diff --git a/.github/workflows/gate.yml b/.github/workflows/gate.yml
index 54db751..af2711f 100644
--- a/.github/workflows/gate.yml
+++ b/.github/workflows/gate.yml
@@ -108,12 +108,16 @@ jobs:
     runs-on: ubuntu-latest
     needs: [gate, pr-body]
     if: always()
+    # The edited path may wait up to 20 minutes for running legs (tools/ci_legs_verdict.sh's
+    # bound); the job's own ceiling sits just above it so a hang is a FAIL, not a stall.
+    timeout-minutes: 25
     # The edited path reads the legs' check runs for the head SHA: a read, granted here and
     # nowhere wider.
     permissions:
       contents: read
       checks: read
     steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
       - name: every leg of the matrix passed, and the pull-request body where there is one
         env:
           GH_TOKEN: ${{ github.token }}
@@ -126,18 +130,12 @@ jobs:
             # the REQUIRED context, and a fresh green under that name on a commit whose legs
             # this run never read would let a title edit stand a green beside a red (a review
             # lane's finding). So the legs' verdicts for this head SHA are read from the checks
-            # API -- the latest run of each leg name -- and all three must be success.
+            # API -- the latest run of each leg name -- and all three must be success. The
+            # read WAITS for legs still running (G-079: an edit during the synchronize run read
+            # null conclusions, failed, and the ruleset blocked on that red beside the later
+            # green); tools/ci_legs_verdict.sh polls to a bound, and its probe runs offline.
             [ "${{ needs.pr-body.result }}" = "success" ] || exit 1
-            # The expanded leg names only: the skipped matrix job of THIS run registers one
-            # unexpanded name ("Gate - leg (OTP ${{ '{{' }} matrix.otp ...", skipped), which is
-            # not a verdict (the first edited run measured exactly that).
-            legs=$(gh api "repos/${GITHUB_REPOSITORY}/commits/${HEAD_SHA}/check-runs?filter=latest&per_page=100" \
-              --jq '.check_runs[] | select(.name | test("^Gate - leg \\(OTP [0-9]+, Elixir [0-9.]+, [a-z]+\\)$")) | "\(.name): \(.conclusion)"')
-            printf '%s\n' "$legs"
-            n=$(printf '%s\n' "$legs" | grep -c . || true)
-            bad=$(printf '%s\n' "$legs" | grep -vc ': success$' || true)
-            [ "$n" -ge 3 ] || { echo "FAIL: only $n leg verdict(s) exist for ${HEAD_SHA}; the tree is unmeasured"; exit 1; }
-            [ "$bad" -eq 0 ] || { echo "FAIL: a leg's latest verdict for ${HEAD_SHA} is not success"; exit 1; }
+            bash tools/ci_legs_verdict.sh "${GITHUB_REPOSITORY}" "${HEAD_SHA}"
           else
             [ "${{ needs.gate.result }}" = "success" ]
             case "${{ needs.pr-body.result }}" in success|skipped) ;; *) exit 1 ;; esac
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 4e1fb7d..170c7de 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -11,6 +11,42 @@ All notable changes to this project are documented here. The format follows
 
 ## [Unreleased]
 
+## [0.8.0] — 2026-09-19
+
+The quiet minor: **no public entry added, removed, renamed, hidden or changed in arity** —
+`docs/public-api.txt` is line for line `0.7.0`'s, `release_markers!("0.8.0")` wrote nothing,
+and the census holds an empty Unreleased section to that. No wire byte and no envelope byte
+moves. Instruments and pages only, below. This is the "full minor release unchanged" the
+README names as `1.0.0`'s condition for the public API; `1.0.0` follows once it has stood.
+
+### Changed — instruments (no public entry moves)
+
+- **The gate diffs `docs/public-api.txt` against `origin/main`** (a sixteenth step,
+  `baseline`; `tools/baseline_diff.sh`, probed by `tools/probe_baseline_diff.sh`): a public
+  entry's line deleted by hand, or a `since=`/`deprecated_since=`/`removed_in=` marker arriving
+  with a release number not above the CHANGELOG's highest heading — one it lists, or a phantom
+  between two releases — is a FAIL. The census reads the tree
+  alone and passed both edits green (a review lane measured it); this step reads git. Where
+  `origin/main` does not resolve the line says so and is not evidence.
+- **The pull-request summary waits for running legs** on a body edit (`tools/ci_legs_verdict.sh`,
+  polling the head SHA's leg check runs until each is completed, bounded at 20 minutes, a
+  failed API read a poll spent rather than a verdict, probed offline by
+  `tools/probe_ci_legs_verdict.sh` with a scripted `gh`): an edit during a synchronize run no
+  longer fails on null conclusions and blocks the merge beside a later green.
+- `tools/probe_gate_honesty.sh`'s "other steps not pass" count reads `pass (<detail>)` as pass
+  (it counted every detailed line: 7 of its 13 names on a green gate, 10 of the 16 the gate
+  prints now; 0 now).
+
+### Changed — pages (copy)
+
+- `docs/governance.md`'s Scorecard table carries the check's own figure beside each row
+  (read 2026-09-19; aggregate 7), with the rule behind each low one — Maintained is 0 for any
+  repository under 90 days old; Signed-Releases and Packaging read GitHub Releases and a
+  publishing workflow, neither of which a Hex release has — and two rows the first result
+  added (Packaging, CII-Best-Practices).
+- `docs/provenance.md`'s example commands name `0.7.0`; `docs/connectome-canonical.md`'s
+  `schema_version` history no longer says `0.6.0` "carries this note".
+
 ## [0.7.0] — 2026-09-19
 
 The signer seam, and nothing else: three public entries added — `BeamMCP.Signer` (the
diff --git a/HANDOFF.md b/HANDOFF.md
index d950769..a0919b3 100644
--- a/HANDOFF.md
+++ b/HANDOFF.md
@@ -3,10 +3,10 @@ SPDX-FileCopyrightText: 2026 Sudo Apt Holdings LLC
 SPDX-License-Identifier: Apache-2.0
 -->
 
-# HANDOFF — beam_mcp, release 0.7.0 prepared; publish and tag are the owner's
+# HANDOFF — beam_mcp, release 0.8.0 prepared; publish and tag are the owner's
 
 Tag and publish are owner steps — never `mix hex.publish`, never push a tag, never bump the
-version in `mix.exs`. For 0.7.0 the version bump is this release commit, reviewed like any
+version in `mix.exs`. For 0.8.0 the version bump is this release commit, reviewed like any
 other change; publishing and tagging remain the owner's, in the order the runbook below gives.
 
 The slice records — plans, findings, lane reports, signoffs, archived gate runs — live in the
@@ -15,7 +15,16 @@ happened.
 
