diff --git a/.agents/scripts/awake.mjs b/.agents/scripts/awake.mjs
new file mode 100644
index 000000000..7a05767e5
--- /dev/null
+++ b/.agents/scripts/awake.mjs
@@ -0,0 +1,16 @@
+// Milliseconds of awake time since boot, from a clock that stops while the computer sleeps
+// (time.monotonic(): mach_absolute_time on macOS, CLOCK_MONOTONIC on Linux); null when python3
+// is missing. clock.mjs reports sleep with it, and budget-guard.mjs counts active time with it.
+import { execFileSync } from "node:child_process";
+
+export function awakeMs() {
+  try {
+    const out = execFileSync("python3", ["-I", "-c", "import time; print(time.monotonic())"], {
+      encoding: "utf8",
+      timeout: 2_000,
+    });
+    return Number(out) * 1000;
+  } catch {
+    return null;
+  }
+}
diff --git a/.agents/scripts/budget-guard.mjs b/.agents/scripts/budget-guard.mjs
index ab298e36d..2388495e1 100755
--- a/.agents/scripts/budget-guard.mjs
+++ b/.agents/scripts/budget-guard.mjs
@@ -16,7 +16,9 @@
 //                 group is killed.
 //
 // Run time counts only active time: a gap of more than 30 minutes between two tool calls (a
-// session waiting overnight for a person) is not counted.
+// session waiting overnight for a person) is not counted, and neither is time the computer
+// slept, measured on the same monotonic clock clock.mjs uses (a laptop with its lid closed
+// wakes for seconds at a time, so wall-clock gaps stay short while hours pass).
 //
 // Limits come from the environment (set them in settings.local.json "env"):
 //   CLAUDE_BUDGET_AGENT_TOKENS  default 15000000   (0 disables the agent token limit)
@@ -27,6 +29,7 @@ import fs from "node:fs";
 import os from "node:os";
 import path from "node:path";
 import { execFileSync } from "node:child_process";
+import { awakeMs } from "./awake.mjs";
 import { emptyState, scan, sessionFiles } from "./usage.mjs";
 
 const LIMIT_AGENT_TOKENS = Number(process.env.CLAUDE_BUDGET_AGENT_TOKENS ?? 15_000_000);
@@ -44,7 +47,7 @@ if (!input.transcript_path || !input.session_id) process.exit(0);
 const { main, subDir, files } = sessionFiles(input.transcript_path);
 
 const stateFile = path.join(os.tmpdir(), `claude-budget-${input.session_id}.json`);
-let state = { activeMs: 0, lastSeen: null, files: {} };
+let state = { activeMs: 0, lastSeen: null, lastAwake: null, files: {} };
 try {
   state = JSON.parse(fs.readFileSync(stateFile, "utf8"));
 } catch {}
@@ -55,9 +58,16 @@ for (const file of files) {
   scan(file, s);
 }
 const now = Date.now();
-const gap = typeof state.lastSeen === "number" ? now - state.lastSeen : 0;
+const awake = awakeMs();
+const wallGap = typeof state.lastSeen === "number" ? now - state.lastSeen : 0;
+// why: the awake gap excludes sleep; the wall gap is the fallback when python3 is missing.
+const gap =
+  awake !== null && typeof state.lastAwake === "number"
+    ? Math.max(0, awake - state.lastAwake)
+    : wallGap;
 if (gap > 0 && gap <= IDLE_GAP_MS) state.activeMs = (state.activeMs ?? 0) + gap;
 state.lastSeen = now;
+state.lastAwake = awake;
 fs.writeFileSync(stateFile, JSON.stringify(state));
 
 const fmt = (n) => `${(n / 1e6).toFixed(1)}M`;
diff --git a/.agents/scripts/cli-args.mjs b/.agents/scripts/cli-args.mjs
new file mode 100644
index 000000000..a3ef8b4ac
--- /dev/null
+++ b/.agents/scripts/cli-args.mjs
@@ -0,0 +1,6 @@
+// The one-line option parser the measuring scripts share: `option("--pr")` is the word after
+// `--pr` on the command line, or null when it is absent.
+export function option(name, args = process.argv.slice(2)) {
+  const at = args.indexOf(name);
+  return at === -1 ? null : (args[at + 1] ?? null);
+}
diff --git a/.agents/scripts/clock.mjs b/.agents/scripts/clock.mjs
index 267564fae..62146a929 100644
--- a/.agents/scripts/clock.mjs
+++ b/.agents/scripts/clock.mjs
@@ -9,10 +9,10 @@
 // while the computer sleeps (mach_absolute_time on macOS, CLOCK_MONOTONIC on Linux), Date.now()
 // does not. A gap under a minute is noise and counts as none. State lives in
 // $TMPDIR/claude-clock-<session>.json. Sub-agents get nothing: they print no status lines.
-import { execFileSync } from "node:child_process";
 import fs from "node:fs";
 import os from "node:os";
 import path from "node:path";
+import { awakeMs } from "./awake.mjs";
 
 const NOISE_MS = 60_000;
 
@@ -54,18 +54,6 @@ console.log(
   }),
 );
 
