          matrix=$(bun run scripts/test-free-shards.ts --ci-plan "$RUNNER_TEMP/free-plan.json" --shards "$shards" --shard-jobs "$jobs")
           echo "matrix=$matrix" >> "$GITHUB_OUTPUT"
           echo "runner=$runner" >> "$GITHUB_OUTPUT"
       - name: Seed growth ratchet (60 s ceiling, shrink-only allowlist)
@@ -186,7 +200,8 @@ jobs:
   # code that compares a realpath with an unresolved path passes on Linux's
   # /tmp and fails here (#2982). TMPDIR is deliberately left at the runner's
   # default, and the job fails if that layout is not symlinked. It also sources
-  # the Codex probe from zsh, macOS's default shell (#3024; no paid call).
+  # the Codex probe from zsh, macOS's default shell (#3024; no paid call), and
+  # runs ubi-runner under macOS's bsdtar and one-true-awk (#3680, #3679).
   macos-named-regressions:
     needs: reuse
     if: ${{ !cancelled() && (needs.reuse.result != 'success' || needs.reuse.outputs.reused != 'true') }}
@@ -229,7 +244,7 @@ jobs:
           resolved="$(realpath "$tmp")"
           echo "temp root: $tmp -> $resolved"
           test "$resolved" != "${tmp%/}"
-          files=(test/gstack-design-detect.test.ts browse/test/chromium-profile-isolation.test.ts test/cso-git-hardening.test.ts test/plan-count-file-permission.test.ts test/spec-quality-gate-secret-sink.test.ts test/gstack-memory-ingest.test.ts test/regression-issue3024-zsh-sourced-helpers.test.ts test/codex-probe-subcommands.test.ts browse/test/proxy-env-daemon.test.ts browse/test/proxy-direct-launch-args.test.ts)
+          files=(test/gstack-design-detect.test.ts browse/test/chromium-profile-isolation.test.ts test/cso-git-hardening.test.ts test/plan-count-file-permission.test.ts test/spec-quality-gate-secret-sink.test.ts test/gstack-memory-ingest.test.ts test/regression-issue3024-zsh-sourced-helpers.test.ts test/codex-probe-subcommands.test.ts browse/test/proxy-env-daemon.test.ts browse/test/proxy-direct-launch-args.test.ts test/regression-issue3680-ubicloud-bsd-userland.test.ts)
           for file in "${files[@]}"; do test -f "$file"; done
           bun test "${files[@]}"
 
diff --git a/.github/workflows/windows-free-tests.yml b/.github/workflows/windows-free-tests.yml
index 6146e927d..d94d1d6f2 100644
--- a/.github/workflows/windows-free-tests.yml
+++ b/.github/workflows/windows-free-tests.yml
@@ -42,6 +42,10 @@ on:
         description: Record per-file Windows durations (artifact) instead of running the lane
         type: boolean
         default: false
+      windows_jobs:
+        description: 'Sub-shard experiments only: sub-shards per Windows shard (--shard-jobs), 1-4 (empty keeps WINDOWS_FREE_JOBS). Never cite such a run as a PR result.'
+        type: string
+        default: ''
   # free-durations-record.yml calls the recording path weekly.
   workflow_call:
     inputs:
@@ -49,6 +53,11 @@ on:
         type: boolean
         default: false
 
+# k in `--ci-plan --shard-jobs k`: each Windows shard runs min(k, its parallel
+# files) sandboxed sub-shards side by side (scripts/lib/shard-jobs.ts).
+env:
+  WINDOWS_FREE_JOBS: '1'
+
 concurrency:
   group: windows-free-${{ github.event.pull_request.number || github.run_id }}
   cancel-in-progress: true
@@ -105,8 +114,18 @@ jobs:
         # 4-vCPU Windows runner, so the lane lands near 4 min with setup.
         env:
           GSTACK_FREE_TEST_DURATIONS: scripts/free-test-durations-windows.json
+          JOBS_OVERRIDE: ${{ inputs.windows_jobs }}
         run: |
-          matrix=$(bun run scripts/test-free-shards.ts --windows-only --ci-plan "$RUNNER_TEMP/windows-plan.json" --shards 6)
+          set -euo pipefail
+          jobs="${JOBS_OVERRIDE:-$WINDOWS_FREE_JOBS}"
+          case "$jobs" in
+            ''|*[!0-9]*) echo "::error::windows_jobs '$jobs' is not a whole number"; exit 1 ;;
+          esac
+          if [ "$jobs" -lt 1 ] || [ "$jobs" -gt 4 ]; then echo "::error::windows_jobs $jobs is outside 1-4"; exit 1; fi
+          if [ "$jobs" != "$WINDOWS_FREE_JOBS" ]; then
+            echo "- Sub-shard experiment: $jobs sub-shards per Windows shard (default $WINDOWS_FREE_JOBS); not a PR Windows Free Tests result" >> "$GITHUB_STEP_SUMMARY"
+          fi
+          matrix=$(bun run scripts/test-free-shards.ts --windows-only --ci-plan "$RUNNER_TEMP/windows-plan.json" --shards 6 --shard-jobs "$jobs")
           echo "matrix=$matrix" >> "$GITHUB_OUTPUT"
       - uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7
         with:
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 0c95b691a..bf0437f2c 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,17 @@
 # Changelog
 
+## [1.92.77.0] - 2026-11-02
+
+**The weekly marathon evals can run /design-motion again: the CI eval image now ships ffmpeg (#3687).** The `design-motion-showcase` case joined the marathon tier on 2026-10-26, and its first scheduled run (Actions run 41200316994, main @ 088bfc5) failed 0/3. Every trial built and captured the page, then stopped at `FFMPEG_MISSING`, because the image had no ffmpeg and the skill rightly refuses to export before its checks run. The model did nothing wrong; the case could not pass on that image.
+
+### Fixed
+- **`.github/docker/Dockerfile.ci`** installs Ubuntu's `ffmpeg` (6.1.1, with `ffprobe`). The build fails if that ffmpeg lacks a filter or encoder the motion checks and exports use (`psnr`, `gblur`, `palettegen`, `paletteuse`, `signalstats`, `tile`, `libx264`, `gif`).
+- **`test/ci-image-cli-pin.test.ts`** fails the free suite if the image stops installing ffmpeg or stops checking it.
+
+The macOS `macos-named-regressions` CI job now runs `test/regression-issue3680-ubicloud-bsd-userland.test.ts`, so the ubi-runner cases for #3680 (bsdtar) and #3679 (macOS awk) run on a real Mac against `/usr/bin/tar` and `/usr/bin/awk`. Before, the file ran only on Linux CI, where bsdtar and one-true-awk aren't installed, so every case skipped and a regression in either could merge green. A new wiring test fails if the macOS job stops naming the file.
+
+The Linux free suite now runs on 10 `ubicloud-standard-8` machines, each running its tests as 4 sandboxed sub-shards side by side, instead of main's 20 `ubicloud-standard-4` machines (set by #3710) running one file at a time. Both use 80 vCPUs per run. In a 3-run A/B at one commit against that main setup, the slowest shard dropped from 389–450 s to 224–258 s and the Ubicloud cost per run dropped from about $0.39 to $0.32 at premium prices, with the same per-file test counts and no new failures. The CI plan now stores each shard's sub-shard count (`--ci-plan --shard-jobs k`), keeps the exclusive host-state shard at one child, and `--ci-verify` rejects a shard that ran at a different count. The nightly Windows suite stays at one child per shard, because two sub-shards made Windows file-lock cleanup failures show up in two of three runs. Dispatch-only `free_jobs` and `windows_jobs` inputs exist for future experiments. See docs/TESTING_INTERNALS.md#sub-shards-inside-one-ci-job.
+
 ## [1.92.76.0] - 2026-11-02
 
 **/retro now remembers its history across git worktrees, so a run in a linked worktree compares against last week instead of announcing "First retro recorded".** Retro snapshots used to live in `.context/retros/` inside the working tree. A worktree (Conductor, Claude Code) saw none of the main checkout's history, and its own snapshot was deleted along with the worktree (#3749).
diff --git a/VERSION b/VERSION
index 730cd3e00..463685b58 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-1.92.76.0
+1.92.77.0
diff --git a/agents-digest/gstack-AGENTS.md b/agents-digest/gstack-AGENTS.md
index f297fb296..ac6c1e65e 100644
--- a/agents-digest/gstack-AGENTS.md
+++ b/agents-digest/gstack-AGENTS.md
@@ -1,4 +1,4 @@
-# gstack digest v1.92.76.0 — regenerate/re-copy after upgrading gstack
+# gstack digest v1.92.77.0 — regenerate/re-copy after upgrading gstack
 
 Behavioral rules from gstack (https://github.com/garrytan/gstack), compressed
 for agent hosts without a full skill install. The full skills add workflows,
diff --git a/docs/TESTING_INTERNALS.md b/docs/TESTING_INTERNALS.md
index 91c400897..441525762 100644
--- a/docs/TESTING_INTERNALS.md
+++ b/docs/TESTING_INTERNALS.md
@@ -195,15 +195,18 @@ fallback; unknown files get 99th-percentile pessimism, and both full-suite and
 the long pole. Packed
 shards get duration-aware walls (`max(base, predicted × 3, files × 5s)`). The
 legacy `--shards N --shard i` path keeps stable hash indices. Required CI uses
-one duration-packed `--ci-plan`, 20 ordinary `--ci-run` shards plus a separate
+one duration-packed `--ci-plan`, 10 ordinary `--ci-run` shards plus a separate
 exclusive-fixture shard, and a `--ci-verify` aggregate. CI shards still run on
-independent machines without ordering unrelated jobs. Two workflow-level
+independent machines without ordering unrelated jobs. Three workflow-level
 constants in `free-tests.yml` set that size for every event: `FREE_SUITE_RUNNER`
-(`ubicloud-standard-4`) and `FREE_SUITE_SHARDS` (`'20'`), which `free-plan`
-resolves into the shards' runner and shard count. A `workflow_dispatch` run may
-override them with `free_runner` (`ubicloud-standard-2`, `-4` or `-8`) and
-`free_shards` (1–40) for runner-sizing experiments only; its job summary says
-so, and it is never cited as a PR or train Free Tests result. Each shard prints
+(`ubicloud-standard-8`), `FREE_SUITE_SHARDS` (`'10'`) and `FREE_SUITE_JOBS`
+(`'4'`, the [sub-shards](#sub-shards-inside-one-ci-job) each shard runs side by
+side), which `free-plan` resolves into the shards' runner, shard count and plan.
+GSTA-203 chose them by a 3-run A/B against 20 × `ubicloud-standard-4` at k = 1
+(the PR that set them has the table). A `workflow_dispatch` run may
+override them with `free_runner` (`ubicloud-standard-2`, `-4` or `-8`),
+`free_shards` (1–40) and `free_jobs` (1–8) for runner-sizing experiments only;
+its job summary says so, and it is never cited as a PR or train Free Tests result. Each shard prints
 `RESOURCES: peak_used_mib=… min_free_disk_mib=… samples=…` from a memory and
 disk sample every 2 seconds, even when the suite fails. The public
 `TREE_MUTATING` map now classifies exclusive host-state fixtures; its sole
@@ -1553,7 +1556,7 @@ refresh to a maintainer.
   `scripts/ubicloud/setup-free-suite.sh` and the Windows seed through
   `windows-free-tests.yml`'s `record_durations` path. The recorder stays on
   `ubicloud-standard-8` while the free-suite shards run on `FREE_SUITE_RUNNER`
-  (`ubicloud-standard-4`); both are Ubicloud standard runners. When either recording adds
+  (`ubicloud-standard-8`); both are Ubicloud standard runners. When either recording adds
   or drops a test file, it force-updates the `bot/free-duration-seeds` branch,
   opens or refreshes one seed PR whose body states the ratchet result, and
   dispatches Free Tests, Windows Smoke and Quality gate on it. Those
@@ -1575,14 +1578,24 @@ k sub-shards side by side on the same machine (`scripts/lib/shard-jobs.ts`). Eac
 sub-shard is an ordinary free shard child: one `bun test` process, serial
 inside, with its own sandbox, process-group kill and log. Bun's worker
 `--parallel` mode is still not used (decision V3/D6 in
-`scripts/test-free-shards.ts`). The default is `k = 1`, which runs today's one
-child, so no workflow changes behavior until it passes `--jobs`.
+`scripts/test-free-shards.ts`). Without `--jobs`, a shard runs at the plan's
+per-shard k (below); a shard with no planned k runs one child (`k = 1`).
 
-- `--jobs` applies only to `--ci-run`. It takes a positive whole number up to the
-  shard's file count (an empty shard takes 1); anything else stops with
-  `FREE_JOBS_INVALID`. `GSTACK_FREE_JOBS` sets the local full-suite process count
-  and never changes `--ci-run`. The exclusive host-state shard has one file, so
-  a workflow passes it `--jobs 1` or no `--jobs`.
+- Workflows set k in the plan, per shard: `--ci-plan <plan> --shards N
+  --shard-jobs k` stores `jobs = min(k, the shard's files that may run in
+  parallel)` on each shard, and only when that is above 1. The exclusive
+  host-state shard holds only `TREE_MUTATING` files, so it stays one child, and
+  a plan made at k = 1 is byte-identical to one made without `--shard-jobs`.
+  `--ci-run` runs the shard at the plan's k, and `--ci-verify` rejects any
+  shard result that ran at a different k. `free-tests.yml` reads k from
+  `FREE_SUITE_JOBS` and `windows-free-tests.yml` from `WINDOWS_FREE_JOBS`; the
+  `free_jobs` and `windows_jobs` dispatch inputs override them for experiments
+  only.
+- `--jobs` applies only to `--ci-run` and overrides the plan's k, for local
+  reproduction. It takes a positive whole number up to the shard's file count
+  (an empty shard takes 1); anything else stops with `FREE_JOBS_INVALID`, as
+  does a bad `--shard-jobs` or one without `--ci-plan`. `GSTACK_FREE_JOBS` sets
+  the local full-suite process count and never changes `--ci-run`.
 - Files are split into disjoint sub-shards packed by the duration seed. Each
   sub-shard has an id (`<plan id prefix>:<shard>.<n>`) bound to its file list.
   Files in `TREE_MUTATING` or `SLOW_UNDER_LOAD` run after the parallel
@@ -1606,6 +1619,13 @@ reproduce command. If the file passes serially, add it to `SLOW_UNDER_LOAD` in
 `scripts/lib/shard-jobs.ts` with that run id and a reason. Never raise the
 timeout. `SLOW_UNDER_LOAD` is empty today.
 
+The nightly Windows lane stays at `WINDOWS_FREE_JOBS: '1'`. In GSTA-203's A/B, k = 2
+made the slowest Windows shard faster, but two of three runs needed the flaky
+retry for file-lock cleanup errors (`EBUSY` in
+`browse/test/stealth-webdriver.test.ts` and a native fixture cleanup in
+`browse/test/cookie-import-native-job.test.ts`), which k = 1 never showed. Fix
+those cleanups before raising it.
+
 Every `--ci-run` result, at any k, records per-file test and `expect()` counts
 in `outcome.counts` (`tests`, `pass`, `fail`, `skip`, `todo`, `expects` per
 file, plus shard totals). They come from bun's JUnit reporter, which the runner
@@ -1624,8 +1644,9 @@ test list in its last step, where each file's reason sits beside it.
 `test/windows-native-workflows.test.ts` keeps that list Windows-safe and fast.
 
 The curated suite below runs nightly on main and on dispatch, never on a PR.
-`windows-free-tests.yml` plans on Linux (`--windows-only --ci-plan --shards 6`),
-runs one strict serial shard per `windows-latest` job
+`windows-free-tests.yml` plans on Linux (`--windows-only --ci-plan --shards 6
+--shard-jobs $WINDOWS_FREE_JOBS`), runs one strict serial shard per
+`windows-latest` job
 (`bun run test:windows --ci-run`), and verifies every shard receipt on Linux
 (`--windows-only --ci-verify`). Packing uses Windows-measured durations from
 `scripts/free-test-durations-windows.json`; the weekly
diff --git a/package.json b/package.json
index c5271f2da..8d40c09f7 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "gstack",
-  "version": "1.92.76",
+  "version": "1.92.77",
   "description": "Garry's Stack — Claude Code skills + fast headless browser. One repo, one install, entire AI engineering workflow.",
   "license": "MIT",
   "type": "module",
diff --git a/scripts/lib/shard-jobs.ts b/scripts/lib/shard-jobs.ts
index c96100e01..d0510fb77 100644
--- a/scripts/lib/shard-jobs.ts
+++ b/scripts/lib/shard-jobs.ts
@@ -45,16 +45,39 @@ export const FREE_JOBS_INVALID = 'FREE_JOBS_INVALID';
 export const FREE_SUBSHARD_TIMEOUT = 'FREE_SUBSHARD_TIMEOUT';
 export const FREE_SUBSHARD_REJECTED = 'FREE_SUBSHARD_REJECTED';
 const DOCS = 'docs/TESTING_INTERNALS.md#sub-shards-inside-one-ci-job';
-const JOBS_FIX = `--jobs applies only to --ci-run and defaults to 1; GSTACK_FREE_JOBS sets the local full-suite process count and never changes --ci-run. See ${DOCS}`;
+const JOBS_FIX = `--jobs applies only to --ci-run and defaults to the plan's per-shard k (1 when the plan has none); --shard-jobs applies only to --ci-plan; GSTACK_FREE_JOBS sets the local full-suite process count and never changes --ci-run. See ${DOCS}`;
 
-/** `--jobs <k>` at parse time: digits only, at least 1. */
-export function parseJobsFlag(raw: string | undefined): number {
+/** `--jobs <k>` or `--shard-jobs <k>` at parse time: digits only, at least 1. */
+export function parseJobsFlag(raw: string | undefined, flag = '--jobs'): number {
   if (raw === undefined || !/^\d+$/.test(raw) || Number(raw) < 1) {
-    throw new Error(`${FREE_JOBS_INVALID}: --jobs needs a positive whole number; got ${JSON.stringify(raw ?? '')}. ${JOBS_FIX}`);
+    throw new Error(`${FREE_JOBS_INVALID}: ${flag} needs a positive whole number; got ${JSON.stringify(raw ?? '')}. ${JOBS_FIX}`);
   }
   return Number(raw);
 }
 
+export function rejectShardJobsWithoutCiPlan(): never {
+  throw new Error(`${FREE_JOBS_INVALID}: --shard-jobs needs --ci-plan. ${JOBS_FIX}`);
+}
+
+/**
+ * One planned shard's k under `--ci-plan --shard-jobs k`: at most one parallel
+ * sub-shard per file that may run in parallel, so a shard of only serial files
+ * (the exclusive TREE_MUTATING shard) keeps k = 1. Undefined means k = 1, and
+ * the plan then carries no `jobs` key, so its id matches a plan without k.
+ */
+export function plannedShardJobs(files: readonly string[], shardJobs: number, isSerial: (file: string) => boolean): number | undefined {
+  const jobs = Math.min(shardJobs, files.filter(file => !isSerial(file)).length);
+  return jobs > 1 ? jobs : undefined;
+}
+
+/** Why a plan's per-shard k cannot stand; null when absent or valid. */
+export function plannedShardJobsProblem(shard: { shard: number; files: readonly string[]; jobs?: unknown }, isSerial: (file: string) => boolean): string | null {
+  if (shard.jobs === undefined) return null;
+  const parallel = shard.files.filter(file => !isSerial(file)).length;
+  return Number.isInteger(shard.jobs) && (shard.jobs as number) >= 2 && (shard.jobs as number) <= parallel ? null
+    : `CI shard ${shard.shard} plans --jobs ${JSON.stringify(shard.jobs)} for ${parallel} parallel file(s)`;
+}
+
 export function rejectJobsWithoutCiRun(): never {
   throw new Error(`${FREE_JOBS_INVALID}: --jobs needs --ci-run. ${JOBS_FIX}`);
 }
diff --git a/scripts/test-free-shards.ts b/scripts/test-free-shards.ts
index deda761cf..d61ca6112 100755
--- a/scripts/test-free-shards.ts
+++ b/scripts/test-free-shards.ts
@@ -77,11 +77,15 @@
  *   bun run scripts/test-free-shards.ts --attribute-home           # name real-home writers: each file alone, private HOME
  *   bun run scripts/test-free-shards.ts --ci-plan plan.json --shards 20
  *   bun run scripts/test-free-shards.ts --ci-run plan.json --shard 1 --result result.json
+ *   bun run scripts/test-free-shards.ts --ci-plan plan.json --shards 20 --shard-jobs 2
+ *       # the plan stores a per-shard k: min(k, the shard's files that may run in parallel),
+ *       # kept only when above 1, so the exclusive TREE_MUTATING shard stays one child.
  *   bun run scripts/test-free-shards.ts --ci-run plan.json --shard 1 --result result.json --jobs 4
- *       # k sub-shards side by side inside one CI job (default 1 = one child, today's behavior;
- *       # scripts/lib/shard-jobs.ts). --jobs applies only to --ci-run: a positive whole number up
- *       # to the shard's file count, else FREE_JOBS_INVALID. GSTACK_FREE_JOBS sets the local
- *       # full-suite process count and never changes --ci-run; --jobs is its only knob.
+ *       # k sub-shards side by side inside one CI job (default: the plan's per-shard k, else 1 =
+ *       # one child, today's behavior; scripts/lib/shard-jobs.ts). --jobs applies only to
+ *       # --ci-run and overrides the plan: a positive whole number up to the shard's file count,
+ *       # else FREE_JOBS_INVALID. --ci-verify requires every result to have run at the plan's k.
+ *       # GSTACK_FREE_JOBS sets the local full-suite process count and never changes --ci-run.
  *   bun run scripts/test-free-shards.ts --ci-verify plan.json --results results/
  */
 
