elanous 0.2.37 → 0.2.40

package/src/task-orchestrator/surfaces/self-implement-pod.ts

@@ -33,6 +33,7 @@ import { mintRunId, normalizeRunId } from '../../harness/harness-space.js';
 import { TA_LAND_ENV } from '../../task-agent/task-hand.js';
 import { appendRunLedgerEntry, loadRunLedger, runLedgerDir, runLedgerPath, type RunLedgerEntry } from '../../self-implement/run-ledger.js';
 import { recordRunStop } from '../../self-implement/run-stop.js';
+import { gateFailureReceipt, gateFailureReceiptLines } from '../../self-implement/gate-failure-signature.js';
 import { runSuffixKey } from '../../self-implement/salvage-branches.js';
 import type { GoalExecutionRecord } from '../../self-implement/orchestrator.js';
 import { hostRepositoryUnavailable, runHostRegate, type HostRegateResult } from '../../self-implement/host-regate.js';
@@ -46,7 +47,7 @@ import { executeSiWarmTransport, type SiWarmFixedRunner,
   type SiWarmPodIdentity } from './pod-si-warm-transport.js';
 
 export const POD_CONTROL_INBOX_DIR = '/tmp/elanous-control.inbox';
-/** Pod Job 수명 상한(`activeDeadlineSeconds`) — 대표 2026-09-26 90분 → 180분(90분에 `DeadlineExceeded` 로 죽은 런 둘). */
+/** Pod Job 수명 상한(`activeDeadlineSeconds`) — 2026-09-26 대표 지시로 90분 → 180분(90분에 `DeadlineExceeded` 로 죽은 런 둘). */
 /** Newest pod log bytes returned when the export passes the 5MB artifact limit (headroom under it). */
 export const POD_LOGS_KEEP_BYTES = 4_500_000;
 
@@ -60,6 +61,34 @@ export const POD_RESULT_FLATTEN = `ELANOUS_CHILD_RC="$rc" bun -e ' const fs=requ
 
 export const POD_JOB_DEADLINE_SECONDS = 10_800;
 
+/** Record the source-stage IO marker emitted by podSourceScript. */
+export function recordPodSourceIoSummary(logs: string, job: string, runId: string | null | undefined,
+  runIdSource: 'explicit-child-id' | 'pod-launch-child-run-id',
+  log: typeof debug.log = (category, event, data, options) => debug.log(category, event, data, options),
+  hostOriginCalls: number | null = 0): void {
+  const ioSummary = /^ELANOUS_POD_SOURCE_IO (\{[^\r\n]+\})\r?$/m.exec(logs)?.[1];
+  if (!ioSummary) return;
+  try {
+    const measured = JSON.parse(ioSummary) as Record<string, unknown>;
+    log('pod.source', 'io-summary', { job, runId, runIdSource, ...measured, hostOriginCalls });
+  } catch (error) {
+    log('pod.source', 'io-summary-invalid', { job, runId, runIdSource, reason: error instanceof Error ? error.message : String(error) }, { level: 'warn' });
+  }
+}
+
+/** Select a timestamp for a DeadlineExceeded observation without making detector time look like Job completion.
+ *  `podFinishedAt` must come from readPodTermination for the same verified Job uid. */
+export function podJobFinishedTiming(input: { jobCreatedAt: string | null; jobCompletedAt: string | null;
+  podFinishedAt: string | null; observedAt: string }): { finishedAt: string; finishedAtSource: 'job-completion' | 'pod-termination' | 'observed-at'; elapsedMs: number | null } {
+  const valid = (value: string | null): value is string => typeof value === 'string' && Number.isFinite(Date.parse(value));
+  const selected = valid(input.jobCompletedAt) ? { finishedAt: input.jobCompletedAt, finishedAtSource: 'job-completion' as const }
+    : valid(input.podFinishedAt) ? { finishedAt: input.podFinishedAt, finishedAtSource: 'pod-termination' as const }
+      : { finishedAt: input.observedAt, finishedAtSource: 'observed-at' as const };
+  const elapsedMs = valid(input.jobCreatedAt) && Number.isFinite(Date.parse(selected.finishedAt))
+    ? Math.max(0, Date.parse(selected.finishedAt) - Date.parse(input.jobCreatedAt)) : null;
+  return { ...selected, elapsedMs };
+}
+
 /** 자식 컨테이너 «요청» — 🩸 2026-09-27: limits 만 두면 k8s 가 requests=limits(cpu 4)로 잡아, 32코어 노드에
  *  «자리 요청»이 88% 차서 잡이 Pending 인데 실사용은 14%였다(풀 25 자리 중 ~8 만 떴다). 상한(limits)은 그대로 두고
  *  예약만 실사용에 맞춘다. */
@@ -86,20 +115,21 @@ import { enterJourneyNode, exitJourneyNode, passJourneyNode } from '../../self-d
 import { routeJourneyEdge } from '../../self-dev/graph-journey-route.js';
 import { selectLiveDetail } from '../../live/detail-switch.js';
 import { authStorePath } from '../../oauth/store.js';
-import { grokAuthFilePath, isGrokSubscriptionExpiring, refreshGrokSubscriptionToken, resolveGrokCredential, type GrokRefreshOutcome } from '../../grok/credential.js';
+import { grokAuthFilePath, isGrokSubscriptionExpiring, readSelectedGrokScope, refreshGrokSubscriptionToken, resolveGrokCredential, type GrokRefreshOutcome } from '../../grok/credential.js';
 import { defaultGrokModel } from '../../grok/models.js';
 import { resolveHostId } from '../../platform/host-id.js';
 import { BUDGET_SNAPSHOT_ENV, encodeBudgetSnapshot, harnessLaunchBudgetDecision, type BudgetSnapshot } from '../../execution-loop/launch-gate.js';
 import type { ReadBudgetInputsDeps } from '../../self-implement/budget-gate.js';
 import { envLiteral } from '../../platform/env-literal.js';
 import { LLM_TIER_MAP_BY_PROVIDER, lookupLlmTierSpec, type LlmTierProvider } from '../../model-tier/llm-tier-map.js';
-import { parseSelfImplementJson, type SelfImplementJobDone, type SelfImplementJobSpawn } from './self-implement.js';
+import { parseSelfImplementJson, type SelfImplementDisposition, type SelfImplementJobDone, type SelfImplementJobSpawn } from './self-implement.js';
 import { codexQuotaPolicyFromConfig } from '../../oauth/codex-account-store.js';
 import { normalizeFallbackChain, type FallbackStep } from '../../oauth/fallback-chain.js';
 import { goalTypeOf, type GoalType } from '../../../scripts/measure-pod-memory-by-goal.js';
 import { declaredGoalType, type GoalType as DeclaredGoalType } from '../../self-implement/goal-author.js';
 import { readPodMemoryAdvice, type PodMemoryAdvice } from '../../cli/pod-memory-advice.js';
 import { DEFAULT_BUDGET_GATE_MAX_USED_PERCENT, getUserConfig, roleLlmEnvName, type UserConfig } from '../../user-config.js';
+import { worktreeAddTimeoutSec } from '../../git-fs/worktree.js';
 import { podChildProviderOf, podChildProviderRefusal, type PodChildProvider } from './pod-child-providers.js';
 import { BEDROCK_POD_EGRESS_LABEL, bedrockPodEnv, checkBedrockPodSecret, type BedrockPodSpec } from './pod-bedrock-child.js';
 import { resolveBedrockRegion, validBedrockRegion } from '../../llm/bedrock.js';
@@ -107,7 +137,7 @@ import { ZAI_POD_EGRESS_LABEL, checkZaiPodSecret, countActiveZaiJobs, decideZaiP
 import { ZAI_CODING_DEFAULT_MODEL, ZAI_OPENROUTER_FALLBACK_MODEL, zaiSettingsFromConfig, type ZaiCodingPlanSettings } from '../../llm/zai-coding-plan.js';
 import { normalizePlanningFallbackReason, selectEnhanceLlm } from '../../prompt-enhance/enhance.js';
 import { inferProviderFromModel } from '../../llm.js';
-import { classifyPodExit, extractPodFailureReason, podChildFailureAttribution, podTerminalRow, POD_TERMINATION_JOB_JSONPATH, POD_TERMINATION_PODS_JSONPATH, readPodTermination, type PodExitClass, type PodTermination } from './pod-failure-reason.js';
+import { classifyPodExit, extractPodFailureReason, podChildFailureAttribution, podDeadlineResultMarker, podJobConditionReason, podTerminalRow, POD_TERMINATION_JOB_JSONPATH, POD_TERMINATION_PODS_JSONPATH, readPodTermination, type PodExitClass, type PodTermination } from './pod-failure-reason.js';
 import { kubectlTransientReason, retryKubectlTransient } from './kubectl-transient-retry.js';
 import { OLD_DOOR_STAMP_ENV } from '../../self-dev/old-door.js';
 import { resolveDefaultSourceHead, seedMirrorOnce, type MirrorSeedResult } from './pod-mirror-seed.js';
@@ -199,7 +229,7 @@ export interface PodSpawnOptions {
   warmPodRequest?: WarmPodRequestHook;
   /** SI-SEED-DEFAULT — resolves a default source's branch tip for the mirror seed (test seam · default
    *  {@link resolveDefaultSourceHead}; none with an injected kubectl unless given). */
-  resolveSourceHead?: (o: { repoUrl: string; base?: string; cwd: string }) => Promise<{ sha?: string; reason?: string }>;
+  resolveSourceHead?: (o: { repoUrl: string; base?: string; cwd: string }) => Promise<{ sha?: string; reason?: string; originCalls?: number }>;
   /** 설정 읽기(시험 주입). kubectl 주입 시험에서는 기본 사용자 설정을 읽지 않는다. */
   configHostMirror?: () => string | undefined;
   /** AUTHOR-POD2 host config seam; Pod itself has no host config. */
@@ -511,7 +541,7 @@ export function hostCredentials(account: string, storePath: string = authStorePa
  *  (POD_GROK_MIN_JOB_SECONDS) plus a margin left, and that launch's `activeDeadlineSeconds` is cut to the token's
  *  remaining life minus the margin (`fitPodDeadlineToGrok`) — so a long run ends as a classified DeadlineExceeded
  *  instead of a silent mid-run 401. With the host credential relay (pod.groundingUrl) the Pod renews mid-run and
- *  nothing is cut. 90 min = the Job cap before 대표 09-26 raised it (only two runs had reached it). */
+ *  nothing is cut. 90 min = the Job cap before the 09-26 decision raised it (only two runs had reached it). */
 export const POD_GROK_MIN_JOB_SECONDS = 90 * 60;
 /** Slack between the fitted Job end and the token expiry (clock skew · Pod scheduling). */
 export const POD_GROK_EXPIRY_MARGIN_MS = 15 * 60_000;
@@ -526,15 +556,6 @@ export function formatGrokRemaining(ms: number | null): string {
   return m >= 60 ? `${Math.floor(m / 60)}h${String(m % 60).padStart(2, '0')}m` : `${m}m`;
 }
 
-/** Latest `expires_at` among scopes that carry an access key (the token the Pod would get). */
-function grokSubscriptionExpiresAtMs(home?: string): number | null {
-  try {
-    const scopes = JSON.parse(readFileSync(grokAuthFilePath(home), 'utf8')) as Record<string, { key?: unknown; expires_at?: unknown }>;
-    const all = Object.values(scopes).flatMap((s) => (s && typeof s.key === 'string' && s.key && typeof s.expires_at === 'string' && Number.isFinite(Date.parse(s.expires_at)) ? [Date.parse(s.expires_at)] : []));
-    return all.length ? Math.max(...all) : null;
-  } catch { return null; }
-}
-
 /** POD-GROK-VALIDITY-FIT — this launch's Job deadline for a refresh-less grok subscription copy.
