v1.91.314.0 feat(autoplan): carry the owner's answer forward, gate ELI5 field, stacked approvals import (#3583)

* refactor: share the decision-trail append rules (lib/decision-trail.ts) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * feat(autoplan): carry the owner's answer to a later gate that asks nothing new, gate.json eli5, stacked approvals import A later gate revision with nothing new is answered with the owner's recorded words (CARRIED); a 'do it' also covers new auto items that change no scope, cost or direction; an unreviewed post-cap delta gets one cross-family reviewer first (CARRY_CHECK, gstack-autoplan check). gate.json gains eli5 and the Captain brief requires ## ELI5 on runs that carry it. gstack-approvals import --stack writes every branch or none. The first next prints REVIEW_ESTIMATE. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: carry-forward, ELI5 section and stacked approvals import Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix(autoplan): carry only fully owner-answered gates; void stale delta checks - a pending delta check is voided (reservation released) when the owner answers, the plan is revised, or a new gate is written; a check bound to an older delta is void - carry refuses a gate with any approval item still pending or answered in item tokens - "do it" never covers an option that changes the plan; isDoIt ignores Captain-relayed prefixes - captain brief rebuilds ELI5 from the current plan and lists items taken under "do it" - gstack-approvals import --stack locates every branch before writing any - DRIVER.md back under its 162-line budget Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: carry-forward conditions, delta-check voiding and import --stack match the final code Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * chore: changelog for autoplan carry-forward (unreleased; stamped at merge) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * chore: stamp v1.91.314.0 Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> --------- Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
2026-10-31 22:14:25 +01:00 · 2026-10-29 09:58:34 +00:00
1 parent 81a82727b3
commit 0cab97447c
@@ -1,5 +1,20 @@
 # Changelog

+## [1.91.314.0] - 2026-10-29
+
+**A later autoplan gate that asks nothing new is now answered with the owner's earlier words instead of another round trip.** Captains were carrying approvals across gate revisions by hand (GSTA-140 reused a rev-1 "Do it" on rev 2; ALPH-50 lost about 27 minutes to one), and a replay over 23 recorded runs found 14 later gates, 8 of them holding only post-cap delta items.
+
+### Added
+- **Carry-forward** (`lib/autoplan-carry.ts`): when gate_rev N+1 asks nothing new and gate_rev N was fully answered by the owner with `rec` or "do it", gstack answers N+1 with those recorded words and prints `CARRIED: gate_rev=<n> <- gate_rev=<m> …`; anything else prints `CARRY: no (<why>)` and goes to the owner. `--no-carry-forward` on the first `next` turns it off.
+- **"do it" covers new low-risk items.** When the owner's own words were "do it", a new item that changes no scope, cost, direction or plan text takes its recommendation, is marked `carried_under: do it`, and is listed in the approval record and the Captain brief as not shown to the owner. A Captain's "do it" never counts.
+- **Delta check.** A "plan changed after cap" item gets one reviewer from another model family on the delta first (`CARRY_CHECK:`, `gstack-autoplan check`), within a per-gate spend cap (`--carry-check-cap`, default $5). High or above, no reviewer, an owner answer, or a plan edit after dispatch leaves the gate for the owner and releases the reservation. The verdict is recorded on `gate.json`, the decision rows and `decisions.tsv`.
+- **`eli5` in gate.json**: the plan's `## ELI5` and the four approvals-shape sections; `gstack-gate render` opens page 1 with them, and the Captain brief requires `## ELI5` (older runs render as before).
+- **`gstack-approvals import --autoplan <dir> --stack b1,b2`** writes the same rows to every branch of a stack, or to none when one branch is unknown.
+- **`REVIEW_ESTIMATE:`** on the first `next`: phases, voices, rounds, attempts and the reservation total.
+
+### Changed
+- `lib/decision-trail.ts` holds the trail append rules shared by `gstack-decision-trail` and the carry.
+
 ## [1.91.313.0] - 2026-10-29

 **The PII scanner stops flagging two kinds of ordinary text: a bare power-of-two or hash constant in code, and a package@version-style filename.** Eval artifacts and PR bodies that print a PRNG range or name a versioned results file no longer raise a MEDIUM phone or email finding, and real phone numbers and addresses next to them still report.
@@ -1 +1 @@
-1.91.313.0
+1.91.314.0
@@ -1,4 +1,4 @@
-# gstack digest v1.91.313.0 — regenerate/re-copy after upgrading gstack
+# gstack digest v1.91.314.0 — regenerate/re-copy after upgrading gstack

 Behavioral rules from gstack (https://github.com/garrytan/gstack), compressed
 for agent hosts without a full skill install. The full skills add workflows,
@@ -26,7 +26,7 @@ line yourself; gstack never writes the plan.

 1. `next` prints one `ATTEMPT: <id> phase=<p> voice=native|outside prompt=<file> result=<file>`
   per voice. Dispatch the native prompt to a subagent on your model and the outside prompt to
-   a subagent on another model family (or `gstack-outside-voice run`). Each writes `result=`.
+   a subagent on another model family (or `gstack-outside-voice run`). Each writes `result=`. The first `next` also prints `REVIEW_ESTIMATE:` (most attempts and reservations the review can take).
 2. Submit each voice exactly once (`BOUND:`). Both voices in closes the phase (`PHASE_CLOSED:`).
 3. Repeat `next` until `status=gate_pending` (`GATE: gate_rev=<n>`); render it with `gstack-gate render <dir>/gate.json --all` and relay every page with the two cost lines printed under `GATE:`: `GATE_SPEND:` (spent, worst case, cap, headroom, attempts; host-subagent attempts are priced unknown and counted at their reservation, so quote their real figure from the host's cost view) and `GATE_REC_COST:` (the phases `rec` can reopen and the estimate for them, an upper bound; `gate_cost` in `--json`).
   You never approve the gate for the owner. Offer `rec` (every recommendation, plan changes included) as the default.
@@ -38,6 +38,7 @@ line yourself; gstack never writes the plan.
 5. `REOPENED: <phases>` means loop `next` again (Eng last). `APPROVED:` means `export` (`DISPATCHES: <n> est_input_tokens=<k>`, relative).
 6. Rounds: one by default; a second only for High or Critical findings; cap 2 (`--max-rounds`). A `revise` whose decisions all resolve below
   High prints `APPLIED: <ids> without re-review (below High)`. At the cap with High open, `ESCALATE: evidence sufficient, decision needed`: relay it, never ask for another round.
+7. Carry-forward ([docs/unattended.md#carry-forward](../docs/unattended.md#carry-forward); `--no-carry-forward` on the first `next` opts out): a later gate that asks nothing new takes the owner's words from the gate before it (`CARRIED:`, then `APPROVED:`); `CARRY: no (<why>)` leaves it for the owner; `CARRY_CHECK:` means one reviewer from another model family reads the printed delta prompt, then `check --out <dir> --attempt <id> --result <file> --model <id>` (or `--unavailable "<reason>"`).

 ### On Capy

@@ -51,9 +52,8 @@ line yourself; gstack never writes the plan.
  with `gstack-config set autoplan_checkpoint drive`) so a run survives machine replacement: a
  generation is published at rest, after each bound voice and after each phase transition; restore
  with `resume --out <new dir> --from <dir>/<run>`, which never redispatches a bound voice.
- Plans carry `## Before → after`, `## Why it matters`, `## What caused this work`, `## Big decisions`;
-  relay the printed `RELAY: gstack-owner-brief render --captain` output to your Captain (or the person),
-  then `gstack-approvals import --autoplan <dir>` after `answer` (`capy/skill/gstack/captain.md`).
+- Plans carry `## ELI5` (5-10 plain sentences), `## Before → after`, `## Why it matters`, `## What caused this work`, `## Big decisions`;
+  relay the printed `RELAY: gstack-owner-brief render --captain` output to your Captain (or the person), then `gstack-approvals import --autoplan <dir>` after `answer` (`--stack <b1,b2,...>` for a stack; `capy/skill/gstack/captain.md`).

 ### Commands

@@ -23,7 +23,7 @@ line yourself; gstack never writes the plan.

 1. `next` prints one `ATTEMPT: <id> phase=<p> voice=native|outside prompt=<file> result=<file>`
   per voice. Dispatch the native prompt to a subagent on your model and the outside prompt to
-   a subagent on another model family (or `gstack-outside-voice run`). Each writes `result=`.
+   a subagent on another model family (or `gstack-outside-voice run`). Each writes `result=`. The first `next` also prints `REVIEW_ESTIMATE:` (most attempts and reservations the review can take).
 2. Submit each voice exactly once (`BOUND:`). Both voices in closes the phase (`PHASE_CLOSED:`).
 3. Repeat `next` until `status=gate_pending` (`GATE: gate_rev=<n>`); render it with `gstack-gate render <dir>/gate.json --all` and relay every page with the two cost lines printed under `GATE:`: `GATE_SPEND:` (spent, worst case, cap, headroom, attempts; host-subagent attempts are priced unknown and counted at their reservation, so quote their real figure from the host's cost view) and `GATE_REC_COST:` (the phases `rec` can reopen and the estimate for them, an upper bound; `gate_cost` in `--json`).
   You never approve the gate for the owner. Offer `rec` (every recommendation, plan changes included) as the default.
@@ -35,6 +35,7 @@ line yourself; gstack never writes the plan.
 5. `REOPENED: <phases>` means loop `next` again (Eng last). `APPROVED:` means `export` (`DISPATCHES: <n> est_input_tokens=<k>`, relative).
 6. Rounds: one by default; a second only for High or Critical findings; cap 2 (`--max-rounds`). A `revise` whose decisions all resolve below
   High prints `APPLIED: <ids> without re-review (below High)`. At the cap with High open, `ESCALATE: evidence sufficient, decision needed`: relay it, never ask for another round.
+7. Carry-forward ([docs/unattended.md#carry-forward](../docs/unattended.md#carry-forward); `--no-carry-forward` on the first `next` opts out): a later gate that asks nothing new takes the owner's words from the gate before it (`CARRIED:`, then `APPROVED:`); `CARRY: no (<why>)` leaves it for the owner; `CARRY_CHECK:` means one reviewer from another model family reads the printed delta prompt, then `check --out <dir> --attempt <id> --result <file> --model <id>` (or `--unavailable "<reason>"`).

 ### On Capy

@@ -48,9 +49,8 @@ line yourself; gstack never writes the plan.
  with `gstack-config set autoplan_checkpoint drive`) so a run survives machine replacement: a
  generation is published at rest, after each bound voice and after each phase transition; restore
  with `resume --out <new dir> --from <dir>/<run>`, which never redispatches a bound voice.
- Plans carry `## Before → after`, `## Why it matters`, `## What caused this work`, `## Big decisions`;
-  relay the printed `RELAY: gstack-owner-brief render --captain` output to your Captain (or the person),
-  then `gstack-approvals import --autoplan <dir>` after `answer` (`capy/skill/gstack/captain.md`).
+- Plans carry `## ELI5` (5-10 plain sentences), `## Before → after`, `## Why it matters`, `## What caused this work`, `## Big decisions`;
+  relay the printed `RELAY: gstack-owner-brief render --captain` output to your Captain (or the person), then `gstack-approvals import --autoplan <dir>` after `answer` (`--stack <b1,b2,...>` for a stack; `capy/skill/gstack/captain.md`).

 ### Commands

@@ -5,7 +5,7 @@
 *   gstack-approvals add --ask <text> --words <exact words> --by <who> --scope <what it covers>
 *                        [--pending <text>] [--outcome approved|rejected|pending|standing]
 *                        [--date YYYY-MM-DD] [--source <ref>] [--branch <name>] [--json]
- *   gstack-approvals import --autoplan <run dir> [--by <whose words>] [--branch <name>] [--json]
+ *   gstack-approvals import --autoplan <run dir> [--by <whose words>] [--branch <name> | --stack <b1,b2,...>] [--json]
 *   gstack-approvals render [--no-heading] [--branch <name>]
 *   gstack-approvals list [--json] [--branch <name>]
 *   gstack-approvals sync [--branch <name>]
@@ -20,14 +20,15 @@
 */
 import * as path from 'node:path';
 import { EXIT, exitTableHelp } from '../lib/headless-artifacts';
-import { appendRows, buildRow, importAutoplan, locate, readRecord, renderRecord, syncRecord, type RecordError, type StoreLocation } from '../lib/approval-record';
+import { appendRows, branchExists, buildRow, importAutoplan, locate, readRecord, renderRecord, syncRecord, type RecordError, type StoreLocation } from '../lib/approval-record';
 import { resultLine } from '../lib/result-codes';

 const HELP = `gstack-approvals — the per-thread approval record (capy/skill/gstack/captain.md)

  add     --ask <text> --words <exact words> --by <who> --scope <what it covers>
          [--pending <text>] [--outcome approved|rejected|pending|standing] [--date YYYY-MM-DD] [--source <ref>]
-  import  --autoplan <run dir> [--by <whose words, e.g. Garry>]
+  import  --autoplan <run dir> [--by <whose words, e.g. Garry>] [--stack <b1,b2,...>]
+          --stack writes the same rows to every branch of a stack; each branch is checked first, and one unknown branch writes none
  render  [--no-heading]          the ## Approvals section for a PR body or report
  list    [--json]
  sync                            re-mirror the record (GSTACK_APPROVALS_DIR, the Captain's drive approvals/, or the project-user Drive)
@@ -40,7 +41,7 @@ Example:

 ${exitTableHelp()}`;

-const VALUE_FLAGS = new Set(['--ask', '--words', '--by', '--scope', '--pending', '--outcome', '--date', '--source', '--branch', '--autoplan']);
+const VALUE_FLAGS = new Set(['--ask', '--words', '--by', '--scope', '--pending', '--outcome', '--date', '--source', '--branch', '--autoplan', '--stack']);

 function parse(args: string[]): { flags: Record<string, string>; bools: Set<string>; error?: string } {
  const flags: Record<string, string> = {};
@@ -64,11 +65,45 @@ function fail(cmd: string, e: RecordError): number {
  return EXIT.refused;
 }