@@ -97,8 +101,8 @@ import { attributeFreeHomeWriters, guardFreeHome, sharedFreeHome, type FreeHomeG
 import { appendStepSummary, ciHealthSummary, SEED_REFRESH_COMMAND, unseededWarning, windowsCurationLine } from './lib/free-ci-health';
 import { curateWindowsSafe, type CurationResult } from './lib/windows-curation';
 import {
-  assertSoloSeedSample, ciRunContext, parseJobsFlag, parseJunitCounts, readJunitReport, rejectJobsWithoutCiRun, runCiShard, SLOW_UNDER_LOAD, subShardRecordProblems,
-  validateJobsForShard, type ShardTestCounts, type SubShardRecord,
+  assertSoloSeedSample, ciRunContext, parseJobsFlag, parseJunitCounts, plannedShardJobs, plannedShardJobsProblem, readJunitReport, rejectJobsWithoutCiRun,
+  rejectShardJobsWithoutCiPlan, runCiShard, SLOW_UNDER_LOAD, subShardRecordProblems, validateJobsForShard, type ShardTestCounts, type SubShardRecord,
 } from './lib/shard-jobs';
 export { curateWindowsSafe, detectWindowsFragility, KNOWN_WINDOWS_INCOMPATIBLE, type CurationResult } from './lib/windows-curation';
 import {
@@ -424,19 +428,27 @@ export interface FreeCiPlan {
   version: 1;
   revision: string;
   id: string;
-  shards: Array<{ shard: number; files: string[]; predictedMs: number }>;
+  /** `jobs` is the shard's k from --shard-jobs, present only when above 1. */
+  shards: Array<{ shard: number; files: string[]; predictedMs: number; jobs?: number }>;
 }
 
+/** Files that run in the trailing serial sub-shard under k > 1, never beside another child. */
+const runsSerialUnderJobs = (file: string) => file in TREE_MUTATING || file in SLOW_UNDER_LOAD;
+
 const planDigest = (plan: Omit<FreeCiPlan, 'id'>): string =>
   createHash('sha256').update(JSON.stringify(plan)).digest('hex');
 
 /** One immutable plan is shared by isolated CI machines; never repack per job. */
-export function createFreeCiPlan(files: string[], count: number, durations: Record<string, number>, revision: string): FreeCiPlan {
+export function createFreeCiPlan(files: string[], count: number, durations: Record<string, number>, revision: string, shardJobs = 1): FreeCiPlan {
   const readers = files.filter(file => !(file in TREE_MUTATING));
   const exclusive = files.filter(file => file in TREE_MUTATING).sort();
   const packed = packShardsByDuration(readers, count, durations);
-  const shards = packed.shards.map((files, index) => ({ shard: index + 1, files, predictedMs: packed.predictedMs[index] }));
-  if (exclusive.length) shards.push({ shard: shards.length + 1, files: exclusive, predictedMs: exclusive.reduce((ms, file) => ms + (durations[file] ?? 0), 0) });
+  const planned = packed.shards.map((files, index) => ({ files, predictedMs: packed.predictedMs[index] }));
+  if (exclusive.length) planned.push({ files: exclusive, predictedMs: exclusive.reduce((ms, file) => ms + (durations[file] ?? 0), 0) });
+  const shards = planned.map(({ files, predictedMs }, index) => {
+    const jobs = plannedShardJobs(files, shardJobs, runsSerialUnderJobs);
+    return { shard: index + 1, files, predictedMs, ...(jobs ? { jobs } : {}) };
+  });
   const body = { version: 1 as const, revision, shards };
   return { ...body, id: planDigest(body) };
 }
@@ -447,6 +459,8 @@ export function validateFreeCiPlan(plan: FreeCiPlan, files: string[], revision:
     || id !== planDigest({ version, revision: plan.revision, shards })) throw new Error('CI plan identity or revision mismatch');
   if (shards.some((shard, index) => shard.shard !== index + 1 || !Array.isArray(shard.files)
     || !Number.isFinite(shard.predictedMs) || shard.predictedMs < 0)) throw new Error('Invalid CI shard plan');
+  const jobsProblems = shards.map(shard => plannedShardJobsProblem(shard, runsSerialUnderJobs)).filter(Boolean);
+  if (jobsProblems.length) throw new Error(`Invalid CI shard plan: ${jobsProblems.join('; ')}`);
   const planned = shards.flatMap(shard => shard.files).sort();
   if (new Set(planned).size !== planned.length || JSON.stringify(planned) !== JSON.stringify([...files].sort())) throw new Error('CI plan must cover every free file exactly once');
 }
@@ -486,6 +500,7 @@ export function verifyFreeCiResults(plan: FreeCiPlan, results: FreeCiResult[]):
     if (!hasCompleteCiSummary(outcome)) throw new Error('Missing or incomplete CI execution summary');
     const subShardProblems = subShardRecordProblems(result);
     if (subShardProblems.length) throw new Error(`CI shard ${outcome.shard} sub-shards rejected: ${subShardProblems.join('; ')}`);
+    if ((result.jobs ?? 1) !== (shard.jobs ?? 1)) throw new Error(`CI shard ${outcome.shard} ran at --jobs ${result.jobs ?? 1}; the plan says ${shard.jobs ?? 1}`);
     if (outcome.status === 'passed') {
       if (outcome.exitCode !== 0 || outcome.failingFiles.length || outcome.unattributedFailures || result.retry) throw new Error('Inconsistent passing CI result');
     } else {
@@ -556,8 +571,10 @@ type CliOptions = {
   result: string | null;
   results: string | null;
   filesFrom: string | null;
-  /** Set only by --jobs (with --ci-run); absent means k = 1. */
+  /** Set only by --jobs (with --ci-run); absent means the plan's per-shard k. */
   jobs?: number;
+  /** Set only by --shard-jobs (with --ci-plan); absent means k = 1 in every shard. */
+  shardJobs?: number;
 };
 
 export function parseCliOptions(argv: string[]): CliOptions {
@@ -573,6 +590,7 @@ export function parseCliOptions(argv: string[]): CliOptions {
   let quick = false;
   let attributeHome = false;
   let jobs: number | undefined;
+  let shardJobs: number | undefined;
   const paths: Record<'ciPlan' | 'ciRun' | 'ciVerify' | 'result' | 'results' | 'filesFrom', string | null> = {
     ciPlan: null, ciRun: null, ciVerify: null, result: null, results: null, filesFrom: null,
   };
@@ -597,6 +615,7 @@ export function parseCliOptions(argv: string[]): CliOptions {
     '--results': pathFlag('--results', 'results'),
     '--files-from': pathFlag('--files-from', 'filesFrom'),
     '--jobs': (next) => { jobs = parseJobsFlag(next()); },
+    '--shard-jobs': (next) => { shardJobs = parseJobsFlag(next(), '--shard-jobs'); },
     '--shards': (next) => {
       const value = next();
       if (!value) throw new Error('Missing value for --shards');
@@ -621,8 +640,9 @@ export function parseCliOptions(argv: string[]): CliOptions {
   if (paths.ciRun && (shardIndex === null || !paths.result)) throw new Error('--ci-run requires --shard and --result');
   if (paths.ciVerify && !paths.results) throw new Error('--ci-verify requires --results');
   if (jobs !== undefined && !paths.ciRun) rejectJobsWithoutCiRun();
+  if (shardJobs !== undefined && !paths.ciPlan) rejectShardJobsWithoutCiPlan();
   if (quick && (recordDurations || windowsOnly || shardIndex !== null)) throw new Error('--quick cannot change recording, Windows or shard selection');
-  return { dryRun, listOnly, recordDurations, windowsOnly, verbose, shardCount, shardIndex, wallTimeoutMs, wallTimeoutExplicit, quick, attributeHome, ...paths, ...(jobs === undefined ? {} : { jobs }) };
+  return { dryRun, listOnly, recordDurations, windowsOnly, verbose, shardCount, shardIndex, wallTimeoutMs, wallTimeoutExplicit, quick, attributeHome, ...paths, ...(jobs === undefined ? {} : { jobs }), ...(shardJobs === undefined ? {} : { shardJobs }) };
 }
 
 const describeLines = (lines: string[]) => `${lines.slice(0, 5).join(', ')}${lines.length > 5 ? ` (+${lines.length - 5} more)` : ''}`;
@@ -1954,7 +1974,7 @@ async function main(): Promise<number> {
         appendStepSummary(line);
       }
       warnUnseededFreeFiles(ciFiles, durations);
-      const plan = createFreeCiPlan(ciFiles, options.shardCount, durations, revision);
+      const plan = createFreeCiPlan(ciFiles, options.shardCount, durations, revision, options.shardJobs);
       validateFreeCiPlan(plan, ciFiles, revision);
       writeJson(options.ciPlan, plan);
       console.log(JSON.stringify({ shard: plan.shards.map(shard => shard.shard) }));
@@ -1975,7 +1995,7 @@ async function main(): Promise<number> {
     }
     const shard = plan.shards[options.shardIndex! - 1];
     if (!shard || shard.shard !== options.shardIndex) throw new Error('CI shard index is outside the plan');
-    const jobs = options.jobs ?? 1;
+    const jobs = options.jobs ?? shard.jobs ?? 1;
     validateJobsForShard(jobs, shard.files.length, shard.shard);
     const durations = loadFreeTestDurations() ?? {};
     const { result, exitCode } = await runCiShard({
@@ -1986,7 +2006,7 @@ async function main(): Promise<number> {
       runShard: (files, shardNumber, totalShards, runOptions) => runFreeShard(files, shardNumber, totalShards, runOptions),
       retry: outcomes => retryFailedFreeFiles(outcomes, plan.shards.length, options),
       pack: (files, count) => packShardsByDuration(files, count, durations).shards,
-      isSerial: file => file in TREE_MUTATING || file in SLOW_UNDER_LOAD,
+      isSerial: runsSerialUnderJobs,
       terminationRequested: isTerminationRequested,
     });
     writeJson(options.result!, result);
diff --git a/test/ci-image-cli-pin.test.ts b/test/ci-image-cli-pin.test.ts
index c4a84f21a..1a94c107c 100644
--- a/test/ci-image-cli-pin.test.ts
+++ b/test/ci-image-cli-pin.test.ts
@@ -256,3 +256,22 @@ describe('CI Node.js installs match the image and are verified (#3354)', () => {
     expect(unpack).toBeGreaterThan(verify);
   });
 });
+
+// Value: protects=the eval image ships an ffmpeg that runs /design-motion's checks and exports;
+//   fails_when=Dockerfile.ci stops installing ffmpeg or stops failing the build on a missing filter or encoder;
+//   why_new=the marathon design-motion-showcase case ran 0/3 on an image without ffmpeg (#3687); seam=none;
+//   no_claim=reads the Dockerfile only; the image build runs the check itself
+describe('ci image ffmpeg for the design-motion evals (#3687)', () => {
+  const lines = logicalLines(fs.readFileSync(DOCKERFILE, 'utf-8')).filter((line) => !/^\s*#/.test(line));
+
+  test('apt installs ffmpeg, which the marathon design-motion-showcase case needs', () => {
+    expect(lines.some((line) => /apt-get install\b[^&|;]*\sffmpeg\b/.test(line))).toBe(true);
+  });
+
+  test('the build fails when that ffmpeg cannot run the motion checks and exports', () => {
+    const verify = lines.find((line) => /\bffmpeg\b.*-filters/.test(line)) ?? '';
+    for (const need of ['ffprobe', 'psnr', 'gblur', 'palettegen', 'paletteuse', 'signalstats', 'tile', 'libx264', 'gif']) {
+      expect(verify, `Dockerfile.ci must check ffmpeg for ${need}`).toContain(need);
+    }
+  });
+});
diff --git a/test/free-tests-workflow-wiring.test.ts b/test/free-tests-workflow-wiring.test.ts
index b4426e205..0b36507d7 100644
--- a/test/free-tests-workflow-wiring.test.ts
+++ b/test/free-tests-workflow-wiring.test.ts
@@ -174,7 +174,8 @@ describe('free-tests workflow wiring', () => {
     expect(workflow.env.FREE_SUITE_SHARDS).toMatch(/^[1-9][0-9]?$/);
     expect(planner.outputs.runner).toBe('${{ steps.plan.outputs.runner }}');
     expect(workflow.jobs['free-suite']['runs-on']).toBe('${{ needs.free-plan.outputs.runner }}');
-    expect(plan.env).toEqual({ RUNNER_OVERRIDE: '${{ inputs.free_runner }}', SHARDS_OVERRIDE: '${{ inputs.free_shards }}' });
+    expect(workflow.env.FREE_SUITE_JOBS).toMatch(/^[1-8]$/);
+    expect(plan.env).toEqual({ RUNNER_OVERRIDE: '${{ inputs.free_runner }}', SHARDS_OVERRIDE: '${{ inputs.free_shards }}', JOBS_OVERRIDE: '${{ inputs.free_jobs }}' });
     expect(plan.run).not.toContain('${{');
     const directory = fs.mkdtempSync(path.join(os.tmpdir(), 'free plan '));
     const bin = path.join(directory, 'bin');
@@ -185,7 +186,7 @@ describe('free-tests workflow wiring', () => {
       const output = path.join(directory, 'output');
       fs.writeFileSync(output, '');
       const result = spawnSync('bash', ['-c', plan.run], {
-        env: { ...process.env, ...workflow.env, RUNNER_OVERRIDE: '', SHARDS_OVERRIDE: '', ...overrides,
+        env: { ...process.env, ...workflow.env, RUNNER_OVERRIDE: '', SHARDS_OVERRIDE: '', JOBS_OVERRIDE: '', ...overrides,
           PATH: `${toShell(bin)}:${process.env.PATH}`, RUNNER_TEMP: toShell(directory),
           GITHUB_OUTPUT: toShell(output), GITHUB_STEP_SUMMARY: toShell(path.join(directory, 'summary')) },
         encoding: 'utf8', timeout: 5000,
@@ -199,12 +200,13 @@ describe('free-tests workflow wiring', () => {
       const unset = resolve({});
       expect(unset.status).toBe(0);
       expect(unset.outputs.runner).toBe(workflow.env.FREE_SUITE_RUNNER);
-      expect(unset.args).toContain(`--shards ${workflow.env.FREE_SUITE_SHARDS}`);
-      const override = resolve({ RUNNER_OVERRIDE: 'ubicloud-standard-4', SHARDS_OVERRIDE: '16' });
+      expect(unset.args).toContain(`--shards ${workflow.env.FREE_SUITE_SHARDS} --shard-jobs ${workflow.env.FREE_SUITE_JOBS}`);
+      const override = resolve({ RUNNER_OVERRIDE: 'ubicloud-standard-4', SHARDS_OVERRIDE: '16', JOBS_OVERRIDE: '3' });
       expect(override.status).toBe(0);
       expect(override.outputs.runner).toBe('ubicloud-standard-4');
-      expect(override.args).toContain('--shards 16');
-      const refusals: Record<string, string>[] = [{ RUNNER_OVERRIDE: 'ubuntu-24.04' }, { RUNNER_OVERRIDE: 'ubicloud-standard-8 x' }, { SHARDS_OVERRIDE: '0' }, { SHARDS_OVERRIDE: '41' }, { SHARDS_OVERRIDE: '16; true' }];
+      expect(override.args).toContain('--shards 16 --shard-jobs 3');
+      const refusals: Record<string, string>[] = [{ RUNNER_OVERRIDE: 'ubuntu-24.04' }, { RUNNER_OVERRIDE: 'ubicloud-standard-8 x' }, { SHARDS_OVERRIDE: '0' }, { SHARDS_OVERRIDE: '41' }, { SHARDS_OVERRIDE: '16; true' },
+        { JOBS_OVERRIDE: '0' }, { JOBS_OVERRIDE: '9' }, { JOBS_OVERRIDE: '2; true' }, { JOBS_OVERRIDE: '-1' }];
       for (const bad of refusals) {
         const refused = resolve(bad);
         expect(refused.status, JSON.stringify(bad)).not.toBe(0);
diff --git a/test/regression-issue3680-ubicloud-bsd-userland.test.ts b/test/regression-issue3680-ubicloud-bsd-userland.test.ts
index a6fd98a86..85561a029 100644
--- a/test/regression-issue3680-ubicloud-bsd-userland.test.ts
+++ b/test/regression-issue3680-ubicloud-bsd-userland.test.ts
@@ -120,3 +120,13 @@ printf 200
     } finally { rmSync(f.root, { recursive: true, force: true }); }
   }, 60_000);
 });
+
+describe('CI runs these cases on the BSD userland itself', () => {
+  // Value: protects=the cases above run where /usr/bin/tar and /usr/bin/awk are the BSD tools; fails_when=macos-named-regressions stops naming this file, so Linux CI (no bsdtar or one-true-awk installed) skips every case and a #3680 or #3679 regression ships green; why_new=the file was only in the Linux free suite; seam=.github/workflows/free-tests.yml; no_claim=the job's runner image or its tool versions
+  test('free-tests.yml macos-named-regressions names this file', () => {
+    const workflow = Bun.YAML.parse(readFileSync(join(ROOT, '.github/workflows/free-tests.yml'), 'utf8')) as any;
+    const mac = workflow.jobs['macos-named-regressions'];
+    expect(mac['runs-on']).toMatch(/^macos-/);
+    expect((mac.steps as any[]).map(step => String(step.run ?? '')).join('\n')).toContain('test/regression-issue3680-ubicloud-bsd-userland.test.ts');
+  });
+});
diff --git a/test/shard-jobs.test.ts b/test/shard-jobs.test.ts
index 83714013d..af18dfa54 100644
--- a/test/shard-jobs.test.ts
+++ b/test/shard-jobs.test.ts
@@ -10,11 +10,11 @@ import * as fs from 'node:fs';
 import * as os from 'node:os';
 import * as path from 'node:path';
 import {
-  collectFreeTestFiles, createFreeCiPlan, eligibleFreeRetryFiles, parseCliOptions, runFreeShard, TREE_MUTATING, verifyFreeCiResults,
+  collectFreeTestFiles, createFreeCiPlan, eligibleFreeRetryFiles, parseCliOptions, runFreeShard, TREE_MUTATING, validateFreeCiPlan, verifyFreeCiResults,
   type FreeCiResult, type FreeShardOutcome,
 } from '../scripts/test-free-shards';
 import {
-  assertSoloSeedSample, assignSubShards, ciRunContext, FREE_JOBS_INVALID, mergeSubShardOutcomes, parseJobsFlag, parseJunitCounts,
+  assertSoloSeedSample, assignSubShards, ciRunContext, FREE_JOBS_INVALID, mergeSubShardOutcomes, parseJobsFlag, parseJunitCounts, plannedShardJobs, plannedShardJobsProblem,
   readJunitReport, runCiShard, SLOW_UNDER_LOAD, subShardRecordProblems, subShardTimeoutText, validateJobsForShard, type CiShardDeps, type RunShardOptions,
 } from '../scripts/lib/shard-jobs';
 
@@ -89,6 +89,52 @@ describe('--jobs validation (FREE_JOBS_INVALID)', () => {
   });
 });
 