  *  · relay (the Pod can renew from the host) or unknown expiry → the requested deadline as is.
  *  · otherwise → min(requested, remaining − margin); below the minimum Job (time passed since the credential check)
@@ -574,38 +595,46 @@ export function podGhLoginRetryBackoffMs(job: string): number {
 export function hostGrokCredentials(opts: {
   home?: string; env?: NodeJS.ProcessEnv; apiKeyOptIn?: boolean; ghToken?: () => string;
   /** Test seams — the host refresh (read-only `grok models`) and the expiry probe. */
-  refresh?: () => void; isExpiring?: () => boolean;
+  refresh?: () => void; isExpiring?: () => boolean; grokPath?: string;
   /** Required remaining token life (default POD_GROK_MIN_VALIDITY_MS) — a caller whose Job is not fitted passes its own deadline. */
   minValidityMs?: number; now?: () => number;
 } = {}): { grokAuth?: string; grokApiKey?: string; ghToken: string } {
   let credential = resolveGrokCredential({ home: opts.home, env: opts.env });
   if (credential?.kind === 'subscription') {
+    const initialScope = readSelectedGrokScope(opts.home);
+    if (!initialScope) throw new Error('grok: 선택된 구독 scope 없음 — 발사를 거부');
     const minValidityMs = opts.minValidityMs ?? POD_GROK_MIN_VALIDITY_MS;
-    const expiring = opts.isExpiring ?? (() => isGrokSubscriptionExpiring({ home: opts.home, bufferMs: minValidityMs, ...(opts.now ? { now: opts.now } : {}) }));
-    if (expiring()) {
-      try { (opts.refresh ?? (() => { refreshGrokSubscriptionToken({ home: opts.home }); }))(); } catch { /* judged below */ }
-      if (expiring()) {
-        const at = grokSubscriptionExpiresAtMs(opts.home);
-        const remainingMs = at === null ? null : at - (opts.now ?? Date.now)();
-        try { debug.log('pod.grok-credentials', 'refused', { remainingMs, minValidityMs, expiresAt: at === null ? null : new Date(at).toISOString() }); } catch { /* fail-soft */ }
-        throw new Error(`grok: 구독 access 토큰 남은 시간 ${formatGrokRemaining(remainingMs)} < Pod 발사 기준 ${formatGrokRemaining(minValidityMs)} — 호스트 갱신(grok models)도 못 늘렸다 · 다음: 호스트에서 grok login --device-auth 후 다시`);
+    const clock = opts.now ?? Date.now;
+    const initialExpiry = initialScope.expiresAt ? Date.parse(initialScope.expiresAt) : NaN;
+    const expiring = opts.isExpiring ? opts.isExpiring()
+      : !Number.isFinite(initialExpiry) || initialExpiry - clock() < minValidityMs;
+    if (expiring) {
+      if (!initialScope.hasRefreshToken) {
+        throw new Error('grok: 선택된 구독 scope에 refresh token 없음 — 다른 계정으로 대체하지 않고 발사를 거부');
       }
-      credential = resolveGrokCredential({ home: opts.home, env: opts.env });
-      if (credential?.kind !== 'subscription') throw new Error('grok: 갱신 뒤 구독 자격을 다시 읽지 못했다');
+      try {
+        if (opts.refresh) opts.refresh();
+        else refreshGrokSubscriptionToken({ home: opts.home, requireRefreshToken: true,
+          requiredValidityMs: minValidityMs, nowMs: clock(), ...(opts.grokPath ? { grokPath: opts.grokPath } : {}) });
+      } catch { /* Final snapshot below is the launch decision. */ }
     }
-    const scopes = JSON.parse(readFileSync(grokAuthFilePath(opts.home), 'utf8')) as Record<string, unknown>;
-    const redacted: Record<string, { key: string; expires_at?: string; user_id?: string }> = {};
-    for (const [name, value] of Object.entries(scopes)) {
-      if (!value || typeof value !== 'object' || Array.isArray(value)) continue;
-      const scope = value as Record<string, unknown>;
-      if (scope.key !== credential.token) continue;
-      redacted[name] = {
-        key: scope.key,
-        ...(typeof scope.expires_at === 'string' ? { expires_at: scope.expires_at } : {}),
-        ...(typeof scope.user_id === 'string' ? { user_id: scope.user_id } : {}),
-      };
+    const selected = readSelectedGrokScope(opts.home);
+    if (!selected || selected.scopeName !== initialScope.scopeName) {
+      throw new Error('grok: 자격 확인 중 선택된 구독 scope가 사라졌거나 바뀜 — 발사를 거부');
+    }
+    const selectedExpiry = selected.expiresAt ? Date.parse(selected.expiresAt) : NaN;
+    if (!Number.isFinite(selectedExpiry) || selectedExpiry - clock() < minValidityMs) {
+      const remainingMs = Number.isFinite(selectedExpiry) ? selectedExpiry - clock() : null;
+      try { debug.log('pod.grok-credentials', 'refused', { remainingMs, minValidityMs, expiresAt: selected.expiresAt }); } catch { /* fail-soft */ }
+      throw new Error(`grok: 구독 access 토큰 남은 시간 ${formatGrokRemaining(remainingMs)} < Pod 발사 기준 ${formatGrokRemaining(minValidityMs)} — 호스트 갱신(grok models)도 못 늘렸다 · 다음: 호스트에서 grok login --device-auth 후 다시`);
     }
-    if (!Object.keys(redacted).length) throw new Error('grok: 구독 access 토큰 없음');
+    const redacted: Record<string, { key: string; expires_at?: string; user_id?: string }> = {
+      [selected.scopeName]: {
+        key: selected.key,
+        ...(selected.expiresAt ? { expires_at: selected.expiresAt } : {}),
+        ...(selected.userId ? { user_id: selected.userId } : {}),
+      },
+    };
     return { grokAuth: JSON.stringify(redacted), ghToken: (opts.ghToken ?? defaultGhToken)() };
   }
   if (credential?.kind === 'api_key' && opts.apiKeyOptIn === true) {
@@ -959,7 +988,7 @@ export function podRunLabels(o: { runId?: string; parentRunId?: string }): Recor
   return { ...(run ? { 'elanous.run': run } : {}), ...(child ? { 'elanous.child-run': child } : {}) };
 }
 
-/** Pod 메모리 «등급» — 대표 09-27: PWA Pod 골이 16Gi 에서 OOM 으로 죽는다 ⇒ «high» 등급을 옵션으로 고른다.
+/** Pod 메모리 «등급» — 09-27 결정: PWA Pod 골이 16Gi 에서 OOM 으로 죽는다 ⇒ «high» 등급을 옵션으로 고른다.
  *  등급 = `standard`(`ELANOUS_POD_MEMORY` 또는 16Gi) · `high`(`ELANOUS_POD_MEMORY_HIGH` 또는 32Gi · standard 보다 작아지지 않는다).