 ## State
 
-- **`0.7.0` carries the signer seam and nothing else**: `BeamMCP.Signer` (one callback,
+- **`0.8.0` is the quiet minor**: no public entry added, removed, renamed, hidden or changed
+  in arity — `docs/public-api.txt` is `0.7.0`'s line for line, `release_markers!("0.8.0")`
+  wrote nothing — and no wire or envelope byte moved. Instruments: the gate's sixteenth step
+  diffs the baseline against `origin/main` (G-076); the pull-request summary waits for running
+  legs on a body edit (G-079); the honesty probe's discriminator reads detailed pass lines
+  (G-078); each with an offline probe. Pages: the governance table carries the Scorecard's
+  measured figures with the rule behind each low one. `mix.exs` says `0.8.0`; the README
+  recommends `~> 0.8.0` and the requirement test refuses `0.7.0` and `0.6.0`; the wire
+  recording's ten version lines are re-taken. `1.0.0` is next, after this minor has stood.
+- **`0.7.0` carried the signer seam and nothing else**: `BeamMCP.Signer` (one callback,
   `sign/2`, two arguments with pinned names), `BeamMCP.Signer.None` (the one no-op under
   `lib/`) and `BeamMCP.Connectome.Canonical.signature/3` (the one call site, over `encode/2`'s
   bytes, moving no envelope byte). Three public entries added, none removed, renamed or
@@ -24,10 +33,9 @@ happened.
   callback list, the one `def sign`, the one call site — and sixteen mutants hold it. The
   signer that holds a key, `BeamMCP.Signer.Ed25519` (Ed25519 through OTP's `:crypto`, the key
   under `opts[:private_key]`), is the separate package `beam_mcp_signer`
-  (github.com/ScriptKittyOS/beam_mcp_signer, unpublished until this release is on hex.pm);
-  this package does not depend on it. `mix.exs` says `0.7.0`; the README recommends
-  `~> 0.7.0` and the requirement test refuses `0.6.0` and `0.5.0`; the wire recording's ten
-  version lines are re-taken.
+  (github.com/ScriptKittyOS/beam_mcp_signer, 0.1.0 on hex.pm, depending on `~> 0.7.0` — so a
+  host on it cannot take `0.8.0` until a signer release admits it; `UPGRADING.md` says so);
+  this package does not depend on it.
 - **`0.6.0` carried everything since `0.5.0`**: the wire hardening after the threat model, the
   threat-model page, the hash-agile canonical envelope (the release's one break), the install-floor slices (the OTP floor at
   compile time, the CI matrix on three pairs, the dependency audit, build provenance, the
@@ -42,18 +50,20 @@ happened.
   `docs/public-api.txt` carried no `Unreleased` marker, so `release_markers!("0.6.0")` wrote
   nothing; at `0.7.0` it wrote three.
 - **No head hash is written here** — a hash written into the file it describes cannot include
-  the commit that writes it. `git log v0.6.0..main` is the authority.
-- Gate on the release commit: fifteen steps, every line `pass` (the 0.5.0 gate had thirteen;
-  the audit step made it fourteen in 024 and Dialyzer fifteen in 027a) — format (the tracked set, not a
-  glob), compile, instruments, test, credo, properties (11 at 1 000 generations), optional
-  deps, bench (the collector's overhead under the 1.5 µs ceiling; **the diff engine's run and
+  the commit that writes it. `git log v0.7.0..main` is the authority.
+- Gate on the release commit: sixteen steps, every line `pass` (the 0.5.0 gate had thirteen;
+  the audit step made it fourteen in 024, Dialyzer fifteen in 027a, the baseline diff sixteen
+  at 0.8.0) — format (the tracked set, not a
+  glob), compile, dialyzer, instruments, test, credo, properties (11 at 1 000 generations),
+  optional deps, audit, bench (the collector's overhead under the 1.5 µs ceiling; **the diff engine's run and
   encode each under its own ceiling now** — 245 ms and 260 ms, medians of five in a fresh
   process after a warm-up, set at roughly double the stable worst of ten runs on the release
   head; the reachability queries' cost recorded and judged by no number, but a query refused
-  on the fixture fails the step by name), docs, reuse, licence files, publication, messages.
-  **11 properties, 705 tests, 0 failures** on the release tree (692 at 0.6.0, 604 at 0.5.0;
-  the differences are the slices' own pins — at 0.7.0 the signer seam's census and behaviour
-  tests; before it the floor, the provenance and security-policy pins, the tracer's session
+  on the fixture fails the step by name), docs, reuse, licence files, publication, baseline,
+  messages.
+  **11 properties, 705 tests, 0 failures** on the release tree (705 at 0.7.0 — 0.8.0 adds no
+  test, its instruments are probed by shell — 692 at 0.6.0, 604 at 0.5.0; the differences are
+  the slices' own pins — at 0.7.0 the signer seam's census and behaviour tests; before it the floor, the provenance and security-policy pins, the tracer's session
   suite, the public-API census on the tree and on fixtures, the governance and export-control
   censuses).
 - The README's four-way split is held by census to the modules compiled from `lib/` (the
@@ -63,10 +73,10 @@ happened.
 
 ## What is next
 
-The road, as `UPGRADING.md` states it and the owner locked it: **`0.7.0`** this release, the
-signer seam (`BeamMCP.Signer`, a behaviour added to the public surface and no authority — the
-last intentional addition), **`0.8.0`** a quiet minor in which no public entry moves,
-**`1.0.0`** after it has stood — the README's condition, that
+The road, as `UPGRADING.md` states it and the owner locked it: **`0.7.0`** the signer seam
+(`BeamMCP.Signer`, a behaviour added to the public surface and no authority — the last
+intentional addition), **`0.8.0`** this release, the quiet minor in which no public entry
+moves, **`1.0.0`** after it has stood — the README's condition, that
 the public API and the stated threat model have each survived a full minor release unchanged.
 The federation seam stays held on another board's answer and is not on that road. A compiler-tracer census
 (module-body code run at compile time, which neither the text censuses nor `:xref` see) is
@@ -112,11 +122,11 @@ refused.
 - The `2025-11-25` revision is served on stdio only; over HTTP the conformance row for it is
   0 / 30 by design, and the README says so beside the number.
 
-## The release steps — the runbook (followed for 0.6.0; the same for 0.7.0)
+## The release steps — the runbook (followed for 0.6.0 and 0.7.0; the same for 0.8.0)
 
 1. The release PR merged to `main` by rebase (the ruleset requires two green checks); `main`
    is then the release commit.
-2. The gate on that commit, fifteen `pass`, output recorded by command and exit code (the
+2. The gate on that commit, sixteen `pass`, output recorded by command and exit code (the
    release PR's own gate run is that record).
 3. Tag the release commit **as it sits on `main` after the rebase-merge** (a new SHA; the
    bytes are a function of the tree, measured) locally, signed (`git tag -s vX.Y.Z`); then
diff --git a/README.md b/README.md
index 2ffb78d..0296a53 100644
--- a/README.md
+++ b/README.md
@@ -14,11 +14,11 @@ and refuses one that is not; what a tool *does* is the host's business.
 
 ```elixir
 def deps do
-  [{:beam_mcp, "~> 0.7.0"}]
+  [{:beam_mcp, "~> 0.8.0"}]
 end
 ```
 