-function awakeMs() {
-  try {
-    const out = execFileSync("python3", ["-I", "-c", "import time; print(time.monotonic())"], {
-      encoding: "utf8",
-      timeout: 2_000,
-    });
-    return Number(out) * 1000;
-  } catch {
-    return null;
-  }
-}
-
 function minutes(ms) {
   return Math.round(ms / 60_000);
 }
diff --git a/.agents/scripts/delivery-counters.mjs b/.agents/scripts/delivery-counters.mjs
new file mode 100644
index 000000000..7fdf32882
--- /dev/null
+++ b/.agents/scripts/delivery-counters.mjs
@@ -0,0 +1,134 @@
+#!/usr/bin/env node
+// `node .agents/scripts/delivery-counters.mjs --since YYYY-MM-DD`: the counters that judge the
+// delivery changes in PROCESS-REVIEW-2026-10-21.md (items 4 to 12), one row per PR merged since
+// the date, from the PR's tagged runs on this machine and its files on GitHub:
+//
+//   ci lines      `ci:` lines the orchestrator put into fix prompts (CI failures sent to a fix run)
+//   syncs         fix runs asked to merge the base or sync the stack
+//   empty builds  build runs that built no cycle, or ended with no report at all
+//   no-change     fix runs whose report fixed nothing (`found 0 fixed 0` on every line)
+//   Tests lines   lines under the issue's `### Tests`
+//   tests added   `it(`/`test(` blocks the PR adds in *.test.ts and *.spec.ts
+//   test/src LOC  lines added to test files over lines added to other .ts files
+//
+// Unknown values print as `-`. Replays (stage=replay) never count. The repo is GH_REPO, else
+// the origin remote's owner/name.
+import { execFileSync } from "node:child_process";
+import { runOf, taggedRuns } from "./transcripts.mjs";
+
+const args = process.argv.slice(2);
+const at = args.indexOf("--since");
+const since = at === -1 ? null : args[at + 1];
+if (since === null || Number.isNaN(Date.parse(since))) {
+  console.error("usage: delivery-counters.mjs --since YYYY-MM-DD");
+  process.exit(64);
+}
+
+const repo =
+  process.env.GH_REPO ??
+  execFileSync("git", ["remote", "get-url", "origin"], { encoding: "utf8" })
+    .trim()
+    .replace(/\.git$/, "")
+    .replace(/.*[/:]([^/:]+\/[^/]+)$/, "$1");
+const api = (path, ...extra) =>
+  JSON.parse(
+    execFileSync("gh", ["api", "-X", "GET", `repos/${repo}/${path}`, ...extra], {
+      encoding: "utf8",
+      maxBuffer: 64 * 1024 * 1024,
+    }),
+  );
+
+const merged = [];
+for (let page = 1; page <= 5; page += 1) {
+  const pulls = api(
+    "pulls",
+    "-f",
+    "state=closed",
+    "-f",
+    "sort=updated",
+    "-f",
+    "direction=desc",
+    "-f",
+    "per_page=100",
+    "-f",
+    `page=${page}`,
+  );
+  for (const p of pulls) if (p.merged_at && p.merged_at >= since) merged.push(p);
+  if (pulls.length < 100 || pulls.every((p) => (p.updated_at ?? "") < since)) break;
+}
+merged.sort((a, b) => a.number - b.number);
+
+const runs = taggedRuns().filter(({ tag }) => tag.stage !== "replay");
+const rows = [];
+for (const p of merged) {
+  const issue = /^(?:bug|feature|task)\/(\d+)$/.exec(p.head.ref)?.[1] ?? null;
+  const mine = runs
+    .filter(({ tag }) => tag.pr === String(p.number) || (issue !== null && tag.issue === issue))
+    .map(({ file, tag }) => ({ tag, ...runOf(file) }));
+  const fixes = mine.filter((r) => r.tag.stage === "fix");
+  const ci = fixes.reduce((n, r) => n + (r.prompt.match(/^\s*-?\s*ci:/gm) ?? []).length, 0);
+  const syncs = fixes.filter((r) =>
+    /gh stack sync|merge (?:the |its )?base|stack: /i.test(r.prompt),
+  ).length;
+  // why: a build that could not start (a base behind main, a missing brief) ends with no report
+  // at all, which is the other way a build run builds nothing.
+  const empty = mine.filter(
+    (r) =>
+      r.tag.stage === "build" && (r.lastText.trim() === "" || /^cycles:\s*0\//m.test(r.lastText)),
+  ).length;
+  const noChange = fixes.filter((r) => {
+    const found = [...r.lastText.matchAll(/-> found (\d+) fixed (\d+)/g)];
+    return found.length > 0 && found.every((m) => m[2] === "0");
+  }).length;
+
+  let testsLines = null;
+  if (issue !== null) {
+    const body = api(`issues/${issue}`).body ?? "";
+    const section = /^### Tests\s*$([\s\S]*?)(?=^#|$(?![\s\S]))/m.exec(body)?.[1] ?? "";
+    testsLines = (section.match(/^- /gm) ?? []).length;
+  }
+  let testsAdded = 0;
+  let testLoc = 0;
+  let srcLoc = 0;
+  for (let page = 1; page <= 3; page += 1) {
+    const files = api(`pulls/${p.number}/files`, "-f", "per_page=100", "-f", `page=${page}`);
+    for (const f of files) {
+      if (!f.filename.endsWith(".ts")) continue;
+      if (/\.(test|spec)\.ts$/.test(f.filename)) {
+        testLoc += f.additions;
+        testsAdded += (f.patch?.match(/^\+\s*(?:it|test)(?:\.each)?\(/gm) ?? []).length;
+      } else srcLoc += f.additions;
+    }
+    if (files.length < 100) break;
+  }
+  rows.push({
+    pr: p.number,
+    issue,
+    runs: mine.length,
+    ci,
+    syncs,
+    empty,
+    noChange,
+    testsLines,
+    testsAdded,
+    testLoc,
+    srcLoc,
+  });
+}
+
+const show = (v) => (v === null ? "-" : String(v));
+const ratio = (a, b) => (b === 0 || b === null ? "-" : (a / b).toFixed(1));
+console.log(`# Delivery counters, PRs merged since ${since} (${rows.length})\n`);
+console.log(
+  "| PR | Issue | Runs | ci lines | syncs | empty builds | no-change fixes | Tests lines | tests added | tests/line | test/src LOC |",
+);