  *  고르는 순서(앞이 이긴다): ① 발사 옵션 `--pod-memory <등급>`(→ `ELANOUS_POD_MEMORY_TIER`) ② 조각 문면 `Pod 메모리: lite|standard|high` ③ 조각의 `apps/pwa/` ④ 부모의 명시 메모리 줄·PWA 경로 ⑤ 실측 권고 ⑥ 권고 없는 research/document 는 lite ⑦ standard.
  *  node-b 노드 할당 가능 약 251Gi. */
@@ -1192,7 +1221,30 @@ export function podSigtermTrapScript(seqFile: string = POD_USAGE_PARTIAL_SEQ_FIL
   return `trap '{ [ -n "$si_pid" ] && kill "$si_pid" 2>/dev/null || true; [ -n "\${usage_partial_pid:-}" ] && kill "$usage_partial_pid" 2>/dev/null || true; si_seq=$(cat ${seqFile} 2>/dev/null || true); case "$si_seq" in ""|*[!0-9]*) si_seq=0;; esac; si_seq=$((10#$si_seq + 1)); { [ -f scripts/usage-rollup.ts ] && bun scripts/usage-rollup.ts --since 12h --partial --seq "$si_seq"; } & si_rollup_pid=$!; for si_i in 1 2 3 4 5 6 7 8; do kill -0 "$si_rollup_pid" 2>/dev/null || break; sleep 1; done; kill "$si_rollup_pid" 2>/dev/null || true; wait "$si_rollup_pid" 2>/dev/null || true; exit 143; }' TERM`;
 }
 
-export function podJobManifest(o: { name: string; namespace: string; image: string; workIdentity?: PodWorkIdentity; boundPrNumber?: number; /** 레지스트리 이미지면 IfNotPresent(노드가 pull) · 반입 이미지면 Never. */ imagePullPolicy?: 'Never' | 'IfNotPresent'; repoUrl: string; source?: PodSource; repairCommitOnly?: { pr: number; baseHead: string }; hostMirror?: string; args: readonly string[]; passEnv: readonly string[]; deadlineSeconds: number; runId?: string; parentRunId?: string; armEnv?: Readonly<Record<string, string>>; hostId?: string; imageCommit?: string | null; skillEnvs?: readonly string[]; memoryLimit?: string; memoryRequest?: string; goalDoc?: string; /** AUTHOR-POD2 — run `harness say` on /creds/feature inside the Pod (authoring happens off the host). */ authorSentence?: boolean; authorGrade?: 'full' | 'lite'; grokCredential?: 'subscription' | 'api_key'; codexAccounts?: readonly string[]; appCredential?: boolean; /** Test seam — default POD_GH_STALE_SECONDS. */ githubStaleSeconds?: number; /** Which goal execution and attempt this Job is — a resumed host verifies it before following the Job (POD9). */ execution?: { key: string; attempt: number }; hostLeaseAdmitted?: boolean; /** POD-ROLE-LLM — OpenRouter 자식 Pod 에 리뷰어용 codex 자격을 깐다. */ reviewerCodexAuth?: boolean; /** BEDROCK-PROVIDER ② — bedrock 자식만: 이름 있는 AWS Secret 의 키(secretKeyRef) ⊕ 리전 ⊕ 이그레스 라벨. */ bedrock?: BedrockPodSpec; /** ZAI-POD-CHILD — glm(ZAI) 자식만: 이름 있는 ZAI Secret 의 키(secretKeyRef) ⊕ 이그레스·동시 상한 라벨(Pod ⊕ Job). */ zai?: ZaiPodSpec; /** Codex Job may use the same named Secret only for an explicitly admitted ZAI fallback. */ zaiFallback?: ZaiPodSpec; fallbackChain?: { steps: readonly FallbackStep[]; source: 'host' | 'default' }; /** TA-POD-WARMBOOT-PARITY — the child's final line (podResultMarkerShellLine), just before `exit $rc`. */ resultMarkerLine?: string; /** POD-LAUNCH-PROVENANCE — the shared provenance set (labels ⊕ annotations ⊕ child env) for this si-task Job. */ provenance?: PodProvenanceMetadata }): Record<string, unknown> {
+/** Leave enough time for the Pod wrapper to print its result and push a salvage branch before Kubernetes kills the Job. */
+export function podDeadlineWatchdogScript(input: { deadlineSeconds: number; job: string; runId?: string; prNumber?: number }): string {
+  const leadSeconds = Math.max(1, Math.min(600, Math.floor(input.deadlineSeconds / 10)));
+  const waitSeconds = Math.max(1, input.deadlineSeconds - leadSeconds);
+  const job = /^[A-Za-z0-9_.:-]+$/.test(input.job) ? input.job : 'unknown-job';
+  const runId = input.runId && /^[A-Za-z0-9_.:-]+$/.test(input.runId) ? input.runId : 'unknown-run';
+  const pr = input.prNumber && Number.isSafeInteger(input.prNumber) && input.prNumber > 0 ? String(input.prNumber) : 'unknown';
+  const jsonPr = pr === 'unknown' ? 'null' : pr;
+  return `printf 'clone\\n' > /tmp/si.stage
+(sleep ${waitSeconds}; [ -f /tmp/si.finished ] && exit 0
+  stage=$(cat /tmp/si.stage 2>/dev/null || true); case "$stage" in clone|setup|run) ;; *) stage=unknown;; esac
+  branch=$(git -C repo branch --show-current 2>/dev/null || true); case "$branch" in ''|*[!A-Za-z0-9_./-]*) branch=unknown;; esac
+  [ -f /tmp/si.finished ] && exit 0
+  si_target=$(cat /tmp/si.pid 2>/dev/null || true)
+  case "$si_target" in ''|*[!0-9]*) ;; *) kill -0 "$si_target" 2>/dev/null || exit 0;; esac
+  printf 'ELANOUS_POD_DEADLINE_RESULT ${job} ${runId} stop-cause=deadline stage=%s pr=${pr} branch=%s\\n' "$stage" "$branch" | tee /tmp/si.deadline-marker
+  printf '{"stage":"deadline","ok":false,"error":"Pod deadline preemption","stopCause":"deadline","prNumber":${jsonPr},"branch":"%s"}\\n' "$branch" > /tmp/si.deadline-result
+  case "$si_target" in ''|*[!0-9]*) ;; *) kill -TERM "$si_target" 2>/dev/null || true;; esac
+) & si_deadline_watch_pid=$!`;
+}
+
+export const POD_DEADLINE_AFTER_WAIT = `if [ "$rc" -ne 0 ] && [ -f /tmp/si.deadline-result ]; then printf '\\n' >> /tmp/si.out; cat /tmp/si.deadline-result >> /tmp/si.out; rc=124; fi`;
+
+export function podJobManifest(o: { name: string; namespace: string; image: string; workIdentity?: PodWorkIdentity; boundPrNumber?: number; /** 레지스트리 이미지면 IfNotPresent(노드가 pull) · 반입 이미지면 Never. */ imagePullPolicy?: 'Never' | 'IfNotPresent'; repoUrl: string; source?: PodSource; repairCommitOnly?: { pr: number; baseHead: string }; hostMirror?: string; args: readonly string[]; passEnv: readonly string[]; deadlineSeconds: number; runId?: string; parentRunId?: string; armEnv?: Readonly<Record<string, string>>; hostId?: string; imageCommit?: string | null; skillEnvs?: readonly string[]; memoryLimit?: string; memoryRequest?: string; goalDoc?: string; /** AUTHOR-POD2 — run `harness say` on /creds/feature inside the Pod (authoring happens off the host). */ authorSentence?: boolean; authorGrade?: 'full' | 'lite'; grokCredential?: 'subscription' | 'api_key'; codexAccounts?: readonly string[]; appCredential?: boolean; /** Test seam — default POD_GH_STALE_SECONDS. */ githubStaleSeconds?: number; /** Which goal execution and attempt this Job is — a resumed host verifies it before following the Job (POD9). */ execution?: { key: string; attempt: number }; hostLeaseAdmitted?: boolean; /** POD-ROLE-LLM — OpenRouter 자식 Pod 에 리뷰어용 codex 자격을 깐다. */ reviewerCodexAuth?: boolean; /** BEDROCK-PROVIDER ② — bedrock 자식만: 이름 있는 AWS Secret 의 키(secretKeyRef) ⊕ 리전 ⊕ 이그레스 라벨. */ bedrock?: BedrockPodSpec; /** ZAI-POD-CHILD — glm(ZAI) 자식만: 이름 있는 ZAI Secret 의 키(secretKeyRef) ⊕ 이그레스·동시 상한 라벨(Pod ⊕ Job). */ zai?: ZaiPodSpec; /** Codex Job may use the same named Secret only for an explicitly admitted ZAI fallback. */ zaiFallback?: ZaiPodSpec; fallbackChain?: { steps: readonly FallbackStep[]; source: 'host' | 'default' }; configSnapshotHash?: string; /** TA-POD-WARMBOOT-PARITY — the child's final line (podResultMarkerShellLine), just before `exit $rc`. */ resultMarkerLine?: string; /** POD-LAUNCH-PROVENANCE — the shared provenance set (labels ⊕ annotations ⊕ child env) for this si-task Job. */ provenance?: PodProvenanceMetadata }): Record<string, unknown> {
   if (o.repairCommitOnly && (o.source?.kind !== 'pr' || o.source.number !== o.repairCommitOnly.pr || !/^[0-9a-f]{40}$/.test(o.repairCommitOnly.baseHead) || o.args.includes('--open-pr') || o.args.includes('--merge-by-host') || o.args.includes('--auto-merge'))) {
     throw new Error('repair commit-only requires matching PR source and no PR/merge flags');
   }
@@ -1223,7 +1275,10 @@ export function podJobManifest(o: { name: string; namespace: string; image: stri
   const script = [
     'set -u',
     ...(o.fallbackChain ? [
-      'mkdir -p "$HOME/.elanous" && install -m 600 /creds/llm-fallback-config.json "$HOME/.elanous/config.json" || exit 6',
+      ...(o.configSnapshotHash ? [
+        'bun -e \'import { createHash } from "node:crypto"; import { readFileSync } from "node:fs"; const e = JSON.parse(readFileSync("/creds/config-snapshot.json", "utf8")); const c = JSON.parse(readFileSync("/creds/llm-fallback-config.json", "utf8")); const h = createHash("sha256").update(JSON.stringify({ version: e.version, keys: e.keys, sources: e.sources })).digest("hex"); if (h !== e.hash || JSON.stringify(c) !== JSON.stringify(e.keys)) process.exit(1);\' || exit 6',
+      ] : []),
+      'mkdir -p "$HOME/.elanous" && { [ -e "$HOME/.elanous/config.json" ] || [ -L "$HOME/.elanous/config.json" ] || install -m 600 /creds/llm-fallback-config.json "$HOME/.elanous/config.json"; } || exit 6',
       `echo "ELANOUS_POD_FALLBACK_CHAIN source=${o.fallbackChain.source} steps=${o.fallbackChain.steps.join(',')}"`,
     ] : []),
     openGatewayModel
@@ -1252,9 +1307,11 @@ export function podJobManifest(o: { name: string; namespace: string; image: stri
     // SI-POD-STAGE-MARKERS — Pod 시계 단계 표지(`ELANOUS_POD_STAGE <stage> <start|end> <epoch초>`): 호스트가 Job 로그에서 읽어
     //   pod.lifecycle stages 의 clone · mirror · run 을 채운다. 실패한 단계는 끝 표지가 없다(→ missing · 지어내지 않는다).