-**`~> 0.7.0`, not the more usual `~> 0.7`.** While this package is `0.x` it documents breaks
+**`~> 0.8.0`, not the more usual `~> 0.8`.** While this package is `0.x` it documents breaks
 at the **minor** position, and it has used that position five times: `0.2.0` removed two
 fields from results for legacy-declared requests, `0.3.0` added the HTTP transport and the
 `ttlMs`/`cacheScope` fields `2026-07-28` requires on `tools/list`, `0.4.0` replaced the
@@ -29,8 +29,9 @@ writes and moves `schema_version` to `2` (in the exported bytes), and requires a
 `resources` and `prompts` lists to hold the package's structs (the host contract), and `0.6.0`
 names the canonical envelope's algorithm in its bytes and moves `schema_version` to `3` (in the
 exported bytes), each with a how-to-tell sentence in the changelog (`0.7.0` is an addition at
-the minor, the signer seam, not a break). `~> 0.7` admits `0.8.0`, so it
-would carry you across the next such break on a routine `mix deps.update`; `~> 0.7.0` does not. The tighter form is deliberate and is not an over-pin to be tidied away. What the pin
+the minor, the signer seam, and `0.8.0` a quiet minor in which no public entry moved; neither
+a break). `~> 0.8` would admit a `0.9.0`, so it
+would carry you across the next such break on a routine `mix deps.update`; `~> 0.8.0` does not. The tighter form is deliberate and is not an over-pin to be tidied away. What the pin
 buys is written down: [`docs/api-stability.md`](docs/api-stability.md) says what is public
 (what ex_doc lists, `docs/public-api.txt` line by line), how a deprecation runs (three steps,
 three minors), and what a `0.x` break must say; a census holds the surface to that record.
@@ -528,7 +529,7 @@ has no stdio server mode). `conformance/README.md` has the rest.
 **Reproduce it:** `tools/conformance.sh` — one command, for anyone with Node ≥ 22 and
 `python3`. **The trade, stated:** this package has two dependencies; producing this number
 costs a second toolchain, so the step runs in a CI job of its own (`conformance.yml`, Node 22
-pinned) and never in the local gate, which stays the fifteen steps a contributor with Elixir
+pinned) and never in the local gate, which stays the sixteen steps a contributor with Elixir
 and Erlang runs green with nothing else installed (Dialyzer among them, on the OTP binary — its
 PLT is built once per machine, about a minute, and kept under `_build`). The gate needs hex.pm (two steps resolve
 dependencies); of those, the dependency audit is the one that refuses an answer hex gives
diff --git a/SECURITY.md b/SECURITY.md
index 976f5a7..c1ddd0f 100644
--- a/SECURITY.md
+++ b/SECURITY.md
@@ -104,7 +104,8 @@ entry says so when it does.
 
 | version | supported |
 |---|---|
-| `0.7.x` | yes |
+| `0.8.x` | yes |
+| `0.7.x` | no — superseded |
 | `0.6.x` | no — superseded |
 | `0.5.x` | no — superseded |
 | `0.4.x` | no — superseded |
diff --git a/UPGRADING.md b/UPGRADING.md
index 0a2f9f6..ec79d9c 100644
--- a/UPGRADING.md
+++ b/UPGRADING.md
@@ -13,8 +13,8 @@ says **BREAKING**, which the census requires.
 
 ## The rule for `0.x`
 
-Breaks land at the **minor** position and nowhere else. Pin `~> 0.7.0` (the current minor,
-three numbers), not `~> 0.7`: the tighter pin stops at the next minor, which is where the
+Breaks land at the **minor** position and nowhere else. Pin `~> 0.8.0` (the current minor,
+three numbers), not `~> 0.8`: the tighter pin stops at the next minor, which is where the
 next documented break can be, so a routine `mix deps.update` never carries you across one.
 To move a minor: read the release's rows below and their CHANGELOG entries, apply each "how
 to tell" sentence to your host, then raise the pin.
@@ -41,18 +41,30 @@ one that does adds the separate package `beam_mcp_signer` and passes its module
 `signature/3`. Raise the pin to `~> 0.7.0` when you take it; `~> 0.6.0` stops before it by the
 rule, not because anything moved.
 
+**`0.8.0` has no row either: it is the quiet minor.** No public entry was added, removed,
+renamed, hidden or changed in arity — `docs/public-api.txt` is line for line `0.7.0`'s, and the
+release step wrote nothing into it — and no wire byte or envelope byte moved. What changed is
+instruments (the gate diffs the baseline against `origin/main`; the pull-request summary waits
+for running legs) and pages (the Scorecard's measured figures on the governance page). Raise
+the pin to `~> 0.8.0`. **A host on `beam_mcp_signer` waits for that package's release that
+admits `0.8.0`**: its `0.1.0` requires `beam_mcp ~> 0.7.0` (three numbers, by the same rule as
+this page's pin), so `{:beam_mcp, "~> 0.8.0"}` beside `{:beam_mcp_signer, "~> 0.1.0"}` does not
+resolve until it does. That is the cost of the three-number pin, paid once per minor, on the
+signer's side.
+
 ## The road to `1.0.0`, in order
 
 Stated here so nobody infers it from a plan's label or a folder's name:
 
 1. **`0.6.0`** — everything since `0.5.0`, the assessability snapshot. One documented break at
    the minor (the exported bytes).
-2. **`0.7.0`** — this release: the signer seam, `BeamMCP.Signer` (a behaviour added to the
-   public surface; no authority passes through it), the last intentional addition before
-   `1.0.0`. An addition, not a break; `~> 0.6.0` stops at it all the same, by the rule.
-3. **`0.8.0`** — a quiet minor: documentation, the Scorecard's rows, instrument leftovers. No
-   public entry added, removed, renamed or hidden — the "full minor release unchanged" the
-   README's `1.0.0` condition requires, measured by `docs/public-api.txt` not moving.
+2. **`0.7.0`** — the signer seam, `BeamMCP.Signer` (a behaviour added to the public
+   surface; no authority passes through it), the last intentional addition before `1.0.0`.
+   An addition, not a break; `~> 0.6.0` stops at it all the same, by the rule.
+3. **`0.8.0`** — this release: the quiet minor. Documentation, the Scorecard's rows,
+   instrument leftovers. No public entry added, removed, renamed or hidden — the "full minor
+   release unchanged" the README's `1.0.0` condition requires, measured by
+   `docs/public-api.txt` not moving (the gate now diffs it against `origin/main` on every run).
 4. **`1.0.0`** — after `0.8.0` has stood: the surface frozen as `docs/api-stability.md` says.
 
 ## What `1.0` will ask
diff --git a/docs/api-stability.md b/docs/api-stability.md
index 395e778..66ac40a 100644
--- a/docs/api-stability.md
+++ b/docs/api-stability.md
@@ -44,7 +44,7 @@ removes a callable arity and is a change to the entry — then markers:
 