+describe('per-shard k in the plan (--ci-plan --shard-jobs)', () => {
+  test('--shard-jobs belongs to --ci-plan only and takes the same positive whole numbers', () => {
+    const base = ['--ci-plan', 'plan.json', '--shards', '4'];
+    expect(parseCliOptions([...base, '--shard-jobs', '2']).shardJobs).toBe(2);
+    expect('shardJobs' in parseCliOptions(base)).toBe(false);
+    for (const raw of ['0', '2.5', 'x']) expect(() => parseCliOptions([...base, '--shard-jobs', raw])).toThrow(`${FREE_JOBS_INVALID}: --shard-jobs needs a positive whole number`);
+    expect(() => parseCliOptions(['--shard-jobs', '2'])).toThrow(`${FREE_JOBS_INVALID}: --shard-jobs needs --ci-plan`);
+    expect(() => parseCliOptions(['--ci-run', 'p.json', '--shard', '1', '--result', 'r.json', '--shard-jobs', '2'])).toThrow(FREE_JOBS_INVALID);
+  });
+
+  test('each shard plans min(k, its parallel files); the exclusive shard and one-file shards stay one child, and k = 1 keeps the plan id', () => {
+    const withExclusive = [...files, EXCLUSIVE];
+    const plain = createFreeCiPlan(withExclusive, 3, durations, 'revision-a');
+    expect(createFreeCiPlan(withExclusive, 3, durations, 'revision-a', 1)).toEqual(plain);
+    expect(plain.shards.every(shard => !('jobs' in shard))).toBe(true);
+    const planned = createFreeCiPlan(withExclusive, 3, durations, 'revision-a', 4);
+    expect(planned.id).not.toBe(plain.id);
+    expect(planned.shards.map(shard => shard.files)).toEqual(plain.shards.map(shard => shard.files));
+    const exclusiveShard = planned.shards.at(-1)!;
+    expect(exclusiveShard.files).toEqual([EXCLUSIVE]);
+    expect('jobs' in exclusiveShard).toBe(false);
+    for (const shard of planned.shards.slice(0, -1)) expect(shard.jobs).toBe(Math.min(4, shard.files.length) > 1 ? Math.min(4, shard.files.length) : undefined);
+    expect(() => validateFreeCiPlan(planned, withExclusive, 'revision-a')).not.toThrow();
+    expect(plannedShardJobs(['a', 'b', 'c'], 2, file => file === 'a')).toBe(2);
+    expect(plannedShardJobs(['a', 'b'], 4, file => file === 'a')).toBeUndefined();
+    expect(plannedShardJobs([], 4, () => false)).toBeUndefined();
+  });
+
+  test('a plan whose per-shard k is out of range is invalid, and every shard result must have run at its planned k', () => {
+    const isSerial = (file: string) => file === EXCLUSIVE;
+    expect(plannedShardJobsProblem({ shard: 1, files }, isSerial)).toBeNull();
+    expect(plannedShardJobsProblem({ shard: 1, files, jobs: 7 }, isSerial)).toBeNull();
+    for (const jobs of [1, 0, 8, 2.5, '2']) expect(plannedShardJobsProblem({ shard: 1, files, jobs }, isSerial)).toContain('CI shard 1 plans --jobs');
+    expect(plannedShardJobsProblem({ shard: 2, files: [EXCLUSIVE, files[0]], jobs: 2 }, isSerial)).toContain('for 1 parallel file(s)');
+    const plan = createFreeCiPlan(files, 1, durations, 'revision-a', 2);
+    expect(plan.shards[0].jobs).toBe(2);
+    const tampered = structuredClone(plan);
+    tampered.shards[0].jobs = 99;
+    expect(() => validateFreeCiPlan(tampered, files, 'revision-a')).toThrow();
+    const k1: FreeCiResult = { planId: plan.id, revision: 'revision-a', outcome: passed(plan.shards[0].files), retry: null };
+    expect(() => verifyFreeCiResults(plan, [k1])).toThrow('CI shard 1 ran at --jobs 1; the plan says 2');
+    const plainPlan = createFreeCiPlan(files, 1, durations, 'revision-a');
+    expect(() => verifyFreeCiResults(plainPlan, [{ ...k1, planId: plainPlan.id }])).not.toThrow();
+  });
+});
+
 describe('k = 1 golden: PR 2 changes nothing in CI', () => {
   test('one child over the whole planned shard with the same wall, one retry pass, and the same result keys', async () => {
     const h = harness(files);
@@ -160,7 +206,8 @@ describe('k > 1: merge, rejection and one retry pass per CI shard', () => {
       summary: { testsRan: 16, filesRan: 8, sawTerminalSummary: true } });
     expect(result.subShards!.map(record => record.status)).toEqual(['passed', 'passed', 'passed', 'passed']);
     expect(h.retries).toHaveLength(1);
-    expect(() => verifyFreeCiResults(h.plan, [result])).not.toThrow();
+    expect(() => verifyFreeCiResults({ ...h.plan, shards: [{ ...h.shard, jobs: 3 }] }, [result])).not.toThrow();
+    expect(() => verifyFreeCiResults(h.plan, [result])).toThrow('CI shard 1 ran at --jobs 3; the plan says 1');
   });
 
   test('failures from every sub-shard are collected first, so k never multiplies the five-file retry cap', async () => {
@@ -263,7 +310,7 @@ describe('k > 1: merge, rejection and one retry pass per CI shard', () => {
   });
 
   test('--ci-verify rejects tampered sub-shard records', () => {
-    const plan = createFreeCiPlan(files, 1, durations, 'revision-a');
+    const plan = createFreeCiPlan(files, 1, durations, 'revision-a', 2);
     const shard = plan.shards[0];
     const assignments = assignSubShards(plan.id, shard, 2, () => false, packFiles => [packFiles.slice(0, 3), packFiles.slice(3)]);
     const merged = mergeSubShardOutcomes(shard, assignments.map(assignment => ({ assignment, outcome: passed(assignment.files) })), 10);
diff --git a/test/windows-native-workflows.test.ts b/test/windows-native-workflows.test.ts
index 48ce50d9d..a111dc605 100644
--- a/test/windows-native-workflows.test.ts
+++ b/test/windows-native-workflows.test.ts
@@ -4,7 +4,9 @@
  * that moved out of it.
  */
 import { expect, test } from 'bun:test';
-import { existsSync, readFileSync } from 'node:fs';
+import { spawnSync } from 'node:child_process';
+import { existsSync, mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from 'node:fs';
+import os from 'node:os';
 import path from 'node:path';
 import { collectFreeTestFiles, createFreeCiPlan, curateWindowsSafe, parseCliOptions, validateFreeCiPlan } from '../scripts/test-free-shards';
 import { windowsCurationLine } from '../scripts/lib/free-ci-health';
@@ -27,7 +29,7 @@ const WINDOWS_CURATED_FLOOR = 795;
 test('the Windows lane runs nightly and by hand, plans, runs one strict shard per job and verifies every result', () => {
   expect(Object.keys(windows.on).sort()).toEqual(['schedule', 'workflow_call', 'workflow_dispatch']);
   expect(windows.on.schedule).toEqual([{ cron: '17 9 * * *' }]);
-  expect(Object.keys(windows.on.workflow_dispatch.inputs)).toEqual(['record_durations']);
+  expect(Object.keys(windows.on.workflow_dispatch.inputs)).toEqual(['record_durations', 'windows_jobs']);
   expect(Object.keys(windows.on.workflow_call.inputs)).toEqual(['record_durations']);
   const plan = windows.jobs['windows-plan'];
   expect(plan['runs-on']).toBe('ubuntu-24.04');
@@ -50,6 +52,36 @@ test('the Windows lane runs nightly and by hand, plans, runs one strict shard pe
   expect(aggregate.steps.find((step: any) => step.name === 'Require exact coverage and complete strict results').run).toContain('--windows-only --ci-verify "$RUNNER_TEMP/windows-plan.json" --results ');
 });
 
+test.skipIf(!Bun.which('bash'))('windows-plan passes the one sub-shard constant, or a validated dispatch override, to --shard-jobs', () => {
+  const planStep = windows.jobs['windows-plan'].steps.find((step: any) => step.id === 'plan');
+  expect(windows.env.WINDOWS_FREE_JOBS).toMatch(/^[1-4]$/);
+  expect(planStep.env.JOBS_OVERRIDE).toBe('${{ inputs.windows_jobs }}');
+  expect(planStep.run).not.toContain('${{');
+  const directory = mkdtempSync(path.join(os.tmpdir(), 'windows plan '));
+  const bin = path.join(directory, 'bin');
+  mkdirSync(bin);
+  writeFileSync(path.join(bin, 'bun'), '#!/bin/sh\nprintf "%s\\n" "$*" > "$RUNNER_TEMP/bun-args"\necho \'{"shard":[1]}\'\n', { mode: 0o755 });
+  const toShell = (p: string) => p.split(path.sep).join('/');
+  const resolve = (override: string) => {
+    rmSync(path.join(directory, 'bun-args'), { force: true });
+    const result = spawnSync('bash', ['-c', planStep.run], {
+      env: { ...process.env, ...windows.env, JOBS_OVERRIDE: override, PATH: `${toShell(bin)}:${process.env.PATH}`, RUNNER_TEMP: toShell(directory),
+        GITHUB_OUTPUT: toShell(path.join(directory, 'output')), GITHUB_STEP_SUMMARY: toShell(path.join(directory, 'summary')) },
+      encoding: 'utf8', timeout: 5000,
+    });
+    return { status: result.status, args: existsSync(path.join(directory, 'bun-args')) ? readFileSync(path.join(directory, 'bun-args'), 'utf8') : '' };
+  };
+  try {
+    expect(resolve('')).toEqual({ status: 0, args: expect.stringContaining(`--shards 6 --shard-jobs ${windows.env.WINDOWS_FREE_JOBS}`) });
+    expect(resolve('2')).toEqual({ status: 0, args: expect.stringContaining('--shards 6 --shard-jobs 2') });
+    for (const bad of ['0', '5', '2; true', 'x']) {
+      const refused = resolve(bad);
+      expect(refused.status, bad).not.toBe(0);
+      expect(refused.args, bad).toBe('');
+    }
+  } finally { rmSync(directory, { recursive: true, force: true }); }
+});
+
 test('a Windows CI plan binds the curated file set, so it never verifies as the full Linux suite', () => {
   for (const argv of [['--windows-only', '--ci-plan', 'p.json'], ['--windows-only', '--ci-run', 'p.json', '--shard', '1', '--result', 'r.json'], ['--windows-only', '--ci-verify', 'p.json', '--results', 'd']]) {
     expect(parseCliOptions(argv).windowsOnly).toBe(true);

From eff19e267854c8fba5a254d8c6d419d9335e9e59 Mon Sep 17 00:00:00 2001
From: Garry Tan <garrytan@gmail.com>
Date: Mon, 2 Nov 2026 05:07:54 -0800
Subject: [PATCH 2/6] v1.92.78.0 test: seven flaky-test repairs and six failure
 diagnostics from the 2026-11-01 flaky pass (GSTA-240) (#3804)

* test: eight flaky-test repairs and five failure diagnostics from the 2026-11-01 flaky pass (GSTA-240)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(free-shards): keep failing a shard on an unseen browser child, and name it and whether it exited (review finding)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* changelog: pregate-regate copies through gitFixtureTemplate after the main merge

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

---------

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---
 CHANGELOG.md                            | 17 +++++++++
 VERSION                                 |  2 +-
 agents-digest/gstack-AGENTS.md          |  2 +-
 bin/gstack-codex-probe                  |  4 +-
 bin/gstack-evidence                     |  2 +-
 package.json                            |  2 +-
 scripts/lib/free-home-guard.ts          |  9 +++--
 scripts/test-free-shards.ts             |  2 +
 test/auq-parallel.test.ts               |  3 +-
 test/autoplan-publication-guard.test.ts |  4 +-
 test/bootstrap-retention-shard.test.ts  |  2 +-
 test/codex-probe-cache.test.ts          | 25 ++++++++++++
 test/evidence.test.ts                   |  8 ++++
 test/free-home-guard.test.ts            | 28 ++++++++++++++
 test/gstack-design-detect.test.ts       |  7 +++-
 test/office-hours-attempt.test.ts       |  3 +-
 test/pregate-regate.test.ts             | 51 +++++++++++++++++++++----
 test/session-update-autostash.test.ts   |  3 +-
 test/shard-engine.test.ts               |  5 ++-
 test/ship-budget.test.ts                |  2 +-
 test/ship-queue-train.test.ts           |  7 +++-
 21 files changed, 162 insertions(+), 26 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index bf0437f2c..517e1d270 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,22 @@
 # Changelog
 
+## [1.92.78.0] - 2026-11-02
+
+**Seven flaky tests from the 2026-11-01 flaky pass stop failing correct runs, and six more now name their cause when they fail.** One product lock broke a fresh lock it had misjudged as abandoned, a /ship lane tripped the free runner's home check with its own log, and the rest were fixtures that raced their own setup, leaned on wall-clock sleeps or read the caller's state root.
+
+### Fixed
+- **`bin/gstack-codex-probe`**: a waiter that cannot read an owner-less lock's age (the holder removed it in that instant, or `stat -f` printed filesystem text on Linux) now treats the lock as fresh instead of abandoned, so it no longer deletes a live owner's lock on a misread age and the probe no longer ends `MODEL_PROBE_INCONCLUSIVE` (Free Tests 41216564482, 41235671867, 41263831361). A stat stub for both race windows fails without the fix; 64 stress runs went from 2 failures to 0.
+- **`scripts/lib/free-home-guard.ts`, `bin/gstack-evidence`**: `gstack-evidence run` names its own lane log in `GSTACK_EVIDENCE_LOG`, and the real-home tripwire ignores exactly that file, so a /ship lane no longer trips over its own output. A foreign write under the home's `~/.gstack` still fails.
+- **`test/autoplan-publication-guard.test.ts`**: the render-root cases clear the state-root variables they don't name, so a `GSTACK_STATE_ROOT` exported by the gstack launch file no longer picks the root. Three cases failed every time under that variable, and the lookalike controls passed for the wrong reason.
+- **`test/gstack-design-detect.test.ts`**: the directory scan proves one batch by counting engine calls instead of sleeping near the 1,500 ms detect timeout (macOS job 124813933194). The product timeout is unchanged.
+- **`test/pregate-regate.test.ts`**: the two fixtures are built once and copied per test through the `gitFixtureTemplate` helper from #3586, so 8 builds become 2.
+- **`test/ship-queue-train.test.ts`**: the planted seam has its own name, so the fixture's own background `git maintenance` lock no longer collides with it (EEXIST, Free Tests 41288253305, 41288417894), and the test now holds a real `maintenance.lock` beside it.
+- **`test/ship-budget.test.ts`**: the fake child writes its args to a temporary file and renames it, so the poll never reads an empty file (Free Tests 41242243205).
+
+### Changed
+- `scripts/test-free-shards.ts` names the recorded browser child it never saw in the daemon's process tree, and whether it was still alive, when it fails a shard for unconfirmed ownership (Free Tests 41264845860 shard 7). The shard still fails.
+- `test/bootstrap-retention-shard.test.ts`, `test/office-hours-attempt.test.ts`, `test/session-update-autostash.test.ts`, `test/shard-engine.test.ts` and `test/auq-parallel.test.ts` print the process state, exit status or cleanup events behind their assertion, so the next red names its cause.
+
 ## [1.92.77.0] - 2026-11-02
 
 **The weekly marathon evals can run /design-motion again: the CI eval image now ships ffmpeg (#3687).** The `design-motion-showcase` case joined the marathon tier on 2026-10-26, and its first scheduled run (Actions run 41200316994, main @ 088bfc5) failed 0/3. Every trial built and captured the page, then stopped at `FFMPEG_MISSING`, because the image had no ffmpeg and the skill rightly refuses to export before its checks run. The model did nothing wrong; the case could not pass on that image.
diff --git a/VERSION b/VERSION
index 463685b58..66231b754 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-1.92.77.0
+1.92.78.0
diff --git a/agents-digest/gstack-AGENTS.md b/agents-digest/gstack-AGENTS.md
index ac6c1e65e..9f27e6a29 100644
--- a/agents-digest/gstack-AGENTS.md
+++ b/agents-digest/gstack-AGENTS.md
@@ -1,4 +1,4 @@
-# gstack digest v1.92.77.0 — regenerate/re-copy after upgrading gstack
+# gstack digest v1.92.78.0 — regenerate/re-copy after upgrading gstack
 
 Behavioral rules from gstack (https://github.com/garrytan/gstack), compressed
 for agent hosts without a full skill install. The full skills add workflows,
diff --git a/bin/gstack-codex-probe b/bin/gstack-codex-probe
index 0292e5674..981ae1499 100755
--- a/bin/gstack-codex-probe
+++ b/bin/gstack-codex-probe
@@ -662,8 +662,8 @@ _gstack_codex_lock_acquire() {
       if [ "$_ohost" = "$_host" ] && ! kill -0 "$_pid" 2>/dev/null; then _dead=1; fi
       [ $((_now - _ts)) -ge "$_stale" ] && _dead=1
     else
-      _ts=$(stat -c %Y "$_dir" 2>/dev/null || stat -f %m "$_dir" 2>/dev/null || echo 0)
-      case "$_ts" in ''|*[!0-9]*) _ts=0 ;; esac
+      _ts=$(stat -c %Y "$_dir" 2>/dev/null || stat -f %m "$_dir" 2>/dev/null || echo "$_now")
+      case "$_ts" in ''|*[!0-9]*) _ts="$_now" ;; esac
       [ $((_now - _ts)) -ge 5 ] && _dead=1
     fi
     [ "$_dead" -eq 1 ] && _gstack_codex_lock_break "$_dir" "$_owner"
diff --git a/bin/gstack-evidence b/bin/gstack-evidence
index 59f8bbe30..35ca01d7b 100755
--- a/bin/gstack-evidence
+++ b/bin/gstack-evidence
@@ -400,7 +400,7 @@ async function cmdRun(argv: string[]): Promise<number> {
   try {
     if (shell && !shell.ok) throw new Error(shell.error);
     const spawnArgv = shell?.ok ? [shell.cmd, ...shell.argv] : cmdArgv;
-    proc = Bun.spawn(spawnArgv, { stdin: "inherit", stdout: "pipe", stderr: "pipe" });
+    proc = Bun.spawn(spawnArgv, { stdin: "inherit", stdout: "pipe", stderr: "pipe", env: log ? { ...process.env, GSTACK_EVIDENCE_LOG: log.path } : process.env });
   } catch (e: any) {
     // Spawn failure (ENOENT on argv-direct form): record exit 127, propagate 127.
     exitCode = 127;
diff --git a/package.json b/package.json
index 8d40c09f7..bce0fb11b 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "gstack",
-  "version": "1.92.77",
+  "version": "1.92.78",
   "description": "Garry's Stack — Claude Code skills + fast headless browser. One repo, one install, entire AI engineering workflow.",
   "license": "MIT",
   "type": "module",
diff --git a/scripts/lib/free-home-guard.ts b/scripts/lib/free-home-guard.ts
index 69b4bfd17..fb07555d3 100644
--- a/scripts/lib/free-home-guard.ts
+++ b/scripts/lib/free-home-guard.ts
@@ -5,8 +5,10 @@
  * the real home before and after each free shard, and names the shard's files
  * when one of those entries changed. It watches only these surfaces; it is not
  * a universal write-containment check. Live agent-session logs that a
- * concurrently running host keeps writing are excluded, and directory mtimes
- * are ignored, so only file, link and directory-set changes count.
+ * concurrently running host keeps writing are excluded, as is the one lane log
+ * `gstack-evidence run` tees this run's own output into (its exact path, from
+ * GSTACK_EVIDENCE_LOG). Directory mtimes are ignored, so only file, link and
+ * directory-set changes count.
  *
  * A shard that runs alone owns its window, so its guard names its files.
  * Concurrent shards share one HOME and cannot tell whose write a change was, so
@@ -104,6 +106,7 @@ export type FreeHomeGuardFactory = (files: string[], env: NodeJS.ProcessEnv, sta
 /** Take the baseline now; verify() returns null or the failure to report. */
 export function guardFreeHome(files: string[], env: NodeJS.ProcessEnv = process.env, scope: FreeHomeScope = { kind: 'shard' }): FreeHomeGuard {
   const home = freeHomeDir(env);
+  const ownLog = env.GSTACK_EVIDENCE_LOG && path.isAbsolute(env.GSTACK_EVIDENCE_LOG) ? path.relative(home, env.GSTACK_EVIDENCE_LOG) : null;
   const unreadable = (error: unknown) => `real home surfaces could not be snapshotted (${(error as NodeJS.ErrnoException).code ?? 'error'}); shard files: ${files.join(', ')}`;
   let before: FreeHomeSnapshot | undefined;
   let baselineError: string | undefined;
@@ -112,7 +115,7 @@ export function guardFreeHome(files: string[], env: NodeJS.ProcessEnv = process.
     verify() {
       if (!before) return baselineError!;
       try {
-        const changed = diffFreeHome(before, snapshotFreeHome(home));
+        const changed = diffFreeHome(before, snapshotFreeHome(home)).filter(entry => entry !== ownLog);
         return changed.length ? formatFreeHomeChange(changed, files, scope) : null;
       } catch (error) { return unreadable(error); }
     },
diff --git a/scripts/test-free-shards.ts b/scripts/test-free-shards.ts
index d61ca6112..b4012a60a 100755
--- a/scripts/test-free-shards.ts
+++ b/scripts/test-free-shards.ts
@@ -1370,6 +1370,8 @@ function trackShardBrowser(stateDir: string, env: NodeJS.ProcessEnv) {
         check(operation);
         if (!Number.isSafeInteger(pid) || (pid as number) <= 1) throw new BrowserCleanupError('invalid browser child identity');
         const identity = identities.get(pid as number);
+        if (!identity && identities.get(daemon as number)?.root) throw new BrowserCleanupError(
+          `browser child ownership is unconfirmed: recorded pid ${pid} (${await inspect(operation, pid as number) ? 'alive' : 'exited'}) was never seen in daemon ${daemon}'s process tree`);
         if (!identity || identity.daemon !== daemon || identity.root) throw new BrowserCleanupError('browser child ownership is unconfirmed');
         if (await live(operation, identity) && (typeof start !== 'string' || !start
           || await nativeStart(operation, identity) !== start.replace(/\s+/g, ' '))) throw new BrowserCleanupError('browser child identity was replaced');
diff --git a/test/auq-parallel.test.ts b/test/auq-parallel.test.ts
index d7296a5b1..5a6facc94 100644
--- a/test/auq-parallel.test.ts
+++ b/test/auq-parallel.test.ts
@@ -111,7 +111,8 @@ test.each([
   expect(facts.receipts.filter((receipt: any) => receipt.outcome === 'timeout')).toHaveLength(3 - scenario.passed);
   expect(facts.receipts.every((receipt: any) => receipt.timeoutMs === 240_000 && !receipt.answered)).toBe(true);
   expect(facts.active).toBe(0); expect(facts.answers).toBe(0); expect(facts.pendingTimers).toBe(0);
-  expect(facts.leftovers).toEqual([]); expect(facts.ownedStateLeftovers).toEqual([]);
+  expect(facts.leftovers).toEqual([]);
+  expect(facts.ownedStateLeftovers, JSON.stringify({ fixtures: facts.fixtures, inputs: facts.inputs.map((input: any) => [input.id, input.config, input.state]), events: facts.events.filter((event: any) => ['query-close', 'capture-settle', 'cleanup'].includes(event.kind)) })).toEqual([]);
   expect(facts.events.filter((event: any) => event.kind === 'query-close')).toHaveLength(scenario.queries);
   expect(facts.events.filter((event: any) => event.kind === 'cleanup')).toHaveLength(3);
 }, 20_000);
diff --git a/test/autoplan-publication-guard.test.ts b/test/autoplan-publication-guard.test.ts
index 44a375db4..d39a6baa2 100644
--- a/test/autoplan-publication-guard.test.ts
+++ b/test/autoplan-publication-guard.test.ts
@@ -1326,7 +1326,9 @@ describe('Autoplan phase entry through the gbrain :user render (#2569)', () => {
       edit(rewriteSectionBase(fs.readFileSync(path.join(ROOT, 'autoplan/sections', name), 'utf8'), linkRoot)));
     return root;
   }
-  function withEnv<T>(env: Record<string, string | undefined>, work: () => T): T {
+  /** The state-root chain the case does not name is cleared, so an ambient GSTACK_STATE_ROOT (every gstack launch file exports one) cannot pick the root. */
+  function withEnv<T>(given: Record<string, string | undefined>, work: () => T): T {
+    const env = { GSTACK_STATE_ROOT: undefined, GSTACK_STATE_DIR: undefined, CLAUDE_PLUGIN_DATA: undefined, ...given };
     const previous = Object.fromEntries(Object.keys(env).map(k => [k, process.env[k]]));
     for (const [k, v] of Object.entries(env)) if (v === undefined) delete process.env[k]; else process.env[k] = v;
     try { return work(); }
diff --git a/test/bootstrap-retention-shard.test.ts b/test/bootstrap-retention-shard.test.ts
index cbeac1275..2b120bd4e 100644
--- a/test/bootstrap-retention-shard.test.ts
+++ b/test/bootstrap-retention-shard.test.ts
@@ -62,7 +62,7 @@ describe.skipIf(process.platform !== 'linux')('bootstrap paid-shard cleanup inte
       fs.writeFileSync(path.join(root, 'controller.stdout.log'), controller.stdout ?? '', { mode: 0o600 });
       fs.writeFileSync(path.join(root, 'controller.stderr.log'), controller.stderr ?? '', { mode: 0o600 });
       expect(controller.error).toBeUndefined();
-      expect(controller.status).toBe(0);
+      expect(controller.status, `controller exited ${controller.status} (signal ${controller.signal}); stderr tail: ${(controller.stderr ?? '').slice(-600).replace(/\s*\n\s*/g, ' | ')}`).toBe(0);
       const outcome = JSON.parse(fs.readFileSync(outcomePath, 'utf8'));
       if (scenario === 'ack-failure') {
         expect(controller.stdout).toContain('(fail) qa-bootstrap');
diff --git a/test/codex-probe-cache.test.ts b/test/codex-probe-cache.test.ts
index 002fdb921..981fde6dd 100644
--- a/test/codex-probe-cache.test.ts
+++ b/test/codex-probe-cache.test.ts
@@ -247,6 +247,31 @@ describe('probe lock recovery and failure', () => {
     expect(calls(f)).toBe(1);
   }));
 