     //   clone 끝 = podSourceScript 전체가 끝난 뒤(갈래마다 마지막 줄이 `ELANOUS_POD_SOURCE` 또는 `cd repo` — 둘 다 소스 받기의 일부).
+    podDeadlineWatchdogScript({ deadlineSeconds: o.deadlineSeconds, job: o.name, runId: o.runId, prNumber: o.boundPrNumber }),
     siTaskPodStageShell('clone', 'start'),
     podSourceScript(o.source ?? { kind: 'default' }, o.repoUrl, undefined, { stageMarkers: true }),
     siTaskPodStageShell('clone', 'end'),
+    `printf 'setup\\n' > /tmp/si.stage`,
     // This image already contains runtime dependencies; there is no child install command.
     // Record that fact before the main command, without treating clone completion as install completion.
     `printf 'ELANOUS_POD_DEPS skipped %s image-prepared-no-install\\n' "$(date +%s)" || true`,
@@ -1278,14 +1335,15 @@ export function podJobManifest(o: { name: string; namespace: string; image: stri
     // 마지막 줄 JSON 이 «맨 끝»이어야 한다(parseSelfImplementJson) — rollup 은 그 앞에.
     // ⛔ `--author-grade` 는 Pod 안 `elanous`(이미지에 깔린 판)가 모를 수 있어 넘기지 않는다 — 10-06 «unknown option» 즉사(#24445 되돌림).
     // AUTHOR-LITE2-POD — 등급은 환경변수 `ELANOUS_AUTHOR_GRADE` 로 싣는다(옛 Pod elanous 는 모르는 환경변수를 무시하고 full 로 돈다).
+    `printf 'run\\n' > /tmp/si.stage`,
     siTaskPodStageShell('run', 'start'),
     o.repairCommitOnly
-      ? `export ${OLD_DOOR_STAMP_ENV}=self-implement; elanous self implement "$(cat /creds/feature)" --json --no-open-pr ${quoted} > /tmp/si.out 2>&1 & si_pid=$!; wait "$si_pid"; rc=$?; ${RUN_END}; if elanous self repair-commit-export '${o.repairCommitOnly.pr}' '${o.repairCommitOnly.baseHead}' /tmp/si.out "$PWD" "$HOME/outbox"; then rc=0; elif [ "$rc" -eq 0 ]; then rc=8; fi`
+      ? `export ${OLD_DOOR_STAMP_ENV}=self-implement; elanous self implement "$(cat /creds/feature)" --json --no-open-pr ${quoted} > /tmp/si.out 2>&1 & si_pid=$!; printf '%s\\n' "$si_pid" > /tmp/si.pid; wait "$si_pid"; rc=$?; touch /tmp/si.finished; ${RUN_END}; rm -f /tmp/si.pid; if elanous self repair-commit-export '${o.repairCommitOnly.pr}' '${o.repairCommitOnly.baseHead}' /tmp/si.out "$PWD" "$HOME/outbox"; then rc=0; elif [ "$rc" -eq 0 ]; then rc=8; fi; ${POD_DEADLINE_AFTER_WAIT}`
       : o.authorSentence
-      ? `echo "ELANOUS_AUTHOR_ON_POD started host=$(hostname) at=$(date -u +%Y-%m-%dT%H:%M:%SZ)"; author_t0=$(date +%s); ${o.authorGrade === 'lite' || o.authorGrade === 'full' ? `ELANOUS_AUTHOR_GRADE=${o.authorGrade} ` : ''}elanous harness say --substrate local --json${quotedAsk}${quotedChildLlm} -- "$(cat /creds/feature)" > /tmp/si.out 2>&1 & si_pid=$!; wait "$si_pid"; rc=$?; ${RUN_END}; echo "ELANOUS_AUTHOR_ON_POD finished host=$(hostname) rc=$rc seconds=$(( $(date +%s) - author_t0 ))"; ${POD_RESULT_FLATTEN}`
+      ? `echo "ELANOUS_AUTHOR_ON_POD started host=$(hostname) at=$(date -u +%Y-%m-%dT%H:%M:%SZ)"; author_t0=$(date +%s); ${o.authorGrade === 'lite' || o.authorGrade === 'full' ? `ELANOUS_AUTHOR_GRADE=${o.authorGrade} ` : ''}elanous harness say --substrate local --json${quotedAsk}${quotedChildLlm} -- "$(cat /creds/feature)" > /tmp/si.out 2>&1 & si_pid=$!; printf '%s\\n' "$si_pid" > /tmp/si.pid; wait "$si_pid"; rc=$?; touch /tmp/si.finished; ${RUN_END}; rm -f /tmp/si.pid; ${POD_DEADLINE_AFTER_WAIT}; echo "ELANOUS_AUTHOR_ON_POD finished host=$(hostname) rc=$rc seconds=$(( $(date +%s) - author_t0 ))"; ${POD_RESULT_FLATTEN}`
       : goalPath
-      ? `elanous harness ask ${goalPath} --json${quotedAsk}${quotedChildLlm} > /tmp/si.out 2>&1 & si_pid=$!; wait "$si_pid"; rc=$?; ${RUN_END}; ${POD_RESULT_FLATTEN}`
-      : `export ${OLD_DOOR_STAMP_ENV}=self-implement; elanous self implement "$(cat /creds/feature)" --json ${quoted} > /tmp/si.out 2>&1 & si_pid=$!; wait "$si_pid"; rc=$?; ${RUN_END}; ${POD_RESULT_FLATTEN}`,
+      ? `elanous harness ask ${goalPath} --json${quotedAsk}${quotedChildLlm} > /tmp/si.out 2>&1 & si_pid=$!; printf '%s\\n' "$si_pid" > /tmp/si.pid; wait "$si_pid"; rc=$?; touch /tmp/si.finished; ${RUN_END}; rm -f /tmp/si.pid; ${POD_DEADLINE_AFTER_WAIT}; ${POD_RESULT_FLATTEN}`
+      : `export ${OLD_DOOR_STAMP_ENV}=self-implement; elanous self implement "$(cat /creds/feature)" --json ${quoted} > /tmp/si.out 2>&1 & si_pid=$!; printf '%s\\n' "$si_pid" > /tmp/si.pid; wait "$si_pid"; rc=$?; touch /tmp/si.finished; ${RUN_END}; rm -f /tmp/si.pid; ${POD_DEADLINE_AFTER_WAIT}; ${POD_RESULT_FLATTEN}`,
     'cat /tmp/si.out',
     ...(o.appCredential && !o.repairCommitOnly ? ['kill "$gh_watch_pid" 2>/dev/null || true'] : []),
     ...(o.repairCommitOnly ? [] : [podSalvageScript()]),
@@ -1352,7 +1410,9 @@ if [ "$found" -eq 0 ]; then echo ELANOUS_RUN_LEDGER_NONE; fi`,
     '[ -n "${usage_partial_pid:-}" ] && kill "$usage_partial_pid" 2>/dev/null || true',
     'wait "$mem_sampler_pid" 2>/dev/null || true',
     '[ -n "${usage_partial_pid:-}" ] && wait "$usage_partial_pid" 2>/dev/null || true',
+    'if [ "$rc" -ne 0 ] && [ -f /tmp/si.deadline-result ]; then rc=124; fi',
     'tail -n 1 /tmp/si.out',
+    'touch /tmp/si.finished; kill "$si_deadline_watch_pid" 2>/dev/null || true; [ -f /tmp/si.deadline-marker ] && cat /tmp/si.deadline-marker || true',
     // TA-POD-WARMBOOT-PARITY: the last line says «the child is done» so the host need not wait for the Job condition.