 `R` is a release number, or the word `Unreleased` while the change waits in the CHANGELOG's
 Unreleased section; the release that ships it writes its number in
-(`MIX_ENV=test mix run -e 'BeamMCP.PublicAPI.release_markers!("0.7.0")'`, one step of cutting
+(`MIX_ENV=test mix run -e 'BeamMCP.PublicAPI.release_markers!("0.8.0")'`, one step of cutting
 a release), and the census refuses a leftover `Unreleased` once that section is empty. That
 step is the one place that knows the release's number, so it holds the three rules the census
 cannot: with a `removed_in=Unreleased` line in the file it refuses a patch number, and on
@@ -83,7 +83,7 @@ From the release this page ships in, a break is a CHANGELOG heading carrying the
 census holds the Unreleased section to that — and `UPGRADING.md` lists them. Earlier releases
 documented their breaks under headings of their own wording (`0.2.0`'s "two fields are
 REMOVED", `0.4.0`'s "BREAKING, and it breaks a host contract"); `UPGRADING.md` names each with
-its CHANGELOG heading. That is why the README recommends the current minor at three numbers (`~> 0.7.0`) rather than two (`~> 0.7`): the
+its CHANGELOG heading. That is why the README recommends the current minor at three numbers (`~> 0.8.0`) rather than two (`~> 0.8`): the
 tighter pin stops at the next minor, which is where the next documented break can be. A patch
 release carries no break to the public surface, the wire, the exported bytes or the host
 contract — a rule the census cannot check (it does not know which number the next release
diff --git a/docs/connectome-canonical.md b/docs/connectome-canonical.md
index 5d529cd..f6b3c53 100644
--- a/docs/connectome-canonical.md
+++ b/docs/connectome-canonical.md
@@ -140,7 +140,7 @@ byte but that value.
 the envelope. **`1`** (0.4.0): the sign values were `allow`, `deny`, `hold`, `unknown`.
 **`2`** (0.5.0): `unset` — no sign was supplied to the package that wrote the bytes — and
 `ungoverned`, a consumer's affirmative "no rule of my policy applies", replace `unknown`;
-nothing else moved. **`3`** (`0.6.0`, which carries this note): the envelope
+nothing else moved. **`3`** (`0.6.0`): the envelope
 names its algorithm as a fourth top-level member, `"algorithm"`, between the version and the
 nodes (rule 1), and the hash is that digest over the bytes (rule 9). The vocabulary is `2`'s,
 unchanged. **At `1`, `unknown` covers both of `2`'s new signs**: a 0.4.0 consumer that looked
diff --git a/docs/governance.md b/docs/governance.md
index 48ba67e..c9c25cf 100644
--- a/docs/governance.md
+++ b/docs/governance.md
@@ -27,10 +27,11 @@ Every change, the maintainer's included, goes the same way:
    force-push, no deletion, linear history, and two required status checks — the DCO sign-off
    and the quality gate. A pull request is the only way onto `main`, and it is rebased, never
    merged, so the history is a line.
-2. **The gate.** `tools/gate.sh` runs the same fifteen steps locally and in CI — format,
+2. **The gate.** `tools/gate.sh` runs the same sixteen steps locally and in CI — format,
    compile with warnings as errors, Dialyzer, the instruments' parse, the suite, Credo, the
    properties, the optional-dependency probe, the dependency audit, the benchmarks, the docs,
-   REUSE, the licence files, the publication census and the commit-message terms — on three
+   REUSE, the licence files, the publication census, the public-API baseline against
+   `origin/main` and the commit-message terms — on three
    OTP/Elixir pairs (the floor, the pinned line, the newest). There is no baseline to hold: a
    non-zero count is a failure.
 3. **Review by tier, then the merge word.** `CONVENTIONS.md` states the tier rule: a contract
@@ -54,24 +55,30 @@ its result. It is a measurement of the tree and the platform, and this page says
 checks this project keeps on purpose, which it cannot, and which it has decided against — so a
 reader does not have to guess whether a low mark is neglect or a decision.
 
-| check | this repository | why |
-| --- | --- | --- |
-| Pinned-Dependencies | every workflow action pinned by commit SHA with its version beside it; Mix dependencies locked in `mix.lock` | a tag can be moved, a SHA cannot; a test holds the pins on every push |
-| Token-Permissions | every workflow declares top-level `permissions:` with no write; the two jobs that write (provenance's attestation, the Scorecard's SARIF upload) hold it at the job, and no other job does | least privilege, held by the same test — the top level, and which jobs may write |
-| Branch-Protection | the ruleset above: no direct push, linear history, required checks | kept; **no required reviewer** — see Code-Review |
-| Code-Review | pull requests, every one since the ruleset (2026-09-06; the eight bootstrap commits before it were pushed directly); the reviewer of record is the maintainer, after the lanes | one maintainer cannot approve their own pull request under GitHub's rules, and there is no second one; the Scorecard will score this low and that is the true state, not an omission |
-| Security-Policy | `SECURITY.md` | the intake, the rubric and the CVE path |
-| License | `LICENSE`, `NOTICE`, `LICENSES/`, REUSE headers on every file, held by the gate | |
-| Dependency-Update-Tool | Dependabot, weekly, Mix and GitHub Actions | |
-| Vulnerabilities | `mix hex.audit` in the gate, OSV-fed, on every push | |
-| CI-Tests | the gate on three OTP/Elixir pairs | |
-| Maintained | commits and releases as the CHANGELOG shows | |
-| Signed-Releases | releases are Hex releases: from `0.6.0` on, the tarball is built by CI on the tag and attested (`docs/provenance.md`; `0.5.0` and earlier carry none); there are no GitHub Releases with assets for this check to read | the attestation binds to the checksum hex.pm shows, which is where consumers fetch from; a GitHub Release would be a copy |
-| SAST | Dialyzer and Credo in the gate; no CodeQL | the gate's analysers are what the language has; a CodeQL workflow is a separate decision and is not taken here |
-| Fuzzing | eleven property-based tests in the gate; no OSS-Fuzz | property tests are the fuzzing the suite does; OSS-Fuzz integration is not taken |
-| Dangerous-Workflow | the pull-request body is read through an environment variable, never interpolated into a script | |
-| Binary-Artifacts | none in the tree | |
-| Contributors | one organization | |
+The figures are the Scorecard's own, read from `api.scorecard.dev` on 2026-09-19 (aggregate
+**7**), and say what the check measured beside what the tree holds; a figure is quoted, not
+promised, and moves when the Scorecard next runs.
+
+| check | this repository | why | measured 2026-09-19 |
+| --- | --- | --- | --- |
+| Pinned-Dependencies | every workflow action pinned by commit SHA with its version beside it; Mix dependencies locked in `mix.lock` | a tag can be moved, a SHA cannot; a test holds the pins on every push | 10 — "all dependencies are pinned" |
+| Token-Permissions | every workflow declares top-level `permissions:` with no write; the two jobs that write (provenance's attestation, the Scorecard's SARIF upload) hold it at the job, and no other job does | least privilege, held by the same test — the top level, and which jobs may write | 10 |
+| Branch-Protection | the ruleset above: no direct push, linear history, required checks | kept; **no required reviewer** — see Code-Review | 4 — "not maximal": the missing tiers are the required reviewer and a second approver, which one maintainer cannot supply |
+| Code-Review | pull requests, every one since the ruleset (2026-09-06; the eight bootstrap commits before it were pushed directly); the reviewer of record is the maintainer, after the lanes | one maintainer cannot approve their own pull request under GitHub's rules, and there is no second one; the Scorecard scores this low and that is the true state, not an omission | 0 — "0/7 approved changesets" |
+| Security-Policy | `SECURITY.md` | the intake, the rubric and the CVE path | 10 |
+| License | `LICENSE`, `NOTICE`, `LICENSES/`, REUSE headers on every file, held by the gate | | 10 |
+| Dependency-Update-Tool | Dependabot, weekly, Mix and GitHub Actions | | 10 |
+| Vulnerabilities | `mix hex.audit` in the gate, OSV-fed, on every push | | 10 |
+| CI-Tests | the gate on three OTP/Elixir pairs | | 10 — "7 out of 7 merged PRs checked" |
+| Maintained | commits and releases as the CHANGELOG shows | the check scores **0 for any repository younger than 90 days**, whatever its activity; this one was created 2026-09-06, so the figure is the rule's until 2026-12-05 and says nothing about the tree | 0 — "created within the last 90 days" |
+| Signed-Releases | releases are Hex releases: from `0.6.0` on, the tarball is built by CI on the tag and attested (`docs/provenance.md`; `0.5.0` and earlier carry none); there are no GitHub Releases with assets for this check to read | the attestation binds to the checksum hex.pm shows, which is where consumers fetch from; a GitHub Release would be a copy | −1 — "no releases found": the check reads GitHub Releases only |
+| Packaging | the package is published to hex.pm by the maintainer from the canonical tarball, on a signed tag; no GitHub Actions publishing workflow | the publish step holds a Hex API key, which stays on the maintainer's seat rather than in a workflow secret — a decision, recorded here; the provenance workflow attests the bytes but does not publish them | −1 — "packaging workflow not detected": the check reads a publishing workflow only |
+| SAST | Dialyzer and Credo in the gate; no CodeQL | the gate's analysers are what the language has; a CodeQL workflow is a separate decision and is not taken here | 0 — the check recognises neither Dialyzer nor Credo |
+| Fuzzing | eleven property-based tests in the gate; no OSS-Fuzz | property tests are the fuzzing the suite does; OSS-Fuzz integration is not taken | 10 — "project is fuzzed": the check reads the property tests as fuzzing |
+| CII-Best-Practices | no OpenSSF Best Practices badge | registration is the maintainer's, and open | 0 |
+| Dangerous-Workflow | the pull-request body is read through an environment variable, never interpolated into a script | | 10 |
+| Binary-Artifacts | none in the tree | | 10 |
+| Contributors | one organization owns the repository; the NOTICE names the owner and the builder | the check counts the companies commit authors declare, and read two | 6 — "2 contributing companies or organizations" |
 