+/** `import --stack`: every branch checked first, then the same rows (ids are branch-independent) written to each branch's record. */
+function importStack(cmd: string, flags: Record<string, string>, json: boolean): number {
+  if (cmd !== 'import' || !flags['--autoplan'] || flags['--branch'] !== undefined) { process.stderr.write(`gstack-approvals: --stack goes with import --autoplan <run dir>, without --branch\n`); return EXIT.usage; }
+  const branches = [...new Set(flags['--stack']!.split(',').map(b => b.trim()).filter(Boolean))];
+  if (!branches.length) { process.stderr.write(`gstack-approvals: --stack needs <b1,b2,...>\n`); return EXIT.usage; }
+  const cwd = process.cwd();
+  const missing = branches.filter(b => !branchExists(cwd, b));
+  if (missing.length) return fail(cmd, { code: 'APPROVAL_RECORD_NO_BRANCH', message: `no local or origin branch ${missing.join(', ')}; nothing was written` });
+  const runDir = path.resolve(flags['--autoplan']);
+  const results: Array<{ branch: string; added: string[]; skipped: number; file: string; mirror: string | null; error?: string }> = [];
+  let firstError: RecordError | undefined;
+  const plans: Array<{ branch: string; loc: StoreLocation; rows: ReturnType<typeof importAutoplan> & unknown[] }> = [];
+  for (const branch of branches) {
+    const loc = locate({ cwd, env: process.env, branch });
+    if ('code' in loc) return fail(cmd, loc);
+    const rows = importAutoplan(runDir, branch, flags['--by']);
+    if ('code' in rows) return fail(cmd, rows);
+    plans.push({ branch, loc, rows });
+  }
+  for (const { branch, loc, rows } of plans) {
+    const res = appendRows(loc, rows);
+    if ('code' in res) { firstError ??= res; results.push({ branch, added: [], skipped: 0, file: loc.file, mirror: loc.mirror ?? null, error: res.message }); continue; }
+    if (res.mirrorError) firstError ??= res.mirrorError;
+    results.push({ branch, added: res.added.map(r => r.id), skipped: res.skipped, file: loc.file, mirror: loc.mirror ?? null, ...(res.mirrorError ? { error: res.mirrorError.message } : {}) });
+  }
+  if (json) process.stdout.write(JSON.stringify({ branches: results }) + '\n');
+  else {
+    for (const r of results) process.stdout.write(`APPROVAL_IMPORTED: branch=${r.branch} n=${r.added.length} skipped=${r.skipped} ${r.file}${r.mirror ? ` mirror=${r.mirror}` : ''}${r.error ? ` error=${r.error}` : ''}\n`);
+    process.stdout.write(`APPROVAL_IMPORTED_STACK: branches=${results.length} added=${results.reduce((n, r) => n + r.added.length, 0)}\n`);
+  }
+  return firstError ? fail(cmd, firstError) : EXIT.ok;
+}
+
 function main(argv: string[]): number {
  const [cmd, ...rest] = argv;
  if (!cmd || cmd === '--help' || cmd === '-h') { process.stdout.write(HELP + '\n'); return cmd ? EXIT.ok : EXIT.usage; }
  const { flags, bools, error } = parse(rest);
  if (error) { process.stderr.write(`gstack-approvals: ${error}\n${HELP}\n`); return EXIT.usage; }
+  if (flags['--stack'] !== undefined) return importStack(cmd, flags, bools.has('--json'));
  const loc = locate({ cwd: process.cwd(), env: process.env, branch: flags['--branch'] });
  if ('code' in loc) return fail(cmd, loc);
  const json = bools.has('--json');
@@ -19,13 +19,14 @@
 *   gstack-autoplan release --out <dir> --attempt <id>  drop a reservation for an attempt that never ran
 *   gstack-autoplan close   --out <dir> --phase <p> --deferred "<reason>" --by <human|agent:<id>>   close a phase accepted-with-findings
 *   gstack-autoplan budget  --out <dir> [--spend-cap <usd>] [--max-rounds <n>] --by <human|agent:<id>>   the owner's journaled cap change
+ *   gstack-autoplan check   --out <dir> --attempt <id> --result <file> --model <id> | --unavailable <reason>   the delta check before a carry
 *   gstack-autoplan checkpoints --from <dir>            list checkpoint generations, newest first, with hash ok|bad
 *   gstack-autoplan contract                            print the coordinator contract every lane prompt inherits (plan E2)
 *
 * Logic: lib/autoplan-run.ts (next/submit/resume), lib/autoplan-gate.ts (close, gate, answer),
 * lib/autoplan-export.ts, lib/autoplan-prompts.ts, lib/spend-ledger.ts. Codes: lib/result-codes.ts.
 * Workflow commands end with GSTACK_RESULT: skill=autoplan status=<s> run=<dir>; the run is never approved here.
- * Mutating verbs (next, submit, answer, recommend, revise, close, budget, export, run) pass admission first (lib/autoplan-admission.ts)
+ * Mutating verbs (next, submit, answer, recommend, revise, close, budget, check, export, run) pass admission first (lib/autoplan-admission.ts)
 * and bring an older run directory forward (lib/autoplan-migrate.ts) under the run lock.
 */
 import * as fs from 'node:fs';
@@ -33,6 +34,7 @@ import * as path from 'node:path';
 import { amendmentSteps, answerGate, closeDeferred, finishGate, recommendGate } from '../lib/autoplan-gate';
 import { gateCost, gateCostPayload, renderGateCost, type GateCost } from '../lib/autoplan-gate-cost';
 import { attemptEstimate, changeBudget, escalateLine, liftStop } from '../lib/autoplan-budget';
+import { DEFAULT_CARRY_CHECK_CAP_USD, carryLines, submitCarryCheck } from '../lib/autoplan-carry';
 import { ANSWERED_BY_RE } from '../lib/gate-list';
 import { checkpointLines, checkpointRun, restoreCheckpoint, type CheckpointMoment } from '../lib/autoplan-checkpoint';
 import { admit, touchInstallLease } from '../lib/autoplan-admission';
@@ -60,6 +62,9 @@ DRIVER: ${driverFile()}
          --estimate-usd: reserved per new attempt; unattended Capy runs reserve a default for host attempts so --spend-cap binds
          --scaffold (first next): a plan missing one of its two sections is reviewed from <out>/plan-source.md with the heading added
          --light loads each phase's checklist plus the sections its scope needs (the unattended default); interactive kinds load everything
+          --no-carry-forward (first next): every later gate revision goes to the owner, even one that asks nothing new
+          --carry-check-cap <usd> (first next): the most one delta check may reserve before a carry (default ${DEFAULT_CARRY_CHECK_CAP_USD.toFixed(2)})
+          the first next prints REVIEW_ESTIMATE: the most the review can reserve at this scope and round cap
          prepare the current phase; one ATTEMPT: line per voice; ends awaiting_result
  submit  --out <dir> --phase <p> --voice native|outside --result <file> | --record <file> [--attest <reason>]
          --model <id> [--attempt <id>] [--runner <r>] [--usage-usd <n>] [--reconcile <jsonl>] [--json]
@@ -84,6 +89,9 @@ DRIVER: ${driverFile()}
  resume  --out <dir> [--from <checkpoint>] [--json]   --from restores a hash-verified checkpoint into an empty --out first
          (a collection or root restores its newest valid generation; GSTACK_AUTOPLAN_CHECKPOINT_DIR set: every command
          leaving the run at rest publishes a generation under <dir>/<run>/ and prints CHECKPOINT: on stderr)
+  check   --out <dir> --attempt <id> --result <file> --model <id> [--same-family "<why>"] | --unavailable "<reason>"
+          bind the one-reviewer delta check a CARRY_CHECK: line asked for: nothing at High or above carries the owner's
+          last answer (CARRIED:); otherwise, or when unavailable, the gate goes to the owner as before
  checkpoints --from <dir>          list checkpoint generations, newest first: run, gate_rev, time, hash ok|bad, current
  close   --out <dir> --phase <p> --deferred "<reason>" --by <human|agent:<id>>
          close the phase accepted-with-findings: open voices skipped (dispatched attempts settle unknown at their
@@ -180,7 +188,10 @@ function doNext(out: string, rest: string[], json: boolean): number {
    const ui = rest.includes('--ui') ? true : rest.includes('--no-ui') ? false : undefined;
    const maxRounds = num(rest, '--max-rounds');
    if (maxRounds !== undefined && (!Number.isInteger(maxRounds) || maxRounds < 1)) throw new RunError('ARTIFACT_SCHEMA', '--max-rounds needs an integer >= 1', 1);
+    const carryCheckCapUsd = num(rest, '--carry-check-cap');
+    if (carryCheckCapUsd !== undefined && !(carryCheckCapUsd >= 0)) throw new RunError('ARTIFACT_SCHEMA', '--carry-check-cap needs a number >= 0', 1);
    initRun({ plan, out, cwd, planHeading: flag(rest, '--plan-heading'), outsideEnabled: !rest.includes('--no-outside'), spendCap: num(rest, '--spend-cap') ?? null, maxRounds,
+      carryForward: !rest.includes('--no-carry-forward'), carryCheckCapUsd,
      deadlineMin: num(rest, '--deadline'), light: rest.includes('--light'), scaffold: rest.includes('--scaffold'), scope: { ui, developerTool: rest.includes('--developer-tool'), agentPrimary: rest.includes('--agent-primary'), dx: rest.includes('--no-dx') ? false : undefined, ...(rest.includes('--docs-only') ? { docsOnly: true } : {}) } }, initLines);
  }
  const redispatch = rest.flatMap((a, i) => (a === '--redispatch' && rest[i + 1] ? [rest[i + 1]!] : []));
@@ -257,7 +268,7 @@ function doClose(out: string, rest: string[], json: boolean): number {
  const r = closeDeferred(state, phase as Phase, { reason: reason.trim(), by, cause: 'owner' });
  moment = 'transition';
  const lines = [...r.lines, ...(stopped ? liftStop(state, `${phase} closed deferred by ${by}`) : [])];
-  if (!currentPhase(state)) { finishGate(state); lines.push(...gateLines(state, out)); }
+  if (!currentPhase(state)) { finishGate(state); lines.push(...carryLines(state), ...(state.status === 'gate_pending' ? gateLines(state, out) : [])); }
  else { state.status = 'running'; lines.push(`NEXT: gstack-autoplan next --out ${out}`); }
  writeState(state);
  lines.push(resultLineFor(state.status, out));
@@ -265,6 +276,19 @@ function doClose(out: string, rest: string[], json: boolean): number {
  return EXIT.ok;
 }

+/** `check`: the delta reviewer's verdict before a carry (lib/autoplan-carry.ts). */
+function doCheck(out: string, rest: string[], json: boolean): number {
+  const unavailable = flag(rest, '--unavailable');
+  const result = flag(rest, '--result');
+  if (unavailable === undefined && (!result || !flag(rest, '--model'))) return usage('check needs --attempt <id> and --result <file> --model <id>, or --unavailable "<reason>"');
+  const r = submitCarryCheck({ out, attempt: flag(rest, '--attempt'), result: result ? path.resolve(result) : undefined, model: flag(rest, '--model'), unavailable, sameFamily: flag(rest, '--same-family') });
+  moment = 'transition';
+  const status = r.state.status;
+  const lines = [...r.lines, ...(status === 'gate_pending' ? gateLines(r.state, out) : []), ...(status === 'amend_pending' || status === 'running' ? [nextLine(r.state, out)] : []), resultLineFor(status, out)];
+  emit(json, lines, { run: r.state.run, status, gate_rev: r.state.gate_rev ?? null, carry: r.state.carry ?? null, out });
+  return EXIT.ok;
+}
+
 function doBudget(out: string, rest: string[], json: boolean): number {
  const state = readState(out);
  const lines = changeBudget(state, { spendCap: num(rest, '--spend-cap'), maxRounds: num(rest, '--max-rounds'), by: flag(rest, '--by') });
@@ -338,6 +362,7 @@ async function main(argv: string[]): Promise<number> {
    if (cmd === 'revise') return mutating(out, json, () => doRevise(out, rest, json));
    if (cmd === 'close') return mutating(out, json, () => doClose(out, rest, json));
    if (cmd === 'budget') return mutating(out, json, () => doBudget(out, rest, json));
+    if (cmd === 'check') return mutating(out, json, () => doCheck(out, rest, json));
    if (cmd === 'run') { const admission = admit(process.env); if (admission.lines.length) process.stdout.write(admission.lines.join('\n') + '\n'); return await doRun(out, rest, json); }
    if (cmd === 'status') {
      const state = readState(out);
@@ -416,7 +441,7 @@ async function main(argv: string[]): Promise<number> {
    if (e instanceof RunError) {
      process.stderr.write(resultLine(`gstack-autoplan ${cmd}`, e.code, e.detail) + '\n');
      if (e.after.lines?.length) process.stdout.write(e.after.lines.join('\n') + '\n');
-      if (['next', 'submit', 'answer', 'resume', 'run', 'revise', 'recommend', 'close', 'budget'].includes(cmd)) process.stdout.write(resultLineFor(e.after.status ?? 'refused', out) + '\n');
+      if (['next', 'submit', 'answer', 'resume', 'run', 'revise', 'recommend', 'close', 'budget', 'check'].includes(cmd)) process.stdout.write(resultLineFor(e.after.status ?? 'refused', out) + '\n');
      return e.exit === 3 ? EXIT.refused : EXIT.fail;
    }
    process.stderr.write(`gstack-autoplan ${cmd}: ${e?.message ?? e}\n`);
@@ -17,7 +17,8 @@
 * and a cell whose first byte a spreadsheet would run as a formula (=, +, -, @)
 * gets a leading apostrophe, so evidence taken from PR titles, filenames or
 * generated text cannot execute when a reviewer opens the file. License text:
- * licenses/MIT-pstack.txt; NOTICE.md carries the entry.
+ * licenses/MIT-pstack.txt; NOTICE.md carries the entry. The append rules live in
+ * lib/decision-trail.ts, shared with the autoplan runner.
 *
 * The start-row rule is gstack's mechanisation of the same skill's prose: a run
 * (one agent conversation, with its later turns) that writes to a trail already
@@ -26,11 +27,8 @@
 * id, and an append with `--run <id>` refuses until that run's start row is the
 * newest start row in the file (DECISION_TRAIL_START_REQUIRED, exit 3).
 */
-import * as fs from "node:fs";
-import * as path from "node:path";
+import { appendTrailRow as appendRow, readTrailRows as readRows, type TrailRow as Row } from "../lib/decision-trail";

-const COLUMNS = ["ts", "phase", "decision", "why", "evidence", "result"] as const;
-const HEADER = COLUMNS.join("\t") + "\n";
 const USAGE = `Usage:
  gstack-decision-trail [--run <id>] <file> <phase> <decision> <why> <evidence> <result>
  gstack-decision-trail start <file> <run-id>
@@ -56,58 +54,6 @@ letters, digits and . _ : @ / - (a Capy jam_ id, a session id, a PR number).
 when the trail has rows and the newest start row is another run's.
 `;

-type Row = { line: number; cells: string[] };
-
-function clean(cell: string): string {
-  const flat = cell.replace(/[\t\n\r]/g, " ");
-  return /^[=+\-@]/.test(flat) ? `'${flat}` : flat;
-}
-
-function readRows(file: string): Row[] {
-  if (!fs.existsSync(file)) return [];
-  const lines = fs.readFileSync(file, "utf8").split("\n");
-  const rows: Row[] = [];
-  lines.forEach((raw, i) => {
-    let text = raw.endsWith("\r") ? raw.slice(0, -1) : raw;
-    if (i === 0 && text.startsWith("\uFEFF")) text = text.slice(1);
-    // A repeated header (the append-on-flaky-mount case) is never a row.
-    if (text === "" || text === HEADER.trimEnd()) return;
-    rows.push({ line: i + 1, cells: text.split("\t") });
-  });
-  return rows;
-}
-
-function appendRow(file: string, cells: readonly string[]): number {
-  const dir = path.dirname(file);
-  if (dir && dir !== "." && !fs.existsSync(dir)) fs.mkdirSync(dir, { recursive: true });
-  // Append the header too, never truncate: on a flaky mount the worst case is
-  // one stray header line, not lost rows.
-  // A file with no header line anywhere (empty, blank, or rows written by hand)
-  // gets one appended once; readRows skips a header line wherever it sits.
-  const size = fs.existsSync(file) ? fs.statSync(file).size : 0;
-  if (size > 0 && !endsWithNewline(file, size)) fs.appendFileSync(file, "\n");
-  if (size === 0 || !hasHeader(file)) fs.appendFileSync(file, HEADER);
-  const ts = new Date().toISOString().replace(/\.\d{3}Z$/, "Z");
-  fs.appendFileSync(file, [ts, ...cells.map(clean)].join("\t") + "\n");
-  return readRows(file).length;
-}
-
-function hasHeader(file: string): boolean {
-  return fs.readFileSync(file, "utf8").replace(/^\uFEFF/, "").split("\n")
-    .some(line => line.replace(/\r$/, "") === HEADER.trimEnd());
-}
-
-function endsWithNewline(file: string, size: number): boolean {
-  const fd = fs.openSync(file, "r");
-  try {
-    const last = Buffer.alloc(1);
-    fs.readSync(fd, last, 0, 1, size - 1);
-    return last[0] === 0x0a;
-  } finally {
-    fs.closeSync(fd);
-  }
-}
-
 const RUN_ID = /^[A-Za-z0-9._:@/-]+$/;
 function checkRunId(runId: string): void {
  if (!RUN_ID.test(runId)) fail(`run id must match ${RUN_ID} (letters, digits, . _ : @ / -), got ${JSON.stringify(runId)}`, 2);
@@ -14,7 +14,7 @@ Spend cap: <focused paid runs per PR, from SPEND>

 1. Plan (when Plan: required). After the session-start install check, install for autoplan
   and follow autoplan/DRIVER.md "On Capy" (launch file sourced, GSTACK_AUTOPLAN_CHECKPOINT_DIR
-   set, run dir under ~/.capy/work). Write the plan with its ## Before → after,
+   set, run dir under ~/.capy/work). Write the plan with its ## ELI5, ## Before → after,
   ## Why it matters, ## What caused this work, ## Big decisions and ## Implementation plan
   sections, then run gstack-autoplan next --out <run dir> --plan <plan.md> <scope flags>
   sized to this brief (a one-file fix: --no-ui --no-dx --max-rounds 1 --estimate-usd <n>
@@ -25,7 +25,8 @@ Spend cap: <focused paid runs per PR, from SPEND>
   (gstack-autoplan answer --out <run dir> --gate-rev <n> --reply <tokens> --rec-hash <h>
   --by agent:<your thread> --owner-reply "<their words>"), incorporate every accepted plan
   change (revise), finish every reopened phase, and start product edits only when the run
-   reports APPROVED. Then gstack-approvals import --autoplan <run dir> --by "<owner>".
+   reports APPROVED (a later gate that asks nothing new prints CARRIED: and needs no second
+   ask). Then gstack-approvals import --autoplan <run dir> --by "<owner>" (--stack b1,b2 for a stack).
   New scope you take on part-way goes back to step 1 with its own plan.
 3. Ship. CHANGELOG entry under [Unreleased]; GUARD_CHECKS, pregate and /ship at the final
   head; one unstamped READY block (captain/templates/READY.md) to me; the PR body carries
@@ -7,6 +7,9 @@ shape from an autoplan run; write it by hand for any other plan.
 ```text
 Decision needed on <thread>: <one-line title>. Reply **rec** to take every recommendation.

+## ELI5
+<5-10 plain sentences: what changes, for whom, and what it costs>
+
 ## Before → after
 <what the user or maintainer sees today; what they will see after; a command and its
 output, a screen, a cost or time number>
@@ -48,6 +48,7 @@ gstack-approvals add --ask "Spend up to \$40 on a paid eval rerun for the ship c
 gstack-approvals add --outcome standing --ask "Merge #3330 when green" \
  --words "yes merge when green" --by "Garry, 2026-10-16 (Captain rule)" --scope "every crew PR, through a train"
 gstack-approvals import --autoplan <run dir> --by Garry   # after gstack-autoplan answer; idempotent
+gstack-approvals import --autoplan <run dir> --by Garry --stack b1,b2   # a stacked PR set: every branch, or none
 gstack-approvals render                                    # the ## Approvals section
 ```

@@ -93,15 +94,15 @@ project can follow it:
   the plan names (docs, trains, cleanups) included, and carry them through to done with no further ask.
   Say in one line what it covers. For an autoplan gate, "do it" answers as `rec`; record the words.