     ...(o.resultMarkerLine ? [o.resultMarkerLine] : []),
     'exit $rc',
@@ -1361,7 +1421,7 @@ if [ "$found" -eq 0 ]; then echo ELANOUS_RUN_LEDGER_NONE; fi`,
     apiVersion: 'batch/v1',
     kind: 'Job',
     metadata: { name: o.name, namespace: o.namespace, labels: { 'elanous.substrate': 'pod', 'elanous.job': o.name, ...podRunLabels(o), ...(o.workIdentity ? podWorkLabels(o.workIdentity) : {}), ...(o.zai || o.zaiFallback ? { [ZAI_POD_EGRESS_LABEL]: 'true' } : {}), ...o.provenance?.labels },
-      ...((o.execution || o.hostLeaseAdmitted || o.provenance) ? { annotations: { ...o.provenance?.annotations, ...(o.execution ? { [POD_EXECUTION_KEY_ANNOTATION]: o.execution.key, [POD_ATTEMPT_ANNOTATION]: String(o.execution.attempt) } : {}), ...(o.hostLeaseAdmitted ? { [POD_HOST_LEASE_ANNOTATION]: 'true' } : {}) } } : {}) },
+      ...((o.execution || o.hostLeaseAdmitted || o.provenance || o.configSnapshotHash) ? { annotations: { ...o.provenance?.annotations, ...(o.configSnapshotHash ? { 'elanous.config-snapshot-sha256': o.configSnapshotHash } : {}), ...(o.execution ? { [POD_EXECUTION_KEY_ANNOTATION]: o.execution.key, [POD_ATTEMPT_ANNOTATION]: String(o.execution.attempt) } : {}), ...(o.hostLeaseAdmitted ? { [POD_HOST_LEASE_ANNOTATION]: 'true' } : {}) } } : {}) },
     spec: {
       backoffLimit: 0,
       ttlSecondsAfterFinished: POD_JOB_TTL_AFTER_FINISHED_SECONDS,
@@ -1388,6 +1448,9 @@ if [ "$found" -eq 0 ]; then echo ELANOUS_RUN_LEDGER_NONE; fi`,
                 ? [{ name: 'ELANOUS_TASK_BOUND_PR', value: String(o.boundPrNumber) }] : []),
               ...(o.workIdentity?.slotId ? [{ name: 'ELANOUS_POD_SLOT_ID', value: o.workIdentity.slotId }] : []),
               { name: 'ELANOUS_SUBSTRATE', value: 'pod' },
+              // POD-INPLACE-WORKTREE: explicit host opt-in only; one Job runs one implementation child.
+              { name: 'ELANOUS_POD_SINGLE_CHILD', value: '1' },
+              { name: 'ELANOUS_POD_INPLACE_WORKTREE', value: process.env.ELANOUS_POD_INPLACE_WORKTREE === '1' ? '1' : '0' },
               ...Object.entries(controlInboxEnv(POD_CONTROL_INBOX_DIR)).map(([name, value]) => ({ name, value })),
               // ⭐ 런 출처(🅣 RFC run-origin §A3 · #20457/#20468 칸 이름) — Pod 는 자기 elanous_id 를 쓰지 않고 «띄운 감독기»의 hostId 를 물려받는다.
               { name: 'ELANOUS_POD_NAME', valueFrom: { fieldRef: { fieldPath: 'metadata.name' } } },
@@ -1910,6 +1973,35 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
       try {
       // Every attempt, including an OOM retry, must acquire a pool slot before applying a Job.
       const stallMs = options.launchStallMs ?? (getUserConfig().harness?.launchStallMinutes ?? 15) * 60_000;
+      const slotNow = options.now ?? Date.now;
+      let slotAttemptNumber = 0;
+      let lastSlotReasonCode: string | undefined;
+      let lastSlotAttempt: { observationKey: string; requestedAtMs: number; requestedAt: string } | undefined;
+      const slotReasonCode = (reason: string | null | undefined): string => {
+        if (!reason) return 'not-measured-yet';
+        if (/no free slot/iu.test(reason)) return 'no-free-slot';
+        if (/memory/iu.test(reason)) return 'memory-capacity';
+        if (/capacity|lease/iu.test(reason)) return 'capacity';
+        return 'pool-slot-unavailable';
+      };
+      const recordSlotObservation = (attempt: { observationKey: string; requestedAtMs: number; requestedAt: string },
+        outcome: 'requested' | 'queued' | 'admitted' | 'rejected', reasonCode: string): void => {
+        const pool = options.pool;
+        let occupants: Record<string, number> | null = null;
+        try { occupants = pool ? pool.snapshot() : null; } catch { /* observation must not change scheduling */ }
+        const poolCapacity = pool?.members?.reduce((total, member) => total + Math.max(0, member.capacity), 0) ?? null;
+        const data = {
+          observationKey: attempt.observationKey, runId: childRunId, host: resolveHostId(env),
+          requestedAt: attempt.requestedAt, waitedMs: Math.max(0, slotNow() - attempt.requestedAtMs),
+          queued: outcome === 'queued', poolCapacity, occupants, reasonCode, parentId: ledgerRunId, outcome,
+        };
+        debug.log('pod.slot', outcome, data, { level: outcome === 'admitted' || outcome === 'requested' ? 'info' : 'warn' });
+        try {
+          appendRunLedgerEntry({ runId: ledgerRunId, event: 'pod-slot-observation', data }, ledgerDir);
+          if (outcome === 'rejected') appendRunLedgerEntry({ runId: ledgerRunId, event: 'pod-slot-incident', data }, ledgerDir);
+        }
+        catch (error) { debug.log('pod.slot', 'ledger-unavailable', { ...data, writeError: error instanceof Error ? error.message : String(error) }, { level: 'warn' }); }
+      };
       // LAUNCH-STALL (10-06): a stage that waits past the cap says so once — today Pods sat 22 min to 5 h with no event.
       const stalledStages = new Set<string>();
       // pool-slot reason = what THIS launch saw on its last empty tryAcquire; admission reason = the pool-wide lease measurement (one pool, one reading).
@@ -1947,25 +2039,42 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
       };
       const acquireSlot = async (): Promise<PodPoolMember | null> => {
         if (!options.pool) return null;
+        // Each actual acquisition is a separate observation, including the OOM retry in this
+        // same child run. The ordinal keeps keys distinct even with a fixed test clock.
+        const requestedAtMs = slotNow();
+        const attempt = { observationKey: `${childRunId}:pod-slot:${requestedAtMs}:${++slotAttemptNumber}`,
+          requestedAtMs, requestedAt: new Date(requestedAtMs).toISOString() };
+        let firstSlotReasonCode: string | undefined;
+        let slotQueuedRecorded = false;
         const stop = watchStall('pool-slot');
         // MINIH-MODE (0.2.37): the location is declared by the launch env — absent/unknown keeps the configured
         // capacities, so this launch path behaves exactly as before unless the declaration is there.
         const minihLocation = minihLocationFrom(env.ELANOUS_MINIH_LOCATION);
         try {
+          recordSlotObservation(attempt, 'requested', 'pending');
           for (;;) {
             if (input.signal?.aborted) return null;
             // The reason comes back inside this call, so a concurrent launch cannot overwrite it.