+console.log("| --- | --- | --: | --: | --: | --: | --: | --: | --: | --: | --: |");
+for (const r of rows) {
+  console.log(
+    `| #${r.pr} | ${r.issue ? `#${r.issue}` : "-"} | ${r.runs} | ${r.ci} | ${r.syncs} | ${r.empty} | ${r.noChange} | ${show(r.testsLines)} | ${r.testsAdded} | ${ratio(r.testsAdded, r.testsLines)} | ${ratio(r.testLoc, r.srcLoc)} |`,
+  );
+}
+const sum = (k) => rows.reduce((n, r) => n + (r[k] ?? 0), 0);
+console.log(
+  `\nTotals: ci lines ${sum("ci")}, syncs ${sum("syncs")}, empty builds ${sum("empty")}, no-change fixes ${sum("noChange")}, tests added ${sum("testsAdded")} over ${sum("testsLines")} Tests lines, test/src LOC ${ratio(sum("testLoc"), sum("srcLoc"))}.`,
+);
diff --git a/.agents/scripts/instruction-reads.mjs b/.agents/scripts/instruction-reads.mjs
new file mode 100644
index 000000000..1c720154e
--- /dev/null
+++ b/.agents/scripts/instruction-reads.mjs
@@ -0,0 +1,66 @@
+#!/usr/bin/env node
+// `node .agents/scripts/instruction-reads.mjs [--since YYYY-MM-DD] [--role <stage/role>]`: how often
+// each role's runs open each instruction file, from Claude Code's transcripts on this machine.
+//
+// A run is a ledger-tagged transcript (`toolchain.md`, "Review measurement"); its role is
+// `<stage>/<role>`. A file counts as opened when any tool call of the run names it: a rule
+// under docs/internal/agent-rules/, a brief under .agents/briefs/ (or the rules/ and briefs/
+// copies a review's inputs directory holds), REVIEW.md, EXPLORATION.md, a flow doc, or a
+// module manual (AGENTS.md). The agent's own definition is not a file it opens: the harness
+// loads it. Prints one Markdown table per role: runs, then each file with the runs that
+// opened it, most opened first. This is the number that judges a rewrite of the rules, the
+// briefs or the router (PROCESS-REVIEW-2026-10-21.md, items 16 to 18).
+import { runOf, taggedRuns } from "./transcripts.mjs";
+import { option } from "./cli-args.mjs";
+
+const since = option("--since");
+const only = option("--role");
+const sinceMs = since === null ? 0 : Date.parse(since);
+if (since !== null && Number.isNaN(sinceMs)) {
+  console.error("usage: instruction-reads.mjs [--since YYYY-MM-DD] [--role <stage/role>]");
+  process.exit(64);
+}
+
+const FILE = new RegExp(
+  String.raw`(?:docs/internal/agent-rules/[\w.-]+\.md|\.agents/briefs/[\w.-]+\.md|(?:^|["'\s/])(?:rules|briefs)/[\w.-]+\.md|review-protocol\.md|docs/internal/REVIEW\.md|docs/internal/EXPLORATION\.md|docs/internal/flows/[\w.-]+\.md|(?:src/[\w/.-]*)?AGENTS\.md)`,
+  "g",
+);
+
+const byRole = new Map();
+for (const { file, tag } of taggedRuns()) {
+  if (tag.stage === "replay") continue;
+  const role = `${tag.stage}/${tag.role}`;
+  if (only !== null && role !== only) continue;
+  const run = runOf(file);
+  if (Date.parse(run.startedAt ?? "") < sinceMs) continue;
+  const opened = new Set();
+  for (const input of run.toolInputs) {
+    for (const m of input.matchAll(FILE)) {
+      opened.add(
+        m[0]
+          .replace(/^["'\s/]/, "")
+          .replace(/^(?:rules|briefs)\//, (p) =>
+            p === "rules/" ? "docs/internal/agent-rules/" : ".agents/briefs/",
+          ),
+      );
+    }
+  }
+  const entry = byRole.get(role) ?? { runs: 0, files: new Map() };
+  entry.runs += 1;
+  for (const f of opened) entry.files.set(f, (entry.files.get(f) ?? 0) + 1);
+  byRole.set(role, entry);
+}
+
+if (byRole.size === 0) {
+  console.log(`No tagged runs${since ? ` since ${since}` : ""}.`);
+  process.exit(0);
+}
+console.log(`# Instruction files opened per role${since ? `, runs since ${since}` : ""}\n`);
+for (const [role, { runs, files }] of [...byRole].sort()) {
+  console.log(`## ${role}: ${runs} runs\n`);
+  console.log("| File | Runs that opened it |\n| --- | --: |");
+  for (const [f, n] of [...files].sort((a, b) => b[1] - a[1] || (a[0] < b[0] ? -1 : 1))) {
+    console.log(`| ${f} | ${n}/${runs} |`);
+  }
+  console.log("");
+}
diff --git a/.agents/scripts/pr-body-check.mjs b/.agents/scripts/pr-body-check.mjs
index 8c8828ce7..ffc17cade 100755
--- a/.agents/scripts/pr-body-check.mjs
+++ b/.agents/scripts/pr-body-check.mjs
@@ -9,7 +9,9 @@
 //     (`skipped` in place of the counts for a review the diff's kind left out) (rule 14);
 //   - a line under `Rejected:` is not tagged `spec:`, `code:` or `claims:` (rule 14);
 //   - the body says "spec needs" anywhere outside a code block (rule 15);
-//   - on a <kind>/<n> branch, `### Status` or `## Assumptions` is missing (coding rule 2).
+//   - on a <kind>/<n> branch, `### Status` or `## Assumptions` is missing (coding rule 2);
+//   - on a <kind>/<n> branch, the Review counts line names a review that ran and the PR has no
+//     comment headed `## Review ledger` (REVIEW.md, "Measurement"); checked only with a PR number.
 //
 // It warns (exit 0) when the Review section has no `Rounds: <n>. Tokens: <n>M.` line, or the
 // prose runs over 200 words: everything outside code blocks, checklist lines and the Status,
@@ -28,6 +30,7 @@ const args = process.argv.slice(2);
 const draftOk = args.includes("--draft-ok");
 let body;
 let branch = "";
+let prNumber = null;
 const fileAt = args.indexOf("--body-file");
 if (fileAt !== -1) {
   body = readFileSync(args[fileAt + 1], "utf8");
@@ -49,6 +52,7 @@ if (fileAt !== -1) {
   );
   body = pull.body ?? "";
   branch = pull.head.ref;
+  prNumber = pr;
 }
 body = body.replace(/\r/g, "");
 