-For an autoplan run, write parts 1 to 4 as `## Before → after`, `## Why it matters`,
-`## What caused this work` and `## Big decisions` sections in the plan. Then, at the gate:
+For an autoplan run, open the plan with `## ELI5` (5-10 plain sentences), then write parts 1 to 4 as
+`## Before → after`, `## Why it matters`, `## What caused this work` and `## Big decisions` sections. Then, at the gate:

 ```bash
 gstack-owner-brief render --captain --out <run dir>
 ```

-It prints a header line ("Decision needed … Reply **rec** …"), the four plan sections (each capped, with a
-pointer to the plan for the rest), and "What we need from you", which lists every choice `rec` takes, in
+It prints a header line ("Decision needed … Reply **rec** …"), the `### ELI5` summary (on a run whose gate
+carries `eli5`), the four plan sections (each capped, with a pointer to the plan for the rest), and "What we need from you", which lists every choice `rec` takes, in
 words, and closes with the "do it" reply, then a `### Cost` section (spent so far, worst case, and the estimate for
 the phases `rec` can reopen). The last line, "For the relaying agent", carries the run, `gate_rev`, `rec_hash` and the exact
 `gstack-autoplan answer` command. Relay everything above that line. When the reply comes back:
@@ -112,6 +113,11 @@ gstack-autoplan answer --out <run dir> --gate-rev <n> --reply rec --rec-hash <h>
 gstack-approvals import --autoplan <run dir> --by <name>
 ```

+A later gate that asks nothing new is answered for you with the words you recorded (`CARRIED:`, then `APPROVED:`);
+the brief says "Nothing pending. Carried forward", so import and build without a second ask. `CARRY_CHECK:` means
+one reviewer from another model family reads the unreviewed delta first (`gstack-autoplan check`). Anything
+else prints `CARRY: no (<why>)` and goes to the person as today (`docs/unattended.md#carry-forward`).
+
 If the person answers in words ("split it into two PRs"), map the words to tokens, answer, and confirm in one
 line what you recorded ("Recorded: split into two PRs; everything else as recommended"). A plan missing a
 section renders "Not ready for approval" with no reply invitation, and exits 1 (`CAPTAIN_BRIEF_INCOMPLETE`):
@@ -138,8 +138,9 @@ this rule shipped cites this section by name.
 **What the lane sends.** The plan and its gate in plain language a person outside the
 project can follow: what changes, why now, what the change does, what it costs, the lane's
 pick, and the recommended one-word reply. `gstack-owner-brief render --captain --out <run dir>`
-prints that from the plan's `## Before → after`, `## Why it matters`,
-`## What caused this work` and `## Big decisions` sections, with the cost lines
+prints that from the plan's `## ELI5`, `## Before → after`, `## Why it matters`,
+`## What caused this work` and `## Big decisions` sections (also saved as `eli5` in
+`gate.json`, so `gstack-gate render` opens with them), with the cost lines
 (`GATE_SPEND:`, `GATE_REC_COST:`) and a last line for the controller; a plan missing a
 section exits `CAPTAIN_BRIEF_INCOMPLETE`, so the lane adds the section and renders again
 ([capy/skill/gstack/captain.md](../capy/skill/gstack/captain.md) § 3 lists them). The lane
@@ -156,10 +157,12 @@ controller and been answered into the run
 every accepted plan change has been incorporated (`revise`), every reopened phase has
 finished, and the run reports `APPROVED` (`status=complete` after `export`). An answered
 gate that still prints `PLAN_AMEND_PENDING` or `REOPENED:` is not a completed plan, and an
-imported approval row alone is not completion.
+imported approval row alone is not completion. A later gate revision that asks nothing new
+is answered with the owner's recorded words and says `CARRIED:`; that is the owner's word,
+not a new ask ([carry-forward](unattended.md#carry-forward)).

-**The record.** `gstack-approvals import --autoplan <run dir> --by "<owner>"` writes the
-approval row (date, the ask, the owner's exact words, the lane that answered as relayer,
+**The record.** `gstack-approvals import --autoplan <run dir> --by "<owner>"` (with
+`--stack <b1,b2,...>` for a stacked PR set) writes the approval row (date, the ask, the owner's exact words, the lane that answered as relayer,
 what it covers); `gstack-approvals render` puts it in the report and the PR body. Before
 training the PR the controller compares that row's words and `gate_rev` with the ask it
 relayed and the reply it carried back; a row that does not match is not READY (shape A).
@@ -1737,8 +1737,11 @@ restate them. Rotate the secret if it was real.

 **Meaning.** The record is kept per branch (one Capy thread ships one branch), and
 HEAD is detached or the working directory is not a git repository. Exit 3.
+From `import --autoplan <dir> --stack <b1,b2,...>` it means a listed branch exists
+neither locally nor on origin; the line names it, and no branch's record was written.

-**Fix.** Check out the branch the work ships from, or pass `--branch <name>`.
+**Fix.** Check out the branch the work ships from, or pass `--branch <name>`. For
+`--stack`, correct or fetch the named branch and import again.

 <a id="approval-record-unavailable"></a>
 ### `gstack-approvals: ... (APPROVAL_RECORD_UNAVAILABLE)`
@@ -1777,7 +1780,9 @@ import again. Imports are idempotent, so a retry never records an answer twice.
 ### `gstack-owner-brief render --captain: ... (CAPTAIN_BRIEF_INCOMPLETE)`

 **Meaning.** The plan is missing at least one section the brief needs: `## Before → after`,
-`## Why it matters`, `## What caused this work`, `## Big decisions`. The brief still prints,
+`## Why it matters`, `## What caused this work`, `## Big decisions`, and `## ELI5` (alias
+`## In plain words`) on a run whose `gate.json` carries an `eli5` field (gstack 1.91.305 and
+later; older runs render as before). The brief still prints,
 but it opens with "Not ready for approval" and has no reply invitation. Exit 1.

 **Fix.** Add the named sections to the plan (`gstack-autoplan revise --plan <file>` if
@@ -495,6 +495,53 @@ by its runner. Result codes: `EXECUTION_UNKNOWN`, `RUN_LOCKED`,
  decision ids are renamed and journaled); one written by a newer gstack is
  refused.

+### Carry-forward
+
+A gate revision after the first (`gate_rev` 2 and later, written after a
+`revise`) is answered with the owner's recorded words from the revision before
+it when nothing new needs their decision. `gstack-autoplan` prints
+`CARRIED: gate_rev=<n> <- gate_rev=<m> words="<their words>" by=<relayer> reason=<why>`,
+then the usual `APPROVED:` lines, and journals `gate_carried`. The carry needs
+all of: the earlier gate approved (`p1a` or `p1b`) with `rec` or "do it"
+(not item tokens), no item there left pending, and the owner's words
+recorded; that gate was not itself carried; every accepted change incorporated
+or settled; the plan file's bytes equal those recorded at the gate or at the
+`revise` that took the changes (any later edit refuses, an ELI5-only one
+included); no urgent or
+outside-plan finding open; and every new gate item one of the cases below.
+Anything else prints `CARRY: no (<why>)` (journal `carry_refused`) and the gate
+goes to the owner as before. `--no-carry-forward` on the first `next` turns it
+off for the run.
+
+- **"do it".** When the owner's words were "do it" (their words, not a
+  `captain:` relay of them), a new `auto` item that changes no scope, cost or
+  direction, whose recommended option changes no plan text, and is no round
+  or budget grant takes its recommendation. The
+  decision row gets `carried_under: do it`, the output prints
+  `CARRIED_UNDER_DO_IT:` per item, and the approval record lists them as
+  "taken under "do it", not seen by the owner". `rec` carries no new item.
+- **Unreviewed post-cap delta.** A "plan changed after <phase> cap" item
+  recommending `a` (accept) is read by one reviewer before it carries. `CARRY_CHECK:` names the attempt and
+  the prompt (the delta against `plan-at-gate.r<m>.md`, the accepted changes,
+  the full plan); the run reserves the estimate in `spend.json` and refuses
+  when it is over `--carry-check-cap <usd>` (default 5) or the run's cap.
+  Dispatch it to a model family other than the plan reviewers', then
+  `check --out <dir> --attempt <id> --result <file> --model <id>`
+  (`--same-family "<why>"` when no other family is available). A High or
+  Critical finding, or `--unavailable "<reason>"`, leaves the gate for the
+  owner. The verdict is saved as `carry_check` on `gate.json`, on the
+  decision rows (`carry_check`) and in `<dir>/decisions.tsv`. A pending
+  check is void, its reservation released and the gate left for the owner,
+  when the owner answers the gate, the plan is revised, a new gate is
+  written, or the delta changed after dispatch.
+
+Carried rows keep `answered_by: agent:gstack-autoplan` and name their source
+(`carried_gate_from`, `carried_gate_rev`, `carried_answered_by`);
+`gstack-approvals import --autoplan` writes the carried gate as its own row,
+and `gstack-owner-brief render --captain` opens with "Nothing pending.
+Carried forward from gate_rev <m>", followed by any items taken under "do it"
+without being shown to the owner.
+
 ## Owner brief

 `gstack-owner-brief render --out <dir>` is what the owner sees at the final
@@ -52,6 +52,11 @@ export type Env = NodeJS.ProcessEnv;

 const git = (cwd: string, args: string[]) => spawnSync('git', args, { cwd, encoding: 'utf8', timeout: 10_000 });

+/** A branch this checkout knows: a local head or an origin remote-tracking ref. */
+export function branchExists(cwd: string, branch: string): boolean {
+  return [`refs/heads/${branch}`, `refs/remotes/origin/${branch}`].some(ref => git(cwd, ['rev-parse', '--verify', '--quiet', ref]).status === 0);
+}
+
 /** One contained path segment for a branch: readable prefix plus a hash of the exact ref. */
 export function branchKey(branch: string): string {
  const readable = branch.replace(/[^A-Za-z0-9._-]+/g, '-').replace(/^[.-]+/, '').slice(0, 60) || 'branch';
@@ -226,13 +231,15 @@ export function renderRecord(read: ReadResult, loc: StoreLocation, o: { heading?
  return [...head, ...lines, '', `Source: gstack-approvals record for ${cell(loc.branch)} (display only; not an approval citation).`, ''].join('\n');
 }

-interface DecisionLike { id?: string; title?: string; status?: string; chosen?: string | null; recommended?: string | null; gate_rev?: number; answered_at?: string; answered_by?: string; reply?: string; owner_reply?: string; carried_from?: string; label?: string; chosen_text?: string }
+interface DecisionLike { id?: string; title?: string; status?: string; chosen?: string | null; recommended?: string | null; gate_rev?: number; answered_at?: string; answered_by?: string; reply?: string; owner_reply?: string; carried_from?: string; label?: string; chosen_text?: string;
+  carried_gate_from?: string; carried_gate_rev?: number; carried_answered_by?: string; carried_under?: string; carry_check?: string }

 const stripRef = (t: string) => t.replace(/\s*\[[^\]]*\]\s*$/, '').trim();

 /**
 * Rows for every answered gate in an autoplan run: one row per answer event (gate_rev, time, answerer), covering
- * the items it settled. Carried-forward answers were recorded when first given and are skipped. Ids are stable,
+ * the items it settled. Item answers carried from an earlier revision were recorded when first given and are skipped;
+ * a whole gate carried forward (lib/autoplan-carry.ts) gets its own row naming the revision its words came from. Ids are stable,
 * so a rerun (or a retry after an interrupted import) adds nothing twice.
 */
 export function importAutoplan(runDir: string, branch: string, by?: string): ApprovalRow[] | RecordError {
@@ -265,10 +272,16 @@ export function importAutoplan(runDir: string, branch: string, by?: string): App
    const changed = items.filter(r => r.chosen && r.recommended && r.chosen !== r.recommended);
    const stillPending = rows.filter(r => r.gate_rev === rev && r.status === 'pending');
    const answerer = first.answered_by ?? 'unknown';
-    const who = answerer === 'human' ? (by ? `${by}, typed at the gate` : 'the owner, typed at the gate') : `${by ?? 'the owner'}, relayed by ${answerer}`;
-    const scope = `the reviewed plan (sha256 ${planSha}), ${changed.length ? `${changed.length} recommendation(s) changed: ${changed.map(r => stripRef(r.title ?? '')).join('; ')}` : 'every recommendation taken'}`;
+    const carried = p1?.carried_gate_from ? p1 : undefined;
+    const relayed = carried?.carried_answered_by && carried.carried_answered_by !== 'human' ? `, relayed by ${carried.carried_answered_by}` : '';
+    const who = carried ? `${by ?? 'the owner'} (words from gate_rev ${carried.carried_gate_rev}${relayed}), carried forward by gstack-autoplan`
+      : answerer === 'human' ? (by ? `${by}, typed at the gate` : 'the owner, typed at the gate') : `${by ?? 'the owner'}, relayed by ${answerer}`;
+    const unseen = items.filter(r => r.carried_under).map(r => stripRef(r.title ?? ''));
+    const checked = [...new Set(items.map(r => r.carry_check).filter((c): c is string => !!c))];
+    const scope = [`the reviewed plan (sha256 ${planSha}), ${changed.length ? `${changed.length} recommendation(s) changed: ${changed.map(r => stripRef(r.title ?? '')).join('; ')}` : 'every recommendation taken'}`,
+      unseen.length ? `taken under "do it", not seen by the owner: ${unseen.join('; ')}` : '', checked.length ? `post-cap delta checked by ${checked.join(', ')}` : ''].filter(Boolean).join('; ');
    const built = buildRow({
-      ask: `Approve the plan "${planTitle}" (autoplan run ${runId}, ${items.length + (p1 ? 1 : 0)} decisions)`.slice(0, CAPS.ask),
+      ask: `Approve the plan "${planTitle}" (autoplan run ${runId}, ${carried ? `gate_rev ${rev} carried forward from gate_rev ${carried.carried_gate_rev}, ` : ''}${items.length + (p1 ? 1 : 0)} decisions)`.slice(0, CAPS.ask),
      words: first.owner_reply ?? first.reply ?? '(no words recorded)',
      by: who.slice(0, CAPS.by),
      scope: scope.slice(0, CAPS.scope),
@@ -0,0 +1,252 @@
+/**
+ * autoplan-carry — carry the owner's last answer to a later gate revision (GSTA-155,
+ * docs/unattended.md#carry-forward). When a gate_rev N+1 asks nothing the owner has not
+ * already decided, the runner answers it with the owner's recorded gate_rev N words instead
+ * of a second round trip, and records that it did (decisions.jsonl `carried_gate_from`,
+ * journal `gate_carried`, `<out>/decisions.tsv`). It never invents an answer: every
+ * carried answer is the owner's own earlier words, and every refusal says which condition failed.
+ *
+ * A carry needs all of: carry not turned off (`--no-carry-forward`); gate_rev N approved (p1 a or b)
+ * with the owner's words recorded, and not itself a carry; every change accepted at N
+ * incorporated or settled; the plan file unchanged since the gate or the revise that took those
+ * changes, and the reviewed plan equal to it; no open urgent finding; and every new item already
+ * carried. Two kinds of new item do not block: items a "do it" covers (the owner's words are
+ * "do it": an auto item that changes no scope, cost or direction and is no round grant), and a
+ * capped-delta item recommending "accept", which first gets one reviewer on just the delta
+ * (`CARRY_CHECK:`, answered with `gstack-autoplan check`); a High or Critical finding there
+ * means no carry, and the gate goes to the owner as before.
+ */
+import { createHash } from 'node:crypto';
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+import { attemptEstimate } from './autoplan-budget';
+import { lineDiff } from './autoplan-confirm';
+import { answerGate } from './autoplan-gate';
+import { driftLine } from './autoplan-plan-text';
+import { BOUNDARY, RESULT_FORMAT, REVIEWER_TRAILER } from './autoplan-prompts';
+import { parseReviewResult } from './autoplan-reconcile';
+import { FILES, RunError, journal, now, readDecisions, readJournal, readState, writeDecisions, writeState, type RunState } from './autoplan-state';
+import { appendTrailRow } from './decision-trail';
+import { decisionId, parseReply, type GateItem, type GateList } from './gate-list';
+import { readJsonl } from './headless-artifacts';
+import { modelFamily } from './outside-voice-runner';
+import { release, reserve, settle } from './spend-ledger';
+
+export const DEFAULT_CARRY_CHECK_CAP_USD = 5;
+const HIGH = ['Critical', 'High'];
+const DO_IT = /^\s*(?:(?![^:]*\bcaptain\b)[a-z]+(?: [a-z]+){0,2}:\s*)?do it[\s.!]*$/i;