             const kind = podPlacementKind(oomRetried ? retryTier : memoryTier, selectedGoalType, requestKind);
-            const acquired = await options.pool.tryAcquire(unfitContexts, (reason) => { stageReason['pool-slot'] = reason; },
+            const acquired = await options.pool.tryAcquire(unfitContexts, (reason) => {
+              stageReason['pool-slot'] = reason;
+              firstSlotReasonCode ??= slotReasonCode(reason);
+            },
               { kind, ...(minihLocation ? { minihLocation } : {}), onPlacement: (placement) => { lastPlacement = placement; } });
             if (acquired) {
+              recordSlotObservation(attempt, 'admitted', firstSlotReasonCode ?? 'admitted');
               debug.log('self-implement.pod', 'pool-slot', { spaceId: input.spaceId, context: acquired.context, inflight: options.pool.snapshot(), ...(lastPlacement ? { placement: lastPlacement } : {}) });
               if (lastPlacement) console.warn(`[pod] 배치: ${lastPlacement.reason}`);
               return acquired;
             }
+            firstSlotReasonCode ??= slotReasonCode(stageReason['pool-slot']);
+            if (!slotQueuedRecorded) {
+              slotQueuedRecorded = true;
+              recordSlotObservation(attempt, 'queued', firstSlotReasonCode);
+            }
             await sleep(options.pollMs ?? 15_000);
           }
-        } finally { stop(); }
+        } finally { lastSlotReasonCode = firstSlotReasonCode; lastSlotAttempt = attempt; stop(); }
       };
       const acquireAdmission = async (): Promise<boolean> => {
         if (!options.pool || typeof options.pool.acquireAdmission !== 'function') return true;
@@ -2001,8 +2110,9 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
       let member: PodPoolMember | null;
       try { member = await acquireSlot(); } catch (error) { podAdmitError(error); throw error; }
       if (options.pool && !member) {
+        if (lastSlotAttempt) recordSlotObservation(lastSlotAttempt, 'rejected', lastSlotReasonCode ?? 'aborted');
         exitJourneyNode('pool-admit', { provenance: 'self-implement-pod', outcome: 'refused', data: { spaceId: input.spaceId, reason: 'no-slot' } });
-        return { exitCode: null, output: '', error: { code: 'aborted', message: 'aborted before a pool slot opened' } };
+        return { exitCode: null, output: '', error: { code: 'aborted', message: 'aborted before a pool slot opened', ...(lastSlotReasonCode ? { reasonCode: lastSlotReasonCode } : {}) } };
       }
       exitJourneyNode('pool-admit', { provenance: 'self-implement-pod', outcome: 'admitted', data: { spaceId: input.spaceId, ...(member ? { context: member.context } : {}) } });
       const currentContext = member ? null : baseKubectl(['config', 'current-context']);
@@ -2030,6 +2140,8 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
       const cleanupSecret = () => { kubectl(['-n', namespace, 'delete', 'secret', `${name}-creds`, '--ignore-not-found']); };
       let repairResultAnnounced = false;
       let repairArtifactReady = false;
+      let hostOriginCalls: number | null = 0;
+      let sourceBaseSha: string | undefined;
       const collectFullLogs = (unavailableEvent: string, replace?: ReadonlySet<string>, onEvent?: (event: string, data: Record<string, unknown>) => void, childId?: string): { status: 'ok' | 'unavailable' | 'incomplete'; childLedgerIncomplete: boolean; samples: ReturnType<typeof parseMemSamples> } => {
         let full: ReturnType<Kubectl>;
         try {
@@ -2046,6 +2158,8 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
         const imageSkip = /^ELANOUS_POD_SOURCE_IMAGE_SKIP ([^\r\n]+)\r?$/m.exec(full.stdout)?.[1];
         const delta = /^ELANOUS_POD_SOURCE_DELTA (mirror|origin)\r?$/m.exec(full.stdout)?.[1];
         if (via) debug.log('self-implement.pod', 'source-via', { job: name, via, ...(mirrorSkip ? { mirrorSkip } : {}), ...(imageSkip ? { imageSkip } : {}), ...(delta ? { delta } : {}) });
+        recordPodSourceIoSummary(full.stdout, name, childId ?? childRunId, childId ? 'explicit-child-id' : 'pod-launch-child-run-id',
+          (category, event, data, logOptions) => debug.log(category, event, data, logOptions), hostOriginCalls);
         let ledgerIncomplete = false;
         let childLedgerIncomplete = false;
         let artifactIncomplete = false;
@@ -2323,20 +2437,43 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
         }
         const now = options.now ?? Date.now;
         const issueVerifiedAppCredential = async (fresh = false): Promise<{ token: string; expiresAt: number } | null> => {
-          if (!repository) return null;
+          if (!repository) {
+            debug.log('self-implement.pod', 'github-app-decision', { job: name, reason: 'repository-unavailable', fresh });
+            return null;
+          }
           // This is the same repository-scoped, contents+pull_requests installation request on launch and refresh.
-          const candidate = (options.githubInstallation ?? ((repo, o) => coalescedInstallationCredential({ repository: repo.split('/')[1]! }, { fresh: o?.fresh })))(repository, { fresh });
-          if (!candidate?.token) return null;
+          let candidate: ReturnType<NonNullable<typeof options.githubInstallation>>;
+          try { candidate = (options.githubInstallation ?? ((repo, o) => coalescedInstallationCredential({ repository: repo.split('/')[1]! }, { fresh: o?.fresh })))(repository, { fresh }); }
+          catch {
+            debug.log('self-implement.pod', 'github-app-decision', { job: name, repository, reason: 'mint-threw', fresh });
+            return null;
+          }
+          if (!candidate?.token) {
+            debug.log('self-implement.pod', 'github-app-decision', { job: name, repository, reason: 'mint-unavailable', fresh });
+            return null;
+          }
           const expiry = candidate.expires_at ?? ('expiresAt' in candidate ? candidate.expiresAt : undefined);
           const expiresAt = typeof expiry === 'number' ? expiry : Date.parse(expiry ?? '');
-          if (!Number.isFinite(expiresAt) || expiresAt <= now()) return null;
-          const available = await (options.githubRepositories ?? installationRepositories)(candidate.token);
-          return available?.length === 1 && available[0]?.toLowerCase() === repository.toLowerCase()
-            ? { token: candidate.token, expiresAt } : null;
+          if (!Number.isFinite(expiresAt) || expiresAt <= now()) {
+            debug.log('self-implement.pod', 'github-app-decision', { job: name, repository, reason: Number.isFinite(expiresAt) ? 'expired' : 'expiry-unavailable', expiresAt: Number.isFinite(expiresAt) ? new Date(expiresAt).toISOString() : null, fresh });
+            return null;
+          }
+          let available: Awaited<ReturnType<NonNullable<typeof options.githubRepositories>>>;
+          try { available = await (options.githubRepositories ?? installationRepositories)(candidate.token); }
+          catch {
+            debug.log('self-implement.pod', 'github-app-decision', { job: name, repository, reason: 'verification-threw', expiresAt: new Date(expiresAt).toISOString(), fresh });
+            return null;
+          }
+          if (available?.length !== 1 || available[0]?.toLowerCase() !== repository.toLowerCase()) {
+            debug.log('self-implement.pod', 'github-app-decision', { job: name, repository, reason: 'repository-verification-failed', availableCount: available?.length ?? null, expiresAt: new Date(expiresAt).toISOString(), fresh });
+            return null;
+          }
+          debug.log('self-implement.pod', 'github-app-decision', { job: name, repository, reason: 'verified', expiresAt: new Date(expiresAt).toISOString(), fresh });
+          return { token: candidate.token, expiresAt };
         };
         let appCredential: Awaited<ReturnType<typeof issueVerifiedAppCredential>> = null;
         try { appCredential = await issueVerifiedAppCredential(ghLoginRetried); }
-        catch { debug.log('self-implement.pod', 'github-app-unavailable', { job: name, reason: 'mint-or-verification-failed' }); }
+        catch { debug.log('self-implement.pod', 'github-app-unavailable', { job: name, repository, reason: 'mint-or-verification-threw' }); }
         // PODCRED1: a short token is still shipped — the host renews it in the Pod on the first live poll — but it is
         // observed, so an exit-7 burst can be checked against the remaining life the Pods actually started with.
         if (appCredential && appCredential.expiresAt - now() < POD_APP_TOKEN_MIN_START_MS) {
@@ -2352,6 +2489,7 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
         // The host polling loop is a refresh path even when the Pod cannot reach the relay.
         const hostRefresh = Boolean(appCredential);
         const podGhToken = hostRefresh ? appCredential!.token : creds.ghToken;
+        debug.log('self-implement.pod', 'github-credential-source', { job: name, repository, source: hostRefresh ? 'verified-app' : 'human-fallback', expiresAt: hostRefresh ? new Date(appCredential!.expiresAt).toISOString() : null });
         const githubPassKeys = hostRefresh ? passKeys.filter((key) => !['GH_TOKEN', 'GITHUB_TOKEN', 'GH_CONFIG_DIR'].includes(key)) : passKeys;
         let ghTokenExpiresAt = hostRefresh ? appCredential!.expiresAt : null;
         const requestedGoalDoc = env.ELANOUS_POD_GOAL_DOC;
@@ -2483,6 +2621,12 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
           if (!zaiFallbackSpec) fallbackSteps = fallbackSteps.filter(step => step !== 'zai');
         }
         const fallbackChain = { steps: fallbackSteps, source: normalizedFallback.usedDefault ? 'default' as const : 'host' as const };
+        const { buildPodConfigSnapshot } = await import('./pod-config-snapshot.js');
+        const configSnapshot = buildPodConfigSnapshot(hostConfig, {
+          steps: fallbackChain.steps, overage: zaiOverage,
+          worktreeAddTimeoutSec: worktreeAddTimeoutSec(hostConfig.harness?.worktreeAddTimeoutSec) ?? 1200,
+        });
+        debug.log('self-implement.pod', 'config-snapshot', { job: name, hash: configSnapshot.hash, keyCount: configSnapshot.keyCount });
         debug.log('self-implement.pod', 'fallback-chain-carried', {
           job: name, source: fallbackChain.source, steps: fallbackChain.steps,
           ...(normalizedFallback.dropped.length ? { droppedCount: normalizedFallback.dropped.length } : {}),
@@ -2501,7 +2645,8 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
                 ? Object.fromEntries(codexCredentials.map((candidate, i) => [`codex-${i}.json`, JSON.stringify({ elanousAuth: candidate.elanousAuth, codexAuth: candidate.codexAuth })]))
                 : { 'elanous-auth.json': (creds as ReturnType<typeof hostCredentials>).elanousAuth, 'codex-auth.json': (creds as ReturnType<typeof hostCredentials>).codexAuth }),
             'gh-token': podGhToken, feature: input.feature,
-            'llm-fallback-config.json': JSON.stringify({ llm: { fallbackChain: fallbackChain.steps, zaiCodingPlan: { overage: zaiOverage } } }),
+            'llm-fallback-config.json': configSnapshot.configJson,
+            'config-snapshot.json': configSnapshot.envelopeJson,
             ...(goalDocument !== undefined ? { 'goal-doc': goalDocument } : {}),
             ...Object.fromEntries(Object.entries(skillEnvs).map(([n, text]) => [`skillenv-${n}`, text])),
             ...(grounding ? { [`env-${GROUNDING_TOKEN_ENV}`]: grounding.token } : {}),
@@ -2515,7 +2660,7 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
         if (missing.length) debug.log('self-implement.pod', 'pass-env-missing', { job: name, missing }, { level: 'warn' });
         // ☸️ 재개(P3 · 2026-09-26): 같은 이름의 Job 이 이미 있으면 «지우지 않고» 붙는다 — 호스트가 끊긴 동안에도 원격은 계속 돌았다.