@@ -110,6 +114,41 @@ if (/^(bug|feature|task)\/\d+$/.test(branch)) {
   }
 }
 
+// why: only a delivery PR (<kind>/<n>) has a ledger; a routine's or a person's chore PR gets
+// the reviewer's findings comment instead.
+if (
+  prNumber !== null &&
+  /^(bug|feature|task)\/\d+$/.test(branch) &&
+  lines.some(
+    (l) =>
+      REVIEW_COUNTS.test(l) &&
+      !/^(Spec|Code|Claims) review: skipped\. (Code|Claims) review: skipped\. Claims review: skipped\.$/.test(
+        l,
+      ),
+  )
+) {
+  const comments = JSON.parse(
+    execFileSync(
+      "gh",
+      [
+        "api",
+        "-X",
+        "GET",
+        `repos/${repository()}/issues/${prNumber}/comments`,
+        "-f",
+        "per_page=100",
+      ],
+      {
+        encoding: "utf8",
+        maxBuffer: 64 * 1024 * 1024,
+      },
+    ),
+  );
+  if (!comments.some((c) => /^## Review ledger\s*$/m.test(c.body ?? ""))) {
+    failures.push("no comment headed `## Review ledger` (REVIEW.md, Measurement)");
+  }
+}
+
 const words = budgetedWords(lines);
 if (words > BUDGET) {
   warnings.push(
diff --git a/.agents/scripts/replay-freeze.mjs b/.agents/scripts/replay-freeze.mjs
index 7afa93331..b301811ba 100755
--- a/.agents/scripts/replay-freeze.mjs
+++ b/.agents/scripts/replay-freeze.mjs
@@ -18,12 +18,8 @@ import { execFileSync } from "node:child_process";
 import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
 import { dirname, join } from "node:path";
 import { fileURLToPath } from "node:url";
+import { option } from "./cli-args.mjs";
 
-const args = process.argv.slice(2);
-const option = (name) => {
-  const at = args.indexOf(name);
-  return at === -1 ? null : args[at + 1];
-};
 const pr = option("--pr");
 const round = option("--round");
 const head = option("--head");
@@ -69,10 +65,23 @@ const comments = JSON.parse(
   gh("api", "-X", "GET", `repos/${nwo}/issues/${pr}/comments`, "-f", "per_page=100"),
 );
 let findings = null;
+const recordedRounds = {};
 for (const c of comments) {
   if (!/^## Review ledger/m.test(c.body ?? "")) continue;
   const m = c.body.match(/```json\s*(\{[\s\S]*?\})\s*```/);
   if (m) findings = JSON.parse(m[1]).findings;
+  for (const [, r, sha] of c.body.matchAll(/^Round (\S+): ([0-9a-f]{7,40})/gm))
+    recordedRounds[r] = sha;
+}
+if (
+  recordedRounds[round] &&
+  !recordedRounds[round].startsWith(head.slice(0, 7)) &&
+  !head.startsWith(recordedRounds[round].slice(0, 7))
+) {
+  console.error(
+    `the ledger comment says round ${round} reviewed ${recordedRounds[round]}, not ${head}`,
+  );
+  process.exit(65);
 }
 if (findings === null)
   console.error(
diff --git a/.agents/scripts/review-inputs.sh b/.agents/scripts/review-inputs.sh
index d3a168eb7..bebe26f48 100755
--- a/.agents/scripts/review-inputs.sh
+++ b/.agents/scripts/review-inputs.sh
@@ -2,7 +2,12 @@
 # Builds the blind inputs for the pre-merge reviews of a PR (delivery rule 14) and prints the
 # directory's path as the last line on stdout.
 #
-#   .agents/scripts/review-inputs.sh <pr> [<previous commit>]
+#   .agents/scripts/review-inputs.sh <pr> [<previous commit> [<previous base>]]
+#
+# A previous commit that is no ancestor of the head (the branch was rebased or the stack synced)
+# is mapped to its rebased equivalent when its base is given: the same number of commits from
+# the new base, each with the same patch-id as its old counterpart. Any difference (a conflict
+# resolved, a commit dropped) stays incomplete, as a rebased history nobody can map is.
 #
 # The directory holds, when each exists:
 #
@@ -10,6 +15,7 @@
 #   commit                 the PR head the directory was built from
 #   base, merge-base       immutable actual PR base tip and fork
 #   previous               validated ancestor for targeted checks
+#   previous-mapped-from   the previous commit as given, when it was mapped across a rebase
 #   review-protocol.md     review policy from the pinned base, when present
 #   kind                   what .agents/scripts/diff-kind.mjs calls the diff: adr-only,
 #                          docs-only, comments-only, tests-only, full or large; it sets which
@@ -43,9 +49,10 @@ set -eu
 
 pr=${1:-}
 previous=${2:-}
+previous_base=${3:-}
 case $pr in
   '' | *[!0-9]*)
-    echo "usage: review-inputs.sh <pr number> [<previous commit>]" >&2
+    echo "usage: review-inputs.sh <pr number> [<previous commit> [<previous base>]]" >&2
     exit 64
     ;;
 esac
@@ -106,10 +113,28 @@ if [ -n "$previous" ]; then
     say "previous commit is unavailable; assessment incomplete"
     exit 65
   }
-  git merge-base --is-ancestor "$previous" "$head" || {
-    say "previous commit is not an ancestor of head; assessment incomplete"
-    exit 65
-  }
+  mapped_from=
+  if ! git merge-base --is-ancestor "$previous" "$head"; then
+    if [ -z "$previous_base" ]; then
+      say "previous commit is not an ancestor of head (rebased?); pass its base to map it, else assessment incomplete"
+      exit 65
+    fi
+    previous_base=$(git rev-parse --verify "$previous_base^{commit}" 2>/dev/null) || {
+      say "previous base is unavailable; assessment incomplete"
+      exit 65
+    }
+    n=$(git rev-list --count --no-merges "$previous_base..$previous")
+    old_ids=$(git rev-list --reverse --no-merges "$previous_base..$previous" | while read -r c; do git show "$c" | git patch-id --stable | cut -d' ' -f1; done)
+    new_ids=$(git rev-list --reverse --no-merges "$base..$head" | head -n "$n" | while read -r c; do git show "$c" | git patch-id --stable | cut -d' ' -f1; done)
+    if [ "$n" -gt 0 ] && [ "$old_ids" = "$new_ids" ]; then
+      mapped_from=$previous
+      previous=$(git rev-list --reverse --no-merges "$base..$head" | sed -n "${n}p")
+      say "previous $mapped_from mapped across the rebase to $previous ($n commits, same patch-ids)"
+    else
+      say "previous commit is not an ancestor of head and its $n commits do not match the rebased history by patch-id; assessment incomplete"
+      exit 65
+    fi
+  fi
 fi
 