 ## What is deliberately not done
 
diff --git a/docs/provenance.md b/docs/provenance.md
index 4ae1a5a..00e76a4 100644
--- a/docs/provenance.md
+++ b/docs/provenance.md
@@ -32,7 +32,7 @@ So the release tarball is built by one script on every seat — CI, the publishe
 reproducing it:
 
 ```sh
-tools/release_tarball.sh v0.6.0 beam_mcp-0.6.0.tar
+tools/release_tarball.sh v0.7.0 beam_mcp-0.7.0.tar
 ```
 
 It `git archive`s the ref with `tar.umask=022` (every file `644` whatever the machine's umask,
@@ -53,7 +53,7 @@ failure, not as a pass.
 ## Verify a published tarball
 
 ```sh
-v=0.6.0
+v=0.7.0
 curl -fsSLO "https://repo.hex.pm/tarballs/beam_mcp-${v}.tar"
 gh attestation verify "beam_mcp-${v}.tar" --repo ScriptKittyOS/beam_mcp
 ```
diff --git a/mix.exs b/mix.exs
index a815ddf..d043a0b 100644
--- a/mix.exs
+++ b/mix.exs
@@ -4,7 +4,7 @@
 defmodule BeamMCP.MixProject do
   use Mix.Project
 
-  @version "0.7.0"
+  @version "0.8.0"
   @source_url "https://github.com/ScriptKittyOS/beam_mcp"
 
   # The oldest OTP this project supports. Mix has an `elixir:` key but none for OTP, so the
diff --git a/test/beam_mcp/boundary/no_signature_test.exs b/test/beam_mcp/boundary/no_signature_test.exs
index 0a82655..15d7e3d 100644
--- a/test/beam_mcp/boundary/no_signature_test.exs
+++ b/test/beam_mcp/boundary/no_signature_test.exs
@@ -99,7 +99,7 @@ defmodule BeamMCP.Boundary.NoSignatureTest do
     {:ok, callbacks} = Code.Typespec.fetch_callbacks(BeamMCP.Signer)
     assert [{{:sign, 2}, [spec]}] = callbacks
 
-    {:type, _, :fun, [{:type, _, :product, args}, ret]} = spec
+    {:type, _, :fun, [{:type, _, :product, args}, _ret]} = spec
 
     names =
       Enum.map(args, fn
diff --git a/test/beam_mcp/readme_claims_test.exs b/test/beam_mcp/readme_claims_test.exs
index 5cf3207..7882bdd 100644
--- a/test/beam_mcp/readme_claims_test.exs
+++ b/test/beam_mcp/readme_claims_test.exs
@@ -137,7 +137,7 @@ defmodule BeamMCP.ReadmeClaimsTest do
 
   describe "the dependency requirement the README hands a consumer" do
     test "it does not span the wire break this release documents" do
-      requirement = "~> 0.7.0"
+      requirement = "~> 0.8.0"
       claims("{:beam_mcp, \"#{requirement}\"}")
 
       version = Mix.Project.config()[:version]
@@ -145,11 +145,14 @@ defmodule BeamMCP.ReadmeClaimsTest do
       assert Version.match?(version, requirement),
              "the README's requirement must admit the version being shipped"
 
-      refute Version.match?("0.6.0", requirement),
-             "0.6.0 is the minor before this one; 0.7.0 adds the signer seam and breaks " <>
-               "nothing, and the pin still stops at the current minor by the 0.x rule -- " <>
+      refute Version.match?("0.7.0", requirement),
+             "0.7.0 is the minor before this one; 0.8.0 is a quiet minor that moves no public " <>
+               "entry, and the pin still stops at the current minor by the 0.x rule -- " <>
                "the next minor is where the next documented break can be."
 