+  test('a lock released while a waiter reads its age is not abandoned; the next owner\'s fresh lock survives', withFixture((f) => {
+    const sig = sigOf(f);
+    const lock = path.join(f.locks, sig);
+    const once = path.join(f.home, 'vanish-once');
+    const real = JSON.stringify(Bun.which('stat'));
+    fs.mkdirSync(lock, { recursive: true });
+    fs.writeFileSync(path.join(f.home, 'bin/stat'), `#!/usr/bin/env bash
+if [ "$3" = "$RACE_LOCK" ] && [ -f "$RACE_ONCE" ]; then
+  case "$1" in
+    -c) rmdir "$RACE_LOCK"; [ "$RACE_PHASE" = early ] && mkdir "$RACE_LOCK"; exit 1 ;;
+    -f) ${real} "$@"; rc=$?; rm -f "$RACE_ONCE"; [ -d "$RACE_LOCK" ] || mkdir "$RACE_LOCK"; exit "$rc" ;;
+  esac
+fi
+exec ${real} "$@"
+`, { mode: 0o755 });
+    for (const phase of ['early', 'late']) {
+      fs.writeFileSync(once, 'armed');
+      const r = probe(f, { _GSTACK_CODEX_LOCK_WAIT: '1', RACE_LOCK: lock, RACE_ONCE: once, RACE_PHASE: phase });
+      expect(fs.existsSync(once)).toBe(false);
+      expect(r.stdout).toContain('MODEL_PROBE_INCONCLUSIVE (another probe of this model still holds its lock)');
+      expect(fs.existsSync(lock)).toBe(true);
+      expect(calls(f)).toBe(0);
+    }
+  }));
+
   test('an unreadable cache or uncreatable lock is unverified and launches no paid probe', withFixture((f) => {
     fs.mkdirSync(f.cache);
     const dir = probe(f);
diff --git a/test/evidence.test.ts b/test/evidence.test.ts
index e40d6769c..0617ef8ed 100644
--- a/test/evidence.test.ts
+++ b/test/evidence.test.ts
@@ -74,6 +74,14 @@ describe('gstack-evidence run', () => {
     expect(fs.readFileSync(rec.log_path, 'utf-8')).toContain('ok');
   });
 
+  test('the lane command sees its own log path as GSTACK_EVIDENCE_LOG (the free runner\'s home tripwire excludes exactly that file)', () => {
+    const r = run(['run', '--label', 'focused', '--', 'printf "%s" "$GSTACK_EVIDENCE_LOG"']);
+    expect(r.status).toBe(0);
+    const rec = records().pop();
+    expect(path.isAbsolute(rec.log_path)).toBe(true);
+    expect(r.stdout).toBe(rec.log_path);
+  });
+
   test('propagates a failing exit code and records it', () => {
     const r = run(['run', '--label', 'tests', '--', 'exit 3']);
     expect(r.status).toBe(3);
diff --git a/test/free-home-guard.test.ts b/test/free-home-guard.test.ts
index d01c8c979..4f387b5bd 100644
--- a/test/free-home-guard.test.ts
+++ b/test/free-home-guard.test.ts
@@ -78,6 +78,34 @@ describe('free-suite home-write tripwire (#2895)', () => {
     expect(failure).toContain('only');
   });
 
+  test.skipIf(process.platform === 'win32')('the wrapping gstack-evidence lane log is excluded by its exact path; a foreign write beside it still trips', async () => {
+    const home = fakeHome();
+    const logs = path.join(home, '.gstack/projects/p/logs');
+    fs.mkdirSync(logs, { recursive: true });
+    const laneLog = path.join(logs, 'focused.log');
+    fs.writeFileSync(laneLog, 'lane start\n');
+    const unexcluded = guardFreeHome(['test/a.test.ts'], { HOME: home });
+    const options = (script: string, name: string) => ({ env: { ...process.env, HOME: home, GSTACK_EVIDENCE_LOG: laneLog }, quiet: true, log: () => {},
+      logFilePath: path.join(home, `${name}.log`),
+      commandFor: () => ({ command: process.execPath, args: ['-e', `${script}console.log('Ran 1 tests across 1 files. [1.00ms]')`] }) });
+    const grow = `require('fs').appendFileSync(${JSON.stringify(laneLog)}, 'lane output\\n');`;
+    const errors: string[] = [];
+    const spy = spyOn(console, 'error').mockImplementation((...args: unknown[]) => { errors.push(args.join(' ')); });
+    try {
+      expect((await runFreeShard(['test/a.test.ts'], 1, 1, options(grow, 'lane'))).status).toBe('passed');
+      expect(unexcluded.verify()).toContain('~/.gstack/projects/p/logs/focused.log');
+      const foreign = `${grow}require('fs').writeFileSync(${JSON.stringify(path.join(logs, 'other.log'))}, 'x');`;
+      expect((await runFreeShard(['test/a.test.ts'], 1, 1, options(foreign, 'foreign'))).status).toBe('failed');
+      expect(errors.join('\n')).toContain('~/.gstack/projects/p/logs/other.log');
+      expect(errors.join('\n')).not.toContain('focused.log');
+    } finally {
+      spy.mockRestore();
+      for (const retained of errors.flatMap(line => /retained (\S+)$/.exec(line)?.[1] ?? [])) {
+        if (/^gstack-free-shard-/.test(path.basename(retained))) fs.rmSync(retained, { recursive: true, force: true });
+      }
+    }
+  });
+
   test('a concurrent phase guard names no shard and points at attribution', () => {
     const home = fakeHome();
     const guard = guardFreeHome(['test/a.test.ts', 'test/b.test.ts'], { HOME: home }, { kind: 'concurrent', shards: 3 });
diff --git a/test/gstack-design-detect.test.ts b/test/gstack-design-detect.test.ts
index 6349ec7d9..39388ac41 100644
--- a/test/gstack-design-detect.test.ts
+++ b/test/gstack-design-detect.test.ts
@@ -1460,8 +1460,13 @@ describe('scan: option-like bases and page-controlled inline ignores', () => {
     fs.mkdirSync(many, { recursive: true });
     for (let i = 0; i < 1100; i++) fs.writeFileSync(path.join(many, `f${i}.css`), 'a{}');
     try {
-      const r = run(['scan', '--format', 'gstack', 'src/many'], { env: { IMPECCABLE_BIN: FAKE, IMPECCABLE_FAKE_SLEEP_MS: '1000', GSTACK_DESIGN_DETECT_TIMEOUT_MS: '1500' } });
+      const log = path.join(SANDBOX, 'argv-many.log');
+      fs.rmSync(log, { force: true });
+      const r = run(['scan', '--format', 'gstack', 'src/many'], { env: { IMPECCABLE_BIN: FAKE, IMPECCABLE_FAKE_LOG: log } });
       expect(r.code).toBe(2); // a directory target is one batch (the fake reports findings): the budget test needs files
+      const calls = fs.readFileSync(log, 'utf-8').trim().split('\n').map(l => JSON.parse(l).argv as string[]);
+      expect(calls).toHaveLength(1);
+      expect(calls[0].some(x => x.endsWith(path.join('src', 'many')))).toBe(true);
       const files = fs.readdirSync(many).map(f => path.join('src', 'many', f));
       const r2 = run(['scan', '--format', 'gstack', ...files], { env: { IMPECCABLE_BIN: FAKE, IMPECCABLE_FAKE_SLEEP_MS: '1000', GSTACK_DESIGN_DETECT_TIMEOUT_MS: '1500' } });
       expect(r2.err).toMatch(/DETECT_TIMEOUT: whole-scan budget 7500ms exceeded, \d+ of 11 batches not run/);
diff --git a/test/office-hours-attempt.test.ts b/test/office-hours-attempt.test.ts
index c36da4486..c96bc8889 100644
--- a/test/office-hours-attempt.test.ts
+++ b/test/office-hours-attempt.test.ts
@@ -312,7 +312,8 @@ describe('Office Hours real session runner with fake processes', () => {
         expect(captured.output).toBe('captured output');
         const diagnostic = JSON.parse(fs.readFileSync(path.join(dir, '.gstack/test-transcripts/drain-failure.json'), 'utf8'));
         expect(diagnostic.stderr).toContain('fixture auth failure');
-        expect(running(Number(fs.readFileSync(path.join(dir, 'child.pid'), 'utf8')))).toBe(false);
+        const child = fs.readFileSync(path.join(dir, 'child.pid'), 'utf8').trim();
+        expect(running(Number(child)), `held-pipe child after the runner returned (group ${fs.readFileSync(path.join(dir, 'parent.pid'), 'utf8').trim()}): ${spawnSync('ps', ['-o', 'pid=,ppid=,pgid=,stat=,etime=', '-p', child], { encoding: 'utf8', timeout: 5_000 }).stdout.trim() || 'gone now'}`).toBe(false);
       });
     }, 10_000);
   }
diff --git a/test/pregate-regate.test.ts b/test/pregate-regate.test.ts
index 2e31a4c5f..09f01a004 100644
--- a/test/pregate-regate.test.ts
+++ b/test/pregate-regate.test.ts
@@ -6,11 +6,12 @@
  * the base branch and every other open PR, and a lookup that cannot finish is
  * `unknown`, which withholds `REGATE: not-needed`.
  */
-import { afterEach, describe, expect, test } from 'bun:test';
+import { afterAll, afterEach, beforeAll, describe, expect, test } from 'bun:test';
 import * as fs from 'node:fs';
 import * as path from 'node:path';
 import { EXIT } from '../lib/headless-artifacts';
-import { BASE_POLICY, cleanup, makeFixtureRepo, makePr, recordAndBundle, restamp, runBin, type FixtureRepo } from './helpers/restamp-fixture';
+import { gitFixtureTemplate, rebindFile, rebindGitDir, type GitFixtureTemplate } from './helpers/git-fixture-template';
+import { BASE_POLICY, cleanup, fixtureRepoAt, makePr, recordAndBundle, restamp, runBin, seedFixtureRepo, type FixtureRepo } from './helpers/restamp-fixture';
 
 const repos: FixtureRepo[] = [];
 afterEach(() => { while (repos.length) cleanup(repos.pop()!); });
@@ -39,10 +40,14 @@ console.error("gh shim: unsupported "+a.join(" ")); process.exit(1)' "${file}" "
 }
 const ghCalls = (repo: FixtureRepo) => fs.readFileSync(path.join(repo.root, 'gh-regate', 'calls.log'), 'utf8');
 
+type Lanes = Array<{ label: string; command: string }>;
+const TESTS_LANE: Lanes = [{ label: 'tests', command: 'true' }];
+const DIGEST_LANES: Lanes = [...TESTS_LANE, { label: 'digest', command: 'cat generated/digest.md' }];
+
 /** pr-a adds migration v240, is stamped 1.2.4.0 and gated (lanes recorded and bundled); then restamped to 1.2.5.0. */
-function gatedThenRestamped(lanes: Array<{ label: string; command: string }> = [{ label: 'tests', command: 'true' }]) {
-  const repo = makeFixtureRepo({ prefix: 'pregate-regate-', policy: POLICY });
-  repos.push(repo);
+function buildGated(root: string, lanes: Lanes) {
+  seedFixtureRepo(root, { policy: POLICY });
+  const repo = fixtureRepoAt(root);
   makePr(repo, 'pr-a', { expects: '1.2.4.0' });
   fs.mkdirSync(path.join(repo.work, 'migrations'), { recursive: true });
   fs.writeFileSync(path.join(repo.work, 'migrations', 'v240-add-index.sql'), 'create index a;\n');
@@ -54,7 +59,39 @@ function gatedThenRestamped(lanes: Array<{ label: string; command: string }> = [
   const head = repo.git('rev-parse', 'HEAD');
   expect(restamp(repo, ['--version', '1.2.5.0']).status).toBe(EXIT.ok);
   repo.git('commit', '-qam', 'stamp 1.2.5.0');
-  return { repo, head, bundle, current: repo.git('rev-parse', 'HEAD') };
+  return { head, current: repo.git('rev-parse', 'HEAD') };
+}
+
+// Each fixture costs four gstack bin runs (two restamps, an evidence run and a bundle); build each lane set once
+// and give every test its own copy instead of rebuilding it per test.
+const templates = new Map<Lanes, GitFixtureTemplate<ReturnType<typeof buildGated>>>();
+beforeAll(() => {
+  for (const lanes of [TESTS_LANE, DIGEST_LANES]) {
+    const template = gitFixtureTemplate({
+      prefix: 'pregate-regate-',
+      build: root => buildGated(root, lanes),
+      rebind: (copy, tpl) => {
+        rebindGitDir(path.join(copy, 'work', '.git'), tpl, copy);
+        rebindFile(path.join(copy, 'bundle.json'), tpl, copy);
+        const projects = path.join(copy, 'home', 'projects');
+        for (const slug of fs.existsSync(projects) ? fs.readdirSync(projects) : []) {
+          const dir = path.join(projects, slug.split(path.basename(tpl)).join(path.basename(copy)));
+          if (dir !== path.join(projects, slug)) fs.renameSync(path.join(projects, slug), dir);
+          for (const file of fs.readdirSync(dir, { recursive: true, withFileTypes: true })) if (file.isFile()) rebindFile(path.join(file.parentPath, file.name), tpl, copy);
+        }
+        fs.rmSync(path.join(copy, 'home', '.capy'), { recursive: true, force: true });
+      },
+    });
+    template.copy().root && templates.set(lanes, template);
+  }
+}, 60_000);
+afterAll(() => { for (const t of templates.values()) t.dispose(); });
+
+function gatedThenRestamped(lanes: Lanes = TESTS_LANE) {
+  const { root, meta: { head, current } } = templates.get(lanes)!.copy();
+  const repo = fixtureRepoAt(root);
+  repos.push(repo);
+  return { repo, head, bundle: path.join(repo.root, 'bundle.json'), current };
 }
 const regate = (repo: FixtureRepo, args: string[], env: Record<string, string>) => runBin(repo, 'gstack-pregate', ['regate', '--base', 'main', ...args], env);
 
@@ -97,7 +134,7 @@ describe('gstack-pregate regate: evidence, stamps and lane verdicts', () => {
   });
 
   test('a stamp-sensitive lane reruns, so a regate is needed; a code edit is not stamp-only and names the file and line', () => {
-    const { repo, head, bundle } = gatedThenRestamped([{ label: 'tests', command: 'true' }, { label: 'digest', command: 'cat generated/digest.md' }]);
+    const { repo, head, bundle } = gatedThenRestamped(DIGEST_LANES);
     const env = ghShim(repo, { list: [] });
     const r = regate(repo, ['--against', head, '--bundle', bundle], env);
     expect(r.status).toBe(EXIT.fail);
diff --git a/test/session-update-autostash.test.ts b/test/session-update-autostash.test.ts
index f2ab223d9..d5a0cff24 100644
--- a/test/session-update-autostash.test.ts
+++ b/test/session-update-autostash.test.ts
@@ -284,7 +284,8 @@ describe('gstack-session-update lock identity + TTL (#2613)', () => {
       const r = runScript(base, install, state);
       expect(r.status).toBe(0);
       const log = await waitForLog(state, /UP_TO_DATE|UPDATING/);
-      expect(log).toMatch(/UP_TO_DATE|UPDATING/);
+      const lockPid = fs.existsSync(path.join(lockDir, 'pid')) ? fs.readFileSync(path.join(lockDir, 'pid'), 'utf8').trim() : 'no lock';
+      expect(log, `planted dead pid ${dead.pid} (alive now: ${isAlive(dead.pid)}); lock pid after 15 s: ${lockPid} (alive: ${isAlive(Number(lockPid))}); script exit ${r.status}, stderr: ${r.stderr}`).toMatch(/UP_TO_DATE|UPDATING/);
     } finally {
       fs.rmSync(base, { recursive: true, force: true });
     }
diff --git a/test/shard-engine.test.ts b/test/shard-engine.test.ts
index 272d5d740..60c99de61 100644
--- a/test/shard-engine.test.ts
+++ b/test/shard-engine.test.ts
@@ -171,7 +171,10 @@ describe('shard engine: process lifecycle', () => {
     });
     expect(paid.status).toBe('timed-out');
     expect(summaryExitCode(summarize([paid]))).toBe(1);
-    expect(await waitDead(Number(fs.readFileSync(pidFile, 'utf8')))).toBe(true);
+    const paidChild = Number(fs.readFileSync(pidFile, 'utf8'));
+    const paidChildDead = await waitDead(paidChild);
+    const procStat = (pid: number) => { try { return fs.readFileSync(`/proc/${pid}/stat`, 'utf8').trim(); } catch { return 'no /proc entry'; } };
+    expect(paidChildDead, `paid grandchild ${paidChild} of group ${paid.groupPid} 2 s after the timeout kill: ${procStat(paidChild)}`).toBe(true);
   }, 30_000);
 
   test('win32 semantics: no process group and never a negative-pid kill', async () => {
diff --git a/test/ship-budget.test.ts b/test/ship-budget.test.ts
index 2ce43a144..5043da5a1 100644
--- a/test/ship-budget.test.ts
+++ b/test/ship-budget.test.ts
@@ -308,7 +308,7 @@ describe('Claude Code child launch', () => {
     const file = ho.out.match(/^HANDOFF: (\S+)/m)![1]!;
     const argsFile = path.join(h.root, 'claude-args.txt');
     const fake = path.join(h.root, 'fake-claude');
-    fs.writeFileSync(fake, `#!/bin/sh\nprintf '%s\\n' "$@" > "${argsFile}"\n`, { mode: 0o755 });
+    fs.writeFileSync(fake, `#!/bin/sh\nprintf '%s\\n' "$@" > "${argsFile}.tmp" && mv "${argsFile}.tmp" "${argsFile}"\n`, { mode: 0o755 });
     const r = h.sb(['launch', '--in', file], { GSTACK_CLAUDE_BIN: fake });
     expect(r.code).toBe(0);
     const session = h.field(r.out, 'CHILD', 'session');
diff --git a/test/ship-queue-train.test.ts b/test/ship-queue-train.test.ts
index ad5c056a3..9a46e65be 100644
--- a/test/ship-queue-train.test.ts
+++ b/test/ship-queue-train.test.ts
@@ -35,13 +35,16 @@ const result = (out: string) => line(out, 'GSTACK_RESULT:');
 const next = (out: string) => line(out, 'NEXT:');
 const remoteRefs = (fx: TrainFixture) => fx.repo.git('ls-remote', 'origin');
 
-// fails_when=coordinatorClone copies origin.git/objects file by file again, so an entry that appears or vanishes there mid-copy (the detached maintenance.lock every push into the fixture origin leaves) fails the clone, as in two Free Tests reds on 2026-10-28 (GSTA-139 ledger); why_new=no other test clones while origin's objects directory changes; seam=a dangling symlink planted under objects/, which a local copy refuses on every run; no_claim=which process removed the loose object those runs named
+// fails_when=coordinatorClone copies origin.git/objects file by file again, so an entry that appears or vanishes there mid-copy (the detached maintenance.lock every push into the fixture origin leaves) fails the clone, as in two Free Tests reds on 2026-10-28 (GSTA-139 ledger); why_new=no other test clones while origin's objects directory changes; seam=a dangling symlink planted under objects/ beside a held maintenance.lock, which a local copy refuses on every run; the seam has its own name because the fixture push's own detached maintenance can hold objects/maintenance.lock at that moment (EEXIST in Free Tests 41288253305 and 41288417894, 2026-11-01); no_claim=which process removed the loose object those runs named
 test('the coordinator clone reads origin through git, never by copying its objects directory (GSTA-139)', () => {
   const fx = make();
-  fs.symlinkSync(path.join(fx.repo.root, 'gone'), path.join(fx.repo.origin, 'objects', 'maintenance.lock'));
+  const objects = path.join(fx.repo.origin, 'objects');
+  fs.writeFileSync(path.join(objects, 'maintenance.lock'), '');
+  fs.symlinkSync(path.join(fx.repo.root, 'gone'), path.join(objects, 'gstack-dangling-seam'));
   const coord = coordinatorClone(fx.repo);
   expect(spawnSync('git', ['-C', coord, 'rev-parse', 'origin/main'], { encoding: 'utf8', timeout: 30_000, env: { ...process.env, ...fx.repo.env } }).stdout.trim()).toBe(fx.repo.git('rev-parse', 'origin/main'));
   expect(fs.existsSync(path.join(coord, '.git', 'objects', 'maintenance.lock'))).toBe(false);
+  expect(fs.readdirSync(path.join(coord, '.git', 'objects'))).not.toContain('gstack-dangling-seam');
 });
 
 describe('the train command surface', () => {

From 322906ab2b0975c6b009ae4a2804259be9e8ed70 Mon Sep 17 00:00:00 2001
From: Garry Tan <garrytan@gmail.com>
Date: Mon, 2 Nov 2026 05:17:17 -0800
Subject: [PATCH 3/6] v1.92.79.0 fix: /plan-tune stops reporting mismatches
 with no observed answers behind them (#3436) (#3856)