         //   🩸 종전엔 여기서 delete 부터 해서, 재개가 원격에서 멀쩡히 돌던 Job 을 죽였다. 실패로 끝난 Job 만 지우고 다시 만든다.
-        // ☸️ 노드 레지스트리에 이 판이 있으면 그것을 pull 한다(델타 · 대표 2026-09-26) — 라벨 판정은 로컬 이미지 이름으로 한다.
+        // ☸️ 노드 레지스트리에 이 판이 있으면 그것을 pull 한다(델타 · 2026-09-26) — 라벨 판정은 로컬 이미지 이름으로 한다.
         const jobImage = member?.imageRef ?? image;
         // A pool member may pull a different image than the host's local one.
         // The local Docker label cannot prove that member image's commit.
@@ -2697,12 +2842,21 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
           slotId: env.ELANOUS_POD_SLOT_ID,
           work: !goalDoc && env.ELANOUS_POD_AUTHOR_ON_POD === '1' && !shard ? 'authoring' : (oomRetried ? retryTier : memoryTier) === 'lite' ? 'lite' : 'harness',
         };
-        // TA-POD-WARMBOOT-PARITY shadow only: record the decision on the existing fresh-Job path.
-        // Until a shared warm-candidate contract exists, evidence stays unmeasured and this never changes execution.
+        // TA-POD-WARMBOOT-PARITY shadow only: read injected evidence, then keep the fresh-Job path.
+        // This boundary must not open host configuration or credential files for a warm decision.
         if (workIdentity.taskCard) {
-          const warmReuseConfig = options.warmReuseConfig?.() ?? (options.kubectl ? undefined : getUserConfig().pod?.warmReuse);
-          const decision = decideWarmReuse('ta', warmReuseConfig, options.warmReuseEvidence?.());
-          debug.log('pod.warm', 'ta-shadow-observation', { ...decision, execution: 'fresh-job' });
+          let stage: 'config' | 'evidence' | 'decision' = 'config';
+          try {
+            const config = options.warmReuseConfig?.();
+            stage = 'evidence';
+            const evidence = options.warmReuseEvidence?.();
+            stage = 'decision';
+            const decision = decideWarmReuse('ta', config, evidence);
+            debug.log('pod.warm', 'ta-shadow-observation', { ...decision, execution: 'fresh-job' });
+          } catch {
+            debug.log('pod.warm', 'ta-shadow-unavailable', { spaceId: input.spaceId,
+              reason: stage === 'decision' ? 'warm-decision-failed' : `injected-${stage}-failed`, execution: 'fresh-job' }, { level: 'warn' });
+          }
         }
         // TA-POD-WARMBOOT-PARITY (GATE-WARM-BOOT parity · #27805): a commit-pinned source on a pool host — put that commit into
         // the host mirror first (once per host·commit in this process) so podSourceScript finds it instead of fetching
@@ -2721,12 +2875,13 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
         else if (seedSource.kind !== 'default') seedSkip = `source-${seedSource.kind}`;
         else {
           const resolveHead = options.resolveSourceHead ?? (options.kubectl ? undefined : resolveDefaultSourceHead);
-          let head: { sha?: string; reason?: string } | undefined;
+          let head: { sha?: string; reason?: string; originCalls?: number } | undefined;
           try { head = resolveHead ? await resolveHead({ repoUrl, ...(seedSource.base ? { base: seedSource.base } : {}), cwd: seedCwd }) : undefined; }
           catch { head = { reason: 'resolve-threw' }; }
           if (!resolveHead) seedSkip = 'no-head-resolver';
-          else if (head?.sha && /^[0-9a-f]{40}$/.test(head.sha)) seedCommit = head.sha;
-          else seedSkip = `default-head-${head?.reason ?? 'unresolved'}`;
+          else if (head) hostOriginCalls = typeof head.originCalls === 'number' && Number.isSafeInteger(head.originCalls) && head.originCalls >= 0 ? head.originCalls : null;
+          if (head?.sha && /^[0-9a-f]{40}$/.test(head.sha)) { seedCommit = head.sha; sourceBaseSha = head.sha; }
+          else if (resolveHead) seedSkip = `default-head-${head?.reason ?? 'unresolved'}`;
         }
         if (seedMirror && seedCommit && member?.sshHost) {
           const seedHost = member.sshHost;
@@ -2737,7 +2892,7 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
           } catch (error) {
             seed = { host: seedHost, ok: false, ms: 0, detail: String(error).slice(0, 300) };
           }
-          debug.log('self-implement.pod', 'mirror-seed', { job: name, host: seedHost, commit: seedCommit, sourceKind: seedSource.kind, ok: seed?.ok ?? false, ms: seed?.ms ?? 0, ...(seed?.detail ? { detail: seed.detail } : {}) }, seed?.ok ? undefined : { level: 'warn' });
+          debug.log('self-implement.pod', 'mirror-seed', { job: name, host: seedHost, commit: seedCommit, sourceKind: seedSource.kind, ...(sourceBaseSha ? { sourceBaseSha } : {}), ok: seed?.ok ?? false, ms: seed?.ms ?? 0, ...(seed?.detail ? { detail: seed.detail } : {}) }, seed?.ok ? undefined : { level: 'warn' });
         } else {
           debug.log('self-implement.pod', 'mirror-seed-skipped', { job: name, reason: seedSkip ?? 'unknown', sourceKind: seedSource.kind, ...(member?.sshHost ? { host: member.sshHost } : {}) });
         }
@@ -2772,7 +2927,7 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
           cleanupSecret();
           return { exitCode: 1, output: '', error: { code: 'pod-provenance', message } };
         }
-        const job = podJobManifest({ provenance, name, namespace, fallbackChain, ...(resultMarkerLine ? { resultMarkerLine } : {}), image: jobImage, workIdentity, ...(env.ELANOUS_TASK_BOUND_PR && /^[1-9]\d*$/.test(env.ELANOUS_TASK_BOUND_PR) && Number.isSafeInteger(Number(env.ELANOUS_TASK_BOUND_PR)) ? { boundPrNumber: Number(env.ELANOUS_TASK_BOUND_PR) } : {}), ...(member?.imageRef ? { imagePullPolicy: 'IfNotPresent' as const } : {}), repoUrl, ...(options.source ? { source: options.source } : {}), ...(options.repairCommitOnly ? { repairCommitOnly: options.repairCommitOnly } : {}), ...(hostMirror ? { hostMirror } : {}), args: jobArgs, passEnv: jobPassEnv, deadlineSeconds: jobDeadlineSeconds, ...(goalDoc ? { goalDoc } : env.ELANOUS_POD_AUTHOR_ON_POD === '1' && !shard ? { authorSentence: true, ...(env.ELANOUS_POD_AUTHOR_GRADE === 'lite' || env.ELANOUS_POD_AUTHOR_GRADE === 'full' ? { authorGrade: env.ELANOUS_POD_AUTHOR_GRADE } : {}) } : {}), runId: launchRunId, ...(parentRunId ? { parentRunId } : {}), ...(jobArmEnv ? { armEnv: jobArmEnv } : {}), hostId: resolveHostId(env), skillEnvs: Object.keys(skillEnvs), memoryLimit, memoryRequest, imageCommit: jobImageCommit, ...(grok ? { grokCredential: 'grokAuth' in creds && creds.grokAuth ? 'subscription' as const : 'api_key' as const } : {}), ...(codexAccounts ? { codexAccounts } : {}), ...(hostRefresh ? { appCredential: true } : {}), execution: { key: executionKey, attempt: oomRetried ? 2 : 1 }, hostLeaseAdmitted: !!releaseAdmission, ...(roleCodexCredential ? { reviewerCodexAuth: true } : {}), ...(bedrockSpec ? { bedrock: bedrockSpec } : {}), ...(zaiSpec ? { zai: zaiSpec } : {}), ...(zaiFallbackSpec ? { zaiFallback: zaiFallbackSpec } : {}) });   // kubectl 주입(=시험)이면 docker 를 부르지 않는다
+        const job = podJobManifest({ provenance, name, namespace, fallbackChain, configSnapshotHash: configSnapshot.hash, ...(resultMarkerLine ? { resultMarkerLine } : {}), image: jobImage, workIdentity, ...(env.ELANOUS_TASK_BOUND_PR && /^[1-9]\d*$/.test(env.ELANOUS_TASK_BOUND_PR) && Number.isSafeInteger(Number(env.ELANOUS_TASK_BOUND_PR)) ? { boundPrNumber: Number(env.ELANOUS_TASK_BOUND_PR) } : {}), ...(member?.imageRef ? { imagePullPolicy: 'IfNotPresent' as const } : {}), repoUrl, ...((seedSource.kind === 'default' && sourceBaseSha) ? { source: { ...seedSource, baseSha: sourceBaseSha } } : options.source ? { source: options.source } : {}), ...(options.repairCommitOnly ? { repairCommitOnly: options.repairCommitOnly } : {}), ...(hostMirror ? { hostMirror } : {}), args: jobArgs, passEnv: jobPassEnv, deadlineSeconds: jobDeadlineSeconds, ...(goalDoc ? { goalDoc } : env.ELANOUS_POD_AUTHOR_ON_POD === '1' && !shard ? { authorSentence: true, ...(env.ELANOUS_POD_AUTHOR_GRADE === 'lite' || env.ELANOUS_POD_AUTHOR_GRADE === 'full' ? { authorGrade: env.ELANOUS_POD_AUTHOR_GRADE } : {}) } : {}), runId: launchRunId, ...(parentRunId ? { parentRunId } : {}), ...(jobArmEnv ? { armEnv: jobArmEnv } : {}), hostId: resolveHostId(env), skillEnvs: Object.keys(skillEnvs), memoryLimit, memoryRequest, imageCommit: jobImageCommit, ...(grok ? { grokCredential: 'grokAuth' in creds && creds.grokAuth ? 'subscription' as const : 'api_key' as const } : {}), ...(codexAccounts ? { codexAccounts } : {}), ...(hostRefresh ? { appCredential: true } : {}), execution: { key: executionKey, attempt: oomRetried ? 2 : 1 }, hostLeaseAdmitted: !!releaseAdmission, ...(roleCodexCredential ? { reviewerCodexAuth: true } : {}), ...(bedrockSpec ? { bedrock: bedrockSpec } : {}), ...(zaiSpec ? { zai: zaiSpec } : {}), ...(zaiFallbackSpec ? { zaiFallback: zaiFallbackSpec } : {}) });   // kubectl 주입(=시험)이면 docker 를 부르지 않는다