+      refute Version.match?("0.6.0", requirement),
+             "0.6.0 is two minors back; 0.7.0 added the signer seam and broke nothing."
+
       refute Version.match?("0.5.0", requirement),
              "0.5.0 is on the far side of the break 0.6.0 documented in the exported " <>
                "bytes: a verifier that pins schema_version 2, or hashes without reading the " <>
diff --git a/test/fixtures/wire/pre-017.json b/test/fixtures/wire/pre-017.json
index 2ba254d..fd03254 100644
--- a/test/fixtures/wire/pre-017.json
+++ b/test/fixtures/wire/pre-017.json
@@ -7,7 +7,7 @@
         "_meta": {
           "io.modelcontextprotocol/serverInfo": {
             "name": "fixture",
-            "version": "0.7.0"
+            "version": "0.8.0"
           }
         },
         "cacheScope": "private",
@@ -27,7 +27,7 @@
         "_meta": {
           "io.modelcontextprotocol/serverInfo": {
             "name": "fixture",
-            "version": "0.7.0"
+            "version": "0.8.0"
           }
         },
         "cacheScope": "private",
@@ -50,7 +50,7 @@
         "_meta": {
           "io.modelcontextprotocol/serverInfo": {
             "name": "fixture",
-            "version": "0.7.0"
+            "version": "0.8.0"
           }
         },
         "cacheScope": "private",
@@ -71,7 +71,7 @@
         "_meta": {
           "io.modelcontextprotocol/serverInfo": {
             "name": "fixture",
-            "version": "0.7.0"
+            "version": "0.8.0"
           }
         },
         "cacheScope": "private",
@@ -95,7 +95,7 @@
         "_meta": {
           "io.modelcontextprotocol/serverInfo": {
             "name": "fixture",
-            "version": "0.7.0"
+            "version": "0.8.0"
           }
         },
         "cacheScope": "private",
@@ -116,7 +116,7 @@
         "_meta": {
           "io.modelcontextprotocol/serverInfo": {
             "name": "fixture",
-            "version": "0.7.0"
+            "version": "0.8.0"
           }
         },
         "cacheScope": "private",
@@ -140,7 +140,7 @@
         "_meta": {
           "io.modelcontextprotocol/serverInfo": {
             "name": "fixture",
-            "version": "0.7.0"
+            "version": "0.8.0"
           }
         },
         "cacheScope": "private",
@@ -171,7 +171,7 @@
         "_meta": {
           "io.modelcontextprotocol/serverInfo": {
             "name": "fixture",
-            "version": "0.7.0"
+            "version": "0.8.0"
           }
         },
         "cacheScope": "private",
@@ -204,7 +204,7 @@
         "_meta": {
           "io.modelcontextprotocol/serverInfo": {
             "name": "fixture",
-            "version": "0.7.0"
+            "version": "0.8.0"
           }
         },
         "cacheScope": "private",
@@ -251,7 +251,7 @@
         "_meta": {
           "io.modelcontextprotocol/serverInfo": {
             "name": "fixture",
-            "version": "0.7.0"
+            "version": "0.8.0"
           }
         },
         "cacheScope": "private",
diff --git a/tools/baseline_diff.sh b/tools/baseline_diff.sh
new file mode 100755
index 0000000..ecdc5df
--- /dev/null
+++ b/tools/baseline_diff.sh
@@ -0,0 +1,60 @@
+#!/usr/bin/env bash
+# SPDX-FileCopyrightText: 2026 Sudo Apt Holdings LLC
+# SPDX-License-Identifier: Apache-2.0
+#
+# The two hand edits to docs/public-api.txt the census cannot see, because it reads the tree
+# alone (G-076, measured by a review lane: the whole gate green on each):
+#
+#   1. a baseline line deleted outright -- the writer never deletes a line, so a non-comment
+#      line present at the base and absent at the head, whose entry (the text before its
+#      markers) does not reappear, is a hand deletion of a public entry;
+#   2. a marker back-dated -- a `since=`, `deprecated_since=` or `removed_in=` value ARRIVING
+#      in this change as a release number that is not above the base CHANGELOG's highest
+#      `## [N]` heading: one it already lists, or a phantom between two releases (a lane
+#      measured `removed_in=0.6.5` passing the first draft) -- a change dressed as one that
+#      already shipped. The release commit writes its own, higher number, and passes.
+#
+# Pure: three files in, a verdict out, so a probe can plant on copies. The gate feeds it the
+# base from `git show origin/main:...` and the head from the tree.
+#
+#     tools/baseline_diff.sh <base public-api.txt> <head public-api.txt> <base CHANGELOG.md>
+# exit 0: neither edit; 1: one or both (each named); 2: usage.
+set -uo pipefail
+[ $# -eq 3 ] || { echo "usage: tools/baseline_diff.sh <base-baseline> <head-baseline> <base-changelog>" >&2; exit 2; }
+base=$1 head=$2 changelog=$3
+for f in "$base" "$head" "$changelog"; do [ -r "$f" ] || { echo "unreadable: $f" >&2; exit 2; }; done
+
+entries() { grep -v '^#' "$1" | grep -v '^[[:space:]]*$' | sed -E 's/( (since|deprecated_since|removed_in)=[^ ]+)+$//' | sort -u; }
+# Per entry, so a NEW line arriving with a number an old line already carries is still an arrival.
+markers() { grep -v '^#' "$1" | awk '{for (i=2;i<=NF;i++) if ($i ~ /^(since|deprecated_since|removed_in)=/) print $1" "$2" "$3" "$i}' | sort -u; }
+
+fail=0
+# 1. Deleted entries.
+deleted=$(comm -23 <(entries "$base") <(entries "$head"))
+if [ -n "$deleted" ]; then
+  echo "FAIL: public entries deleted from the baseline by hand (the writer never deletes a line; a removal is marked removed_in= and stays):"
+  printf '%s\n' "$deleted" | sed 's/^/  /'
+  fail=1
+fi
+# 2. Markers arriving with an already-released number.
+highest=$(grep -oE '^## \[[0-9]+\.[0-9]+\.[0-9]+\]' "$changelog" | tr -d '#[] ' | sort -V | tail -1)
+arriving=$(comm -13 <(markers "$base") <(markers "$head"))
+backdated=""
+while IFS= read -r m; do
+  [ -n "$m" ] || continue
+  v=${m##*=}
+  [ "$v" = "Unreleased" ] && continue
+  # Not above the highest released heading: equal to it, below it, or between two releases.
+  if [ -z "$highest" ] || [ "$v" = "$highest" ] || [ "$(printf '%s\n%s\n' "$highest" "$v" | sort -V | tail -1)" != "$v" ]; then
+    backdated="${backdated}${m}"$'\n'
+  fi
+done <<< "$arriving"
+if [ -n "$backdated" ]; then
+  echo "FAIL: markers arriving with a release number not above the base CHANGELOG's highest heading ($highest) -- a change dressed as one that shipped:"
+  printf '%s' "$backdated" | sort -u | sed 's/^/  /'
+  fail=1
+fi
+if [ "$fail" -eq 0 ]; then
+  echo "pass ($(entries "$head" | wc -l | tr -d ' ') entries at the head, none deleted; $(printf '%s\n' "$arriving" | grep -c . || true) marker(s) arriving, none back-dated)"
+fi
+exit "$fail"
diff --git a/tools/ci_legs_verdict.sh b/tools/ci_legs_verdict.sh
new file mode 100755
index 0000000..cb3623a
--- /dev/null
+++ b/tools/ci_legs_verdict.sh
@@ -0,0 +1,56 @@
+#!/usr/bin/env bash
+# SPDX-FileCopyrightText: 2026 Sudo Apt Holdings LLC
+# SPDX-License-Identifier: Apache-2.0
+#
+# The edited-summary read, out of the workflow and into a script a probe can run with a fake
+# `gh`. On a pull-request body edit the legs do not run (the tree did not move) but the
+# summary job carries the REQUIRED context, so it reads the legs' latest check runs for the
+# head SHA and requires all three to be success. G-079 measured the hole: a body edit while the
+# same SHA's legs were still running read their conclusions as null and failed "by design",
+# and the ruleset then blocked the pull request on that red although the legs finished green a
+# minute later -- a failed run of the required name blocks even beside a later green one. So
+# this waits: it polls until every leg check-run for the SHA is `completed`, bounded, and only
+# then judges. A tree whose legs never finish is a FAIL at the bound, named as such.
+#
+#     GH_TOKEN=... tools/ci_legs_verdict.sh <owner/repo> <head sha>
+# env: LEGS_MIN (default 3), POLL_SECONDS (default 30), POLL_LIMIT (default 40 -- 20 minutes).
+# exit 0: every leg completed and success; 1: a leg not success, too few legs, or the bound hit.
+set -uo pipefail
+[ $# -eq 2 ] || { echo "usage: tools/ci_legs_verdict.sh <owner/repo> <head-sha>" >&2; exit 2; }
+repo=$1 sha=$2
+min=${LEGS_MIN:-3} every=${POLL_SECONDS:-30} limit=${POLL_LIMIT:-40}
+# The expanded leg names only: the skipped matrix job of the edited run registers one
+# unexpanded name ("Gate - leg (OTP ${{ matrix.otp }} ...", skipped), which is not a verdict.
+read_legs() {
+  gh api "repos/${repo}/commits/${sha}/check-runs?filter=latest&per_page=100" \
+    --jq '.check_runs[] | select(.name | test("^Gate - leg \\(OTP [0-9]+, Elixir [0-9.]+, [a-z]+\\)$")) | "\(.name): \(.status)/\(.conclusion)"'
+}
+i=0
+while :; do
+  # A failed read is a poll spent, not a verdict: a transient API error during the wait must
+  # not fail the required check (a red run of this name blocks beside a later green).
+  if legs=$(read_legs); then read_ok=1; else read_ok=0; legs=""; fi
+  if [ -n "$legs" ]; then
+    n=$(printf '%s\n' "$legs" | grep -c . || true)
+    pending=$(printf '%s\n' "$legs" | grep -vc ': completed/' || true)
+  else
+    n=0; pending=0
+  fi
+  if [ "$read_ok" -eq 1 ] && [ "$n" -ge "$min" ] && [ "$pending" -eq 0 ]; then break; fi
+  i=$((i + 1))
+  if [ "$i" -gt "$limit" ]; then
+    printf '%s\n' "$legs"
+    echo "FAIL: after $((limit * every)) s, $n leg verdict(s) exist for ${sha} and $pending are not completed (last read $([ "$read_ok" -eq 1 ] && echo ok || echo failed)); the tree is unmeasured"
+    exit 1
+  fi
+  if [ "$read_ok" -eq 1 ]; then
+    echo "waiting: $n leg check-run(s), $pending not completed (poll $i of $limit, every ${every}s)"
+  else
+    echo "waiting: the checks API could not be read (poll $i of $limit, every ${every}s)"
+  fi
+  sleep "$every"
+done
+printf '%s\n' "$legs"
+bad=$(printf '%s\n' "$legs" | grep -vc ': completed/success$' || true)
+[ "$bad" -eq 0 ] || { echo "FAIL: a leg's latest verdict for ${sha} is not success"; exit 1; }
+echo "pass: $n leg(s) completed, all success, for ${sha}"
diff --git a/tools/gate.sh b/tools/gate.sh
index 7308928..7dfb3a9 100755
--- a/tools/gate.sh
+++ b/tools/gate.sh
@@ -469,6 +469,26 @@ else
   fail=1
 fi
 