* fix(plan-tune): compare declared and inferred only with 3+ observed answers (#3436)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* test: derive keeps the gap at 3 events; changelog fragment for #3436

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(plan-tune): profile view and design doc honor the 3-answer floor (#3436)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(plan-tune): keep the 10-answer floor in --check-mismatch beside the per-dimension floor

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

---------

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---
 CHANGELOG.md                          | 11 ++++
 VERSION                               |  2 +-
 agents-digest/gstack-AGENTS.md        |  2 +-
 bin/gstack-developer-profile          | 23 +++++--
 docs/designs/PLAN_TUNING_V0.md        |  2 +-
 package.json                          |  2 +-
 plan-tune/SKILL.md                    |  5 ++
 plan-tune/SKILL.md.tmpl               |  5 ++
 test/gstack-developer-profile.test.ts | 92 +++++++++++++++++++++++++++
 9 files changed, 135 insertions(+), 9 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 517e1d270..1fc172340 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,16 @@
 # Changelog
 
+## [1.92.79.0] - 2026-11-02
+
+**/plan-tune no longer reports a mismatch between your stated and observed preferences when it has no observed answers behind it (#3436).** `--check-mismatch` compared a declared value against an inferred value that had never moved from its 0.5 starting point, so a strong declaration with no matching answers printed `scope_appetite: declared 0.9 vs inferred 0.5 (gap 0.4)` even though no logged answer touched that dimension.
+
+### Fixed
+- **`bin/gstack-developer-profile`**: `--check-mismatch`, `--gap` and the gap that `--derive` stores compare a dimension only when at least 3 logged answers moved it. `--check-mismatch` still needs 10 logged answers overall before it reports anything. Below the 3-answer floor, `--check-mismatch` prints `no observed evidence` or `too little evidence (n of 3 events)` instead of a gap, and `--gap` lists the dimension under a new `low_evidence` field. A profile derived before `contributing_events` existed keeps the old behavior until its next derive.
+- **/plan-tune "Show gap"**: says "not enough observed answers yet" for those dimensions instead of calling them drift.
+
+### Tests
+- `test/gstack-developer-profile.test.ts`: a declared dimension with no evidence is not flagged by `--check-mismatch` or `--gap` (both failed on main), 1 or 2 events report too little evidence, 3 events are compared, three same-day answers that agree with the declaration still report not enough data, and `--derive` stores no gap without evidence.
+
 ## [1.92.78.0] - 2026-11-02
 
 **Seven flaky tests from the 2026-11-01 flaky pass stop failing correct runs, and six more now name their cause when they fail.** One product lock broke a fresh lock it had misjudged as abandoned, a /ship lane tripped the free runner's home check with its own log, and the rest were fixtures that raced their own setup, leaned on wall-clock sleeps or read the caller's state root.
diff --git a/VERSION b/VERSION
index 66231b754..dfb847afe 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-1.92.78.0
+1.92.79.0
diff --git a/agents-digest/gstack-AGENTS.md b/agents-digest/gstack-AGENTS.md
index 9f27e6a29..192027e8e 100644
--- a/agents-digest/gstack-AGENTS.md
+++ b/agents-digest/gstack-AGENTS.md
@@ -1,4 +1,4 @@
-# gstack digest v1.92.78.0 — regenerate/re-copy after upgrading gstack
+# gstack digest v1.92.79.0 — regenerate/re-copy after upgrading gstack
 
 Behavioral rules from gstack (https://github.com/garrytan/gstack), compressed
 for agent hosts without a full skill install. The full skills add workflows,
diff --git a/bin/gstack-developer-profile b/bin/gstack-developer-profile
index bc489c5c1..409c4f56e 100755
--- a/bin/gstack-developer-profile
+++ b/bin/gstack-developer-profile
@@ -54,6 +54,7 @@ case "$(uname -s)" in
 esac
 PROFILE_FILE="$GSTACK_HOME/developer-profile.json"
 LEGACY_FILE="$GSTACK_HOME/builder-profile.jsonl"
+MIN_EVIDENCE=3
 eval "$("$SCRIPT_DIR/gstack-slug" 2>/dev/null || true)"
 SLUG="${SLUG:-unknown}"
 
@@ -512,14 +513,19 @@ do_gap() {
     const p = JSON.parse(await Bun.stdin.text());
     const declared = p.declared || {};
     const inferred = (p.inferred && p.inferred.values) || {};
+    const contributing = p.inferred && p.inferred.contributing_events;
     const dims = ['scope_appetite','risk_tolerance','detail_preference','autonomy','architecture_care'];
     const gap = {};
+    const low_evidence = {};
     for (const d of dims) {
-      if (declared[d] !== undefined && inferred[d] !== undefined) {
-        gap[d] = +(Math.abs(declared[d] - inferred[d])).toFixed(3);
+      if (declared[d] === undefined || inferred[d] === undefined) continue;
+      if (contributing && (contributing[d] || 0) < $MIN_EVIDENCE) {
+        low_evidence[d] = contributing[d] || 0;
+        continue;
       }
+      gap[d] = +(Math.abs(declared[d] - inferred[d])).toFixed(3);
     }
-    console.log(JSON.stringify({ declared, inferred, gap }, null, 2));
+    console.log(JSON.stringify({ declared, inferred, gap, low_evidence }, null, 2));
   "
 }
 
@@ -657,6 +663,7 @@ do_derive() {
       // Recompute gap.
       const gap = {};
       for (const d of Object.keys(values)) {
+        if ((contributing[d] || 0) < $MIN_EVIDENCE) continue;
         if (profile.declared && profile.declared[d] !== undefined) {
           gap[d] = +(Math.abs(profile.declared[d] - values[d])).toFixed(3);
         }
@@ -727,9 +734,8 @@ do_check_mismatch() {
     const declared = p.declared || {};
     const inferred = (p.inferred && p.inferred.values) || {};
     const sampleSize = (p.inferred && p.inferred.sample_size) || 0;
-    const diversity = (p.inferred && p.inferred.diversity) || {};
+    const contributing = p.inferred && p.inferred.contributing_events;
 
-    // Require enough data before reporting mismatch.
     if (sampleSize < 10) {
       console.log('MISMATCH: not enough data (' + sampleSize + ' events; need 10+)');
       process.exit(0);
@@ -737,8 +743,14 @@ do_check_mismatch() {
 
     const THRESHOLD = 0.3;
     const flagged = [];
+    const thin = [];
     for (const d of Object.keys(declared)) {
       if (inferred[d] === undefined) continue;
+      const n = contributing ? (contributing[d] || 0) : undefined;
+      if (n !== undefined && n < $MIN_EVIDENCE) {
+        thin.push('  ' + d + ': ' + (n === 0 ? 'no observed evidence' : 'too little evidence (' + n + ' of $MIN_EVIDENCE events)'));
+        continue;
+      }
       const gap = Math.abs(declared[d] - inferred[d]);
       if (gap > THRESHOLD) {
         flagged.push({ dim: d, declared: declared[d], inferred: inferred[d], gap: +gap.toFixed(3) });
@@ -753,6 +765,7 @@ do_check_mismatch() {
         console.log('  ' + f.dim + ': declared ' + f.declared + ' vs inferred ' + f.inferred + ' (gap ' + f.gap + ')');
       }
     }
+    for (const line of thin) console.log(line);
   "
 }
 