 tmp=${TMPDIR:-/tmp}
@@ -121,6 +146,7 @@ echo "$head" >"$out/commit"
 echo "$base" >"$out/base"
 echo "$fork" >"$out/merge-base"
 [ -z "$previous" ] || echo "$previous" >"$out/previous"
+[ -z "${mapped_from:-}" ] || echo "$mapped_from" >"$out/previous-mapped-from"
 if git cat-file -e "$rules_ref:docs/internal/REVIEW.md" 2>/dev/null; then
   git show "$rules_ref:docs/internal/REVIEW.md" >"$out/review-protocol.md"
 fi
diff --git a/.agents/scripts/review-metrics.mjs b/.agents/scripts/review-metrics.mjs
index 9cb340345..20939624c 100644
--- a/.agents/scripts/review-metrics.mjs
+++ b/.agents/scripts/review-metrics.mjs
@@ -15,11 +15,11 @@
 // left out, so a replay never counts as the PR's delivery work.
 //
 // --findings validates the coordinator's findings file (`{"findings": [...]}`, each with id,
-// role, class, outcome, path) and adds the outcome and class counts. --json prints the ledger
+// role, class, outcome, path and, when present, round) and adds the outcome and class counts. --json prints the ledger
 // (runs, totals, findings) instead of the summary lines.
-import { existsSync, openSync, readFileSync, readSync, closeSync, readdirSync } from "node:fs";
-import { homedir } from "node:os";
-import { join } from "node:path";
+import { readFileSync } from "node:fs";
+import { parsed, projectsRoot, tagOf, transcripts } from "./transcripts.mjs";
+import { option } from "./cli-args.mjs";
 
 // $ per million tokens: input, output, cache read, 5-minute cache write, 1-hour cache write.
 // https://platform.claude.com/docs/en/about-claude/pricing, read 2026-10-19.
@@ -29,12 +29,9 @@ const PRICES = {
   "claude-haiku-5-5": [0.1, 0.5, 0.01, 0.125, 0.2],
 };
 const OUTCOMES = ["open", "fixed", "duplicate", "incorrect", "intentional", "nonblocking"];
+const ROUNDS = ["1", "2", "fix-check"];
 
 const args = process.argv.slice(2);
-const option = (name) => {
-  const at = args.indexOf(name);
-  return at === -1 ? null : args[at + 1];
-};
 const pr = option("--pr");
 const issue = option("--issue");
 const findingsFile = option("--findings");
@@ -45,7 +42,7 @@ if (pr === null || !/^\d+$/.test(pr) || (issue !== null && !/^\d+$/.test(issue))
   process.exit(64);
 }
 
-const root = join(process.env.CLAUDE_CONFIG_DIR ?? join(homedir(), ".claude"), "projects");
+const root = projectsRoot();
 const runs = transcripts(root)
   .map((file) => ({ file, tag: tagOf(file) }))
   .filter(({ tag }) => tag !== null && (tag.pr === pr || (issue !== null && tag.issue === issue)))
@@ -100,36 +97,6 @@ if (findings !== null) {
   console.log(`Finding outcomes: ${count("outcome")}. Classes: ${count("class")}.`);
 }
 
-function transcripts(dir) {
-  if (!existsSync(dir)) return [];
-  const files = [];
-  for (const entry of readdirSync(dir, { withFileTypes: true })) {
-    const path = join(dir, entry.name);
-    if (entry.isDirectory()) files.push(...transcripts(path));
-    else if (entry.name.endsWith(".jsonl")) files.push(path);
-  }
-  return files;
-}
-
-// why: transcripts run to hundreds of megabytes; the tag is in the first user record, near the top.
-function tagOf(file) {
-  const fd = openSync(file, "r");
-  const head = Buffer.alloc(256 * 1024);
-  const read = readSync(fd, head, 0, head.length, 0);
-  closeSync(fd);
-  for (const line of head.subarray(0, read).toString("utf8").split("\n")) {
-    const rec = parsed(line);
-    if (rec?.type !== "user") continue;
-    const first = promptOf(rec).split("\n")[0].trim();
-    if (!first.startsWith("ledger:")) return null;
-    const tag = { pr: "-", issue: "-", stage: "-", round: "-", role: "-", lens: "-" };
-    for (const [, key, value] of first.matchAll(/(\w+)=(\S+)/g))
-      if (key in tag) tag[key] = value.replace(/^#/, "");
-    return tag;
-  }
-  return null;
-}
-
 function run(file, tag) {
   const { messages, startedAt, finishedAt } = read(file);
   const tokens = { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 };
@@ -202,23 +169,11 @@ function checkFinding(f, list) {
   const missing = ["role", "class", "path"].find((k) => typeof f[k] !== "string" || f[k] === "");
   if (missing !== undefined) throw new Error(`finding ${f.id} has no ${missing}`);
   if (!OUTCOMES.includes(f.outcome)) throw new Error(`finding ${f.id} has an invalid outcome`);
+  if (f.round !== undefined && !ROUNDS.includes(String(f.round))) {
+    throw new Error(`finding ${f.id} has an invalid round (1, 2 or fix-check)`);
+  }
   const canonical = list.some((x) => x.id === f.canonicalId && x.outcome !== "duplicate");
   if (f.outcome === "duplicate" && !canonical) {
     throw new Error(`duplicate ${f.id} must name a canonical finding`);
   }
 }
-
-function promptOf(rec) {
-  const content = rec.message?.content;
-  if (typeof content === "string") return content;
-  return Array.isArray(content) ? (content.find((b) => b.type === "text")?.text ?? "") : "";
-}
-
-function parsed(line) {
-  if (line === "") return null;
-  try {
-    return JSON.parse(line);
-  } catch {
-    return null;
-  }
-}
diff --git a/.agents/scripts/review-score.mjs b/.agents/scripts/review-score.mjs
index 54849fba9..e10d8118c 100644
--- a/.agents/scripts/review-score.mjs
+++ b/.agents/scripts/review-score.mjs
@@ -16,12 +16,8 @@ import { execFileSync } from "node:child_process";
 import { appendFileSync, readFileSync } from "node:fs";
 import { dirname, join } from "node:path";
 import { fileURLToPath } from "node:url";
+import { option } from "./cli-args.mjs";
 