+/** The owner's verbatim words are a "do it" (optionally after their name), not a Captain's words or an item token (`p3b: do it`) naming one. */
+export const isDoIt = (words: string | undefined): boolean => !!words && DO_IT.test(words);
+
+export interface CarryCheck {
+  attempt: string; prompt: string; result: string; delta_sha256: string; items: string[]; estimate_usd: number;
+  status: 'dispatched' | 'passed' | 'failed' | 'unavailable'; model?: string; max_severity?: string; findings?: number; at: string;
+}
+export interface CarryRecord {
+  gate_rev: number; from_gate_rev?: number; verdict: 'carried' | 'refused' | 'check_pending';
+  reason: string; words?: string; under_do_it?: string[]; check?: CarryCheck; lines: string[]; at: string;
+}
+
+const sha256 = (text: string | Buffer) => createHash('sha256').update(text).digest('hex');
+const planFile = (state: RunState) => state.plan_input ?? state.plan_source;
+const fileSha = (file: string) => (fs.existsSync(file) ? sha256(fs.readFileSync(file)) : undefined);
+export const atGateCopy = (out: string, rev: number) => path.join(out, `plan-at-gate.r${rev}.md`);
+
+/** What finishGate journals beside gate_written: the plan file's bytes and a copy of the reviewed plan the owner is asked about. */
+export function gateEvidence(state: RunState, rev: number): { source_file_sha256?: string } {
+  if (fs.existsSync(state.active_plan)) fs.copyFileSync(state.active_plan, atGateCopy(state.out, rev));
+  return { source_file_sha256: fileSha(planFile(state)) };
+}
+
+/** The lines the latest carry decision printed, for the command that wrote this gate revision. */
+export function carryLines(state: RunState): string[] {
+  const c = state.carry;
+  return c && c.gate_rev === state.gate_rev ? c.lines : [];
+}
+
+function record(state: RunState, c: Omit<CarryRecord, 'at'>): CarryRecord {
+  const full = { ...c, at: now() };
+  state.carry = full;
+  journal(state.out, { event: c.verdict === 'carried' ? 'gate_carried' : c.verdict === 'check_pending' ? 'carry_check_dispatched' : 'carry_refused', gate_rev: c.gate_rev, from_gate_rev: c.from_gate_rev ?? null, reason: c.reason, ...(c.under_do_it?.length ? { under_do_it: c.under_do_it } : {}), ...(c.check ? { check: c.check.attempt, check_status: c.check.status } : {}) });
+  appendTrailRow(path.join(state.out, 'decisions.tsv'), ['gate', c.verdict === 'carried' ? `carried gate_rev ${c.gate_rev} from gate_rev ${c.from_gate_rev}` : c.verdict === 'check_pending' ? `delta check before carrying gate_rev ${c.gate_rev}` : `no carry for gate_rev ${c.gate_rev}`,
+    c.reason, c.check ? path.basename(c.check.result) : 'decisions.jsonl', c.check?.status ?? c.verdict]);
+  return full;
+}
+const refuse = (state: RunState, rev: number, from: number | undefined, reason: string) => record(state, { gate_rev: rev, from_gate_rev: from, verdict: 'refused', reason, lines: [`CARRY: no (${reason})`] });
+
+/** Drop a delta check still waiting for its reviewer (the owner answered, the plan was revised, a new gate was written) and release its reservation. */
+export function voidPendingCheck(state: RunState, why: string): string[] {
+  const c = state.carry;
+  if (c?.verdict !== 'check_pending' || !c.check) return [];
+  release(path.join(state.out, FILES.spend), c.check.attempt);
+  const out = refuse(state, c.gate_rev, c.from_gate_rev, `delta check ${c.check.attempt} void: ${why}`);
+  out.check = { ...c.check, status: 'unavailable' };
+  return out.lines;
+}
+
+/** The reviewed plan's changes since the copy taken when the owner was asked at `rev`, and their hash. */
+function deltaSince(state: RunState, rev: number): { diff: string; sha: string } | undefined {
+  const before = atGateCopy(state.out, rev);
+  if (!fs.existsSync(before)) return undefined;
+  const diff = lineDiff(fs.readFileSync(before, 'utf8'), fs.readFileSync(state.active_plan, 'utf8')).filter(l => l.op !== ' ').map(l => `${l.op} ${l.text}`).join('\n');
+  return { diff, sha: sha256(diff) };
+}
+
+interface Source { id: string; gate_rev: number; words: string; by?: string }
+type Plan = { source: Source; doIt: GateItem[]; deltas: GateItem[] } | { reason: string; from?: number };
+
+/** Every condition but the delta check; returns why not, or what the carry takes. */
+function assess(state: RunState, prev: GateList, list: GateList): Plan {
+  const from = prev.gate_rev;
+  if (state.carry_forward === false) return { reason: 'carry-forward is off for this run (--no-carry-forward)', from };
+  const p1 = readDecisions(state.out).find(r => r.gate_rev === from && (r.label ?? '').toLowerCase() === 'p1');
+  if (!p1 || p1.status === 'pending' || !p1.answered_at) return { reason: `gate_rev ${from} was not answered`, from };
+  const unanswered = readDecisions(state.out).find(r => r.gate_rev === from && r.status === 'pending');
+  if (unanswered) return { reason: `gate_rev ${from} left ${unanswered.label ?? unanswered.id} unanswered`, from };
+  if (p1.chosen !== 'a' && p1.chosen !== 'b') return { reason: `gate_rev ${from} answer was p1${p1.chosen}, not an approval`, from };
+  if (p1.carried_gate_from) return { reason: `gate_rev ${from} was itself carried; the owner answers this one`, from };
+  if (parseReply(p1.reply ?? '', prev).mode !== 'rec') return { reason: `gate_rev ${from} was answered with item tokens, not rec or "do it"`, from };
+  const words = p1.owner_reply ?? (p1.answered_by === 'human' ? p1.reply : undefined);
+  if (!words?.trim()) return { reason: `no owner words recorded at gate_rev ${from}`, from };
+  const amendments = (state.amendments ?? []).filter(a => a.gate_rev === from);
+  const open = amendments.find(a => a.status !== 'incorporated' && a.status !== 'settled');
+  if (open) return { reason: `accepted change ${open.decision} is ${open.status}`, from };
+  const journalRows = readJournal(state.out);
+  const at = journalRows.map(r => r.event === 'gate_written' && r.gate_rev === from).lastIndexOf(true);
+  let expected: string | undefined = at >= 0 ? journalRows[at]!.source_file_sha256 : undefined;
+  for (const r of journalRows.slice(at + 1)) if (r.event === 'plan_revised') expected = r.source_file_sha256;
+  if (!expected) return { reason: 'no plan evidence recorded for the gate or the revise (a run written before carry-forward)', from };
+  if (fileSha(planFile(state)) !== expected) return { reason: `the plan file ${path.basename(planFile(state))} changed after the owner's answer or the revise that took it`, from };
+  if (driftLine(state, 'PLAN_SOURCE_CHANGED')) return { reason: 'the reviewed plan differs from the plan file', from };
+  const urgent = readJsonl(path.join(state.out, 'findings.jsonl')).rows.find(r => (r.urgent === true || r.outside_plan === true) && !r.settled_by);
+  if (urgent) return { reason: `urgent finding open: ${String(urgent.title ?? urgent.id)}`, from };
+  const doIt: GateItem[] = []; const deltas: GateItem[] = [];
+  for (const item of list.items) {
+    if (item.id === 'p1' || item.carried) continue;
+    const option = item.options.find(o => o.key === item.recommended);
+    if (item.cap?.kind === 'delta' && !option?.changes_plan && item.recommended === 'a') { deltas.push(item); continue; }
+    if (isDoIt(words) && item.kind === 'auto' && !item.cap && !option?.changes_plan && !['scope', 'cost', 'direction'].includes(item.changes ?? '')) { doIt.push(item); continue; }
+    return { reason: `new item ${item.id}: ${item.title}`, from };
+  }
+  return { source: { id: p1.id, gate_rev: from, words, by: p1.answered_by }, doIt, deltas };
+}
+
+/** Called by finishGate once gate_rev N+1 is written (status gate_pending): carry, set up the delta check, or say why not. */
+export function evaluateCarry(state: RunState, prev: GateList | undefined, list: GateList): CarryRecord | undefined {
+  voidPendingCheck(state, `gate_rev ${list.gate_rev} was written`);
+  if (!prev || list.gate_rev < 2) return undefined;
+  try {
+    const plan = assess(state, prev, list);
+    if ('reason' in plan) return refuse(state, list.gate_rev, plan.from, plan.reason);
+    if (plan.deltas.length) return dispatchCheck(state, prev, list, plan);
+    return carry(state, list, plan.source, plan.doIt, plan.deltas);
+  } catch (e: any) {
+    if (readDecisions(state.out).some(r => r.gate_rev === list.gate_rev && r.answered_at && !r.carried_from)) throw e;
+    return refuse(state, list.gate_rev, prev.gate_rev, `error: ${e instanceof RunError ? `${e.detail} (${e.code})` : e?.message ?? e}`);
+  }
+}
+
+function dispatchCheck(state: RunState, prev: GateList, list: GateList, plan: Extract<Plan, { source: Source }>): CarryRecord {
+  const rev = list.gate_rev;
+  const cap = state.carry_check_cap_usd ?? DEFAULT_CARRY_CHECK_CAP_USD;
+  const estimate = attemptEstimate(state, undefined, process.env).usd;
+  if (estimate > cap) return refuse(state, rev, prev.gate_rev, `delta check estimate ${estimate.toFixed(2)} is over the per-gate cap ${cap.toFixed(2)} (--carry-check-cap)`);
+  const delta = deltaSince(state, prev.gate_rev);
+  if (!delta) return refuse(state, rev, prev.gate_rev, `no copy of the plan the owner answered at gate_rev ${prev.gate_rev}`);
+  const current = fs.readFileSync(state.active_plan, 'utf8');
+  const { diff, sha: deltaSha } = delta;
+  const attempt = `${state.run}-c${rev}`;
+  const prompt = path.join(state.out, `carry-check-g${rev}-prompt.md`);
+  const result = path.join(state.out, `carry-check-g${rev}.md`);
+  try { reserve(path.join(state.out, FILES.spend), attempt, estimate, { cap: state.spend_cap_usd, label: 'carry-check' }); }
+  catch (e: any) { return refuse(state, rev, prev.gate_rev, `delta check does not fit the run's spend cap: ${e?.message ?? e}`); }
+  const accepted = (state.amendments ?? []).filter(a => a.gate_rev === prev.gate_rev).map(a => `- ${a.title} -> ${a.text}`).join('\n') || '- (none recorded)';
+  fs.writeFileSync(prompt, [BOUNDARY, '',
+    `You are checking one edit to a reviewed plan before gstack reuses the owner's approval of gate_rev ${prev.gate_rev} for gate_rev ${rev}. The owner accepted these changes:`, accepted, '',
+    'The plan was edited after that answer, but the phases that would re-review the edit were at their round cap, so no reviewer has read it. Read the delta and say whether it implements the accepted changes, adds nothing the owner did not accept (scope, cost or direction), and leaves the plan correct. Rate a finding High or Critical only when the edit adds an unaccepted change, contradicts an accepted one, or would make the plan fail a test, break production, or lose data or secrets; anything else is Medium or lower.', '',
+    `Delta (reviewed plan at gate_rev ${prev.gate_rev} -> now; + added, - removed; SHA-256 ${deltaSha}):`, '```diff', diff, '```', '',
+    'The complete current plan follows as context.', '', current,
+    REVIEWER_TRAILER, RESULT_FORMAT('carry-check', deltaSha, result)].join('\n'));
+  const check: CarryCheck = { attempt, prompt, result, delta_sha256: deltaSha, items: plan.deltas.map(i => i.id), estimate_usd: estimate, status: 'dispatched', at: now() };
+  const family = state.native_model ? `not ${modelFamily(state.native_model)}` : 'differs from the plan reviewers where available';
+  return record(state, { gate_rev: rev, from_gate_rev: prev.gate_rev, verdict: 'check_pending', words: plan.source.words, under_do_it: plan.doIt.map(i => i.id), check,
+    reason: `only an unreviewed post-cap delta is new (${check.items.join(',')}); one reviewer reads it first`,
+    lines: [`CARRY_CHECK: attempt=${attempt} items=${check.items.join(',')} prompt=${prompt} result=${result} model_family=${family} estimate=${estimate.toFixed(2)} cap=${cap.toFixed(2)}`,
+      `NEXT: dispatch one reviewer on the prompt, then gstack-autoplan check --out ${state.out} --attempt ${attempt} --result ${result} --model <id> (or --unavailable "<reason>" to relay the gate as today)`] });
+}
+
+function carry(state: RunState, list: GateList, source: Source, doIt: GateItem[], deltas: GateItem[], check?: CarryCheck): CarryRecord {
+  const rev = list.gate_rev;
+  const p1 = list.items.find(i => i.id === 'p1')!;
+  const reply = list.items.map(i => `${i.id}${i.id === 'p1' ? p1.recommended : i.carried?.chosen ?? i.recommended}`).join(' ');
+  const reason = [doIt.length || deltas.length ? '' : 'no new choice', doIt.length ? `"do it" covers ${doIt.map(i => i.id).join(',')}` : '', deltas.length ? `delta check passed on ${deltas.map(i => i.id).join(',')}` : '',
+    'plan unchanged since the accepted changes'].filter(Boolean).join('; ');
+  const answered = answerGate(state, rev, reply, 'agent:gstack-autoplan', { carry: { from: source.id, gate_rev: source.gate_rev, words: source.words, by: source.by, reason } });
+  const marks = new Map<string, Record<string, string>>([...doIt.map(i => [decisionId(list.run, rev, i.id), { carried_under: 'do it' }] as const), ...deltas.map(i => [decisionId(list.run, rev, i.id), { carry_check: check?.attempt ?? '' }] as const)]);
+  if (marks.size) writeDecisions(state.out, readDecisions(state.out).map(r => (marks.has(r.id) ? { ...r, ...marks.get(r.id) } : r)));
+  const lines = [`CARRIED: gate_rev=${rev} <- gate_rev=${source.gate_rev} words=${JSON.stringify(source.words)} by=${source.by ?? 'unknown'} reason=${reason}`,
+    ...doIt.map(i => `CARRIED_UNDER_DO_IT: ${i.id} ${i.title} -> ${i.options.find(o => o.key === i.recommended)?.text ?? i.recommended} (the owner has not seen this item)`),
+    ...answered.lines];
+  return record(state, { gate_rev: rev, from_gate_rev: source.gate_rev, verdict: 'carried', reason, words: source.words, under_do_it: doIt.map(i => i.id), ...(check ? { check } : {}), lines });
+}
+
+export interface CheckOptions { out: string; attempt?: string; result?: string; model?: string; unavailable?: string; sameFamily?: string }
+
+/** `gstack-autoplan check`: bind the delta reviewer's result (or its absence) and carry, or leave the gate for the owner. */
+export function submitCarryCheck(o: CheckOptions): { state: RunState; lines: string[] } {
+  const state = readState(path.resolve(o.out));
+  const c = state.carry;
+  if (!c || c.verdict !== 'check_pending' || !c.check || state.status !== 'gate_pending' || c.gate_rev !== state.gate_rev) throw new RunError('PHASE_NOT_AWAITING', `no delta check is pending (run is ${state.status}, gate_rev=${state.gate_rev ?? 'none'})`, 3);
+  const check = c.check;
+  if (o.attempt !== check.attempt) throw new RunError('ATTEMPT_MISMATCH', `pass --attempt ${check.attempt}`, 3);
+  const stale = readDecisions(state.out).some(r => r.gate_rev === c.gate_rev && r.answered_at && !r.carried_from) ? `the owner answered gate_rev ${c.gate_rev}`
+    : deltaSince(state, c.from_gate_rev ?? 0)?.sha !== check.delta_sha256 ? 'the plan changed after the delta check was dispatched' : undefined;
+  if (stale) {
+    const lines = voidPendingCheck(state, stale);
+    writeState(state);
+    return { state, lines };
+  }
+  const gateFile = path.join(state.out, 'gate.json');
+  const list = JSON.parse(fs.readFileSync(gateFile, 'utf8')) as GateList;
+  const spend = path.join(state.out, FILES.spend);
+  let out: CarryRecord;
+  if (o.unavailable !== undefined) {
+    release(spend, check.attempt);
+    list.carry_check = { attempt: check.attempt, model: 'none', verdict: 'unavailable', max_severity: 'none', findings: 0, at: now() };
+    fs.writeFileSync(gateFile, JSON.stringify(list, null, 2) + '\n');
+    out = refuse(state, c.gate_rev, c.from_gate_rev, `delta check unavailable: ${o.unavailable}`);
+    out.check = { ...check, status: 'unavailable' };
+  } else {
+    if (!o.result || !o.model) throw new RunError('RESULT_RECEIPT_MISSING', 'check needs --result <file> and --model <id>, or --unavailable "<reason>"', 3);
+    if (!fs.existsSync(o.result)) throw new RunError('RESULT_RECEIPT_MISSING', `result file not found: ${o.result}`);
+    const family = modelFamily(o.model);
+    const author = state.native_model ? modelFamily(state.native_model) : undefined;
+    if (!o.sameFamily && (family === 'unknown' || family === author)) throw new RunError('MODEL_FAMILY_CONFLICT', `${o.model} is ${family === 'unknown' ? 'of no recognized family' : `the plan reviewers' family (${family})`}; use another family, or pass --same-family "<why none is available>"`, 3);
+    const text = fs.readFileSync(o.result, 'utf8');
+    const parsed = parseReviewResult(text, { phase: 'carry-check', sha256: check.delta_sha256 });
+    if (!parsed.receipt.ok) throw new RunError('RESULT_RECEIPT_MISSING', parsed.errors[0]!);
+    if (!parsed.fenceFound || parsed.errors.length) throw new RunError('RESULT_FINDINGS_MISSING', parsed.errors.join('; '));