diff --git a/docs/designs/PLAN_TUNING_V0.md b/docs/designs/PLAN_TUNING_V0.md
index e9bf2b440..77bf95e8d 100644
--- a/docs/designs/PLAN_TUNING_V0.md
+++ b/docs/designs/PLAN_TUNING_V0.md
@@ -105,7 +105,7 @@ After weighing Codex's argument, we chose to roll back CEO EXPANSION and ship an
 }
 ```
 
-`signal_events` counts the logged answers that moved any dimension, and `contributing_events` counts, per dimension, the logged answers that moved it (an answer that writes two deltas to one dimension counts once), so a dimension with no evidence (0) is visible beside one whose evidence balances out; both print 0.5 (#2623). Each value is the sigmoid of its accumulated deltas, and each UTC day's sum per dimension (by the row's `ts`) is clamped to the largest single delta in the signal map (currently ±0.06), so one burst of answers can't pin a dimension; `--trace <dim>` lists every contributing answer, each clamped day and the total. The bound keys on the day, not on `session_id`: the PostToolUse hook and the agent log different ids for the same review (the host's conversation id and the preamble's run id), and an agent can leave the `SESSION_ID` placeholder unsubstituted, so no logged id reliably marks one review. The cost is that several reviews on one day count as one. `contributing_events` still counts an answer logged by both writers twice. Only an ISO 8601 date or date-time `ts` counts as dated (a time without a zone is read as UTC), every other row shares one undated bucket, and `diversity.days_span` counts the same UTC days. Nothing displays `contributing_events` yet (#3436).
+`signal_events` counts the logged answers that moved any dimension, and `contributing_events` counts, per dimension, the logged answers that moved it (an answer that writes two deltas to one dimension counts once), so a dimension with no evidence (0) is visible beside one whose evidence balances out; both print 0.5 (#2623). Each value is the sigmoid of its accumulated deltas, and each UTC day's sum per dimension (by the row's `ts`) is clamped to the largest single delta in the signal map (currently ±0.06), so one burst of answers can't pin a dimension; `--trace <dim>` lists every contributing answer, each clamped day and the total. The bound keys on the day, not on `session_id`: the PostToolUse hook and the agent log different ids for the same review (the host's conversation id and the preamble's run id), and an agent can leave the `SESSION_ID` placeholder unsubstituted, so no logged id reliably marks one review. The cost is that several reviews on one day count as one. `contributing_events` still counts an answer logged by both writers twice. Only an ISO 8601 date or date-time `ts` counts as dated (a time without a zone is read as UTC), every other row shares one undated bucket, and `diversity.days_span` counts the same UTC days. `--check-mismatch`, `--gap`, the stored `gap` and /plan-tune's profile view compare a dimension only when its `contributing_events` is at least 3; `--gap` lists the others under `low_evidence` and the stored `gap` omits them (#3436).
 
 **Diversity check** (Codex #13): `inferred` is considered "enough data" only when `sample_size >= 20 AND skills_covered >= 3 AND question_ids_covered >= 8 AND days_span >= 7`. Below this, `/plan-tune profile` shows "not enough observed data yet" instead of a potentially-misleading inferred value.
 
diff --git a/package.json b/package.json
index bce0fb11b..b8a73016c 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "gstack",
-  "version": "1.92.78",
+  "version": "1.92.79",
   "description": "Garry's Stack — Claude Code skills + fast headless browser. One repo, one install, entire AI engineering workflow.",
   "license": "MIT",
   "type": "module",
diff --git a/plan-tune/SKILL.md b/plan-tune/SKILL.md
index 86b2cbdb7..481d7764e 100644
--- a/plan-tune/SKILL.md
+++ b/plan-tune/SKILL.md
@@ -341,6 +341,8 @@ Parse the JSON. Present in **plain English**, not raw floats:
   the inferred column next to declared:
   "**scope_appetite:** declared 0.8 (boil the ocean) ↔ observed 0.72 (close)"
   Use words for the gap: 0.0-0.1 "close", 0.1-0.3 "drift", 0.3+ "mismatch".
+  When `inferred.contributing_events[dim]` is below 3, show "not enough
+  observed answers yet" for that dimension instead of an observed value or gap word.
 
   This display gate is intentionally lower than the E1 **promotion gate**
   (90+ days stable across 3+ skills, per `docs/designs/PLAN_TUNING_V0.md`).
@@ -491,6 +493,9 @@ Parse the JSON. For each dimension where both declared and inferred exist:
   Consider updating your declared value, or reflect on whether your behavior
   is actually what you want."
 
+For each dimension in `low_evidence` (fewer than 3 observed answers moved it),
+say "not enough observed answers yet" instead of a gap.
+
 Never auto-update declared based on the gap. In v1 the gap is reporting only —
 the user decides whether declared is wrong or behavior is wrong.
 
diff --git a/plan-tune/SKILL.md.tmpl b/plan-tune/SKILL.md.tmpl
index 748dfad62..88b7a8eb3 100644
--- a/plan-tune/SKILL.md.tmpl
+++ b/plan-tune/SKILL.md.tmpl
@@ -272,6 +272,8 @@ Parse the JSON. Present in **plain English**, not raw floats:
   the inferred column next to declared:
   "**scope_appetite:** declared 0.8 (boil the ocean) ↔ observed 0.72 (close)"
   Use words for the gap: 0.0-0.1 "close", 0.1-0.3 "drift", 0.3+ "mismatch".
+  When `inferred.contributing_events[dim]` is below 3, show "not enough
+  observed answers yet" for that dimension instead of an observed value or gap word.
 
   This display gate is intentionally lower than the E1 **promotion gate**
   (90+ days stable across 3+ skills, per `docs/designs/PLAN_TUNING_V0.md`).
@@ -422,6 +424,9 @@ Parse the JSON. For each dimension where both declared and inferred exist:
   Consider updating your declared value, or reflect on whether your behavior
   is actually what you want."
 
+For each dimension in `low_evidence` (fewer than 3 observed answers moved it),
+say "not enough observed answers yet" instead of a gap.
+
 Never auto-update declared based on the gap. In v1 the gap is reporting only —
 the user decides whether declared is wrong or behavior is wrong.
 
diff --git a/test/gstack-developer-profile.test.ts b/test/gstack-developer-profile.test.ts
index df23a9c3b..cffaadebc 100644
--- a/test/gstack-developer-profile.test.ts
+++ b/test/gstack-developer-profile.test.ts
@@ -622,6 +622,98 @@ describe('gstack-developer-profile --check-mismatch', () => {
     expect(r.stdout).toContain('scope_appetite');
     expect(r.stdout).toContain('autonomy');
   });
+
+  const deriveWithOneSignal = () => {
+    logQuestion({ skill: 'plan-eng-review', question_id: 'plan-eng-review-arch-finding', question_summary: 'arch?', user_choice: 'fix-now', session_id: 's0' });
+    for (let i = 1; i < 13; i++) {
+      logQuestion({ skill: 'plan-eng-review', question_id: `adhoc-unregistered-${i}`, question_summary: 'x', user_choice: 'anything', session_id: `s${i}` });
+    }
+    expect(runDev('--derive').status).toBe(0);
+    const file = path.join(tmpHome, 'developer-profile.json');
+    const p = readProfile() as any;
+    expect(p.inferred.contributing_events.scope_appetite).toBe(0);
+    p.declared = { scope_appetite: 0.9 };
+    fs.writeFileSync(file, JSON.stringify(p));
+  };
+
+  test('a declared dimension with no observed evidence is not flagged (#3436)', () => {
+    deriveWithOneSignal();
+    const r = runDev('--check-mismatch');
+    expect(r.status).toBe(0);
+    expect(r.stdout).not.toContain('scope_appetite: declared');
+    expect(r.stdout).toContain('scope_appetite: no observed evidence');
+  });
+
+  test('--gap leaves out a dimension with no observed evidence (#3436)', () => {
+    deriveWithOneSignal();
+    const out = JSON.parse(runDev('--gap').stdout);
+    expect(out.gap.scope_appetite).toBeUndefined();
+    expect(out.low_evidence).toEqual({ scope_appetite: 0 });
+  });
+
+  test('a dimension with 1 or 2 contributing events reports too little evidence', () => {
+    runDev('--read');
+    const file = path.join(tmpHome, 'developer-profile.json');
+    const p = readProfile() as any;
+    p.declared = { scope_appetite: 0.9 };
+    p.inferred.values.scope_appetite = 0.45;
+    p.inferred.sample_size = 10;
+    p.inferred.contributing_events = { scope_appetite: 2 };
+    fs.writeFileSync(file, JSON.stringify(p));
+    const r = runDev('--check-mismatch');
+    expect(r.stdout).toContain('MISMATCH: none');
+    expect(r.stdout).toContain('scope_appetite: too little evidence (2 of 3 events)');
+    expect(JSON.parse(runDev('--gap').stdout).low_evidence).toEqual({ scope_appetite: 2 });
+  });
+
+  test('a dimension with 3 contributing events is compared', () => {
+    runDev('--read');
+    const file = path.join(tmpHome, 'developer-profile.json');
+    const p = readProfile() as any;
+    p.declared = { scope_appetite: 0.9 };
+    p.inferred.values.scope_appetite = 0.45;
+    p.inferred.sample_size = 10;
+    p.inferred.contributing_events = { scope_appetite: 3 };
+    fs.writeFileSync(file, JSON.stringify(p));
+    expect(runDev('--check-mismatch').stdout).toContain('scope_appetite: declared 0.9 vs inferred 0.45 (gap 0.45)');
+    expect(JSON.parse(runDev('--gap').stdout).gap.scope_appetite).toBeCloseTo(0.45, 2);
+  });
+
+  test('three same-day answers that agree with the declaration are not a mismatch', () => {
+    for (let i = 0; i < 3; i++) {
+      logQuestion({ skill: 'plan-ceo-review', question_id: 'plan-ceo-review-mode', question_summary: 'mode?', user_choice: 'expand', session_id: `s${i}`, ts: '2026-10-20T12:00:00Z' });
+    }
+    expect(runDev('--derive').status).toBe(0);
+    const file = path.join(tmpHome, 'developer-profile.json');
+    const p = readProfile() as any;
+    expect(p.inferred.contributing_events.scope_appetite).toBe(3);
+    p.declared = { scope_appetite: 0.9 };
+    fs.writeFileSync(file, JSON.stringify(p));
+    const r = runDev('--check-mismatch');
+    expect(r.stdout).toContain('not enough data');
+    expect(r.stdout).not.toContain('scope_appetite: declared');
+  });
+
+  test('derive stores the gap for a dimension with 3 contributing events', () => {
+    for (let i = 0; i < 3; i++) {
+      logQuestion({ skill: 'plan-eng-review', question_id: 'plan-eng-review-arch-finding', question_summary: 'arch?', user_choice: 'fix-now', session_id: `s${i}`, ts: `2026-10-2${i}T12:00:00Z` });
+    }
+    expect(runDev('--derive').status).toBe(0);
+    const file = path.join(tmpHome, 'developer-profile.json');
+    const p = readProfile() as any;
+    expect(p.inferred.contributing_events.architecture_care).toBe(3);
+    p.declared = { architecture_care: 0.1 };
+    fs.writeFileSync(file, JSON.stringify(p));
+    expect(runDev('--derive').status).toBe(0);
+    expect((readProfile() as any).gap.architecture_care).toBeGreaterThan(0.3);
+  });
+
+  test('derive stores no gap for a dimension without evidence', () => {
+    deriveWithOneSignal();
+    expect(runDev('--derive').status).toBe(0);
+    const p = readProfile() as any;
+    expect(p.gap.scope_appetite).toBeUndefined();
+  });
 });
 
 // -----------------------------------------------------------------------

From 0542230e632b5e6b2d776aadbe21fb4b8109d0d5 Mon Sep 17 00:00:00 2001
From: Garry Tan <garrytan@gmail.com>
Date: Mon, 2 Nov 2026 05:28:00 -0800
Subject: [PATCH 4/6] test(fixtures): leak scan skips Capy's git shim log in
 fixture homes (#3860)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---
 CHANGELOG.md                         |  8 ++++++++
 VERSION                              |  2 +-
 agents-digest/gstack-AGENTS.md       |  2 +-
 package.json                         |  2 +-
 test/git-fixture-template.test.ts    | 17 +++++++++++++++++
 test/helpers/git-fixture-template.ts |  8 ++++++--
 6 files changed, 34 insertions(+), 5 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 1fc172340..f32c13f1e 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,13 @@
 # Changelog
 
+## [1.92.80.0] - 2026-11-02
+
+**On a Capy machine, the git fixture tests no longer fail locally because of Capy's git shim log.** Capy's `/usr/bin/git` shim appends to `$HOME/.capy/git-shim.log`, so each fixture home that ran git during its template build ended up with a log naming the template. The leak scan in `test/helpers/git-fixture-template.ts` flagged it, and 34 tests in `test/train-backstop.test.ts`, `test/restamp-mentions.test.ts` and `test/pregate-user-policy.test.ts` failed on every Capy lane, though they passed on the remote gate.
+
+### Tests
+- `templateReferences` skips `.capy/git-shim.log`. Nothing reads that file back, and every other file is still scanned.
+- `test/git-fixture-template.test.ts` plants the shim log in a template's `home/.capy/` (the copy succeeds), and plants `home/.capy/state.json` as a control (the copy still throws).
+
 ## [1.92.79.0] - 2026-11-02
 
 **/plan-tune no longer reports a mismatch between your stated and observed preferences when it has no observed answers behind it (#3436).** `--check-mismatch` compared a declared value against an inferred value that had never moved from its 0.5 starting point, so a strong declaration with no matching answers printed `scope_appetite: declared 0.9 vs inferred 0.5 (gap 0.4)` even though no logged answer touched that dimension.
diff --git a/VERSION b/VERSION
index dfb847afe..f17ba7e50 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-1.92.79.0
+1.92.80.0
diff --git a/agents-digest/gstack-AGENTS.md b/agents-digest/gstack-AGENTS.md
index 192027e8e..835c2c05f 100644
--- a/agents-digest/gstack-AGENTS.md
+++ b/agents-digest/gstack-AGENTS.md
@@ -1,4 +1,4 @@
-# gstack digest v1.92.79.0 — regenerate/re-copy after upgrading gstack
+# gstack digest v1.92.80.0 — regenerate/re-copy after upgrading gstack
 
 Behavioral rules from gstack (https://github.com/garrytan/gstack), compressed
 for agent hosts without a full skill install. The full skills add workflows,
diff --git a/package.json b/package.json
index b8a73016c..2da268dcb 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "gstack",
-  "version": "1.92.79",
+  "version": "1.92.80",
   "description": "Garry's Stack — Claude Code skills + fast headless browser. One repo, one install, entire AI engineering workflow.",
   "license": "MIT",
   "type": "module",
diff --git a/test/git-fixture-template.test.ts b/test/git-fixture-template.test.ts
index 76375466c..f63a6c002 100644
--- a/test/git-fixture-template.test.ts
+++ b/test/git-fixture-template.test.ts
@@ -71,6 +71,23 @@ describe('git fixture template', () => {
     expect(() => leaky.copy()).toThrow(/still references its template [^\n]*\n.*config/);
   });
 
+  test("Capy's git shim log in a fixture home (home/.capy/git-shim.log) is not a leak; any other file naming the template still is", () => {
+    const plant = (rel: string) => gitFixtureTemplate({ prefix: 'gstack-fixture-template-shim-', build: (root) => {
+      const meta = build(root);
+      fs.mkdirSync(path.dirname(path.join(root, rel)), { recursive: true });
+      fs.writeFileSync(path.join(root, rel), '# capy git shim v2: ts cwd ppid parent verb\n2026-11-02T02:15:30Z\t' + path.join(root, 'work') + '\t1\tbun\tcat-file\n');
+      return meta;
+    }, rebind: (copy, tpl) => rebindGitDir(path.join(copy, 'work', '.git'), tpl, copy) });
+    const shim = plant(path.join('home', '.capy', 'git-shim.log'));
+    const other = plant(path.join('home', '.capy', 'state.json'));
+    try {
+      const c = shim.copy();
+      roots.push(c.root);
+      expect(git(path.join(c.root, 'work'), 'remote', 'get-url', 'origin')).toBe(path.join(c.root, 'origin.git'));
+      expect(() => other.copy()).toThrow(/still references its template [^\n]*\n.*state\.json/);
+    } finally { shim.dispose(); other.dispose(); }
+  });
+
   test('native, forward-slash, backslash-escaped and Git Bash spellings are each rewritten, and each one left behind is found', () => {
     const to = fs.mkdtempSync(path.join(os.tmpdir(), 'gstack-fixture-spellings-'));
     roots.push(to);
diff --git a/test/helpers/git-fixture-template.ts b/test/helpers/git-fixture-template.ts
index a646747ab..5fb9f72eb 100644
--- a/test/helpers/git-fixture-template.ts
+++ b/test/helpers/git-fixture-template.ts
@@ -52,7 +52,11 @@ export function rebindGitDir(gitDir: string, from: string, to: string): void {
   }
 }
 
-/** Files and symlinks under dir (and the template's meta value) that still name templateRoot; zlib-compressed object storage is skipped. */
+/**
+ * Files and symlinks under dir (and the template's meta value) that still name templateRoot. Skipped: zlib-compressed object
+ * storage, and the log Capy's /usr/bin/git shim appends to in whatever HOME it runs under (.capy/git-shim.log), which names
+ * the template whenever a fixture home ran git during build; nothing reads it back.
+ */
 export function templateReferences(dir: string, templateRoot: string, meta?: unknown): string[] {
   const needles = pathSpellings(templateRoot).map(s => Buffer.from(s));
   const names = (buf: Buffer) => needles.some(n => buf.includes(n));
@@ -62,7 +66,7 @@ export function templateReferences(dir: string, templateRoot: string, meta?: unk
       const p = path.join(d, entry.name);
       if (entry.isSymbolicLink()) { if (names(Buffer.from(fs.readlinkSync(p)))) hits.push(p); }
       else if (entry.isDirectory()) { if (!(path.basename(d) === 'objects' && /^(pack|[0-9a-f]{2})$/.test(entry.name))) walk(p); }
-      else if (entry.isFile() && names(fs.readFileSync(p))) hits.push(p);
+      else if (entry.isFile() && !(entry.name === 'git-shim.log' && path.basename(d) === '.capy') && names(fs.readFileSync(p))) hits.push(p);
     }
   };
   walk(dir);

From 1022f4fc51a4771e3838ce6bbf7c19b7e94e67fe Mon Sep 17 00:00:00 2001
From: Garry Tan <garrytan@gmail.com>
Date: Mon, 2 Nov 2026 05:38:02 -0800
Subject: [PATCH 5/6] v1.92.81.0 feat(ship-release): finish accepts the owner's
 edit after a fix-check (GST-965) (#3861)

* feat(ship-release): finish accepts the owner's edit after a fix-check (GST-965)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(ship-release): fix_check.files narrows the accepted scope; gate-edited and over-cap paths stay behavior (GST-965)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* docs(review-policy): the gate-edited and over-cap exclusions hold with --reviewed-base (GST-965)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(ship-release): a gate-edited test is never owner-accepted; an empty fix_check.files stops (GST-965)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* docs(review-policy): the over-cap exclusion holds with --reviewed-base (GST-965 fix-check note)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

---------

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---
 CHANGELOG.md                        | 10 ++++
 VERSION                             |  2 +-
 agents-digest/gstack-AGENTS.md      |  2 +-
 docs/review-policy.md               |  9 +++-
 lib/ship-release.ts                 | 57 ++++++++++++++++------
 package.json                        |  2 +-
 test/ship-rereview-triggers.test.ts | 76 +++++++++++++++++++++++++++++
 7 files changed, 140 insertions(+), 18 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index f32c13f1e..209155077 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,15 @@
 # Changelog
 
+## [1.92.81.0] - 2026-11-02
+
+**`gstack-ship-release finish` now accepts the owner's edit after a fix-check instead of asking for another review round (GST-965).** On gbrain #8425, the fix-check found a blocker, the owner's pre-agreed fallback was applied and "then ship" was granted. Finish still classed the edit as behavior and stopped with `insert 5–11.5 → 12–14 → 16`, a third round the R51 cap forbids, so /ship could not get through Step 16 without running its stages by hand.
+
+### Added
+- **`finish --review-loop-exception <json|@file>`** (`lib/ship-release.ts`) takes the same record `gstack-ship-receipt write --review-loop-exception` takes. Paths changed since `fix_check.head` that the branch already touched at that commit (narrowed to the record's `fix_check.files` when it lists them) route as `accepted=` on the `ROUTE:` line, a `REVIEW_LOOP_EXCEPTION:` line names them with the owner, date and channel, and the docs stage carries them (`accepted=` in the DOCS line). A path outside that scope and a gate-edited test (with or without `--reviewed-base`) still route as behavior, and with `--reviewed-base` so does a file too large for the scoped review. A malformed record, or a `fix_check.head` that is not HEAD or an ancestor of it, stops before the release commit. Without the flag nothing changes.
+
+### Tests
+- `test/ship-rereview-triggers.test.ts`: an owner fallback edit to a source file and a doc after the fix-check stops without the record and passes Step 16 with it; a new file outside the scope (even when `fix_check.files` lists it), a fix-check head off the branch, a malformed record, an unreadable record file and an empty `fix_check.files` stop; a `fix_check.files` list narrows what is accepted; a test weakened after the fix-check stops with and without `--reviewed-base`.
+
 ## [1.92.80.0] - 2026-11-02
 
 **On a Capy machine, the git fixture tests no longer fail locally because of Capy's git shim log.** Capy's `/usr/bin/git` shim appends to `$HOME/.capy/git-shim.log`, so each fixture home that ran git during its template build ended up with a log naming the template. The leak scan in `test/helpers/git-fixture-template.ts` flagged it, and 34 tests in `test/train-backstop.test.ts`, `test/restamp-mentions.test.ts` and `test/pregate-user-policy.test.ts` failed on every Capy lane, though they passed on the remote gate.
diff --git a/VERSION b/VERSION
index f17ba7e50..9a1e28b7f 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-1.92.80.0
+1.92.81.0
diff --git a/agents-digest/gstack-AGENTS.md b/agents-digest/gstack-AGENTS.md
index 835c2c05f..45e52d15a 100644
--- a/agents-digest/gstack-AGENTS.md
+++ b/agents-digest/gstack-AGENTS.md
@@ -1,4 +1,4 @@
-# gstack digest v1.92.80.0 — regenerate/re-copy after upgrading gstack
+# gstack digest v1.92.81.0 — regenerate/re-copy after upgrading gstack
 
 Behavioral rules from gstack (https://github.com/garrytan/gstack), compressed
 for agent hosts without a full skill install. The full skills add workflows,
diff --git a/docs/review-policy.md b/docs/review-policy.md
index db10c3753..16e93e891 100644
--- a/docs/review-policy.md
+++ b/docs/review-policy.md
@@ -429,7 +429,14 @@ rules hash, base and registry commits, the gate scan bound to the head and the r
 the queue refuses a receipt below the depth its head needs
 ([ship-policy.md#ship-receipt](ship-policy.md#ship-receipt)). A stop the owner
 accepted after a fix-check edit goes in `review_loop_exception`, never in prose
-under `REVIEW_LOOP:`.
+under `REVIEW_LOOP:`. Step 16 takes the same record:
+`gstack-ship-release finish --review-loop-exception <json|@file>` routes the
+paths the branch touched at `fix_check.head` (narrowed to the record's
+`fix_check.files` when it lists them) and changed since as `accepted=` in
+`ROUTE:`, prints them on a `REVIEW_LOOP_EXCEPTION:` line, and carries them
+through the docs stage. A path outside that scope and a gate-edited test still
+route as behavior, with `--reviewed-base` (as /ship passes it) so does a file
+too large for the scoped review, and without the record nothing changes.
 
 The receipt's summary line prints `passes=<n> blocking_found=<k> deferred=<m>`,
 which is what a scorecard tracks (target: a median of 1 pass per PR). A
diff --git a/lib/ship-release.ts b/lib/ship-release.ts
index bdce14414..02f5218a7 100644
--- a/lib/ship-release.ts
+++ b/lib/ship-release.ts
@@ -15,6 +15,7 @@ import { mergeChangelogs, normalizeChangelog } from './changelog-check';
 import { versionAtRef } from './restamp';
 import { classifyPath } from './review-depth';
 import { comparePath, contextPacket, gateEditedPaths, isTestCode, PACKET_SECTION_CAP, readScopedReview, touchedPaths, type PatchSides } from './ship-patch';
+import { parseReviewLoopException, type ReviewLoopException } from './ship-receipt-depth';
 import { parseReceiptDocs, type ReceiptDocs } from './ship-receipt-docs';
 import { resolveStateRoot } from './state-root';
 import { bumpVersion, cmpVersion, fmtVersion, parseVersion, versionWidth, type Bump } from './version-source';
@@ -27,10 +28,11 @@ export const RELEASE_HELP = `gstack-ship-release — /ship Steps 12 and 15-17 (s
       --keep-current (or --version <current>) keeps VERSION; under bump ask a session no person answers needs --answer before a bump is written
   finish --base <base> --reviewed-tree <tree> [--reviewed-base <sha>] --docs-post <post.json> [--lane <label>=<command>]... [--build <command>]...
          [--trailer <text>] [--docs-rechecked] [--no-lanes-approved] [--context-reviewed <record.json>]
-         [--docs-requested <path>... --docs-result <the audit's result JSON>]
+         [--docs-requested <path>... --docs-result <the audit's result JSON>] [--review-loop-exception <json|@file>]
       Step 15's release commit, Step 16 stages 1-4 (your builds, the reviewed-tree route, docs freshness,
       each lane's receipt run once if stale) and the pregate on the stamped tree; --reviewed-base (the merge base
-      saved with the reviewed tree) routes paths only the base moved as base=, and an adjacent base edit as context=
+      saved with the reviewed tree) routes paths only the base moved as base=, and an adjacent base edit as context=;