-const args = process.argv.slice(2);
-const option = (name) => {
-  const at = args.indexOf(name);
-  return at === -1 ? null : args[at + 1];
-};
 const pr = option("--pr");
 const round = option("--round");
 const file = option("--findings");
@@ -140,8 +136,15 @@ for (const r of replay) {
     matches.push({ replay: r.id, truth: hit.id, truthOutcome: hit.outcome });
   } else unmatched.push(r);
 }
-// why: a single lens is judged on the findings its role owned in the real review.
-const defects = truth.filter((t) => t.defect && (role === null || t.role === role));
+// why: a single lens is judged on the findings its role owned in the real review, and a later
+// round on the findings that round raised, when the ledger says which round that was.
+const rounded = truth.some((t) => t.round !== undefined);
+const defects = truth.filter(
+  (t) =>
+    t.defect &&
+    (role === null || t.role === role) &&
+    (!rounded || String(t.round) === String(round)),
+);
 const found = defects.filter((t) => taken.has(t.id));
 const missed = defects.filter((t) => !taken.has(t.id));
 const nonDefectHits = matches.filter((m) => !["fixed", "open"].includes(m.truthOutcome));
diff --git a/.agents/scripts/transcripts.mjs b/.agents/scripts/transcripts.mjs
new file mode 100644
index 000000000..6ea0b676c
--- /dev/null
+++ b/.agents/scripts/transcripts.mjs
@@ -0,0 +1,86 @@
+// Claude Code's transcripts, for the scripts that measure agent runs: review-metrics.mjs,
+// instruction-reads.mjs and delivery-counters.mjs. A run is a transcript whose first user
+// message starts with the ledger tag line (`toolchain.md`, "Review measurement").
+import { closeSync, existsSync, openSync, readFileSync, readSync, readdirSync } from "node:fs";
+import { homedir } from "node:os";
+import { join } from "node:path";
+
+export const projectsRoot = () =>
+  join(process.env.CLAUDE_CONFIG_DIR ?? join(homedir(), ".claude"), "projects");
+
+export function transcripts(dir) {
+  if (!existsSync(dir)) return [];
+  const files = [];
+  for (const entry of readdirSync(dir, { withFileTypes: true })) {
+    const path = join(dir, entry.name);
+    if (entry.isDirectory()) files.push(...transcripts(path));
+    else if (entry.name.endsWith(".jsonl")) files.push(path);
+  }
+  return files;
+}
+
+// why: transcripts run to hundreds of megabytes; the tag is in the first user record, near the top.
+export function tagOf(file) {
+  const fd = openSync(file, "r");
+  const head = Buffer.alloc(256 * 1024);
+  const read = readSync(fd, head, 0, head.length, 0);
+  closeSync(fd);
+  for (const line of head.subarray(0, read).toString("utf8").split("\n")) {
+    const rec = parsed(line);
+    if (rec?.type !== "user") continue;
+    const first = promptOf(rec).split("\n")[0].trim();
+    if (!first.startsWith("ledger:")) return null;
+    const tag = { pr: "-", issue: "-", stage: "-", round: "-", role: "-", lens: "-" };
+    for (const [, key, value] of first.matchAll(/(\w+)=(\S+)/g))
+      if (key in tag) tag[key] = value.replace(/^#/, "");
+    return tag;
+  }
+  return null;
+}
+
+export function taggedRuns(root = projectsRoot()) {
+  return transcripts(root)
+    .map((file) => ({ file, tag: tagOf(file) }))
+    .filter(({ tag }) => tag !== null);
+}
+
+export function runOf(file) {
+  const records = readFileSync(file, "utf8").split("\n").map(parsed).filter(Boolean);
+  const first = records.find((r) => r.type === "user");
+  const blocks = assistantBlocks(records);
+  return {
+    prompt: first ? promptOf(first) : "",
+    toolInputs: blocks
+      .filter((b) => b.type === "tool_use")
+      .map((b) => JSON.stringify(b.input ?? {})),
+    lastText: blocks.filter((b) => b.type === "text" && b.text?.trim()).at(-1)?.text ?? "",
+    startedAt: records.map((r) => r.timestamp).find((t) => typeof t === "string") ?? null,
+  };
+}
+
+// why: one API message is written as several records with the same id; its blocks count once.
+function assistantBlocks(records) {
+  const seen = new Set();
+  const blocks = [];
+  for (const r of records) {
+    if (r.type !== "assistant" || !r.message?.id || seen.has(r.message.id)) continue;
+    seen.add(r.message.id);
+    blocks.push(...(r.message.content ?? []));
+  }
+  return blocks;
+}
+
+export function promptOf(rec) {
+  const content = rec.message?.content;
+  if (typeof content === "string") return content;
+  return Array.isArray(content) ? (content.find((b) => b.type === "text")?.text ?? "") : "";
+}
+
+export function parsed(line) {
+  if (line === "") return null;
+  try {
+    return JSON.parse(line);
+  } catch {
+    return null;
+  }
+}
diff --git a/.claude/agents/reviewer.md b/.claude/agents/reviewer.md
index e6acf6ac9..cb5e7f3da 100644
--- a/.claude/agents/reviewer.md
+++ b/.claude/agents/reviewer.md
@@ -14,9 +14,11 @@ impacts, the round cap and metrics. This file adds only how you run it.
 Edit no code. Arguments: PR, round (`1`, `2`, `fix-check`), previous full
 SHA after round 1, prior resolution ledger, and optional `pilot: sonnet`.
 