+    fs.writeFileSync(path.join(state.out, `carry-check-g${c.gate_rev}.bound.md`), text, process.platform === 'win32' ? {} : { mode: 0o444 });
+    settle(spend, check.attempt, 'unknown');
+    const order = ['Informational', 'Low', 'Medium', 'High', 'Critical'];
+    const max = parsed.findings.map(f => String(f.severity)).reduce((a, b) => (order.indexOf(b) > order.indexOf(a) ? b : a), 'none');
+    const blocking = parsed.findings.find(f => HIGH.includes(String(f.severity)));
+    const done: CarryCheck = { ...check, status: blocking ? 'failed' : 'passed', model: o.model, max_severity: max, findings: parsed.findings.length, at: now() };
+    list.carry_check = { attempt: check.attempt, model: o.model + (o.sameFamily ? ` (same family: ${o.sameFamily})` : ''), verdict: blocking ? 'failed' : 'passed', max_severity: max, findings: parsed.findings.length, at: done.at };
+    if (blocking) for (const item of list.items) if (check.items.includes(item.id)) (item.flags ??= []).push(`carry check ${o.model}: ${blocking.severity} ${blocking.title}`);
+    fs.writeFileSync(gateFile, JSON.stringify(list, null, 2) + '\n');
+    if (blocking) {
+      out = refuse(state, c.gate_rev, c.from_gate_rev, `delta check found ${blocking.severity}: ${blocking.title}`);
+      out.check = done;
+    } else {
+      const plan = assess(state, { ...list, gate_rev: c.from_gate_rev! } as GateList, list);
+      if ('reason' in plan) { out = refuse(state, c.gate_rev, c.from_gate_rev, plan.reason); out.check = done; }
+      else {
+        out = carry(state, list, plan.source, plan.doIt, plan.deltas, done);
+        out.lines.unshift(`CARRY_CHECK: passed attempt=${check.attempt} model=${o.model} findings=${done.findings} highest=${max}`);
+      }
+    }
+  }
+  writeState(state);
+  return { state, lines: out.lines };
+}
@@ -24,7 +24,8 @@ import { consensusRecord, reconcile, remedyKey, renderConsensus, SETTLEABLE_CHAN
 import type { Phase } from './autoplan-prompts';
 import { FILES, RunError, journal as journalRow, now, readDecisions, voiceRows, voiceTerminal, writeDecisions, type Amendment, type RunState, type VoiceState } from './autoplan-state';
 import { applyCapChoices, pendingDeltas, releaseUndispatched, roundCap, roundOf } from './autoplan-budget';
-import { capyRelayLine } from './captain-brief';
+import { evaluateCarry, gateEvidence, voidPendingCheck } from './autoplan-carry';
+import { buildEli5, capyRelayLine } from './captain-brief';
 import { previousReview, reviewRows, settledIds } from './autoplan-confirm';
 import { phaseClose } from './autoplan-timing';
 import { ANSWERED_BY_RE, applyReply, decisionId, decisionRows, keepDirection, parseReply, pendingHint, recHash, validateGateList, type GateChoice, type GateItem, type GateList, type GateOption, type ParsedReply } from './gate-list';
@@ -237,7 +238,8 @@ export function buildGate(state: RunState, gateRev: number): GateList {
  }
  items.push(...capItems(state, { d: () => `d${++d}`, uc: () => `uc${++uc}` }));
  items.push(planApproval(state, items));
-  const list: GateList = { schema_version: 1, run: state.run, gate_rev: gateRev, items, title: '/autoplan final approval gate', rec_rev: 0 };
+  const list: GateList = { schema_version: 1, run: state.run, gate_rev: gateRev, items, title: '/autoplan final approval gate', rec_rev: 0,
+    ...(fs.existsSync(state.active_plan) ? { eli5: buildEli5(fs.readFileSync(state.active_plan, 'utf8'), state.active_plan) } : {}) };
  const problems = validateGateList(list);
  if (problems.length) throw new Error(`gate list invalid: ${problems.join('; ')}`);
  return list;
@@ -285,9 +287,29 @@ export function recommendGate(state: RunState, gateRev: number, itemId: string,
    ...[capyRelayLine(state.out)].filter((l): l is string => !!l)];
 }

-/** Every phase closed: write the gate list, decisions, tasks and the exported files; the run ends gate_pending. Unchanged answered proposals carry forward. */
+/** The gate list the owner last answered: gate.json, or its archive (`gate.r<rev>[.v<k>].json`) once a reopen moved it. */
+function previousGate(out: string, rev: number): GateList | undefined {
+  if (rev < 1) return undefined;
+  const re = new RegExp(`^gate\\.r${rev}(?:\\.v(\\d+))?\\.json$`);
+  const archived = fs.readdirSync(out).map(f => [f, re.exec(f)] as const).filter(([, m]) => m).sort((a, b) => Number(b[1]![1] ?? 1) - Number(a[1]![1] ?? 1)).map(([f]) => f);
+  for (const name of ['gate.json', ...archived]) {
+    const file = path.join(out, name);
+    if (!fs.existsSync(file)) continue;
+    const list = JSON.parse(fs.readFileSync(file, 'utf8')) as GateList;
+    if (list.gate_rev === rev) return list;
+  }
+  return undefined;
+}
+
+/**
+ * Every phase closed: write the gate list, decisions, tasks and the exported files; the run ends gate_pending. Unchanged
+ * answered proposals carry forward, and a later revision that asks nothing new carries the owner's last answer
+ * (lib/autoplan-carry.ts; its lines are on `carryLines(state)`).
+ */
 export function finishGate(state: RunState): GateList {
  const gateRev = (state.gate_rev ?? 0) + 1;
+  const gateFile = path.join(state.out, 'gate.json');
+  const prev = previousGate(state.out, state.gate_rev ?? 0);
  const list = buildGate(state, gateRev);
  const kept = readDecisions(state.out).filter(r => r.gate_rev !== gateRev);
  const answered = new Map(kept.filter(r => r.origin && r.chosen != null && r.status !== 'pending').map(r => [r.origin!, r]));
@@ -295,7 +317,7 @@ export function finishGate(state: RunState): GateList {
    const prev = item.origin ? answered.get(item.origin) : undefined;
    if (prev && item.options.some(o => o.key === prev.chosen)) item.carried = { from: prev.carried_from ?? prev.id, chosen: prev.chosen!, gate_rev: prev.gate_rev };
  }
-  fs.writeFileSync(path.join(state.out, 'gate.json'), JSON.stringify(list, null, 2) + '\n');
+  fs.writeFileSync(gateFile, JSON.stringify(list, null, 2) + '\n');
  const rows = decisionRows(list).map(row => {
    const item = list.items.find(i => decisionId(list.run, gateRev, i.id) === row.id)!;
    if (!item.carried) return row;
@@ -308,7 +330,8 @@ export function finishGate(state: RunState): GateList {
  state.status = 'gate_pending';
  state.gate = { rev: gateRev, items: list.items.length, written_at: now() };
  exportRun(state);
-  journalRow(state.out, { event: 'gate_written', gate_rev: gateRev, items: list.items.length, carried: list.items.filter(i => i.carried).map(i => i.id) });
+  journalRow(state.out, { event: 'gate_written', gate_rev: gateRev, items: list.items.length, carried: list.items.filter(i => i.carried).map(i => i.id), ...gateEvidence(state, gateRev) });
+  evaluateCarry(state, prev, list);
  return list;
 }

@@ -321,6 +344,8 @@ export interface AnswerOptions {
  recHash?: string;
  /** The owner's verbatim words; required for every `--by agent:<id>` answer. */
  ownerReply?: string;
+  /** A whole-gate carry (lib/autoplan-carry.ts): the earlier p1 decision and the owner's words it reuses. Replaces ownerReply and its re-parse. */
+  carry?: { from: string; gate_rev: number; words: string; by?: string; reason: string };
 }

 /** How an accepted plan change is discharged: a packaging/ordering change settles without a plan edit. */
@@ -361,7 +386,7 @@ export function answerGate(state: RunState, gateRev: number, reply: string, answ
  if (state.status !== 'gate_pending' || !fs.existsSync(gateFile)) throw new RunError('PHASE_NOT_AWAITING', `run is ${state.status}, not gate_pending`, 3);
  if (answeredBy !== undefined && !ANSWERED_BY_RE.test(answeredBy)) throw new RunError('ANSWER_BY_REQUIRED', `--by ${JSON.stringify(answeredBy)} is neither human nor agent:<id>`, 3);
  if (answeredBy === undefined && state.session_kind === 'unattended') throw new RunError('ANSWER_BY_REQUIRED', 'an unattended run records who answered; pass --by human or --by agent:<id>', 3);
-  if (answeredBy?.startsWith('agent:') && o.ownerReply === undefined) throw new RunError('OWNER_REPLY_REQUIRED', `${answeredBy} relayed or decided this reply; pass the owner's words verbatim with --owner-reply "<text>" (or --owner-reply - / --owner-reply-file <path>)`, 3);
+  if (answeredBy?.startsWith('agent:') && o.ownerReply === undefined && !o.carry) throw new RunError('OWNER_REPLY_REQUIRED', `${answeredBy} relayed or decided this reply; pass the owner's words verbatim with --owner-reply "<text>" (or --owner-reply - / --owner-reply-file <path>)`, 3);
  const list = JSON.parse(fs.readFileSync(gateFile, 'utf8')) as GateList;
  const rows = readDecisions(state.out);
  const prior = Object.fromEntries(rows.filter(r => r.gate_rev === list.gate_rev && r.chosen != null && r.status !== 'pending' && r.label).map(r => [r.label!.toLowerCase(), r.chosen!]));
@@ -372,12 +397,13 @@ export function answerGate(state: RunState, gateRev: number, reply: string, answ
  if (o.recHash !== undefined && o.recHash !== currentHash) throw new RunError('GATE_REC_STALE', `--rec-hash ${o.recHash} was rendered before the recommendations changed; current rec_hash=${currentHash}`, 3);
  if (parsed.mode === 'rec' && list.rec_rev === undefined && o.recHash === undefined) throw new RunError('GATE_REC_STALE', `this gate was written before \`rec\`/\`all\` took plan changes (gate.json has no rec_rev), so the owner saw a different meaning; render it again (gstack-gate render ${gateFile} --all), relay it, and answer with --rec-hash ${currentHash}, or reply with explicit <id><option> tokens`, 3);
  if (parsed.mode === 'rec' && (list.rec_rev ?? 0) > 0 && o.recHash === undefined) throw new RunError('GATE_REC_STALE', `recommend changed this gate (rec_rev=${list.rec_rev}); pass --rec-hash from the render the owner saw (current rec_hash=${currentHash})`, 3);
-  const ownerParsed = o.ownerReply !== undefined ? checkOwnerReply(list, gateRev, prior, parsed, o.ownerReply) : undefined;
-  const provenance = o.ownerReply !== undefined ? { owner_reply: o.ownerReply, owner_reply_parsed: ownerParsed } : {};
+  const ownerParsed = o.ownerReply !== undefined && !o.carry ? checkOwnerReply(list, gateRev, prior, parsed, o.ownerReply) : undefined;
+  const provenance = o.carry ? { owner_reply: o.carry.words, carried_gate_from: o.carry.from, carried_gate_rev: o.carry.gate_rev, carry_reason: o.carry.reason, ...(o.carry.by ? { carried_answered_by: o.carry.by } : {}) }
+    : o.ownerReply !== undefined ? { owner_reply: o.ownerReply, owner_reply_parsed: ownerParsed } : {};
  const itemOf = (id: string) => list.items.find(i => i.id === id)!;
  const optionOf = (c: GateChoice) => itemOf(c.id).options.find(o => o.key === c.chosen)!;
  const plan = parsed.choices.find(c => c.id === 'p1');
-  const lines: string[] = [];
+  const lines: string[] = o.carry ? [] : voidPendingCheck(state, `the owner answered gate_rev ${gateRev}`);
  const answeredAt = now();
  const record = () => writeDecisions(state.out, applyReply(rows, parsed, answeredAt, answeredBy, provenance).map(r => {
    const c = r.gate_rev === list.gate_rev && r.chosen && !r.chosen_text ? itemOf((r.label ?? '').toLowerCase())?.options.find(o => o.key === r.chosen) : undefined;
@@ -399,7 +425,8 @@ export function answerGate(state: RunState, gateRev: number, reply: string, answ
  const reopenSet = new Set<Phase>(plan.chosen === 'd' ? state.phase_order : []);
  record();
  state.gate = { ...(state.gate ?? {}), answered_at: answeredAt, reply, gate_rev: gateRev, plan: plan.chosen, overrides: tokens(overrides), ...(answeredBy ? { answered_by: answeredBy } : {}), ...provenance };
-  journalRow(state.out, { event: 'answered', gate_rev: gateRev, reply, plan: plan.chosen, overrides: tokens(overrides), changes: tokens(changing), reopen: [...reopenSet], ...(answeredBy ? { answered_by: answeredBy } : {}), ...provenance });
+  const answerClass = plan.chosen !== 'a' && plan.chosen !== 'b' ? `p1${plan.chosen}` : changing.length ? 'plan-changes' : overrides.length ? 'overrides' : 'as-recommended';
+  journalRow(state.out, { event: 'answered', gate_rev: gateRev, reply, plan: plan.chosen, answer_class: answerClass, overrides: tokens(overrides), changes: tokens(changing), reopen: [...reopenSet], ...(answeredBy ? { answered_by: answeredBy } : {}), ...provenance });
  if (plan.chosen === 'c') {
    lines.push(`GATE_PENDING: interrogate recorded; the gate stays at gate_rev=${gateRev}`);
    return { state, applied: true, pending: ['p1'], reopened: [], outcome: 'gate_pending', lines };
@@ -17,10 +17,12 @@
 * written and `revise_intent` journaled before anything moves; `resume`
 * (recoverRevise) completes an interrupted revise from the stage.
 */
+import { createHash } from 'node:crypto';
 import * as fs from 'node:fs';
 import * as path from 'node:path';
 import { exportRun } from './autoplan-export';
 import { applyRenames, approve, closeDraftDirections, finishGate, planReopen } from './autoplan-gate';
+import { carryLines, voidPendingCheck } from './autoplan-carry';
 import { atCap, escalateLine, pendingDeltas, roundCap, roundOf } from './autoplan-budget';
 import { diffStat, lineDiff } from './autoplan-confirm';
 import { SETTLEABLE_CHANGES } from './autoplan-reconcile';
@@ -109,8 +111,10 @@ export function decisionSeverity(state: RunState, decision: string): string {
 function approveOrRegate(state: RunState, at: string): string[] {
  if (pendingDeltas(state).length) {
    finishGate(state);
+    const carried = carryLines(state);
+    if (state.status !== 'gate_pending') return carried;
    const escalate = escalateLine(state);
-    return [`GATE: gate_rev=${state.gate_rev} (a plan change after a round cap needs the owner) render with: gstack-gate render ${path.join(state.out, 'gate.json')}`, ...(escalate ? [escalate] : [])];
+    return [...carried, `GATE: gate_rev=${state.gate_rev} (a plan change after a round cap needs the owner) render with: gstack-gate render ${path.join(state.out, 'gate.json')}`, ...(escalate ? [escalate] : [])];
  }
  return [...approve(state, at), `APPROVED: gate_rev=${state.gate_rev} (every accepted change is in the plan or settled)`];
 }
@@ -176,6 +180,7 @@ export function reviseRun(o: ReviseOptions): ReviseResult {
  const delta = capped.length || quiet ? diffStat(lineDiff(fs.readFileSync(state.active_plan, 'utf8'), fs.readFileSync(path.join(staging, FILES.active), 'utf8'))) : { added: 0, removed: 0 };
  const atGate = state.status === 'gate_pending' ? state.gate_rev : undefined;
  journal(out, { event: 'revise_intent', k, staging: path.relative(out, staging), plan: file, old_plan_sha256: oldHash, new_plan_sha256: newHash, at_gate_rev: atGate ?? null, incorporates, withdraw, settle });
+  voidPendingCheck(state, 'the plan was revised');

  const renames = reopen.length ? planReopen(state, reopen, state.gate_rev ?? 0, `revise v${k}`) : [];
  for (const phase of capped) {
@@ -199,7 +204,7 @@ export function reviseRun(o: ReviseOptions): ReviseResult {
  if (amending) state.status = pendingAmendments(state).length ? 'amend_pending' : state.phase_order.some(p => !state.phases[p]?.closed_at) ? 'running' : pendingDeltas(state).length ? 'gate_pending' : 'complete';
  // A quiet revise that ends at gate_pending still has an earlier capped delta waiting: the owner gets a new gate for it.
  const regate = (capped.length > 0 || quiet) && state.status === 'gate_pending';
-  const committed = { event: 'plan_revised', k, plan: file, old_plan_sha256: oldHash, new_plan_sha256: newHash, reopened: reopen, added, at_gate_rev: atGate ?? null };
+  const committed = { event: 'plan_revised', k, plan: file, old_plan_sha256: oldHash, new_plan_sha256: newHash, source_file_sha256: createHash('sha256').update(fs.readFileSync(file)).digest('hex'), reopened: reopen, added, at_gate_rev: atGate ?? null };
  fs.writeFileSync(path.join(staging, FILES.run), JSON.stringify({ state, renames, committed, settled }, null, 2) + '\n');
  publish(state, staging, renames, committed, settled);
  const lines = [`REVISED: plan=${newHash.slice(0, 12)} reopened=${reopen.join(',') || 'none'}${added.length ? ` added=${added.join(',')}` : ''}${capped.length ? ` capped=${capped.join(',')}` : ''}`,
@@ -20,6 +20,7 @@ import * as os from 'node:os';
 import * as path from 'node:path';
 import { canonicalRows, parseReceipt, parseRecordFindings, parseReviewResult, type CanonicalFinding } from './autoplan-reconcile';
 import { amendmentSteps, applyGuidance, closeDeferred, closePhase, finishGate } from './autoplan-gate';
+import { carryLines } from './autoplan-carry';
 import { GSTACK_VERSION } from './autoplan-migrate';
 import { driftLine, normalizeFindingFile, reviewHash, scaffold } from './autoplan-plan-text';