+# The public-API baseline against origin/main: no entry's line deleted by hand, no marker
+# arriving with a number the CHANGELOG already lists (G-076: the census reads the tree alone,
+# so both edits passed it green; tools/baseline_diff.sh is the pure check, tools/probe_baseline_diff.sh
+# its probe). Where origin/main does not resolve the line says so and is not evidence.
+if git cat-file -e origin/main:docs/public-api.txt 2>/dev/null; then
+  bl_dir=$(mktemp -d "${TMPDIR:-/tmp}/beam_mcp-gate-baseline.XXXXXX")
+  git show origin/main:docs/public-api.txt > "$bl_dir/base.txt"
+  git show origin/main:CHANGELOG.md > "$bl_dir/changelog.md"
+  if bl_out=$(bash tools/baseline_diff.sh "$bl_dir/base.txt" docs/public-api.txt "$bl_dir/changelog.md" 2>&1); then
+    note "baseline" "$bl_out"
+  else
+    note "baseline" "FAIL (docs/public-api.txt against origin/main)"
+    printf '%s\n' "$bl_out" | sed 's/^/      /'
+    fail=1
+  fi
+  rm -rf "$bl_dir"
+else
+  note "baseline" "pass (origin/main does not resolve here; the baseline was not diffed -- this line is not evidence)"
+fi
+
 # Commit messages: the branch's own commits carry no attribution trailer, no session link, no
 # board identifier and no consumer name.
 #
diff --git a/tools/probe_baseline_diff.sh b/tools/probe_baseline_diff.sh
new file mode 100755
index 0000000..b2a7482
--- /dev/null
+++ b/tools/probe_baseline_diff.sh
@@ -0,0 +1,42 @@
+#!/usr/bin/env bash
+# SPDX-FileCopyrightText: 2026 Sudo Apt Holdings LLC
+# SPDX-License-Identifier: Apache-2.0
+#
+# Probe: does tools/baseline_diff.sh go red on the two hand edits G-076 names -- a public
+# entry's line deleted outright, and a marker arriving with a number the CHANGELOG already
+# lists -- and stay green on the edits the writer and the release step make? Plants are on
+# copies in a temp dir; the tree is not touched.
+#
+#     tools/probe_baseline_diff.sh
+set -uo pipefail
+here=$(cd "$(dirname "$0")/.." && pwd)
+work=$(mktemp -d "${TMPDIR:-/tmp}/beam_mcp-probe-baseline.XXXXXX"); trap 'rm -rf "$work"' EXIT
+cp "$here/docs/public-api.txt" "$work/base.txt"
+cp "$here/CHANGELOG.md" "$work/changelog.md"
+released=$(grep -oE '^## \[[0-9]+\.[0-9]+\.[0-9]+\]' "$work/changelog.md" | head -1 | tr -d '#[] ')
+[ -n "$released" ] || { echo "PROBE NOT MEASURED: no released heading in CHANGELOG.md"; exit 2; }
+entry=$(grep -v '^#' "$work/base.txt" | grep -m1 -v 'since=' || true)
+[ -n "$entry" ] || { echo "PROBE NOT MEASURED: no unmarked entry in the baseline to plant on"; exit 2; }
+bad=0
+run() { # run <case> <expected exit> <head file>
+  out=$(bash "$here/tools/baseline_diff.sh" "$work/base.txt" "$3" "$work/changelog.md" 2>&1); rc=$?
+  printf '  %-58s exit %s (want %s): %s\n' "$1" "$rc" "$2" "$(printf '%s\n' "$out" | head -1 | cut -c1-70)"
+  [ "$rc" -eq "$2" ] || bad=1
+}
+cp "$work/base.txt" "$work/same.txt";                                      run "P0 unchanged" 0 "$work/same.txt"
+grep -vxF "$entry" "$work/base.txt" > "$work/deleted.txt";                run "P1 an entry's line deleted outright" 1 "$work/deleted.txt"
+{ cat "$work/base.txt"; echo "$entry"; } | awk -v e="$entry" '$0==e && !s {print e" removed_in=Unreleased"; s=1; next} $0!=e' > "$work/marked.txt"
+                                                                          run "P2 the same entry marked removed_in=Unreleased" 0 "$work/marked.txt"
+sed "s|^$entry\$|$entry removed_in=$released|" "$work/base.txt" > "$work/backdated.txt"
+                                                                          run "P3 removed_in= with the last release's number" 1 "$work/backdated.txt"
+sed "s|^$entry\$|$entry deprecated_since=$released|" "$work/base.txt" > "$work/backdated2.txt"
+                                                                          run "P4 deprecated_since= with the last release's number" 1 "$work/backdated2.txt"
+sed "s|^$entry\$|$entry since=9.9.9|" "$work/base.txt" > "$work/release.txt"
+                                                                          run "P5 a marker with a number not yet released (the release step)" 0 "$work/release.txt"
+{ cat "$work/base.txt"; echo "BeamMCP.Probe function added/1 since=Unreleased"; } > "$work/added.txt"
+                                                                          run "P6 a new entry marked Unreleased (the writer)" 0 "$work/added.txt"
+{ cat "$work/base.txt"; echo "BeamMCP.Probe function added/1 since=$released"; } > "$work/added-backdated.txt"
+                                                                          run "P7 a new entry marked since= the last release (back-dated addition)" 1 "$work/added-backdated.txt"
+sed "s|^$entry\$|$entry removed_in=0.0.1|" "$work/base.txt" > "$work/phantom.txt"
+                                                                          run "P8 removed_in= with a phantom number below the last release" 1 "$work/phantom.txt"
+if [ "$bad" -eq 0 ]; then echo "PROBE OK: the two hand edits are red; the writer's and the release step's edits are green"; else echo "PROBE FAILED"; exit 1; fi
diff --git a/tools/probe_ci_legs_verdict.sh b/tools/probe_ci_legs_verdict.sh
new file mode 100755
index 0000000..37c03f1
--- /dev/null
+++ b/tools/probe_ci_legs_verdict.sh
@@ -0,0 +1,74 @@
+#!/usr/bin/env bash
+# SPDX-FileCopyrightText: 2026 Sudo Apt Holdings LLC
+# SPDX-License-Identifier: Apache-2.0
+#
+# Probe: does tools/ci_legs_verdict.sh wait for legs still running and then judge them, fail a
+# red leg, fail too few legs, and fail at its bound -- with a fake `gh` on PATH that answers a
+# scripted sequence of check-run pages (G-079's in_progress-then-success case first)? Nothing