-1. **Inputs.** Run `.agents/scripts/review-inputs.sh <PR> [<previous SHA>]`
-   and record its base, merge-base and commit; a failed build is
-   incomplete. A `replay: true` argument comes with an inputs directory
+1. **Inputs.** Run `.agents/scripts/review-inputs.sh <PR> [<previous SHA>
+[<previous base>]]` and record its base, merge-base and commit; a
+   failed build is incomplete. A `previous-base:` argument comes with a
+   rebased branch: pass it, and report the mapping the script prints;
+   never check a rebase by hand. A `replay: true` argument comes with an inputs directory
    already built by `.agents/scripts/review-replay.sh`: use it, do not run
    `review-inputs.sh`, ignore the PR's state on GitHub (it may be merged),
    post nothing anywhere, never open `.agents/replay/`, and write the
diff --git a/.claude/agents/spec-checker.md b/.claude/agents/spec-checker.md
index 21bc40b7b..284f648a9 100644
--- a/.claude/agents/spec-checker.md
+++ b/.claude/agents/spec-checker.md
@@ -91,8 +91,9 @@ _note_:
 - **Note**: anything else, above all a gap the implementer may close as an
   `Assumption:` line (`coding.md` rule 2): a wording, an order, a bound
   the spec implies. A test for a case the spec does not name, defensive
-  handling, a needless Depends on, a task slightly over size. At most five
-  notes, the ones most likely to matter; drop the rest.
+  handling, a needless Depends on, a task slightly over size. A note only
+  when the implementer would act on it; none is the usual count, and a
+  report never pads to a number.
 
 1. **contradiction.** A line that breaks an agent rule or an accepted ADR,
    or two lines that cannot both hold. A Done when line the task's own
diff --git a/.claude/skills/deliver/SKILL.md b/.claude/skills/deliver/SKILL.md
index 9022644f0..2178410d1 100644
--- a/.claude/skills/deliver/SKILL.md
+++ b/.claude/skills/deliver/SKILL.md
@@ -159,7 +159,10 @@ they name; never build again or restart review at round 1.
      is neither a defect nor a pass.
    - Fix lines with `Rerun: round 2` or `fix-check`: one consolidated fix
      run with the immutable base/merge-base, `previous <Commit>`, ledger
-     and findings, then `reviewer` with the round `Rerun` names.
+     and findings, then `reviewer` with the round `Rerun` names. After a
+     rebase or stack sync, the review also gets `previous-base: <the
+base-sha the previous check used>`, so the inputs script can map the
+     previous commit across the rebase.
    - `Rerun: none` with unresolved nonmechanical findings or unchecked fixes
      at the cap: park. Mechanical changes get a coordinator diff check.
    - Commits after reviews passed, including conflict resolutions and CI
@@ -206,7 +209,9 @@ they name; never build again or restart review at round 1.
    the counts for a review the diff's kind left out (for an ADR-only diff:
    `Code review: skipped. Claims review: skipped.`). `<n>` is the number of
    legacy full/targeted rounds, 1 or 2. Post the `## Review ledger` comment
-   as `docs/internal/REVIEW.md` "Measurement" says, fix-checks included. Add
+   as `docs/internal/REVIEW.md` "Measurement" says, fix-checks included:
+   one `Round <r>: <full commit>` line per check first, and `round` on
+   every finding. Add
    `Assessment: complete` only when coverage is complete and no finding is
    open; otherwise `Assessment: incomplete` and park. `<t>` is
    `tokens=` from `node .agents/scripts/usage.mjs`, this session's usage so
diff --git a/docs/internal/DELIVERY.md b/docs/internal/DELIVERY.md
index 6c45455d0..1598e63d7 100644
--- a/docs/internal/DELIVERY.md
+++ b/docs/internal/DELIVERY.md
@@ -131,8 +131,12 @@ Two transitions stay manual on purpose, one click each: `bug:new` →
 last five: blocking findings per review raised and fixed, notes, PRs
 needing a second round, median hours to merge, rounds and tokens,
 handoffs. `--since <date>` lists PRs; `--calibration` prints the recent
-incorrect findings every reviewer reads. Change one thing in the pipeline
-at a time and compare the rows before and after.
+incorrect findings every reviewer reads. `delivery-counters.mjs --since
+<date>` prints the per-PR counters the process review judges delivery
+changes by (CI lines sent to fix runs, stack syncs, empty builds, fix runs
+that changed nothing, tests per Tests line), and `instruction-reads.mjs`
+how often each role opens each rule, brief and manual. Change one thing in
+the pipeline at a time and compare the rows before and after.
 