 import { PHASES, SNAPSHOT_TOOL, attemptFile, bindPromptResult, detectScope, driverFile, extractHeading, phaseOrder, reviewSkillFile, scopeLines, scopeSourceLabel, sectionLoadFor, snapshotTool, writePrompts, type Phase, type Scope, type ScopeFlags } from './autoplan-prompts';
@@ -71,7 +72,7 @@ export function withRunLock<T>(out: string, fn: () => T, opts: { steal?: boolean
 // ---------------------------------------------------------------------------
 // init
 // ---------------------------------------------------------------------------
-export interface InitOptions { plan: string; out: string; cwd: string; scope?: ScopeFlags; planHeading?: string; outsideEnabled?: boolean; spendCap?: number | null; sessionKind?: string; deadlineMin?: number; light?: boolean; runId?: string; scaffold?: boolean; maxRounds?: number }
+export interface InitOptions { plan: string; out: string; cwd: string; scope?: ScopeFlags; planHeading?: string; outsideEnabled?: boolean; spendCap?: number | null; sessionKind?: string; deadlineMin?: number; light?: boolean; runId?: string; scaffold?: boolean; maxRounds?: number; carryForward?: boolean; carryCheckCapUsd?: number }

 /** `lines` receives what init itself reports (SCAFFOLDED:). A plan with one section heading missing refuses, or with `scaffold` is reviewed from a scaffolded copy. */
 export function initRun(o: InitOptions, lines: string[] = []): RunState {
@@ -121,6 +122,7 @@ export function initRun(o: InitOptions, lines: string[] = []): RunState {
    ...(o.light ? { light: true } : {}),
    gstack_version: GSTACK_VERSION, plan_input: input, scope_flags: { ...(o.scope ?? {}) },
    ...(o.planHeading ? { plan_heading: o.planHeading } : {}), ...(scaffolded ? { scaffolded: true } : {}),
+    ...(o.carryForward === false ? { carry_forward: false } : {}), ...(o.carryCheckCapUsd !== undefined ? { carry_check_cap_usd: o.carryCheckCapUsd } : {}),
  };
  fs.writeFileSync(path.join(out, 'timing.json'), JSON.stringify({ schema_version: 1, run, phases: [], started_at: now(), session_kind: state.session_kind }, null, 2) + '\n');
  journal(out, { event: 'init', run, plan_source: source, scope, phase_order: order, ...(consentMissing ? { outside: 'not_run', reason: 'consent_missing' } : {}) });
@@ -176,6 +178,17 @@ function openPhase(state: RunState, phase: Phase, cwd: string): void {
  journal(state.out, { event: 'phase_opened', phase, snapshot: snap.sha256, prompts: Object.fromEntries(Object.entries(prompts).map(([k, p]) => [k, path.relative(state.out, p)])) });
 }

+/** The first next's price line: the most the review can reserve at this scope and round cap (a reservation total, not a spend ceiling). */
+export function reviewEstimateLine(state: RunState, perVoiceUsd: number): string {
+  const voices = state.phase_order.map(p => (!state.outside_enabled || (p === 'ceo' && state.scope.docs_only) ? 1 : 2));
+  const perRound = voices.reduce((a, b) => a + b, 0);
+  const rounds = state.max_rounds ?? null;
+  const per = perVoiceUsd > 0 ? perVoiceUsd.toFixed(2) : 'unknown';
+  const attempts = rounds === null ? `${perRound} per round, rounds uncapped` : `<=${perRound * rounds}`;
+  const total = rounds === null || perVoiceUsd <= 0 ? 'unknown' : `<=${(perRound * rounds * perVoiceUsd).toFixed(2)}`;
+  return `REVIEW_ESTIMATE: phases=${state.phase_order.join(',')} voices=${voices.join('+')} rounds${rounds === null ? '=uncapped' : `<=${rounds}`} attempts${attempts.startsWith('<=') ? attempts : `=${attempts}`} per_voice=${per} reservations${total === 'unknown' ? '=unknown' : total} (a reservation total, not a spend ceiling; actual host spend stays unknown; round grants at the gate add more)`;
+}
+
 export function nextStep(o: NextOptions): NextResult {
  const state = readState(path.resolve(o.out));
  const lines: string[] = [];
@@ -193,7 +206,7 @@ export function nextStep(o: NextOptions): NextResult {
    lines.push(...closeDeferred(state, phase, { reason: `round cap: round ${roundOf(state, phase)} would pass max ${roundCap(state, phase)}`, by: 'runner', cause: 'cap' }).lines);
    phase = currentPhase(state);
  }
-  if (!phase) { finishGate(state); writeState(state); return result; }
+  if (!phase) { finishGate(state); lines.push(...carryLines(state)); writeState(state); return result; }
  const first = !state.phase_order.some(p => state.phases[p]?.snapshot);
  openPhase(state, phase, o.cwd);
  const ps = state.phases[phase]!;
@@ -208,6 +221,7 @@ export function nextStep(o: NextOptions): NextResult {
  const prompts = ps.prompts!;
  const estimate = attemptEstimate(state, o.estimateUsd, process.env);
  if (estimate.line) lines.push(estimate.line);
+  if (first) lines.push(reviewEstimateLine(state, estimate.usd));
  const fresh = (['native', 'outside'] as const).filter(voice => { const v = voiceState(state, phase!, voice); const open = openAttempt(v); return !voiceTerminal(v) && (!open || (open.state === 'execution_unknown' && o.redispatch?.includes(open.attempt))); });
  const before = summarize(readLedger(path.join(state.out, FILES.spend), state.spend_cap_usd));
  if (state.spend_cap_usd !== null && fresh.length && before.worst_case_usd + fresh.length * estimate.usd > state.spend_cap_usd + 1e-9) {
@@ -369,8 +383,9 @@ export function submitResult(o: SubmitOptions): SubmitResult {

 function closeAndAdvance(state: RunState, phase: Phase, reconcile?: string): string[] {
  const close = closePhase(state, phase, { overridesFile: reconcile });
-  if (!currentPhase(state)) finishGate(state); else state.status = 'running';
-  return [`PHASE_CLOSED: ${phase} confirmed=${close.consensus.confirmed}${close.consensus.unverified ? ` unverified=${close.consensus.unverified}` : ''} disagree=${close.consensus.disagree} new=${close.consensus.new} native_only=${close.consensus.native_only} coverage=${close.consensus.coverage}`];
+  const gated = !currentPhase(state);
+  if (gated) finishGate(state); else state.status = 'running';
+  return [`PHASE_CLOSED: ${phase} confirmed=${close.consensus.confirmed}${close.consensus.unverified ? ` unverified=${close.consensus.unverified}` : ''} disagree=${close.consensus.disagree} new=${close.consensus.new} native_only=${close.consensus.native_only} coverage=${close.consensus.coverage}`, ...(gated ? carryLines(state) : [])];
 }

 /**
@@ -451,14 +466,15 @@ export function resumeRun(out: string): ResumeResult {
    if (ps.native?.status === 'completed' && voiceTerminal(ps.outside)) closePhase(state, phase, {});
  }
  const stopped = resumeStop(state);
-  if (!currentPhase(state) && !['gate_pending', 'amend_pending', 'complete', 'incomplete', 'refused'].includes(state.status)) finishGate(state);
+  const gated = !currentPhase(state) && !['gate_pending', 'amend_pending', 'complete', 'incomplete', 'refused'].includes(state.status);
+  if (gated) finishGate(state);
  else if (state.status === 'interrupted' || state.status === 'running' || state.status === 'awaiting_result') {
    const anyOpen = state.phase_order.some(p => (['native', 'outside'] as const).some(v => openAttempt(voiceState(state, p, v))));
    state.status = anyOpen ? 'awaiting_result' : 'running';
  }
  journal(out, { event: 'resumed', unknown, rebound, status: state.status });
  writeState(state);
-  const lines = [...recovered, ...stopped, `RESUMED: run=${state.run} status=${state.status} execution_unknown=${unknown.length} rebound=${rebound.length}`];
+  const lines = [...recovered, ...stopped, ...(gated ? carryLines(state) : []), `RESUMED: run=${state.run} status=${state.status} execution_unknown=${unknown.length} rebound=${rebound.length}`];
  for (const a of unknown) lines.push(`EXECUTION_UNKNOWN: ${a}`);
  return { state, unknown, rebound, lines };
 }
@@ -70,6 +70,8 @@ export interface RunState extends RunManifest {
  stop?: RunStop; stops?: RunStop[]; capped_deltas?: CappedDelta[];
  /** Written by export from attempts.jsonl; est_input_tokens is dispatched prompt bytes / 4, a relative estimate only. Absent on older runs. */
  dispatches?: DispatchCounts; est_input_tokens?: number;
+  /** Carry-forward (lib/autoplan-carry.ts): false with --no-carry-forward; the per-gate cap on the delta check; the latest carry decision. */
+  carry_forward?: boolean; carry_check_cap_usd?: number; carry?: import('./autoplan-carry').CarryRecord;
 }

 export const TERMINAL_VOICE_STATUSES = ['completed', 'disabled', 'unavailable', 'failed', 'skipped'] as const;
@@ -14,7 +14,7 @@ import { createHash } from 'node:crypto';
 import * as fs from 'node:fs';
 import * as path from 'node:path';
 import { gateCost, gateCostSentence, priorChoices } from './autoplan-gate-cost';
-import { parseReply, recHash, type GateItem } from './gate-list';
+import { parseReply, recHash, type GateEli5, type GateItem } from './gate-list';
 import { renderUrgentBlock } from './headless-artifacts';
 import { findSection, firstTitle } from './markdown-sections';
 import type { BriefInputs } from './owner-brief';
@@ -25,6 +25,8 @@ export const BRIEF_SECTIONS = [
  { title: 'What caused this work', aliases: ['What caused this work', 'What caused us to work on this', 'What caused this', 'Trigger'] },
  { title: 'Big decisions', aliases: ['Big decisions', 'Big decisions we made', 'Key decisions'] },
 ] as const;
+/** The 5-10 plain sentences a plan opens its ask with; required in a captain brief only when the gate carries `eli5` (written by gstack 1.91.305+). */
+export const ELI5_SECTION = { title: 'ELI5', aliases: ['ELI5', 'In plain words'] } as const;
 const SECTION_LINES = 24;
 const SECTION_CHARS = 3000;

@@ -60,6 +62,13 @@ function capped(body: string, planPath: string): string {
  return hidden > 0 || out.length < body.length ? `${out}\n\n_+ more in the plan (${path.basename(planPath)})._` : out;
 }

+/** gate.json `eli5`: the plan's ELI5 and the four approvals-shape sections, capped as the brief caps them. */
+export function buildEli5(planText: string, planFile: string): GateEli5 {
+  const body = (aliases: readonly string[]) => { const s = findSection(planText, [...aliases]); return s?.body.trim() ? capped(s.body.trim(), planFile) : ''; };
+  const [before, why, cause, decisions] = BRIEF_SECTIONS.map(s => body(s.aliases));
+  return { summary: body(ELI5_SECTION.aliases), before_after: before!, why: why!, cause: cause!, decisions: decisions!, plan_file_sha256: createHash('sha256').update(planText).digest('hex') };
+}
+
 export interface CaptainBrief { text: string; missing: string[]; pending: boolean; recHash: string }

 export function renderCaptainBrief(inputs: BriefInputs, o: { outDir: string }): CaptainBrief {
@@ -71,8 +80,10 @@ export function renderCaptainBrief(inputs: BriefInputs, o: { outDir: string }):
  const rowOf = (id: string) => inputs.decisions.filter(d => d.gate_rev === list.gate_rev).find(d => (d.label ?? '').toLowerCase() === id);
  const p1Row = p1 ? rowOf('p1') : undefined;
  const pending = !p1Row || p1Row.status === 'pending';
-  const sections = BRIEF_SECTIONS.map(s => ({ title: s.title, found: findSection(plan.text, [...s.aliases]) }));
-  const missing = sections.filter(s => !s.found || !s.found.body.trim()).map(s => s.title);
+  const eli5 = list.eli5 ? buildEli5(plan.text, plan.file ?? 'plan.md') : undefined;
+  const fromGate = eli5 ? [eli5.before_after, eli5.why, eli5.cause, eli5.decisions] : [];
+  const sections = BRIEF_SECTIONS.map((s, i) => ({ title: s.title, found: eli5 ? { body: fromGate[i]! } : findSection(plan.text, [...s.aliases]) }));
+  const missing = [...(eli5 && !eli5.summary.trim() ? [ELI5_SECTION.title] : []), ...sections.filter(s => !s.found || !s.found.body.trim()).map(s => s.title)];
  const urgent = inputs.findings.some(r => r.urgent === true || r.outside_plan === true) ? [renderUrgentBlock(inputs.findings).trimEnd(), ''] : [];
  const items = list.items.filter(i => i.id !== 'p1');
  const prior = priorChoices(o.outDir, list.gate_rev);
@@ -83,7 +94,11 @@ export function renderCaptainBrief(inputs: BriefInputs, o: { outDir: string }):
  const keptNote = (i: GateItem) => i.carried?.chosen === recKey(i) ? ` (your answer at gate_rev ${i.carried.gate_rev}, carried forward)` : ' (your answer at this gate)';

  let header: string[];
-  if (!pending) {
+  if (!pending && p1Row?.carried_gate_from) {
+    const unseen = inputs.decisions.filter(d => d.gate_rev === list.gate_rev && d.carried_under).map(d => (d.title ?? '').replace(/\s*\[[^\]]*\]\s*$/, ''));
+    header = [`**Nothing pending.** Carried forward from gate_rev ${p1Row.carried_gate_rev}: ${P1_DONE[p1Row?.chosen ?? ''] ?? 'answered'} under the owner's words “${p1Row?.owner_reply ?? ''}”, because nothing new needed deciding.`,
+      ...(unseen.length ? ['', `Taken under "do it" without being shown to the owner: ${unseen.join('; ')}.`] : [])];
+  } else if (!pending) {
    header = [`**Nothing pending.** Already decided${p1Row?.answered_at ? ` on ${p1Row.answered_at.slice(0, 10)}` : ''}: ${P1_DONE[p1Row?.chosen ?? ''] ?? 'answered'}${p1Row?.owner_reply ?? p1Row?.reply ? ` (“${p1Row?.owner_reply ?? p1Row?.reply}”)` : ''}.`];
  } else if (missing.length) {
    header = [`**Not ready for approval:** the plan "${title}" is missing ${missing.join(', ')}. The thread will add them and send this again; do not approve from this draft.`];
@@ -94,7 +109,8 @@ export function renderCaptainBrief(inputs: BriefInputs, o: { outDir: string }):
    ];
  }

-  const body = sections.flatMap(s => [`### ${s.title}`, '', s.found?.body.trim() ? capped(s.found.body.trim(), plan.file ?? 'plan.md') : '_Not written for this plan yet._', '']);
+  const summary = eli5?.summary.trim() ? [`### ${ELI5_SECTION.title}`, '', eli5.summary.trim(), ''] : [];
+  const body = [...summary, ...sections.flatMap(s => [`### ${s.title}`, '', s.found?.body.trim() ? (eli5 ? s.found.body.trim() : capped(s.found.body.trim(), plan.file ?? 'plan.md')) : '_Not written for this plan yet._', ''])];
  const need: string[] = ['### What we need from you', ''];
  if (!pending) need.push('Nothing. This gate is answered.', '');
  else {
@@ -0,0 +1,67 @@
+/**
+ * decision-trail — the append rules of a run's decision trail (docs/unattended.md#the-decision-trail):
+ * a TSV with one line per decision (ts, phase, decision, why, evidence, result). The header is
+ * written once and only ever appended, cells lose their tabs and line breaks, and a cell whose first
+ * byte a spreadsheet would run as a formula (=, +, -, @) gets a leading apostrophe. Shared by
+ * bin/gstack-decision-trail and the autoplan runner (lib/autoplan-carry.ts records its carry verdicts).
+ * The rules are a port of pstack's show-me-your-work/scripts/log.sh (Copyright Lauren Tan, MIT);
+ * licenses/MIT-pstack.txt, NOTICE.md.
+ */
+import * as fs from "node:fs";
+import * as path from "node:path";
+
+export const TRAIL_COLUMNS = ["ts", "phase", "decision", "why", "evidence", "result"] as const;
+export const TRAIL_HEADER = TRAIL_COLUMNS.join("\t") + "\n";
+
+export type TrailRow = { line: number; cells: string[] };
+
+function clean(cell: string): string {
+  const flat = cell.replace(/[\t\n\r]/g, " ");
+  return /^[=+\-@]/.test(flat) ? `'${flat}` : flat;
+}
+
+export function readTrailRows(file: string): TrailRow[] {
+  if (!fs.existsSync(file)) return [];
+  const lines = fs.readFileSync(file, "utf8").split("\n");
+  const rows: TrailRow[] = [];
+  lines.forEach((raw, i) => {
+    let text = raw.endsWith("\r") ? raw.slice(0, -1) : raw;
+    if (i === 0 && text.startsWith("\uFEFF")) text = text.slice(1);
+    // A repeated header (the append-on-flaky-mount case) is never a row.
+    if (text === "" || text === TRAIL_HEADER.trimEnd()) return;
+    rows.push({ line: i + 1, cells: text.split("\t") });
+  });
+  return rows;
+}
+
+/** Append one row (ts stamped in UTC) and return the row count. */
+export function appendTrailRow(file: string, cells: readonly string[]): number {
+  const dir = path.dirname(file);
+  if (dir && dir !== "." && !fs.existsSync(dir)) fs.mkdirSync(dir, { recursive: true });
+  // Append the header too, never truncate: on a flaky mount the worst case is
+  // one stray header line, not lost rows.
+  // A file with no header line anywhere (empty, blank, or rows written by hand)
+  // gets one appended once; readTrailRows skips a header line wherever it sits.
+  const size = fs.existsSync(file) ? fs.statSync(file).size : 0;
+  if (size > 0 && !endsWithNewline(file, size)) fs.appendFileSync(file, "\n");
+  if (size === 0 || !hasHeader(file)) fs.appendFileSync(file, TRAIL_HEADER);
+  const ts = new Date().toISOString().replace(/\.\d{3}Z$/, "Z");
+  fs.appendFileSync(file, [ts, ...cells.map(clean)].join("\t") + "\n");
+  return readTrailRows(file).length;
+}
+
+function hasHeader(file: string): boolean {