+      --review-loop-exception (the record write takes) routes paths the branch touched at fix_check.head (narrowed to fix_check.files when listed) and changed since as accepted=
   push --branch <branch-name> [--gate-open <n>] [--gate-open-approved]
       Step 17 after its guard block: the saved-branch check, the open-gate check, the sealed guard, the idempotency check, git push -u
 
@@ -159,8 +161,8 @@ function commitRelease(cwd: string, release: string[], trailer: string | undefin
 }
 
 /** Step 16 stage 2's route. With --reviewed-base, a path whose branch patch is the reviewed one counts as base (lib/ship-patch.ts). */
-type Routed = Record<'behavior' | 'docs' | 'metadata' | 'tests' | 'base' | 'context', string[]>;
-function route(cwd: string, base: string, reviewed: string, reviewedBase: string | undefined, release: string[], opts: { docsRechecked: boolean; contextReviewed?: string }, r: Report): Routed | null {
+type Routed = Record<'behavior' | 'docs' | 'metadata' | 'tests' | 'base' | 'context' | 'accepted', string[]>;
+function route(cwd: string, base: string, reviewed: string, reviewedBase: string | undefined, release: string[], opts: { docsRechecked: boolean; contextReviewed?: string; loop?: ReviewLoopException }, r: Report): Routed | null {
   const now = bin(cwd, 'gstack-wtree', [], 60_000).stdout.trim();
   if (!/^[0-9a-f]{40,64}$/.test(now)) { r.stop('gstack-wtree printed no tree; the stage 2 comparison is blocked'); return null; }
   const changedSet = touchedPaths(cwd, reviewed, now);
@@ -179,7 +181,7 @@ function route(cwd: string, base: string, reviewed: string, reviewedBase: string
     sides = { reviewedBase: rb, reviewed, currentBase: cb, current: now };
     touched = new Set([...a, ...b]);
   }
-  const by: Routed = { behavior: [], docs: [], metadata: [], tests: [], base: [], context: [] };
+  const by: Routed = { behavior: [], docs: [], metadata: [], tests: [], base: [], context: [], accepted: [] };
   for (const p of [...changedSet].sort()) {
     const verdict = sides ? (touched.has(p) ? comparePath(cwd, sides, p) : 'same') : 'changed';
     by[verdict === 'same' ? 'base' : release.includes(p) ? 'metadata' : sides && isTestCode(p) ? 'tests' : classifyPath(p, cwd) === 'docs' ? 'docs' : verdict === 'context' ? 'context' : 'behavior'].push(p);
@@ -194,7 +196,22 @@ function route(cwd: string, base: string, reviewed: string, reviewedBase: string
     by.context = by.context.filter(p => !packetBody.over.includes(p));
     by.tests = by.tests.filter(p => !packetBody.over.includes(p));
   }
-  r.add(`ROUTE: reviewed=${reviewed.slice(0, 12)} current=${now.slice(0, 12)} behavior=${by.behavior.length} docs=${by.docs.length} metadata=${by.metadata.length} tests=${by.tests.length} base=${sides ? by.base.length : 'unknown'} context=${by.context.length}`);
+  const loop = opts.loop;
+  if (loop) {
+    const head = loop.fix_check.head;
+    const since = touchedPaths(cwd, head, now);
+    const fork = git(cwd, ['merge-base', baseRef, head]).stdout.trim();
+    const atHead = fork ? touchedPaths(cwd, fork, head) : null;
+    if (!since || !atHead) { r.stop(`git could not list the paths changed since fix-check ${head.slice(0, 12)} or the paths the branch touched at it against ${baseRef}: stage 2 is blocked`); return null; }
+    const listed = loop.fix_check.files;
+    const unscanned = [...by.behavior, ...by.docs].filter(isTestCode);
+    const forced = new Set([...gated, ...(packetBody?.over ?? []), ...(gateEditedPaths(cwd, reviewed, now, unscanned) ?? unscanned)]);
+    by.accepted = [...by.behavior, ...by.docs].filter(p => since.has(p) && atHead.has(p) && (!listed || listed.includes(p)) && !forced.has(p)).sort();
+    by.behavior = by.behavior.filter(p => !by.accepted.includes(p));
+    by.docs = by.docs.filter(p => !by.accepted.includes(p));
+  }
+  r.add(`ROUTE: reviewed=${reviewed.slice(0, 12)} current=${now.slice(0, 12)} behavior=${by.behavior.length} docs=${by.docs.length} metadata=${by.metadata.length} tests=${by.tests.length} base=${sides ? by.base.length : 'unknown'} context=${by.context.length}${loop ? ` accepted=${by.accepted.length}` : ''}`);
+  if (loop) r.add(`REVIEW_LOOP_EXCEPTION: accepted ${by.accepted.length ? short(by.accepted) : 'none'} (changed since fix-check ${loop.fix_check.head.slice(0, 12)}; accepted by ${loop.accepted_by} ${loop.date} via ${loop.via})`);
   if (by.behavior.length) { r.stop(`behavior, tests or build inputs changed since review (${short(by.behavior.sort())}): insert 5–11.5 → 12–14 → 16 before Step 17 (stage 2.1)`); return null; }
   const scoped = [...by.context, ...by.tests].sort();
   if (scoped.length && sides) {
@@ -225,7 +242,7 @@ function route(cwd: string, base: string, reviewed: string, reviewedBase: string
  * version, each against the audited bytes; GST-906), and --docs-requested fixes the audit's own result asked for do not
  * stale the audit (GST-391).
  */
-function docsFresh(cwd: string, base: string, postFile: string, moved: readonly string[], requested: string[], resultFile: string | undefined, r: Report): boolean {
+function docsFresh(cwd: string, base: string, postFile: string, moved: readonly string[], requested: string[], resultFile: string | undefined, loopAccepted: readonly string[] | undefined, r: Report): boolean {
   const cmp = bin(cwd, 'gstack-docs-candidate', ['compare', postFile], 120_000);
   let stale: string[] | null = null;
   try { stale = JSON.parse(cmp.stdout).content_changed; } catch { stale = null; }
@@ -277,10 +294,11 @@ function docsFresh(cwd: string, base: string, postFile: string, moved: readonly
     const merged = mergeChangelogs(audited.stdout, atNowBase.stdout, atAuditBase.stdout);
     return kept !== null && merged !== null && normalizeChangelog(kept) === normalizeChangelog(audited.stdout) && normalizeChangelog(merged) === normalizeChangelog(now);
   });
-  const carried = { tests: stale.filter(isTestCode), base: stale.filter(p => !isTestCode(p) && (moved.includes(p) || restamped.includes(p) || (!!auditBase && !!nowBase && !branchAtAudit.has(p) && !branchNow.has(p)))), requested: [] as string[] };
+  const carried = { tests: stale.filter(isTestCode), base: stale.filter(p => !isTestCode(p) && (moved.includes(p) || restamped.includes(p) || (!!auditBase && !!nowBase && !branchAtAudit.has(p) && !branchNow.has(p)))), requested: [] as string[], accepted: [] as string[] };
+  carried.accepted = stale.filter(p => loopAccepted?.includes(p) && !carried.tests.includes(p) && !carried.base.includes(p));
   let audit: any = null;
   try { audit = resultFile ? JSON.parse(fs.readFileSync(resultFile, 'utf8')) : null; } catch { audit = null; }
-  for (const p of requested.filter(q => stale!.includes(q) && !carried.tests.includes(q) && !carried.base.includes(q))) {
+  for (const p of requested.filter(q => stale!.includes(q) && !carried.tests.includes(q) && !carried.base.includes(q) && !carried.accepted.includes(q))) {
     const asked = [...(Array.isArray(audit?.blockers) ? audit.blockers : []), ...(Array.isArray(audit?.decisions) ? audit.decisions : [])].filter((s: unknown) => typeof s === 'string');
     const token = new RegExp(`(^|[\\s\`'"(\\[])${p.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')}(?=$|[\\s\`'")\\]]|[,.:;](?:$|\\s))`);
     const stat = git(cwd, ['diff', '--numstat', '--no-renames', record.head, '--', `:(literal)${p}`]).stdout.trim().split(/\s+/);
@@ -295,8 +313,8 @@ function docsFresh(cwd: string, base: string, postFile: string, moved: readonly
     carried.requested.push(p);
   }
   if (carried.requested.length && bin(cwd, 'gstack-ship-receipt', ['docs-caps', '--base', base], 60_000).status !== 0) { r.stop('a --docs-requested fix breaks a doc cap (gstack-ship-receipt docs-caps): trim the file, then rerun finish'); return false; }
-  const left = stale.filter(p => !carried.tests.includes(p) && !carried.base.includes(p) && !carried.requested.includes(p));
-  const note = left.length || stale.length === 0 ? '' : ` (carried: tests=${carried.tests.length} base=${carried.base.length} requested=${carried.requested.length})`;
+  const left = stale.filter(p => !carried.tests.includes(p) && !carried.base.includes(p) && !carried.requested.includes(p) && !carried.accepted.includes(p));
+  const note = left.length || stale.length === 0 ? '' : ` (carried: tests=${carried.tests.length} base=${carried.base.length} requested=${carried.requested.length}${loopAccepted ? ` accepted=${carried.accepted.length}` : ''})`;
   const exception = record.exception ? ` (blocked; documentation exception accepted by ${record.exception.accepted_by} ${record.exception.date} via ${record.exception.via})` : '';
   r.add(`DOCS: ${left.length ? `stale (${short(left)})` : `fresh against the Step 14.5 post-audit snapshot${note}${exception}`}`);
   if (left.length) { r.stop('documentation inputs changed since the accepted audit: use stage 3 recovery (the remaining re-audit, a fix the audit\'s result asked for via --docs-requested <path> --docs-result <audit-id>-result.json, or a named risk); never refresh hashes'); return false; }
@@ -308,6 +326,17 @@ export async function finish(argv: readonly string[], cwd: string): Promise<Repo
   const base = flag('--base')!;
   const r = new Report();
   const list = lanes(argv) as { label: string; command: string }[];
+  let loop: ReviewLoopException | undefined;
+  const loopArg = flag('--review-loop-exception');
+  if (loopArg !== undefined) {
+    let raw: unknown;
+    try { raw = JSON.parse(loopArg.startsWith('@') ? fs.readFileSync(path.resolve(cwd, loopArg.slice(1)), 'utf8') : loopArg); } catch (e) { return r.stop(`--review-loop-exception: ${firstLine((e as Error).message)}`); }
+    const parsed = parseReviewLoopException(raw);
+    if (!parsed.ok) return r.stop(`--review-loop-exception: ${parsed.message}`);
+    if (Array.isArray((raw as { fix_check?: { files?: unknown } }).fix_check?.files) && !parsed.exception.fix_check.files) return r.stop('--review-loop-exception: fix_check.files lists no paths; omit it to accept the paths the branch touched at fix_check.head');
+    if (git(cwd, ['merge-base', '--is-ancestor', parsed.exception.fix_check.head, 'HEAD']).status !== 0) return r.stop(`--review-loop-exception: fix_check.head ${parsed.exception.fix_check.head.slice(0, 12)} is not HEAD or an ancestor of it`);
+    loop = parsed.exception;
+  }
   const policy = shipPolicy(cwd, base);
   if (!commitRelease(cwd, releasePaths(policy, cwd, 'HEAD', base), flag('--trailer'), r)) return r;
   for (const cmd of all('--build')) {
@@ -317,9 +346,9 @@ export async function finish(argv: readonly string[], cwd: string): Promise<Repo
   }
   if (!all('--build').length) r.add('BUILD: none passed (record the inspected sources, or pass --build per declared command)');
   const release = releasePaths(policy, cwd, flag('--reviewed-tree')!, base);
-  const routed = route(cwd, base, flag('--reviewed-tree')!, flag('--reviewed-base'), release, { docsRechecked: has('--docs-rechecked'), contextReviewed: flag('--context-reviewed') }, r);
+  const routed = route(cwd, base, flag('--reviewed-tree')!, flag('--reviewed-base'), release, { docsRechecked: has('--docs-rechecked'), contextReviewed: flag('--context-reviewed'), loop }, r);
   if (!routed) return r;
-  if (!docsFresh(cwd, base, flag('--docs-post')!, [...routed.base, ...routed.context, ...stampedFiles(policy, cwd, flag('--reviewed-tree')!, base)], all('--docs-requested'), flag('--docs-result'), r)) return r;
+  if (!docsFresh(cwd, base, flag('--docs-post')!, [...routed.base, ...routed.context, ...stampedFiles(policy, cwd, flag('--reviewed-tree')!, base)], all('--docs-requested'), flag('--docs-result'), loop ? routed.accepted : undefined, r)) return r;
   if (!list.length && !has('--no-lanes-approved')) return r.stop('no test lanes: require Step 5\'s explicit untested-scope approval (rerun with --no-lanes-approved) or run Steps 5-15');
   const allow = evidenceExempt(release).join(',');
   const check = (l: { label: string; command: string }) => bin(cwd, 'gstack-evidence', ['check', '--label', l.label, '--expect-cmd', l.command, '--max-age', '24', '--allow-paths', allow], 60_000).stdout;
@@ -343,7 +372,7 @@ export async function finish(argv: readonly string[], cwd: string): Promise<Repo
     r.add(`PREGATE: ${verdict} (rerun on the stamped tree)`);
     if (!/^(pass|warn|requires-remote cleared)/.test(verdict)) return r.stop(`pregate ${verdict} on the stamped tree: fix it${/uncleared/.test(verdict) ? ', or gstack-pregate clear with the remote run URL' : ''}, then rerun finish`);
   }
-  return r.next('Step 16 stage 5: paste these results and the two-line honesty inventory, then run Step 17\'s guard block and gstack-ship-release push');
+  return r.next(`Step 16 stage 5: paste these results and the two-line honesty inventory, then run Step 17's guard block and gstack-ship-release push${loop ? '; Step 18 records the same exception with gstack-ship-receipt write --review-loop-exception' : ''}`);
 }
 
 export function push(argv: readonly string[], cwd: string): Report {
diff --git a/package.json b/package.json
index 2da268dcb..2aaa352ae 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "gstack",
-  "version": "1.92.80",
+  "version": "1.92.81",
   "description": "Garry's Stack — Claude Code skills + fast headless browser. One repo, one install, entire AI engineering workflow.",
   "license": "MIT",
   "type": "module",
diff --git a/test/ship-rereview-triggers.test.ts b/test/ship-rereview-triggers.test.ts
index cf28cc156..581c95713 100644
--- a/test/ship-rereview-triggers.test.ts
+++ b/test/ship-rereview-triggers.test.ts
@@ -662,3 +662,79 @@ describe('a CHANGELOG re-heading after the docs audit (GST-906, GBRA-258, gbrain
     }
   });
 });
+
+describe('finish with an owner-accepted edit after a fix-check (GST-965, GBRA-277, gbrain #8425)', () => {
+  const exceptionFile = (f: ShipFx, head: string, extra: Record<string, unknown> = {}) => {
+    const file = path.join(f.root, 'loop-exception.json');
+    fs.writeFileSync(file, JSON.stringify({ accepted_by: 'GStack Minder (rule 23 grant)', date: '2026-11-02', via: 'GSTA-36', reason: 'the pre-agreed fallback after the fix-check blocker, then ship', fix_check: { head, findings: 1 }, ...extra }));
+    return file;
+  };
+  const afterFixCheck = () => {
+    const f = fx({ 'src/x.ts': 'export const x = 1;\n' }, { 'src/x.ts': 'export const x = 2;\n', 'README.md': '# Fx, with the branch feature\n' });
+    const fixHead = f.git('rev-parse', 'HEAD').stdout.trim();
+    const reviewed = wtree(f);
+    const post = snapshot(f);
+    f.write({ 'src/x.ts': 'export const x = 3;\n', 'README.md': '# Fx, with the fallback\n' }); f.commit('owner fallback after the fix-check');
+    return { f, fixHead, reviewed, post };
+  };
+
+  // Value: protects=an edit the owner accepted after a fix-check ships through Step 16 without another review round (R51 cap);
+  //   fails_when=finish routes the accepted src edit as behavior or stales the docs stage on the accepted doc, or accepts it
+  //   without the exception file; seam=finish's route and docsFresh; no_claim=the receipt and queue checks (exceptionProblems)
+  test('paths changed since fix_check.head route as accepted and the docs stage carries them; without the file it still stops', () => {
+    const { f, fixHead, reviewed, post } = afterFixCheck();
+    const plain = finish(f, reviewed, post);
+    expect([plain.status, plain.stdout.includes('STOP: behavior, tests or build inputs changed since review (src/x.ts)')]).toEqual([3, true]);
+    expect(lineOf(plain.stdout, 'ROUTE')).toMatch(/ behavior=1 docs=1 metadata=0 tests=0 base=unknown context=0$/);
+    const r = finish(f, reviewed, post, ['--review-loop-exception', `@${exceptionFile(f, fixHead)}`]);
+    expect(lineOf(r.stdout, 'ROUTE')).toMatch(/ behavior=0 docs=0 metadata=0 tests=0 base=unknown context=0 accepted=2$/);
+    expect(lineOf(r.stdout, 'REVIEW_LOOP_EXCEPTION')).toBe(`REVIEW_LOOP_EXCEPTION: accepted README.md,src/x.ts (changed since fix-check ${fixHead.slice(0, 12)}; accepted by GStack Minder (rule 23 grant) 2026-11-02 via GSTA-36)`);
+    expect(lineOf(r.stdout, 'DOCS')).toBe('DOCS: fresh against the Step 14.5 post-audit snapshot (carried: tests=0 base=0 requested=0 accepted=2)');
+    expect([r.status, lineOf(r.stdout, 'EVIDENCE')]).toEqual([0, expect.stringMatching(/^EVIDENCE: FRESH label=tests /)]);
+  });
+
+  // Value: protects=the exception covers only the frozen scope and only a real fix-check; fails_when=a path the branch never
+  //   touched at the fix-check, or an exception naming a commit outside the branch or a malformed record, is accepted; seam=finish
+  test('a path outside the frozen scope still routes as behavior; a fix-check head off the branch or a malformed record stops', () => {
+    const { f, fixHead, reviewed, post } = afterFixCheck();
+    f.write({ 'src/new.ts': 'export const n = 1;\n' }); f.commit('a new file after the fix-check');
+    const outside = finish(f, reviewed, post, ['--review-loop-exception', `@${exceptionFile(f, fixHead)}`]);
+    expect(lineOf(outside.stdout, 'ROUTE')).toMatch(/ behavior=1 .* accepted=2$/);
+    expect([outside.status, outside.stdout.includes('STOP: behavior, tests or build inputs changed since review (src/new.ts)')]).toEqual([3, true]);
+    const off = finish(f, reviewed, post, ['--review-loop-exception', `@${exceptionFile(f, 'e'.repeat(40))}`]);
+    expect([off.status, off.stdout]).toEqual([3, expect.stringContaining(`STOP: --review-loop-exception: fix_check.head eeeeeeeeeeee is not HEAD or an ancestor of it`)]);
+    const bad = finish(f, reviewed, post, ['--review-loop-exception', `@${exceptionFile(f, fixHead, { reason: '' })}`]);
+    expect([bad.status, bad.stdout]).toEqual([3, expect.stringContaining('STOP: --review-loop-exception: review_loop_exception.reason must be a non-empty one-line string')]);
+    const missing = finish(f, reviewed, post, ['--review-loop-exception', `@${path.join(f.root, 'no-such-record.json')}`]);
+    expect([missing.status, missing.stdout]).toEqual([3, expect.stringContaining('STOP: --review-loop-exception: ')]);
+  });
+
+  // Value: protects=fix_check.files only narrows the frozen scope; fails_when=a listed path the branch never touched at the
+  //   fix-check routes as accepted, or a touched path the list leaves out is accepted anyway; seam=finish's route
+  test('fix_check.files narrows the accepted scope and cannot add a path the branch had not touched at the fix-check', () => {
+    const { f, fixHead, reviewed, post } = afterFixCheck();
+    f.write({ 'src/new.ts': 'export const n = 1;\n' }); f.commit('a new file after the fix-check');
+    const r = finish(f, reviewed, post, ['--review-loop-exception', `@${exceptionFile(f, fixHead, { fix_check: { head: fixHead, findings: 1, files: ['src/x.ts', 'src/new.ts'] } })}`]);
+    expect(lineOf(r.stdout, 'ROUTE')).toMatch(/ behavior=1 docs=1 metadata=0 tests=0 base=unknown context=0 accepted=1$/);
+    expect(lineOf(r.stdout, 'REVIEW_LOOP_EXCEPTION')).toStartWith('REVIEW_LOOP_EXCEPTION: accepted src/x.ts (');
+    expect([r.status, r.stdout.includes('STOP: behavior, tests or build inputs changed since review (src/new.ts)')]).toEqual([3, true]);
+    const none = finish(f, reviewed, post, ['--review-loop-exception', `@${exceptionFile(f, fixHead, { fix_check: { head: fixHead, findings: 1, files: [] } })}`]);
+    expect([none.status, none.stdout]).toEqual([3, expect.stringContaining('STOP: --review-loop-exception: fix_check.files lists no paths')]);
+  });
+
+  // Value: protects=a test weakened after the fix-check is never owner-accepted, with or without --reviewed-base;
+  //   fails_when=finish accepts a gate-edited test (a .skip, a removed assertion) because no base sides were given; seam=finish's route
+  test('a gate-edited test after the fix-check still routes as behavior under the exception', () => {
+    const f = fx({ 'src/x.ts': 'export const x = 1;\n' }, { 'src/x.ts': 'export const x = 2;\n', 'test/x.test.ts': "import { expect, test } from 'bun:test';\ntest('a', () => { expect(2).toBe(2); });\n" });
+    const fixHead = f.git('rev-parse', 'HEAD').stdout.trim();
+    const reviewed = wtree(f);
+    const reviewedBase = mergeBase(f);
+    const post = snapshot(f);
+    f.write({ 'test/x.test.ts': "import { test } from 'bun:test';\ntest.skip('a', () => {});\n" }); f.commit('weaken the test after the fix-check');
+    for (const extra of [[], ['--reviewed-base', reviewedBase]]) {
+      const r = finish(f, reviewed, post, ['--review-loop-exception', `@${exceptionFile(f, fixHead)}`, ...extra]);
+      expect([r.status, lineOf(r.stdout, 'ROUTE'), r.stdout.includes('STOP: behavior, tests or build inputs changed since review (test/x.test.ts)')]).toEqual([3, expect.stringMatching(/ behavior=1 .* accepted=0$/), true]);
+    }
+  });
+});
+

From 26539fb0d9b0351c5e0c772801af0b9374af0335 Mon Sep 17 00:00:00 2001
From: Garry Tan <garrytan@gmail.com>
Date: Mon, 2 Nov 2026 06:26:18 -0800
Subject: [PATCH 6/6] v1.92.82.0 fix(ship): verify re-derives each merged
 finding from the reader's output (GST-541) (#3717)

* fix(ship): verify re-derives each merged row by index (GST-541)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(ship): verify fails closed on a missing output, a forged merged copy and uncollected rows (GST-541 review)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(ship): a missing recorded output file is dirty even with no findings (GST-541 fix-check)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(ship): a reply collect refused as no review keeps its NOT_REVIEWED row (Rule 25 with #3701)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(ship): only a reply refused for its tree skips verify's count check (GST-541 fix-check)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(ship): verify re-derives a tree refusal from the reply before it skips the count check or voids an output (GST-541 fix-check 2)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(ship): an output with no readable recorded file is dirty even with no ids (GST-541 fix-check 3)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(ship): verify accepts an output as a review only when its reply names the render's tree (GST-541 red team)

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* fix(review-ledger): judge binding compares real paths, so a run reached through a symlink verifies clean

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

---------

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---
 CHANGELOG.md                   |  14 ++++
 VERSION                        |   2 +-
 agents-digest/gstack-AGENTS.md |   2 +-
 docs/review-policy.md          |  13 +++-
 lib/review-ledger.ts           |  70 ++++++++++++++----
 package.json                   |   2 +-
 test/review-dispatch.test.ts   | 127 +++++++++++++++++++++++++++++++++
 7 files changed, 213 insertions(+), 17 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 209155077..a99754d85 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,19 @@
 # Changelog
 
+## [1.92.82.0] - 2026-11-02
+
+**`gstack-review-dispatch verify` now catches a merged findings file edited after collect.** Verify used to compare only output hashes and line counts. So a `merged.jsonl` row whose label was lowered, or whose text was replaced with another line's summary, passed with `VERIFY: clean` once the judge agreed with the edited row. Verify now re-parses each reader's unchanged output and checks every merged row against the finding at its index, label and raw text byte for byte. A mismatch prints `merged_line_differs` on the OUTPUT and JUDGE rows, and verify is dirty. A missing output file or an output event with no file recorded (even for a reply with no findings), a judgment run on a copy of the merged file, merged rows that collect never wrote, and an output whose finding count no longer matches its recorded ids are dirty too. A reply refused for naming the wrong checkout is still allowed to hold findings collect never merged, but only when verify re-checks its tree line and finds it wrong.
+
+### Known
+- Deleting a whole output event together with its merged rows still verifies clean, because the ledger keeps no anchor for it. It is listed for the structured-findings plan.
+
+### Tests
+- `test/review-dispatch.test.ts`: a forged label, a summary copied onto a blocker's id, a deleted output file, a judgment on a forged copy and an extra merged row all verify dirty. Each verified clean at main c830c9b.
+- `test/review-dispatch.test.ts`: an output event edited to look like a wrong-tree refusal (no ids, `review: false`, `tree_ok: false`) verifies dirty, even after a re-sent brief is answered, and so does one edited to drop its file and ids. A real wrong-tree reply stays `NOT_REVIEWED` until a re-sent brief gets a review, then verifies clean.
+
+### Docs
+- `docs/review-policy.md` (verify) describes the per-row check.
+
 ## [1.92.81.0] - 2026-11-02
 
 **`gstack-ship-release finish` now accepts the owner's edit after a fix-check instead of asking for another review round (GST-965).** On gbrain #8425, the fix-check found a blocker, the owner's pre-agreed fallback was applied and "then ship" was granted. Finish still classed the edit as behavior and stopped with `insert 5–11.5 → 12–14 → 16`, a third round the R51 cap forbids, so /ship could not get through Step 16 without running its stages by hand.
diff --git a/VERSION b/VERSION
index 9a1e28b7f..9a90f85b2 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-1.92.81.0
+1.92.82.0
diff --git a/agents-digest/gstack-AGENTS.md b/agents-digest/gstack-AGENTS.md