 Weekly routines keep the suite and the reviews in shape: `harden-tests`,
 `audit-tests`, `garden-comments` and `calibrate-reviews`
diff --git a/docs/internal/REVIEW.md b/docs/internal/REVIEW.md
index f85acfcec..a9cd9874d 100644
--- a/docs/internal/REVIEW.md
+++ b/docs/internal/REVIEW.md
@@ -4,9 +4,12 @@ Delivery rule 14 delegates its review procedure to this document. Review
 one immutable PR snapshot. Use `.agents/scripts/review-inputs.sh <PR>` for
 round 1, and add the previous report's full commit SHA for later checks.
 The inputs record `base` (base tip), `merge-base`, `commit` and `previous`.
-Never substitute `origin/main` for a stacked PR's base. Missing or
-non-ancestor previous commits make assessment incomplete: reconcile the
-history and rebuild inputs, do not silently perform a full review.
+Never substitute `origin/main` for a stacked PR's base. A missing previous
+commit makes assessment incomplete. After a rebase or stack sync the
+previous commit is no ancestor of the head: give the script that round's
+base as well, and it maps the commit to its rebased equivalent when every
+commit's patch-id matches; otherwise assessment is incomplete, and nobody
+silently performs a full review in its place.
 Rules, briefs and this protocol come from the pinned PR base, not a moving
 `main`. No reviewer sees author narration (PR body, commit messages) or
 another lens's findings before synthesis.
@@ -107,9 +110,13 @@ transcripts. Unknown values print as unknown, never zero.
 The coordinator keeps one findings file per PR, `{"findings": [...]}`, one
 entry per canonical finding: `id`, `role`, `class` (the `[class]` the lens
 gave, a rule like `testing 2` or at most 8 words), `outcome`, `path`,
-`fix` (the fix commit, once fixed) and `canonicalId` for a duplicate. Before the gate, post one PR comment headed
-`## Review ledger`: the output of `review-metrics.mjs --pr <M> --issue <N>
---findings <file>`, then the findings file in a `json` code block. The
+`round` (the check that raised it: `1`, `2` or `fix-check`), `fix` (the
+fix commit, once fixed) and `canonicalId` for a duplicate. Before the gate,
+post one PR comment headed `## Review ledger`: one line `Round <r>:
+<full commit>` per check in order, the output of `review-metrics.mjs --pr
+<M> --issue <N> --findings <file>`, then the findings file in a `json`
+code block. The round lines are what lets a check be frozen and replayed
+later (`replay-freeze.mjs`). The
 weekly `recurring-findings` routine reads these comments. Record escaped
 defects after merge. Do not call raised-minus-fixed a false-positive rate.
 
diff --git a/src/pr-body-check.test.ts b/src/pr-body-check.test.ts
index 995e39b65..9abd32734 100644
--- a/src/pr-body-check.test.ts
+++ b/src/pr-body-check.test.ts
@@ -33,14 +33,28 @@ function check(body: string, ...args: string[]) {
   }
 }
 
-function fromGitHub(pull: object, options: { origin?: boolean } = {}) {
+function fromGitHub(
+  pull: object,
+  options: { origin?: boolean; comments?: readonly string[] } = {},
+) {
   const dir = mkdtempSync(join(tmpdir(), "pr-body-gh-"));
   writeFileSync(join(dir, "pull.json"), JSON.stringify(pull));
+  // why: the comments endpoint answers only when the test says so, so a case with no
+  // `comments` proves the script never asked for them (a chore branch has no ledger).
+  if (options.comments !== undefined) {
+    writeFileSync(
+      join(dir, "comments.json"),
+      JSON.stringify(options.comments.map((body) => ({ body }))),
+    );
+  }
   writeFileSync(
     join(dir, "gh"),
     `#!/bin/sh
-[ "$*" = "api -X GET repos/o/r/pulls/7" ] || { echo "fake gh: unexpected: $*" >&2; exit 9; }
-cat ${JSON.stringify(join(dir, "pull.json"))}
+case "$*" in
+  "api -X GET repos/o/r/pulls/7") cat ${JSON.stringify(join(dir, "pull.json"))} ;;
+  "api -X GET repos/o/r/issues/7/comments -f per_page=100") cat ${JSON.stringify(join(dir, "comments.json"))} 2>/dev/null || { echo "fake gh: comments not expected" >&2; exit 9; } ;;
+  *) echo "fake gh: unexpected: $*" >&2; exit 9 ;;
+esac
 `,
   );
   chmodSync(join(dir, "gh"), 0o755);
@@ -158,9 +172,10 @@ describe("pr-body-check.mjs", () => {
     const bare = `## Review\n\n${COUNTS}\nRounds: 1. Tokens: 1.0M.\n\n${FOOTER}\n`;
     expect(fromGitHub({ body: bare, head: { ref: "chore/x" } })).toEqual({ status: 0, lines: [] });
 
-    expect(fromGitHub({ body: bare, head: { ref: "bug/7" } }).lines).toEqual([
+    expect(fromGitHub({ body: bare, head: { ref: "bug/7" } }, { comments: [] }).lines).toEqual([
       "pr-body: no `### Status` section",
       expect.stringContaining("no `## Assumptions` section"),
+      expect.stringContaining("no comment headed `## Review ledger`"),
     ]);
     expect(fromGitHub({ body: null, head: { ref: "chore/x" } }).lines).toEqual([
       expect.stringContaining('the last line is not "*Written by an agent.*"'),
@@ -168,6 +183,24 @@ describe("pr-body-check.mjs", () => {
     ]);
   });
 
+  it("requires the Review ledger comment on a delivery branch whose reviews ran, and only there", () => {
+    const body = valid;
+    const ledger = '## Review ledger\n\nRound 1: abc\n```json\n{"findings": []}\n```';
+    expect(fromGitHub({ body, head: { ref: "task/7" } }, { comments: ["hi", ledger] })).toEqual({
+      status: 0,
+      lines: [],
+    });
+    expect(fromGitHub({ body, head: { ref: "task/7" } }, { comments: ["hi"] }).lines).toEqual([
+      "pr-body: no comment headed `## Review ledger` (REVIEW.md, Measurement)",
+    ]);
+    const skipped = body.replace(
+      COUNTS,
+      "Spec review: skipped. Code review: skipped. Claims review: skipped.",
+    );
+    expect(fromGitHub({ body: skipped, head: { ref: "task/7" } }).status).toBe(0);
+    expect(fromGitHub({ body, head: { ref: "chore/x" } }).status).toBe(0);
+  });
+
   it("takes the repo from the origin remote when GH_REPO is unset", () => {
     const bare = `## Review\n\n${COUNTS}\nRounds: 1. Tokens: 1.0M.\n\n${FOOTER}\n`;
     const result = fromGitHub({ body: bare, head: { ref: "chore/x" } }, { origin: true });