+  return fs.readFileSync(file, "utf8").replace(/^\uFEFF/, "").split("\n")
+    .some(line => line.replace(/\r$/, "") === TRAIL_HEADER.trimEnd());
+}
+
+function endsWithNewline(file: string, size: number): boolean {
+  const fd = fs.openSync(file, "r");
+  try {
+    const last = Buffer.alloc(1);
+    fs.readSync(fd, last, 0, 1, size - 1);
+    return last[0] === 0x0a;
+  } finally {
+    fs.closeSync(fd);
+  }
+}
@@ -42,7 +42,14 @@ export interface GateItem {
 }
 export interface GateList { schema_version: 1; run: string; gate_rev: number; items: GateItem[]; title?: string;
  /** Bumped by every `recommend`, so a `rec` relayed before it no longer matches `recHash`. */
-  rec_rev?: number }
+  rec_rev?: number;
+  /** The plan's plain-language ask, read from the reviewed plan (lib/captain-brief.ts `buildEli5`); printed first. */
+  eli5?: GateEli5;
+  /** The one-reviewer check of an unreviewed post-cap delta before a carry (lib/autoplan-carry.ts). */
+  carry_check?: { attempt: string; model: string; verdict: 'passed' | 'failed' | 'unavailable'; max_severity: string; findings: number; at: string } }
+/** The captain approvals shape (docs/captain.md § Approvals), section text taken from the reviewed plan; '' when absent. */
+export interface GateEli5 { summary: string; before_after: string; why: string; cause: string; decisions: string; plan_file_sha256: string }
+export const ELI5_FIELDS = [['before_after', 'Before → after'], ['why', 'Why it matters'], ['cause', 'What caused this work'], ['decisions', 'Big decisions']] as const;

 export interface GateChoice { id: string; chosen: string; kind: 'auto' | 'approval'; how: 'recommended' | 'override' | 'auto_default'; prior?: boolean }
 export interface ParsedReply {
@@ -64,6 +71,10 @@ export function validateGateList(list: unknown): string[] {
  if (g.schema_version !== 1) problems.push('schema_version must be 1');
  if (typeof g.run !== 'string' || !g.run) problems.push('run is required');
  if (!Number.isInteger(g.gate_rev) || (g.gate_rev as number) < 1) problems.push('gate_rev must be an integer >= 1');
+  if (g.eli5 !== undefined) {
+    if (!g.eli5 || typeof g.eli5 !== 'object') problems.push('eli5 must be an object');
+    else for (const k of ['summary', 'before_after', 'why', 'cause', 'decisions', 'plan_file_sha256'] as const) if (typeof g.eli5[k] !== 'string') problems.push(`eli5.${k} must be a string`);
+  }
  if (!Array.isArray(g.items) || g.items.length === 0) return [...problems, 'items must be a non-empty array'];
  const seen = new Set<string>();
  g.items.forEach((item, i) => {
@@ -103,6 +114,8 @@ export function renderGate(list: GateList, page = 1, pageSize = PAGE_SIZE): stri
  const lines: string[] = [];
  lines.push(`GATE: run=${list.run} gate_rev=${list.gate_rev} items=${list.items.length} page=${current}/${pages}${list.title ? ` — ${list.title}` : ''}`);
  lines.push('');
+  if (list.eli5 && current === 1) lines.push(...renderEli5(list.eli5), '');
+  if (list.carry_check) lines.push(`Carry check ${list.carry_check.attempt}: ${list.carry_check.verdict} (${list.carry_check.model}; ${list.carry_check.findings} finding(s), highest ${list.carry_check.max_severity}).`, '');
  list.items.slice(start, start + pageSize).forEach((item, i) => {
    const n = start + i + 1;
    lines.push(`${n}. [${item.id}] ${item.title} (${item.kind}; recommended: ${item.recommended}${item.recommended_basis ? ` (${item.recommended_basis}${item.recommended_by ? `; set by ${item.recommended_by}` : ''})` : ''}${item.phase ? `; from ${item.phase}` : ''})`);
@@ -121,6 +134,13 @@ export function renderGate(list: GateList, page = 1, pageSize = PAGE_SIZE): stri
  return lines.join('\n') + '\n';
 }

+/** The ELI5 block: the summary, then each approvals-shape section under its title. An empty summary says so (CAPTAIN_BRIEF_INCOMPLETE). */
+export function renderEli5(e: GateEli5): string[] {
+  const lines = e.summary.trim() ? ['ELI5', e.summary.trim()] : [];
+  for (const [key, title] of ELI5_FIELDS) if (e[key].trim()) lines.push(...(lines.length ? [''] : []), `${title}:`, e[key].trim());
+  return lines;
+}
+
 function exampleTokens(list: GateList): string {
  return list.items.slice(0, 2).map(i => `${i.id}${i.options.find(o => o.key !== i.recommended)?.key ?? i.recommended}`).join(' ');
 }
@@ -41,6 +41,8 @@ export interface DecisionRow {
  recommended_basis?: string; changes?: string; origin?: string; carried_from?: string; chosen_text?: string;
  /** An agent-relayed answer: the owner's verbatim words, and whether they parsed under the gate grammar. */
  owner_reply?: string; owner_reply_parsed?: boolean; recommended_by?: string;
+  /** A whole-gate carry (lib/autoplan-carry.ts): the earlier p1 decision whose owner words this answer reuses, its gate_rev, why, who relayed them, items taken under "do it", and the delta check attempt. */
+  carried_gate_from?: string; carried_gate_rev?: number; carry_reason?: string; carried_answered_by?: string; carried_under?: string; carry_check?: string;
 }
 export interface TaskRow {
  schema_version: 2; run: string; id: string; title: string; status: TaskStatus; depends_on: string[];
@@ -170,6 +172,7 @@ export const ARTIFACT_SCHEMAS = {
      default_if_unanswered: nullableStr, gate_rev: { type: 'integer', minimum: 1 }, phase: str, chosen: nullableStr,
      answered_at: str, reply: str, source: str, cost: str,
      recommended_basis: str, changes: str, origin: str, carried_from: str, chosen_text: str,
+      carried_gate_from: str, carried_gate_rev: { type: 'integer', minimum: 1 }, carry_reason: str, carried_answered_by: str, carried_under: str, carry_check: str,
    },
  },
  tasks: {
@@ -41,7 +41,7 @@ export const CAPY_RESULT_CODES = {
  },
  CAPTAIN_BRIEF_INCOMPLETE: {
    anchor: 'captain-brief-incomplete',
-    summary: 'the plan lacks one or more of the sections the Captain brief needs (Before → after, Why it matters, What caused this work, Big decisions), so the brief is not ready for approval',
+    summary: 'the plan lacks one or more of the sections the Captain brief needs (ELI5 on a run whose gate carries it, Before → after, Why it matters, What caused this work, Big decisions), so the brief is not ready for approval',
    fix: 'add the named sections to the plan, revise the run if the plan is under review, and render again; --allow-incomplete prints a preview that still says it is not ready',
  },
  PR_SHOTS_INPUT_NOT_FOUND: {
@@ -1,6 +1,6 @@
 {
  "name": "gstack",
-  "version": "1.91.313",
+  "version": "1.91.314",
  "description": "Garry's Stack — Claude Code skills + fast headless browser. One repo, one install, entire AI engineering workflow.",
  "license": "MIT",
  "type": "module",
@@ -196,3 +196,43 @@ describe('Capy mirror', () => {
    expect(read.state).toBe('present');
  });
 });
+
+describe('import --autoplan --stack', () => {
+  function stackRepo(): string {
+    const d = repo('feature/one');
+    const g = (args: string[]) => expect(spawnSync('git', ['-c', 'user.name=t', '-c', 'user.email=t@example.com', ...args], { cwd: d, timeout: 30_000 }).status).toBe(0);
+    g(['commit', '-q', '--allow-empty', '-m', 'seed']);
+    g(['branch', 'feature/two']);
+    return d;
+  }
+
+  test('writes the same rows to every branch, each in its own record; a retry adds nothing', () => {
+    const run = tmp('appr-run-');
+    autoplanRun(run, [{ by: 'agent:GSTA-36', words: 'Garry: Do it', at: '2026-10-29T03:40:00.000Z', p1: 'a' }]);
+    const cwd = stackRepo(); const state = tmp('appr-state-');
+    const env = { GSTACK_STATE_ROOT: state };
+    const r = bin(['import', '--autoplan', run, '--by', 'Garry', '--stack', 'feature/one,feature/two'], { cwd, env });
+    expect(r.status, r.err).toBe(0);
+    expect(r.out).toMatch(/^APPROVAL_IMPORTED: branch=feature\/one n=1 skipped=0 /m);
+    expect(r.out).toMatch(/^APPROVAL_IMPORTED: branch=feature\/two n=1 skipped=0 /m);
+    expect(r.out).toMatch(/^APPROVAL_IMPORTED_STACK: branches=2 added=2$/m);
+    for (const b of ['feature/one', 'feature/two']) expect(bin(['render', '--branch', b], { cwd, env }).out).toContain('Garry: Do it');
+    expect(bin(['import', '--autoplan', run, '--by', 'Garry', '--stack', 'feature/one,feature/two'], { cwd, env }).out).toMatch(/^APPROVAL_IMPORTED_STACK: branches=2 added=0$/m);
+  });
+
+  test('one unknown branch writes nothing to any branch', () => {
+    const run = tmp('appr-run-');
+    autoplanRun(run, [{ by: 'agent:GSTA-36', words: 'rec', at: '2026-10-29T03:40:00.000Z', p1: 'a' }]);
+    const cwd = stackRepo(); const state = tmp('appr-state-');
+    const r = bin(['import', '--autoplan', run, '--stack', 'feature/one,feature/nope'], { cwd, env: { GSTACK_STATE_ROOT: state } });
+    expect(r.status).toBe(3);
+    expect(r.err).toContain('feature/nope');
+    expect(r.err).toContain('(APPROVAL_RECORD_NO_BRANCH)');
+    expect(bin(['render', '--branch', 'feature/one'], { cwd, env: { GSTACK_STATE_ROOT: state } }).out).not.toContain('Approve the plan');
+  });
+
+  test('--stack goes with import --autoplan and without --branch', () => {
+    const cwd = stackRepo();
+    expect(bin(['import', '--autoplan', tmp('appr-run-'), '--branch', 'feature/one', '--stack', 'feature/one'], { cwd }).status).toBe(2);
+  });
+});
@@ -113,7 +113,7 @@ describe('answer', () => {
  }, 30_000);

  test('a non-recommended choice on a ceo item is recorded without reopening; revise reopens the changed phases, archives their bound files, and the next gate is gate_rev 2', () => {
-    const out = gatedRun('reopen');
+    const out = gatedRun('reopen', ['--no-carry-forward']);
    const r = autoplan(['answer', '--by', 'human', '--out', out, '--gate-rev', '1', '--reply', 'all except d1b', '--json']);
    expect(r.status, r.stderr).toBe(0);
    const runId = readRun(out).run;
@@ -166,7 +166,7 @@ describe('answer', () => {
  }, 30_000);

  test('an eng-only override reopens nothing at answer; revise reruns every phase that reviewed the old plan; p1d reopens every phase; p1c keeps the gate; p1e ends incomplete', () => {
-    const out = gatedRun('eng-only', ['--max-rounds', '3']);
+    const out = gatedRun('eng-only', ['--max-rounds', '3', '--no-carry-forward']);
    const engOnly = autoplan(['answer', '--by', 'human', '--out', out, '--gate-rev', '1', '--reply', 'all except uc2b', '--json']);
    const runId = readRun(out).run;
    expect(JSON.parse(engOnly.stdout)).toMatchObject({ outcome: 'amend_pending', reopened: [], amendments: [`${runId}-uc2`] });
@@ -0,0 +1,327 @@
+/**
+ * Carry-forward (GSTA-155, lib/autoplan-carry.ts): a later gate revision that asks nothing new is answered with the
+ * owner's recorded words from the revision before it, and recorded as a carry; every other case prints why not and
+ * stays at the gate. A "do it" also covers new auto items that change no scope, cost or direction. A capped,
+ * unreviewed delta gets one reviewer on the delta first (CARRY_CHECK, gstack-autoplan check): High or above means no
+ * carry. Runs on the simulated Capy host.
+ */
+import { afterAll, describe, expect, setDefaultTimeout, test } from 'bun:test';
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+import { isDoIt } from '../lib/autoplan-carry';
+import { importAutoplan } from '../lib/approval-record';
+import { renderCaptainBrief } from '../lib/captain-brief';
+import { recHash, renderGate } from '../lib/gate-list';
+import { loadBriefInputs } from '../lib/owner-brief';
+import { SMALL_PLAN } from './helpers/autoplan-run-fixture';
+import { capyRunner, phaseVia, readJsonl, readRunJson, writePlan } from './helpers/autoplan-capy';
+
+setDefaultTimeout(30_000);
+const r = capyRunner();
+afterAll(() => r.cleanup());
+let n = 0;
+const NONE = { native: [], outside: [] };
+const AGREE = {
+  native: [{ id: 'F1', severity: 'High', title: 'Ship the backoff behind a flag first', user_challenge: true, changes: 'scope', fix: 'gate the change behind RETRY_BACKOFF=1' }],
+  outside: [{ id: 'O1', severity: 'High', title: 'Ship the backoff behind a flag first', user_challenge: true, changes: 'scope', fix: 'gate the change behind RETRY_BACKOFF=1' }],
+};
+const V2 = SMALL_PLAN.replace('cap 30 s', 'cap 30 s, behind RETRY_BACKOFF=1');
+const gateOf = (out: string) => JSON.parse(fs.readFileSync(path.join(out, 'gate.json'), 'utf8'));
+const decisions = (out: string) => readJsonl(path.join(out, 'decisions.jsonl'));
+const journal = (out: string) => readJsonl(path.join(out, 'attempts.jsonl'));
+
+/** Gate 1 with one recommended plan change (uc1b), answered with the owner's words, then revised to take it. */
+function answeredAndRevised(args: string[] = [], words = 'Garry: Do it', ceo: { native?: unknown[]; outside?: unknown[] } = AGREE) {
+  const dir = path.join(r.host.base, `carry${++n}`);
+  const plan = writePlan(dir);
+  const out = path.join(dir, 'run');
+  expect(r.ap(['next', '--out', out, '--plan', plan, '--no-ui', ...args]).status).toBe(0);
+  phaseVia(r, out, 'ceo', ceo);
+  expect(phaseVia(r, out, 'eng', NONE).outside.stdout).toContain('status=gate_pending');
+  const run = readRunJson(out).run;
+  const ans = r.ap(['answer', '--out', out, '--gate-rev', '1', '--reply', 'rec', '--rec-hash', recHashOf(out), '--by', 'agent:gstack-minder', '--owner-reply', words]);
+  expect(ans.status, ans.stderr).toBe(0);
+  expect(ans.stdout).toMatch(/^APPLY: /m);
+  const v2 = path.join(dir, 'plan-v2.md');
+  fs.writeFileSync(v2, V2);
+  const rev = r.ap(['revise', '--out', out, '--plan', v2, '--incorporates', 'uc1']);
+  expect(rev.status, rev.stderr).toBe(0);
+  return { dir, out, run, v2, rev };
+}
+const recHashOf = (out: string) => recHash(gateOf(out));
+
+describe('carry-forward', () => {
+  test('a second gate with nothing new carries the owner\'s words: CARRIED, APPROVED, the record names its source', () => {
+    const s = answeredAndRevised();
+    phaseVia(r, s.out, 'ceo', NONE);
+    const eng = phaseVia(r, s.out, 'eng', NONE).outside;
+    expect(eng.status, eng.stderr).toBe(0);
+    expect(eng.stdout).toMatch(/^CARRIED: gate_rev=2 <- gate_rev=1 words="Garry: Do it" by=agent:gstack-minder reason=no new choice; plan unchanged since the accepted changes$/m);
+    expect(eng.stdout).toMatch(/^APPROVED: gate_rev=2/m);
+    expect(eng.stdout).toContain('status=complete');
+    const p1 = decisions(s.out).find(d => d.gate_rev === 2 && d.label === 'P1');
+    expect(p1).toMatchObject({ chosen: 'a', owner_reply: 'Garry: Do it', answered_by: 'agent:gstack-autoplan', carried_gate_from: `${s.run}-p1`, carried_gate_rev: 1, carried_answered_by: 'agent:gstack-minder' });
+    expect(journal(s.out).find(e => e.event === 'gate_carried')).toMatchObject({ gate_rev: 2, from_gate_rev: 1 });
+    expect(journal(s.out).find(e => e.event === 'answered' && e.gate_rev === 1)).toMatchObject({ answer_class: 'plan-changes' });
+    expect(fs.readFileSync(path.join(s.out, 'decisions.tsv'), 'utf8')).toContain('carried gate_rev 2 from gate_rev 1');
+    expect(r.ap(['export', '--out', s.out]).status).toBe(0);
+    expect(r.validate(s.out).status).toBe(0);
+    const rows = importAutoplan(s.out, 'feature/x', 'Garry');
+    if ('code' in rows) throw new Error(rows.message);
+    const carried = rows.find(x => x.source.endsWith('gate_rev=2'))!;
+    expect(carried).toMatchObject({ words: 'Garry: Do it', outcome: 'approved' });
+    expect(carried.ask).toContain('gate_rev 2 carried forward from gate_rev 1');
+    expect(carried.by).toBe('Garry (words from gate_rev 1, relayed by agent:gstack-minder), carried forward by gstack-autoplan');
+    const loaded = loadBriefInputs(s.out);
+    if ('error' in loaded) throw new Error(String(loaded.error));
+    expect(renderCaptainBrief(loaded.inputs, { outDir: s.out }).text).toContain('**Nothing pending.** Carried forward from gate_rev 1: approved as reviewed under the owner\'s words “Garry: Do it”');
+  });
+
+  test('--no-carry-forward: the second gate goes to the owner', () => {
+    const off = answeredAndRevised(['--no-carry-forward']);
+    phaseVia(r, off.out, 'ceo', NONE);
+    const o = phaseVia(r, off.out, 'eng', NONE).outside;
+    expect(o.stdout).toMatch(/^CARRY: no \(carry-forward is off for this run \(--no-carry-forward\)\)$/m);
+    expect(o.stdout).toMatch(/^GATE: gate_rev=2 /m);
+    expect(o.stdout).toContain('status=gate_pending');
+  });
+
+  test('an edit to the plan file after the revise (an ELI5-only line) refuses the carry', () => {
+    const edited = answeredAndRevised();
+    fs.appendFileSync(edited.v2, '\nAn ELI5 sentence added after the revise.\n');
+    phaseVia(r, edited.out, 'ceo', NONE);
+    const e = phaseVia(r, edited.out, 'eng', NONE).outside;