index 45e52d15a..2a13be4cc 100644
--- a/agents-digest/gstack-AGENTS.md
+++ b/agents-digest/gstack-AGENTS.md
@@ -1,4 +1,4 @@
-# gstack digest v1.92.81.0 — regenerate/re-copy after upgrading gstack
+# gstack digest v1.92.82.0 — regenerate/re-copy after upgrading gstack
 
 Behavioral rules from gstack (https://github.com/garrytan/gstack), compressed
 for agent hosts without a full skill install. The full skills add workflows,
diff --git a/docs/review-policy.md b/docs/review-policy.md
index 16e93e891..f7b872281 100644
--- a/docs/review-policy.md
+++ b/docs/review-policy.md
@@ -579,7 +579,18 @@ ones, and the arm lost blockers. `bin/gstack-review-dispatch` closes both ends:
    missing coverage leaves the review incomplete. A mismatched send re-sent
    byte-exact before anything was collected is void and counted (`voided=N` on the
    verdict line); one that produced an output stays `MISMATCH`. Then output hashes
-   against the merged lines they produced, and the judgment rerun against the
+   against the merged lines they produced, and each merged line against the finding collect
+   parses at that id's index from the unchanged output file, label and raw text byte for byte
+   (`merged_line_differs` on the OUTPUT and JUDGE rows otherwise, GST-541), so an edited
+   merged.jsonl cannot lower a finding or swap its text. A recorded output file that is missing
+   (`output_file_missing`), a judgment run on a merged file collect did not write
+   (`judge_merged_not_collected`) and merged rows collect never recorded
+   (`merged_rows_not_collected`) are dirty too, and so is an output whose re-parsed findings
+   outnumber or undercount the ids collect recorded (`merged_count_differs`). The one exception
+   is a reply collect refused for its tree: verify re-checks the reply's `REVIEWED_TREE` line
+   against the render's tree, and such an output stays `NOT_REVIEWED` until a re-sent brief is
+   answered by a reply that passes the same check. Deleting a whole output event with its rows is
+   not caught yet, because the ledger has no anchor for it. The judgment is then rerun against the
    recorded judged set, covering every collected id. A ledger from a run that kept no
    prompts prints `VERIFY: no brief records` (exit 3), never clean.
 6. `pass --base <b> --out <dir> --pass <id> -- <plan flags>` runs one pass from a
diff --git a/lib/review-ledger.ts b/lib/review-ledger.ts
index 5f2973f66..dd3266dee 100644
--- a/lib/review-ledger.ts
+++ b/lib/review-ledger.ts
@@ -16,7 +16,7 @@
  */
 import * as fs from 'node:fs';
 import * as path from 'node:path';
-import { isReviewedTreeLine, promptHash, type TreeCheck } from './review-brief';
+import { checkReviewedTree, isReviewedTreeLine, promptHash, type TreeCheck } from './review-brief';
 
 export interface LedgerEvent { event: 'rendered' | 'sent' | 'output' | 'judgment' | 'judge' | 'program'; [k: string]: unknown }
 export const readerKey = (role: string, name?: string) => (name ? `${role}/${name}` : role);
@@ -236,6 +236,21 @@ export function recordJudgment(ledger: string, merged: MergedLine[], judged: Jud
   return res;
 }
 
+/** A recorded output's file text while it still hashes to what collect recorded. */
+function unchangedOutput(o: LedgerEvent): string | undefined {
+  try {
+    const text = typeof o.file === 'string' ? fs.readFileSync(o.file, 'utf8') : undefined;
+    return text !== undefined && promptHash(text) === o.sha256 ? text : undefined;
+  } catch { return undefined; }
+}
+
+/** Whether a reply names the tree its render recorded (GST-482), checked on the reply itself, never on an output event's editable fields. */
+const treeHolds = (render: LedgerEvent | undefined, reply: string | undefined) => !render || !('tree' in render)
+  || (reply !== undefined && checkReviewedTree(reply, typeof render.tree === 'string' ? render.tree : null).ok);
+
+/** One file however its path is spelled: collect and judge may reach a run directory through a symlink. */
+const realPath = (f: string) => { try { return fs.realpathSync(f); } catch { return path.resolve(f); } };
+
 /**
  * The blocking findings a run's ledger holds (GST-951): every collected line whose label ranks CRITICAL, raised to its judged
  * severity when the newest judge accepted a judged set that is unchanged since; lead judgment's additions count and a folded
@@ -284,15 +299,16 @@ export function ledgerBlockers(ledger: string): LedgerBlocker[] {
 }
 
 /**
- * Whether a recorded output was a review: it has findings, or its file (unchanged since collect) carries the reader's clean marker, so
- * a ledger written before collect recorded the field is judged by the same rule; with no readable file, collect's own review field.
+ * Whether a recorded output was a review: its unchanged reply names its render's tree, and it has findings or carries the reader's
+ * clean marker, so a ledger written before collect recorded the field is judged by the same rule; with no readable file, collect's own
+ * review field.
  */
-function reviewed(o: LedgerEvent): boolean {
+function reviewed(o: LedgerEvent, render?: LedgerEvent): boolean {
   if (o.tree_ok === false || unavailableOutput(o)) return false;
+  const text = unchangedOutput(o);
+  if (!treeHolds(render, text)) return false;
   if (Array.isArray(o.ids) && o.ids.length) return true;
-  let text: string | undefined;
-  try { text = typeof o.file === 'string' ? fs.readFileSync(o.file, 'utf8') : undefined; } catch { /* unreadable */ }
-  return text !== undefined && promptHash(text) === o.sha256 ? hasCleanMarker(String(o.reader), text) : o.review === true;
+  return text !== undefined ? hasCleanMarker(String(o.reader), text) : o.review === true;
 }
 
 /** Every id collect recorded, in order. */
@@ -314,6 +330,7 @@ export function verify(ledger: string): { rows: VerifyRow[]; clean: boolean; bri
     return { at, next, reader, label: nth > 1 ? `${reader}@${nth}` : reader };
   });
   let voided = 0;
+  const differs: string[] = [];
   for (const p of episodes) {
     const r = events[p.at]!;
     const inside = events.map((e, i) => ({ e, i })).filter(({ e, i }) => i > p.at && i < p.next && e.reader === p.reader);
@@ -331,7 +348,7 @@ export function verify(ledger: string): { rows: VerifyRow[]; clean: boolean; bri
     voided += epVoided;
     const lastMatch = [...sent].reverse().find(({ e }) => e.sha256 === r.sha256);
     const outputs = lastMatch ? inside.filter(({ e, i }) => e.event === 'output' && i > lastMatch.i) : [];
-    const collected = outputs.some(({ e }) => reviewed(e));
+    const collected = outputs.some(({ e }) => reviewed(e, r));
     const unavailable = !collected && outputs.some(({ e }) => unavailableOutput(e));
     const status = !sent.length ? 'NOT_SENT' : mismatched ? 'MISMATCH' : collected ? 'MATCH' : unavailable ? 'UNAVAILABLE' : 'NOT_COLLECTED';
     const tampered = fileNow && fileNow !== r.sha256 ? ' file_changed_since_render' : '';
@@ -346,20 +363,38 @@ export function verify(ledger: string): { rows: VerifyRow[]; clean: boolean; bri
     const lastSent = p ? events.slice(p.at + 1, i).filter(e => e.event === 'sent' && e.reader === reader).at(-1) : undefined;
     if (!p) { status = 'MISMATCH'; notes.push('brief_not_rendered'); }
     else if (!lastSent || lastSent.sha256 !== events[p.at]!.sha256) { status = 'MISMATCH'; notes.push('brief_not_sent_verbatim'); }
-    if (typeof o.file === 'string' && fs.existsSync(o.file) && promptHash(fs.readFileSync(o.file, 'utf8')) !== o.sha256) { status = 'MISMATCH'; notes.push('output_file_changed'); }
+    let text: string | undefined;
+    try { text = typeof o.file === 'string' ? fs.readFileSync(o.file, 'utf8') : undefined; } catch { /* missing or unreadable */ }
+    const render = p ? events[p.at]! : undefined;
+    if (text !== undefined && promptHash(text) !== o.sha256) { status = 'MISMATCH'; notes.push('output_file_changed'); }
     const ids = Array.isArray(o.ids) ? (o.ids as string[]) : [];
     let lines: MergedLine[] = [];
     try { lines = typeof o.merged === 'string' ? readJsonl<MergedLine>(o.merged).filter(l => ids.includes(l.id)) : []; } catch { notes.push('merged_unreadable'); status = 'MISMATCH'; }
     if (lines.length !== ids.length) { status = 'MISMATCH'; notes.push(`merged_lines=${lines.length}/${ids.length}`); }
     if (lines.some(l => l.output_sha256 !== o.sha256)) { status = 'MISMATCH'; notes.push('merged_line_from_other_output'); }
-    if (status === 'MATCH' && !reviewed(o)) {
+    // Each merged row must be the finding collect parses at that id's index from the unchanged output, label and raw text byte for byte.
+    if (text === undefined) {
+      status = 'MISMATCH';
+      notes.push(typeof o.file === 'string' ? 'output_file_missing' : 'output_file_unrecorded');
+      differs.push(...ids);
+    } else if (promptHash(text) === o.sha256) {
+      const parsed = parseFindings(reader, text);
+      const off = ids.filter((id, k) => { const m = lines.find(l => l.id === id); return !m || parsed[k]?.label !== m.label || parsed[k]?.raw !== m.raw; });
+      if (unavailableOutput(o) && parsed.length) { status = 'MISMATCH'; notes.push('unavailable_output_has_findings'); }
+      else if (!unavailableOutput(o)) {
+        if (off.length) { status = 'MISMATCH'; notes.push(`merged_line_differs=${JSON.stringify(off)}`); differs.push(...off); }
+        // A reply collect refused for its tree (GST-482) records no ids whatever it holds; the refusal is re-derived from the reply.
+        if (parsed.length !== ids.length && !(o.tree_ok === false && !ids.length && !treeHolds(render, text))) { status = 'MISMATCH'; notes.push(`merged_count_differs=${parsed.length}/${ids.length}`); }
+      }
+    }
+    if (status === 'MATCH' && !reviewed(o, render)) {
       // A non-review the same dispatch later replaced with a review (the brief re-sent verbatim, then answered) is void, counted like a voided send.
       const later = p ? events.slice(i + 1, p.next) : [];
-      const review = later.findIndex(e => e.event === 'output' && e.reader === reader && reviewed(e));
+      const review = later.findIndex(e => e.event === 'output' && e.reader === reader && reviewed(e, render));
       const replaced = review > 0 && later.slice(0, review).some(e => e.event === 'sent' && e.reader === reader && e.sha256 === events[p!.at]!.sha256);
       if (unavailableOutput(o)) { status = 'UNAVAILABLE'; notes.push(`reason=${JSON.stringify(o.unavailable)}`); }
       else if (replaced) { status = 'VOIDED'; voided++; notes.push('replaced_by_a_review'); }
-      else if (o.tree_ok === false) { status = 'NOT_REVIEWED'; notes.push(`wrong_tree reported=${JSON.stringify(o.reported_tree)} expected=${JSON.stringify(o.tree)}`); }
+      else if (o.tree_ok === false || !treeHolds(render, text)) { status = 'NOT_REVIEWED'; notes.push(`wrong_tree reported=${JSON.stringify(o.reported_tree)} expected=${JSON.stringify(o.tree)}`); }
       else { status = 'NOT_REVIEWED'; notes.push('no_finding_no_clean_marker'); }
     }
     rows.push({ kind: 'OUTPUT', reader: p?.label ?? reader, status, detail: `sha=${short(o.sha256)} findings=${ids.length}${notes.length ? ' ' + notes.join(' ') : ''}` });
@@ -373,13 +408,22 @@ export function verify(ledger: string): { rows: VerifyRow[]; clean: boolean; bri
     const judgedIds = new Set(Array.isArray(v.merged_ids) ? (v.merged_ids as string[]) : []);
     const missing = collectedIds(events.slice(0, lastJudge)).filter(id => !judgedIds.has(id));
     if (missing.length) notes.push(`merged_ids_missing=${missing.join(',')}`);
+    // The judged set must come from the merged file collect wrote, holding only rows collect recorded.
+    const outputs = events.slice(0, lastJudge).filter(e => e.event === 'output' && Array.isArray(e.ids) && e.ids.length);
+    if (outputs.some(e => typeof e.merged !== 'string' || realPath(e.merged) !== realPath(String(v.merged)))) notes.push('judge_merged_not_collected');
     let rerun: Judgment | null = null;
     try {
       const judgedText = fs.readFileSync(String(v.judged), 'utf8');
       if (promptHash(judgedText) !== v.judged_sha256) notes.push('judged_changed_since_judge');
-      rerun = judge(readJsonl<MergedLine>(String(v.merged)), readJsonl<JudgedLine>(String(v.judged)));
+      const merged = readJsonl<MergedLine>(String(v.merged));
+      const collected = new Set(collectedIds(events.slice(0, lastJudge)));
+      const extra = merged.filter(m => !collected.has(m.id)).map(m => m.id);
+      if (extra.length) notes.push(`merged_rows_not_collected=${JSON.stringify(extra)}`);
+      rerun = judge(merged, readJsonl<JudgedLine>(String(v.judged)));
     } catch { notes.push('judged_or_merged_unreadable'); }
     if (rerun && !rerun.ok) notes.push(`rerun_rejects dropped=${JSON.stringify(rerun.dropped)} downgraded=${JSON.stringify(rerun.downgraded)}`);
+    const forged = differs.filter(id => judgedIds.has(id));
+    if (forged.length) notes.push(`merged_line_differs=${JSON.stringify(forged)}`);
     const status = !v.ok ? 'REJECTED' : notes.length ? 'MISMATCH' : 'OK';
     rows.push({ kind: 'JUDGE', reader: '-', status, detail: `dropped=${JSON.stringify(v.dropped)} downgraded=${JSON.stringify(v.downgraded)} added=${v.added} promoted=${v.promoted} folded=${v.folded}${notes.length ? ' ' + notes.join(' ') : ''}` });
   } else if (events.some(e => e.event === 'output')) rows.push({ kind: 'JUDGE', reader: '-', status: 'MISSING', detail: 'outputs were collected but no judged set was checked' });
diff --git a/package.json b/package.json
index 2aaa352ae..180c0ad68 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "gstack",
-  "version": "1.92.81",
+  "version": "1.92.82",
   "description": "Garry's Stack — Claude Code skills + fast headless browser. One repo, one install, entire AI engineering workflow.",
   "license": "MIT",
   "type": "module",
diff --git a/test/review-dispatch.test.ts b/test/review-dispatch.test.ts
index d0c01e345..56e140c84 100644
--- a/test/review-dispatch.test.ts
+++ b/test/review-dispatch.test.ts
@@ -683,6 +683,133 @@ describe('gstack-review-dispatch sent, collect, judge and verify: nothing change
     const v = run(['verify', '--ledger', ledger]);
     expect([v.status, v.stdout.trim()]).toEqual([3, "VERIFY: no brief records (this run's ledger kept no prompts; dispatched-vs-rendered cannot be checked)"]);
   });
+
+  // Value: protects=verify re-derives every merged row from the reader's unchanged output by its index, so a merged.jsonl edited
+  //   after collect cannot lower a finding or swap a blocker's text for a summary and still verify clean; fails_when=a forged label
+  //   or a copied raw line on a real blocker's id verifies clean (GST-541: both did at main c830c9b), or a clean run collected through a
+  //   symlinked run directory and judged there with relative paths verifies dirty (GST-541 ship2-p11); why_new=verify compared only
+  //   output hashes and line counts; seam=verify; no_claim=that an output collected without a recorded file can be re-derived
+  test('verify re-derives each merged row by index: a forged label or a summary copied onto a blocker id is dirty', () => {
+    const blocker = '{"severity":"CRITICAL","path":"log.ts","line":3,"category":"security","summary":"bearer token written to a public log","specialist":"testing"}';
+    const summary = '{"severity":"INFORMATIONAL","path":"a.ts","line":9,"category":"style","summary":"I found nothing CRITICAL.","specialist":"testing"}';
+    const setup = () => {
+      const p = pass();
+      run(['sent', '--ledger', p.ledger, '--role', 'specialist', '--name', 'testing', '--sent', p.file]);
+      const out = p.w('spec.out', blocker + '\n' + summary + '\n');
+      expect(run(['collect', '--ledger', p.ledger, '--role', 'specialist', '--name', 'testing', '--output', out, '--merged', p.merged]).status).toBe(0);
+      return p;
+    };
+    const judgeAll = (p: ReturnType<typeof pass>, severity: string) => run(['judge', '--ledger', p.ledger, '--merged', p.merged, '--judged', p.w('judged.jsonl',
+      readJsonl<MergedLine>(p.merged).map(m => JSON.stringify({ id: m.id, severity: m.id.endsWith('#1') ? severity : 'INFORMATIONAL' })).join('\n') + '\n')]);
+    const clean = setup();
+    expect(judgeAll(clean, 'CRITICAL').status).toBe(0);
+    expect(run(['verify', '--ledger', clean.ledger]).stdout).toContain('VERIFY: clean');
+    const forge = (p: ReturnType<typeof pass>, edit: (rows: MergedLine[]) => void) => {
+      const rows = readJsonl<MergedLine>(p.merged);
+      edit(rows);
+      fs.writeFileSync(p.merged, rows.map(r => JSON.stringify(r)).join('\n') + '\n');
+      expect(judgeAll(p, 'INFORMATIONAL').stdout.trim()).toBe('JUDGMENT: ok added=0 promoted=0 folded=0');
+      return run(['verify', '--ledger', p.ledger]);
+    };
+    const label = forge(setup(), rows => { rows[0]!.label = 'INFORMATIONAL'; });
+    expect([label.status, label.stdout]).toEqual([1, expect.stringContaining('merged_line_differs=["specialist/testing#1"]')]);
+    const copied = forge(setup(), rows => { rows[0]!.label = rows[1]!.label; rows[0]!.raw = rows[1]!.raw; });
+    expect([copied.status, copied.stdout]).toEqual([1, expect.stringContaining('merged_line_differs=["specialist/testing#1"]')]);
+    expect(copied.stdout).toMatch(/^JUDGE - MISMATCH .*merged_line_differs/m);
+    const gone = setup();
+    fs.rmSync(path.join(gone.out, 'spec.out'));
+    const missing = forge(gone, rows => { rows[0]!.label = 'INFORMATIONAL'; });
+    expect([missing.status, missing.stdout.includes('output_file_missing'), missing.stdout.includes('merged_line_differs=["specialist/testing#1","specialist/testing#2"]')]).toEqual([1, true, true]);
+    const copy = setup();
+    const forgedCopy = copy.w('merged2.jsonl', readJsonl<MergedLine>(copy.merged).map((r, k) => JSON.stringify(k ? r : { ...r, label: 'INFORMATIONAL' })).join('\n') + '\n');
+    expect(run(['judge', '--ledger', copy.ledger, '--merged', forgedCopy, '--judged', copy.w('judged.jsonl', readJsonl<MergedLine>(copy.merged).map(m => JSON.stringify({ id: m.id, severity: 'INFORMATIONAL' })).join('\n') + '\n')]).status).toBe(0);
+    const viaCopy = run(['verify', '--ledger', copy.ledger]);
+    expect([viaCopy.status, viaCopy.stdout]).toEqual([1, expect.stringMatching(/^JUDGE - MISMATCH .*judge_merged_not_collected/m)]);
+    const relinked = pass();
+    const linkDir = path.join(tmp(), 'run');
+    fs.symlinkSync(relinked.out, linkDir);
+    const ledgerViaLink = path.join(linkDir, 'review-ledger.jsonl');
+    run(['sent', '--ledger', ledgerViaLink, '--role', 'specialist', '--name', 'testing', '--sent', relinked.file]);
+    const linkOut = relinked.w('spec.out', blocker + '\n' + summary + '\n');
+    expect(run(['collect', '--ledger', ledgerViaLink, '--role', 'specialist', '--name', 'testing', '--output', path.join(linkDir, path.basename(linkOut)), '--merged', path.join(linkDir, 'merged.jsonl')]).status).toBe(0);
+    relinked.w('judged.jsonl', readJsonl<MergedLine>(relinked.merged).map(m => JSON.stringify({ id: m.id, severity: m.label })).join('\n') + '\n');
+    expect(run(['judge', '--ledger', 'review-ledger.jsonl', '--merged', 'merged.jsonl', '--judged', 'judged.jsonl'], linkDir).status).toBe(0);
+    const linked = run(['verify', '--ledger', ledgerViaLink]);
+    expect([linked.status, linked.stdout.includes('judge_merged_not_collected'), linked.stdout.includes('VERIFY: clean')]).toEqual([0, false, true]);
+    const quiet = pass();
+    run(['sent', '--ledger', quiet.ledger, '--role', 'specialist', '--name', 'testing', '--sent', quiet.file]);
+    const none = quiet.w('none.out', 'NO FINDINGS\n');
+    expect(run(['collect', '--ledger', quiet.ledger, '--role', 'specialist', '--name', 'testing', '--output', none, '--merged', quiet.merged]).status).toBe(0);
+    fs.writeFileSync(quiet.merged, '');
+    expect(run(['judge', '--ledger', quiet.ledger, '--merged', quiet.merged, '--judged', quiet.w('judged.jsonl', '')]).status).toBe(0);
+    fs.rmSync(none);
+    const noneGone = run(['verify', '--ledger', quiet.ledger]);
+    expect([noneGone.status, noneGone.stdout.includes('output_file_missing')]).toEqual([1, true]);
+    const extra = forge(setup(), rows => { rows[0]!.label = 'INFORMATIONAL'; rows.push({ ...rows[1]!, id: 'specialist/testing#9' }); });
+    expect([extra.status, extra.stdout.includes('merged_rows_not_collected=["specialist/testing#9"]')]).toEqual([1, true]);
+  });
+  // Value: protects=an output event edited to read like a refused reply (no ids, review false, tree_ok false) or to drop its file
+  //   cannot drop its findings and still verify clean, while a reply collect really refused for its tree stays NOT_REVIEWED until a re-sent brief is answered;
+  //   fails_when=verify skips the count check on the editable review or tree_ok field, so a clean marker keeps the output MATCH or a
+  //   later review voids it, or an output with no recorded file passes on its review field (fix-checks at 721539c, 56ed6e3 and 50dc73c: VERIFY: clean); why_new=the GST-482 count exemption trusted collect's
+  //   fields; seam=verify; no_claim=that a deleted output event is caught (#3719)
+  test('an output event edited to look refused, or a refused one edited to look accepted, is dirty; a real wrong-tree reply is NOT_REVIEWED, then voided by a review', () => {
+    const head = spawnSync('git', ['rev-parse', 'HEAD'], { cwd: ROOT, encoding: 'utf8', timeout: 30_000 }).stdout.trim();
+    const findings = '1. **CRITICAL**: auth.ts:3 leaks a token\nBlocks: yes\n\nRecommendation: request changes\n';
+    const voice = (reply: string) => {
+      const out = tmp();
+      const ledger = path.join(out, 'review-ledger.jsonl');
+      const merged = path.join(out, 'merged.jsonl');
+      const brief = field(run(['render', '--role', 'outside-voice', '--out', out, '--diff-base', head]).stdout, 'file');
+      const send = () => run(['sent', '--ledger', ledger, '--role', 'outside-voice', '--sent', brief]);
+      let n = 0;
+      const take = (text: string) => { const f = path.join(out, `ov${++n}.out`); fs.writeFileSync(f, text); return run(['collect', '--ledger', ledger, '--role', 'outside-voice', '--output', f, '--merged', merged]); };
+      send();
+      const c = take(reply);
+      const forge = (edit: Record<string, unknown>) => {
+        fs.writeFileSync(ledger, readJsonl<Record<string, unknown>>(ledger).map(e => JSON.stringify(e.event === 'output' ? { ...e, ...edit } : e)).join('\n') + '\n');
+        fs.writeFileSync(merged, '');
+      };
+      const verify = () => {
+        fs.writeFileSync(path.join(out, 'judged.jsonl'), '');
+        expect(run(['judge', '--ledger', ledger, '--merged', merged, '--judged', path.join(out, 'judged.jsonl')]).status).toBe(0);
+        return run(['verify', '--ledger', ledger]);
+      };
+      return { c, send, take, forge, verify };
+    };
+    const noIds = voice(findings);
+    expect(noIds.c.status).toBe(0);
+    noIds.forge({ ids: [], review: false });
+    const v = noIds.verify();
+    expect([v.status, v.stdout.includes('merged_count_differs=1/0'), v.stdout.includes('VERIFY: dirty')]).toEqual([1, true, true]);
+    const unfiled = voice(findings);
+    unfiled.forge({ file: null, ids: [] });
+    const u = unfiled.verify();
+    expect([u.status, u.stdout.includes('output_file_unrecorded'), u.stdout.includes('VERIFY: dirty')]).toEqual([1, true, true]);
+    const flagged = voice(findings);
+    flagged.forge({ ids: [], review: false, tree_ok: false });
+    flagged.send();
+    expect(flagged.take('Recommendation: approve\n').status).toBe(0);
+    const f = flagged.verify();
+    expect([f.status, f.stdout.includes('merged_count_differs=1/0'), f.stdout.includes('VERIFY: dirty')]).toEqual([1, true, true]);
+    const wrong = voice(`REVIEWED_TREE: ${'b'.repeat(40)} /elsewhere\n` + findings);
+    expect([wrong.c.status, wrong.c.stdout.startsWith('COLLECT_WRONG_TREE: reader=outside-voice')]).toEqual([1, true]);
+    const w = wrong.verify();
+    expect([w.status, /^OUTPUT outside-voice NOT_REVIEWED .*wrong_tree/m.test(w.stdout), w.stdout.includes('merged_count_differs')]).toEqual([1, true, false]);
+    wrong.send();
+    expect(wrong.take('Recommendation: approve\n').status).toBe(0);
+    const r = wrong.verify();
+    expect([r.status, r.stdout.includes('replaced_by_a_review'), r.stdout.includes('VERIFY: clean briefs=1 outputs=2 voided=1')]).toEqual([0, true, true]);
+    const elsewhere = pass();
+    run(['sent', '--ledger', elsewhere.ledger, '--role', 'specialist', '--name', 'testing', '--sent', elsewhere.file]);
+    const refused = run(['collect', '--ledger', elsewhere.ledger, '--role', 'specialist', '--name', 'testing', '--output', elsewhere.w('else.out', `REVIEWED_TREE: ${'b'.repeat(40)} /elsewhere\nNO FINDINGS\n`), '--merged', elsewhere.merged]);
+    expect(refused.stdout.startsWith('COLLECT_WRONG_TREE: reader=specialist/testing')).toBe(true);
+    fs.writeFileSync(elsewhere.ledger, readJsonl<Record<string, unknown>>(elsewhere.ledger).map(e => JSON.stringify(e.event === 'output' ? { ...e, tree_ok: true, review: true, reported_tree: e.tree } : e)).join('\n') + '\n');
+    fs.writeFileSync(elsewhere.merged, '');
+    expect(run(['judge', '--ledger', elsewhere.ledger, '--merged', elsewhere.merged, '--judged', elsewhere.w('judged.jsonl', '')]).status).toBe(0);
+    const e = run(['verify', '--ledger', elsewhere.ledger]);
+    expect([e.status, /^OUTPUT specialist\/testing NOT_REVIEWED .*wrong_tree/m.test(e.stdout), e.stdout.includes('VERIFY: dirty')]).toEqual([1, true, true]);
+  });
 });
 
 describe('the workspace paragraph (GST-482)', () => {