+# reaches GitHub; the probe is offline.
+#
+#     tools/probe_ci_legs_verdict.sh
+set -uo pipefail
+here=$(cd "$(dirname "$0")/.." && pwd)
+work=$(mktemp -d "${TMPDIR:-/tmp}/beam_mcp-probe-legs.XXXXXX"); trap 'rm -rf "$work"' EXIT
+mkdir -p "$work/bin"
+# The fake gh: prints the JSON page named by a counter file, advancing one per call, holding
+# the last. It ignores --jq and applies the real jq filter itself, so the script's filter is
+# the one exercised.
+cat > "$work/bin/gh" <<'GH'
+#!/usr/bin/env bash
+seq_dir=$PROBE_SEQ; n=$(cat "$seq_dir/n"); pages=$(ls "$seq_dir"/page.* | wc -l)
+[ "$n" -lt "$pages" ] && echo $((n + 1)) > "$seq_dir/n"
+page="$seq_dir/page.$n"; [ -f "$page" ] || page="$seq_dir/page.$((pages - 1))"
+filter=""; while [ $# -gt 0 ]; do [ "$1" = "--jq" ] && filter=$2; shift; done
+grep -q '"__fail__"' "$page" && { echo "gh: HTTP 502 (fake)" >&2; exit 1; }
+jq -r "$filter" < "$page"
+GH
+chmod +x "$work/bin/gh"
+command -v jq >/dev/null || { echo "PROBE NOT MEASURED: jq is not installed"; exit 2; }
+leg() { printf '{"name":"Gate - leg (OTP %s, Elixir %s, %s)","status":"%s","conclusion":%s}' "$1" "$2" "$3" "$4" "$5"; }
+page() { printf '{"check_runs":[%s]}' "$(IFS=,; echo "$*")"; }
+skipped='{"name":"Gate - leg (OTP ${{ matrix.otp }}, Elixir ${{ matrix.elixir }}, ${{ matrix.leg }})","status":"completed","conclusion":"skipped"}'
+bad=0
+run() { # run <case> <want> <seqdir>
+  echo 0 > "$3/n"
+  out=$(PATH="$work/bin:$PATH" PROBE_SEQ=$3 POLL_SECONDS=0 POLL_LIMIT=3 bash "$here/tools/ci_legs_verdict.sh" o/r deadbeef 2>&1); rc=$?
+  printf '  %-52s exit %s (want %s); pages advanced %s; last: %s\n' "$1" "$rc" "$2" "$(cat "$3/n")" "$(printf '%s\n' "$out" | tail -1 | cut -c1-60)"
+  [ "$rc" -eq "$2" ] || bad=1
+}
+# S1: G-079 -- two polls in progress (conclusion null), then all success.
+d=$work/s1; mkdir -p "$d"
+page "$(leg 27 1.17 floor in_progress null)" "$(leg 28 1.18 pinned in_progress null)" "$(leg 29 1.20 head completed '"success"')" "$skipped" > "$d/page.0"
+page "$(leg 27 1.17 floor completed '"success"')" "$(leg 28 1.18 pinned in_progress null)" "$(leg 29 1.20 head completed '"success"')" "$skipped" > "$d/page.1"
+page "$(leg 27 1.17 floor completed '"success"')" "$(leg 28 1.18 pinned completed '"success"')" "$(leg 29 1.20 head completed '"success"')" "$skipped" > "$d/page.2"
+run "S1 in_progress twice, then three successes" 0 "$d"
+# S2: a leg red once complete.
+d=$work/s2; mkdir -p "$d"
+page "$(leg 27 1.17 floor in_progress null)" "$(leg 28 1.18 pinned completed '"success"')" "$(leg 29 1.20 head completed '"success"')" > "$d/page.0"
+page "$(leg 27 1.17 floor completed '"failure"')" "$(leg 28 1.18 pinned completed '"success"')" "$(leg 29 1.20 head completed '"success"')" > "$d/page.1"
+run "S2 a leg completes as failure" 1 "$d"
+# S3: a leg never completes -- the bound.
+d=$work/s3; mkdir -p "$d"
+page "$(leg 27 1.17 floor in_progress null)" "$(leg 28 1.18 pinned completed '"success"')" "$(leg 29 1.20 head completed '"success"')" > "$d/page.0"
+run "S3 a leg never completes (bound of 3 polls)" 1 "$d"
+# S4: only two legs exist, both green -- the tree is unmeasured on one pair.
+d=$work/s4; mkdir -p "$d"
+page "$(leg 28 1.18 pinned completed '"success"')" "$(leg 29 1.20 head completed '"success"')" "$skipped" > "$d/page.0"
+run "S4 two legs only, the skipped unexpanded name beside" 1 "$d"
+# S5: all three complete and green on the first read -- no wait.
+d=$work/s5; mkdir -p "$d"
+cp "$work/s1/page.2" "$d/page.0"
+run "S5 three successes on the first read" 0 "$d"
+# S6: the API fails once, then three successes -- a read spent, not a verdict.
+d=$work/s6; mkdir -p "$d"
+echo '{"__fail__":true}' > "$d/page.0"; cp "$work/s1/page.2" "$d/page.1"
+run "S6 the API fails once, then three successes" 0 "$d"
+# S7: the API fails every time -- the bound, named as a failed read.
+d=$work/s7; mkdir -p "$d"
+echo '{"__fail__":true}' > "$d/page.0"
+run "S7 the API fails every time (bound)" 1 "$d"
+# S8: an empty page -- no check runs at all; the count says 0, not 1.
+d=$work/s8; mkdir -p "$d"
+echo '{"check_runs":[]}' > "$d/page.0"
+run "S8 no check runs at all" 1 "$d"
+if [ "$bad" -eq 0 ]; then echo "PROBE OK: the read waits for running legs, then judges; red, too few, and the bound are FAIL"; else echo "PROBE FAILED"; exit 1; fi
diff --git a/tools/probe_gate_honesty.sh b/tools/probe_gate_honesty.sh
index 4ecc277..1a37bd7 100755
--- a/tools/probe_gate_honesty.sh
+++ b/tools/probe_gate_honesty.sh
@@ -93,7 +93,9 @@ run_gate() { # run_gate <probe-id> <what was planted>
   printf '     licence files line:  %s\n' "$(grep -E '^  licence files ' "$W/gate.out" | sed 's/^  *//' || true)"
   printf '     licence files line count: %s   (0 means the step printed NOTHING)\n' \
          "$(grep -c '^  licence files ' "$W/gate.out")"
-  other=$(grep -E '^  (format|compile|instruments|test|credo|properties|optional deps|bench|docs|reuse|licence files|publication|messages) ' "$W/gate.out" | grep -cv ' pass$')
+  # A step line ends in `pass` or `pass (<detail>)`; the earlier `' pass$'` counted every
+  # detailed line as not-pass (G-078: 7 on every probe, P0 included).
+  other=$(grep -E '^  (format|compile|dialyzer|instruments|test|credo|properties|optional deps|audit|bench|docs|reuse|licence files|publication|baseline|messages) ' "$W/gate.out" | grep -cvE ' pass( \(|$)')
   printf '     other steps not pass: %s   (so the objection above is this step, not a side effect)\n' "$other"
   printf '     GATE_EXIT=%s\n' "$rc"
   grep -E '^ +(probe|PROBE)[^ ]*$' "$W/gate.out" | sed 's/^ */     named by the step: /'