+    expect(e.stdout).toMatch(/^CARRY: no \(the plan file plan-v2\.md changed after the owner's answer or the revise that took it\)$/m);
+    expect(readRunJson(edited.out).status).toBe('gate_pending');
+  });
+
+  test('a new scope challenge or an urgent finding refuses the carry', () => {
+    const scope = answeredAndRevised();
+    phaseVia(r, scope.out, 'ceo', NONE);
+    const s2 = phaseVia(r, scope.out, 'eng', { native: [{ id: 'F9', severity: 'High', title: 'Add a dead-letter queue', user_challenge: true, changes: 'scope', fix: 'dlq' }], outside: [] }).outside;
+    expect(s2.stdout).toMatch(/^CARRY: no \(new item \w+: Add a dead-letter queue/m);
+
+    const urgent = answeredAndRevised();
+    phaseVia(r, urgent.out, 'ceo', NONE);
+    const u = phaseVia(r, urgent.out, 'eng', { native: [{ id: 'F8', severity: 'Medium', title: 'Tokens logged in plain text', urgent: true, suggested_owner: 'security' }], outside: [] }).outside;
+    expect(u.stdout).toMatch(/^CARRY: no \(urgent finding open: Tokens logged in plain text\)$/m);
+  });
+
+  test('"do it" also covers a new packaging challenge; "rec" does not', () => {
+    const PKG = { native: [{ id: 'F7', severity: 'Medium', title: 'Split the backoff into its own PR', user_challenge: true, changes: 'packaging', fix: 'own PR' }], outside: [] };
+    const doIt = answeredAndRevised();
+    phaseVia(r, doIt.out, 'ceo', NONE);
+    const d = phaseVia(r, doIt.out, 'eng', PKG).outside;
+    expect(d.stdout).toMatch(/^CARRIED: gate_rev=2 <- gate_rev=1 .*reason="do it" covers \w+; plan unchanged/m);
+    expect(d.stdout).toMatch(/^CARRIED_UNDER_DO_IT: \w+ Split the backoff into its own PR .*\(the owner has not seen this item\)$/m);
+    expect(decisions(doIt.out).find(x => x.gate_rev === 2 && /Split the backoff/.test(x.title))).toMatchObject({ carried_under: 'do it' });
+    const loaded = loadBriefInputs(doIt.out);
+    if ('error' in loaded) throw new Error(String(loaded.error));
+    expect(renderCaptainBrief(loaded.inputs, { outDir: doIt.out }).text).toContain('Taken under "do it" without being shown to the owner: Split the backoff into its own PR.');
+    const rec = answeredAndRevised([], 'rec');
+    phaseVia(r, rec.out, 'ceo', NONE);
+    expect(phaseVia(r, rec.out, 'eng', PKG).outside.stdout).toMatch(/^CARRY: no \(new item \w+: Split the backoff/m);
+  });
+
+test('an earlier answer in item tokens, or one that left an item unanswered, is not carried', () => {
+    const dir = path.join(r.host.base, `carry${++n}`);
+    const out = path.join(dir, 'run');
+    expect(r.ap(['next', '--out', out, '--plan', writePlan(dir), '--no-ui']).status).toBe(0);
+    phaseVia(r, out, 'ceo', AGREE);
+    phaseVia(r, out, 'eng', NONE);
+    const g = gateOf(out);
+    const tokens = g.items.map((i: any) => `${i.id}${i.recommended}`).join(' ');
+    expect(r.ap(['answer', '--out', out, '--gate-rev', '1', '--reply', tokens, '--by', 'human']).status).toBe(0);
+    const v2 = path.join(dir, 'plan-v2.md');
+    fs.writeFileSync(v2, V2);
+    expect(r.ap(['revise', '--out', out, '--plan', v2, '--incorporates', 'uc1']).status).toBe(0);
+    phaseVia(r, out, 'ceo', NONE);
+    expect(phaseVia(r, out, 'eng', NONE).outside.stdout).toMatch(/^CARRY: no \(gate_rev 1 was answered with item tokens, not rec or "do it"\)$/m);
+  });
+
+  test('an answer that left an approval item pending is not carried', () => {
+    const dir = path.join(r.host.base, `carry${++n}`);
+    const out = path.join(dir, 'run');
+    expect(r.ap(['next', '--out', out, '--plan', writePlan(dir), '--no-ui']).status).toBe(0);
+    phaseVia(r, out, 'ceo', AGREE);
+    phaseVia(r, out, 'eng', NONE);
+    const p1 = gateOf(out).items.find((i: any) => i.id === 'p1');
+    const ans = r.ap(['answer', '--out', out, '--gate-rev', '1', '--reply', `p1${p1.recommended}`, '--by', 'human']);
+    expect(ans.status, ans.stderr).toBe(0);
+    expect(decisions(out).some(x => x.gate_rev === 1 && x.status === 'pending')).toBe(true);
+    const v2 = path.join(dir, 'plan-v2.md');
+    fs.writeFileSync(v2, V2);
+    expect(r.ap(['revise', '--out', out, '--plan', v2]).status).toBe(0);
+    phaseVia(r, out, 'ceo', NONE);
+    expect(phaseVia(r, out, 'eng', NONE).outside.stdout).toMatch(/^CARRY: no \(gate_rev 1 left \S+ unanswered\)$/m);
+  });
+
+  test('isDoIt reads the owner\'s words, not a Captain\'s', () => {
+    for (const w of ['Do it', 'do it.', 'Garry: Do it', 'Garry Tan: do it!']) expect(isDoIt(w)).toBe(true);
+    for (const w of ['rec', 'accept all', 'Captain: Do it', 'GStack Captain: do it', 'p3b: do it', "Captain: rec, Highs as build tasks under Garry's 'Do it'", 'Do it and also improve it']) expect(isDoIt(w)).toBe(false);
+  });
+});
+
+describe('the delta check before a carry', () => {
+  function capped() {
+    const s = answeredAndRevised(['--max-rounds', '1']);
+    expect(s.rev.stdout).toMatch(/^CAPPED: ceo used round 1 of 1/m);
+    expect(s.rev.stdout).toMatch(/^CARRY_CHECK: attempt=\S+-c2 items=uc\d+,uc\d+ prompt=\S+ result=\S+ model_family=not anthropic estimate=5\.00 cap=5\.00$/m);
+    expect(s.rev.stdout).toMatch(/^GATE: gate_rev=2 /m);
+    const run = readRunJson(s.out);
+    expect(run.status).toBe('gate_pending');
+    const check = run.carry.check;
+    const prompt = fs.readFileSync(check.prompt, 'utf8');
+    expect(prompt).toContain('+ ');
+    expect(prompt).toContain(`INPUT: carry-check ${check.delta_sha256}`);
+    return { ...s, check };
+  }
+  const result = (file: string, sha: string, findings: unknown[]) => fs.writeFileSync(file, `INPUT: carry-check ${sha}\n\nReviewed the delta.\n\n` + '```gstack-findings\n' + findings.map(f => JSON.stringify(f)).join('\n') + '\n```\n');
+
+  test('a clean delta check carries; the verdict is on the gate, the decisions and the trail', () => {
+    const s = capped();
+    result(s.check.result, s.check.delta_sha256, [{ id: 'F1', severity: 'Low', title: 'Wording' }]);
+    const same = r.ap(['check', '--out', s.out, '--attempt', s.check.attempt, '--result', s.check.result, '--model', 'anthropic/claude-opus-5-5']);
+    expect(same.status).toBe(3);
+    expect(same.stderr).toContain('MODEL_FAMILY_CONFLICT');
+    const ok = r.ap(['check', '--out', s.out, '--attempt', s.check.attempt, '--result', s.check.result, '--model', 'openai/gpt-6.1-sol']);
+    expect(ok.status, ok.stderr).toBe(0);
+    expect(ok.stdout).toMatch(/^CARRY_CHECK: passed attempt=\S+ model=openai\/gpt-6\.1-sol findings=1 highest=Low$/m);
+    expect(ok.stdout).toMatch(/^CARRIED: gate_rev=2 <- gate_rev=1 .*reason=delta check passed on uc\d+,uc\d+; plan unchanged/m);
+    expect(ok.stdout).toContain('status=complete');
+    expect(gateOf(s.out).carry_check).toMatchObject({ verdict: 'passed', model: 'openai/gpt-6.1-sol', max_severity: 'Low', findings: 1 });
+    expect(decisions(s.out).filter(x => x.gate_rev === 2 && x.carry_check).length).toBe(2);
+    expect(fs.readFileSync(path.join(s.out, 'decisions.tsv'), 'utf8')).toContain('delta check before carrying gate_rev 2');
+    expect(r.ap(['export', '--out', s.out]).status).toBe(0);
+    expect(r.validate(s.out).status).toBe(0);
+  });
+
+  test('a High finding, or no reviewer, leaves the gate for the owner', () => {
+    const s = capped();
+    result(s.check.result, s.check.delta_sha256, [{ id: 'F1', severity: 'High', title: 'The edit also drops the jitter' }]);
+    const bad = r.ap(['check', '--out', s.out, '--attempt', s.check.attempt, '--result', s.check.result, '--model', 'openai/gpt-6.1-sol']);
+    expect(bad.status, bad.stderr).toBe(0);
+    expect(bad.stdout).toMatch(/^CARRY: no \(delta check found High: The edit also drops the jitter\)$/m);
+    expect(bad.stdout).toMatch(/^GATE: gate_rev=2 /m);
+    const g = gateOf(s.out);
+    expect(g.carry_check).toMatchObject({ verdict: 'failed', max_severity: 'High' });
+    expect(g.items.filter((i: any) => i.cap?.kind === 'delta').every((i: any) => i.flags.some((f: string) => f.includes('The edit also drops the jitter')))).toBe(true);
+    const again = r.ap(['check', '--out', s.out, '--attempt', s.check.attempt, '--unavailable', 'x']);
+    expect(again.status).toBe(3);
+    const u = capped();
+    const none = r.ap(['check', '--out', u.out, '--attempt', u.check.attempt, '--unavailable', 'no other family online']);
+    expect(none.status, none.stderr).toBe(0);
+    expect(none.stdout).toMatch(/^CARRY: no \(delta check unavailable: no other family online\)$/m);
+    expect(JSON.parse(fs.readFileSync(path.join(u.out, 'spend.json'), 'utf8')).released.map((x: any) => x.attempt)).toContain(u.check.attempt);
+  });
+
+  // Value: protects=a delta check over --carry-check-cap or with the wrong --attempt never carries the owner's answer;
+  // fails_when=the cap comparison or the attempt guard is dropped or inverted; why_new=tests above run at the default cap
+  // with the right attempt only; seam=none; no_claim=real reviewer dispatch (simulated Capy host)
+  test('a check over --carry-check-cap refuses the carry; a wrong --attempt is refused', () => {
+    const over = answeredAndRevised(['--max-rounds', '1', '--carry-check-cap', '1']);
+    expect(over.rev.stdout).toMatch(/^CARRY: no \(delta check estimate 5\.00 is over the per-gate cap 1\.00 \(--carry-check-cap\)\)$/m);
+    expect(over.rev.stdout).not.toMatch(/^CARRY_CHECK:/m);
+    expect(readRunJson(over.out).status).toBe('gate_pending');
+    const s = capped();
+    result(s.check.result, s.check.delta_sha256, []);
+    const wrong = r.ap(['check', '--out', s.out, '--attempt', 'not-the-attempt', '--result', s.check.result, '--model', 'openai/gpt-6.1-sol']);
+    expect(wrong.status).toBe(3);
+    expect(wrong.stderr).toContain('ATTEMPT_MISMATCH');
+    expect(readRunJson(s.out).status).toBe('gate_pending');
+  });
+
+  test('an owner answer at the gate voids a pending delta check and releases its reservation', () => {
+    const s = capped();
+    const ans = r.ap(['answer', '--out', s.out, '--gate-rev', '2', '--reply', 'p1c', '--by', 'human']);
+    expect(ans.status, ans.stderr).toBe(0);
+    expect(ans.stdout).toContain(`CARRY: no (delta check ${s.check.attempt} void: the owner answered gate_rev 2)`);
+    result(s.check.result, s.check.delta_sha256, []);
+    const late = r.ap(['check', '--out', s.out, '--attempt', s.check.attempt, '--result', s.check.result, '--model', 'openai/gpt-6.1-sol']);
+    expect(late.status).toBe(3);
+    expect(late.stderr).toContain('PHASE_NOT_AWAITING');
+    expect(readRunJson(s.out).status).not.toBe('complete');
+    expect(decisions(s.out).some(x => x.gate_rev === 2 && x.carried_gate_from)).toBe(false);
+    expect(JSON.parse(fs.readFileSync(path.join(s.out, 'spend.json'), 'utf8')).released.map((x: any) => x.attempt)).toContain(s.check.attempt);
+  });
+
+  test('a plan edited after the check was dispatched is not carried on that check', () => {
+    const s = capped();
+    fs.appendFileSync(readRunJson(s.out).active_plan, '\nA line the reviewer never saw.\n');
+    result(s.check.result, s.check.delta_sha256, []);
+    const c = r.ap(['check', '--out', s.out, '--attempt', s.check.attempt, '--result', s.check.result, '--model', 'openai/gpt-6.1-sol']);
+    expect(c.status, c.stderr).toBe(0);
+    expect(c.stdout).toContain('void: the plan changed after the delta check was dispatched');
+    expect(c.stdout).not.toMatch(/^CARRIED:/m);
+    expect(readRunJson(s.out).status).toBe('gate_pending');
+  });
+
+  test('--same-family binds a same-family reviewer and records why', () => {
+    const s = capped();
+    result(s.check.result, s.check.delta_sha256, []);
+    const ok = r.ap(['check', '--out', s.out, '--attempt', s.check.attempt, '--result', s.check.result, '--model', 'anthropic/claude-opus-5-5', '--same-family', 'no other family online']);
+    expect(ok.status, ok.stderr).toBe(0);
+    expect(gateOf(s.out).carry_check.model).toBe('anthropic/claude-opus-5-5 (same family: no other family online)');
+  });
+});
+
+describe('the review estimate', () => {
+  test('the first next prints the most the review can reserve', () => {
+    const dir = path.join(r.host.base, 'estimate');
+    const out = path.join(dir, 'run');
+    const first = r.ap(['next', '--out', out, '--plan', writePlan(dir), '--no-ui']);
+    expect(first.stdout).toMatch(/^REVIEW_ESTIMATE: phases=ceo,eng voices=2\+2 rounds<=2 attempts<=8 per_voice=5\.00 reservations<=40\.00 \(a reservation total, not a spend ceiling/m);
+    expect(r.ap(['next', '--out', out]).stdout).not.toContain('REVIEW_ESTIMATE');
+  });
+});
+
+describe('gate eli5', () => {
+  const SECTIONS = '\n## Before → after\n\nBefore: retries hammer. After: they back off.\n\n## Why it matters\n\nOutages.\n\n## What caused this work\n\nIncident 7.\n\n## Big decisions\n\n1. Exponential backoff.\n';
+  function gated(name: string, text: string) {
+    const dir = path.join(r.host.base, name);
+    const out = path.join(dir, 'run');
+    expect(r.ap(['next', '--out', out, '--plan', writePlan(dir, text), '--no-ui']).status).toBe(0);
+    phaseVia(r, out, 'ceo', NONE);
+    expect(phaseVia(r, out, 'eng', NONE).outside.stdout).toContain('status=gate_pending');
+    return out;
+  }
+
+  test('gate.json carries the plan\'s ELI5 and four sections, and page 1 of the render opens with them', () => {
+    const out = gated('eli5', SMALL_PLAN + '\n## ELI5\n\nRetries wait longer each time.\n' + SECTIONS);
+    const gate = gateOf(out);
+    expect(gate.eli5).toMatchObject({ summary: 'Retries wait longer each time.', before_after: 'Before: retries hammer. After: they back off.', why: 'Outages.', cause: 'Incident 7.', decisions: '1. Exponential backoff.' });
+    expect(gate.eli5.plan_file_sha256).toMatch(/^[0-9a-f]{64}$/);
+    const page = renderGate(gate).split('\n');
+    expect(page.slice(2, 4)).toEqual(['ELI5', 'Retries wait longer each time.']);
+    expect(page.indexOf('Before → after:')).toBeGreaterThan(3);
+    expect(page.indexOf('Before → after:')).toBeLessThan(page.findIndex(l => /^1\. \[/.test(l)));
+    const loaded = loadBriefInputs(out);
+    if ('error' in loaded) throw new Error(String(loaded.error));
+    const brief = renderCaptainBrief(loaded.inputs, { outDir: out });
+    expect(brief.missing).toEqual([]);
+    expect(brief.text.indexOf('### ELI5')).toBeLessThan(brief.text.indexOf('### Before → after'));
+  });
+
+  test('a plan without ## ELI5: the Captain brief is not ready and names it; the plain gate render does not ask for it', () => {
+    const out = gated('no-eli5', SMALL_PLAN + SECTIONS);
+    const page = renderGate(gateOf(out));
+    expect(page).not.toContain('CAPTAIN_BRIEF_INCOMPLETE');
+    expect(page).toContain('Before → after:');
+    const loaded = loadBriefInputs(out);
+    if ('error' in loaded) throw new Error(String(loaded.error));
+    const brief = renderCaptainBrief(loaded.inputs, { outDir: out });
+    expect(brief.missing).toEqual(['ELI5']);
+    expect(brief.text).toStartWith('**Not ready for approval:**');
+  });
+
+  test('a gate written before the eli5 field renders and briefs as before', () => {
+    const out = gated('legacy', SMALL_PLAN + SECTIONS);
+    const gate = gateOf(out);
+    delete gate.eli5;
+    fs.writeFileSync(path.join(out, 'gate.json'), JSON.stringify(gate, null, 2) + '\n');
+    expect(renderGate(gate)).not.toContain('ELI5');
+    const loaded = loadBriefInputs(out);
+    if ('error' in loaded) throw new Error(String(loaded.error));
+    expect(renderCaptainBrief(loaded.inputs, { outDir: out }).missing).toEqual([]);
+  });
+});
@@ -157,7 +157,7 @@ describe('gstack-owner-brief render --captain', () => {
 describe('carried decisions at a cap gate (#3546)', () => {
  const r = capyRunner();
  afterAll(() => r.cleanup());
-  const SECTIONS = '\n## Before → after\n\nBefore: retries hammer. After: they back off.\n\n## Why it matters\n\nOutages.\n\n## What caused this work\n\nIncident 7.\n\n## Big decisions\n\n1. Exponential backoff.\n';
+  const SECTIONS = '\n## ELI5\n\nRetries wait longer each time.\n\n## Before → after\n\nBefore: retries hammer. After: they back off.\n\n## Why it matters\n\nOutages.\n\n## What caused this work\n\nIncident 7.\n\n## Big decisions\n\n1. Exponential backoff.\n';
  const PLAN = SMALL_PLAN + SECTIONS;
  const WORDS: Record<string, string> = { 'keep your direction': 'keep the plan as written', 'accept the change': "make the reviewer's change" };

@@ -19,6 +19,10 @@ afterAll(() => r.cleanup());

 const PLAN = `# Retry backoff for the webhook worker

+## ELI5
+
+A failed webhook now waits longer before each retry, so the partner stops dropping them.
+
 ## Before → after

 Before: a failed webhook retries every second until it is dropped. After: retries back off from 500 ms to 30 s.