         const carriedBranch = goalDoc ? 'goal-doc' as const : env.ELANOUS_POD_AUTHOR_ON_POD === '1' && !shard ? 'author' as const : undefined;
         const carriedChildLlm = carriedBranch ? podDelegatedChildLlm(jobArgs) : undefined;
         jobApplyStartedAt = Date.now();
@@ -2780,6 +2935,13 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
         zaiPendingApply.delete(name);   // apply 뒤엔 Job 라벨이 센다(성공) · 실패면 셀 것이 없다
         if (a.status !== 0) { cleanupSecret(); return { exitCode: 1, output: a.stderr, error: { code: 'pod-apply', message: a.stderr.trim() } }; }
         releaseAdmission?.applied?.(name, context, namespace);
+        try { appendRunLedgerEntry({ runId: ledgerRunId, event: 'pod-config-snapshot', data: {
+          job: name, hash: configSnapshot.hash, keyCount: configSnapshot.keyCount,
+        } }, ledgerDir); } catch (error) {
+          debug.log('self-implement.pod', 'config-snapshot-ledger-unavailable', {
+            job: name, reason: error instanceof Error ? error.message : String(error),
+          }, { level: 'warn' });
+        }
         const imageFacts = { job: name, childRunId: liveChildRunId, image: jobImage, imageCommit: jobImageCommit, context };
         // POD-LAUNCH-PROVENANCE — append-only «started» row: who · why · when, on the parent's existing run ledger.
         const startedRow = { job: name, run: launchRunId, attempt: oomRetried ? 2 : 1, provenance: provenanceRowFromAnnotations(provenance.annotations) };
@@ -2859,11 +3021,36 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
         debug.log('self-implement.pod', 'job-applied', { job: name, namespace, ...(member ? { context: member.context } : {}), image: jobImage, spaceId: input.spaceId, passEnv, ...(openrouterCredentialRequired ? { openrouterKeyPresent: passEnv.includes('OPENROUTER_API_KEY') } : {}), ...(openGateway ? { openGatewayKeyPresent: passEnv.includes('OPENGATEWAY_API_KEY') } : {}), ...(anthropic ? { anthropicKeyPresent: passEnv.includes('ANTHROPIC_API_KEY') } : {}), ...(bedrockSpec ? { bedrockSecret: bedrockSpec.secretName, bedrockSecretKeys: bedrockSpec.keys, bedrockRegionFrom: bedrockSpec.region.from } : {}), ...(zaiRoute ? { zaiRoute, ...(zaiSpec ? { zaiSecret: zaiSpec.secretName } : {}) } : {}), extraArgs: options.extraArgs ?? [], ...(options.armEnv?.ELANOUS_ARM_ID ? { armId: options.armEnv.ELANOUS_ARM_ID } : {}) });
         let state: 'complete' | 'failed' | 'aborted' = 'failed';
         let failedReason = '';
+        let deadlineJobCreatedAt: string | null = null;
+        let deadlineJobCompletedAt: string | null = null;
         let containerReason: string | null = null;
         let containerExitCode: number | null = null;
         // JOB-FAILURE-TERMINATION-CAPTURE — this Job's Kubernetes facts, read once at the failure verdict (before GC).
         //   Observation only: failedReason · containerReason · containerExitCode keep driving the classification.
         let podTermination: PodTermination | null = null;
+        const refreshPodTermination = async (): Promise<void> => {
+          if (state !== 'failed' || !podTermination || (podTermination.podLookup === 'terminated'
+            && podTermination.containerReason !== 'unknown' && podTermination.exitCode !== 'unknown'
+            && podTermination.finishedAt !== 'unknown')) return;
+          // The result marker and the Job Failed condition can precede container termination status.
+          // Re-read this launch's Pod once; keep the verdict's Job facts even if the Job has since been GC'd.
+          await sleep(1_000);
+          const retry = readPodTermination({ kubectl, namespace, job: name, path: podTermination.path,
+            jobUid: lifecycleJobUid, jobRead: null });
+          const observed = retry.podLookup === 'unavailable' ? podTermination : retry;
+          const complete = observed.podLookup === 'terminated' && observed.containerReason !== 'unknown'
+            && observed.exitCode !== 'unknown' && observed.finishedAt !== 'unknown';
+          const collectionFailureReason: PodTermination['collectionFailureReason'] | undefined = complete ? undefined
+            : retry.podLookup === 'unavailable' ? 'pod-read-unavailable-after-retry'
+            : retry.podLookup === 'absent' ? 'pod-absent-after-retry'
+            : retry.podLookup === 'running' || retry.podLookup === 'pending' ? 'pod-still-running-after-retry'
+            : retry.podLookup === 'terminated' ? 'terminated-fields-missing-after-retry'
+            : 'pod-state-unknown-after-retry';
+          podTermination = { ...podTermination, observedAt: retry.observedAt,
+            podLookup: observed.podLookup, containerReason: observed.containerReason, exitCode: observed.exitCode,
+            finishedAt: observed.finishedAt, oomKilled: observed.oomKilled, collectionAttempts: 2,
+            ...(collectionFailureReason ? { collectionFailureReason } : {}) };
+        };
         // METER-POD-ROLLUP-GAP — abort 때 Job 삭제 «전»에 거둔 로그 사본(삭제 뒤엔 PARTIAL 줄도 못 읽는다).
         let abortLogs = '';
         // ⭐ 런 «도중» 원장 증분 회수 — 호스트 슈퍼바이저가 Pod 걸음을 실시간으로 본다(🅣 요청 · 힐 루프 runtime 성형의 전제).
@@ -2950,7 +3137,13 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
             let jobReadError: unknown;
             try { jobRead = kubectl(['-n', namespace, 'get', 'job', name, '-o', `jsonpath=${POD_TERMINATION_JOB_JSONPATH}`]); }
             catch (error) { jobReadError = error; }
-            failedReason = jobRead?.status === 0 ? (jobRead.stdout.split('\t')[0] ?? '').trim() : '';
+            const observedJobReason = jobRead?.status === 0 ? podJobConditionReason(jobRead.stdout.split('\t')[0]) : 'unknown';
+            failedReason = observedJobReason === 'unknown' ? '' : observedJobReason;
+            if (jobRead?.status === 0) {
+              const [, , , createdText, completedText] = jobRead.stdout.replace(/\r?\n$/, '').split('\t');
+              deadlineJobCreatedAt = createdText && Number.isFinite(Date.parse(createdText)) ? createdText : null;
+              deadlineJobCompletedAt = completedText && Number.isFinite(Date.parse(completedText)) ? completedText : null;
+            }
             // One Pods read serves both: its first three columns are the classification layout, the rest the termination capture.
             let podsRead: { status: number | null; stdout: string } | null = null;
             try {
@@ -2973,6 +3166,7 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
             captureFailedJob(liveChildRunId);
             if (jobReadError !== undefined) {
               // Today's outcome stays (the throw ends the spawn as pod-error); the capture is in the ledger first.
+              await refreshPodTermination();
               try { appendRunLedgerEntry({ runId: ledgerRunId, event: 'pod-child-evidence', data: { childEvidence: null, childRunId: liveChildRunId, job: name, podTermination } }, ledgerDir); }
               catch (error) { debug.log('self-implement.pod', 'child-evidence-write-failed', { job: name, childRunId: liveChildRunId, reason: error instanceof Error ? error.message : String(error) }, { level: 'warn' }); }
               debug.log('self-implement.pod', 'pod-termination-captured', { job: name, podTermination, jobReadThrew: true }, { level: 'warn' });
@@ -3058,6 +3252,14 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
         if (!logs && abortLogs) logs = abortLogs;
         // TA-POD-WARMBOOT-PARITY: the marker line is not child output — remove it so the last-line result parse sees the same bytes.
         if (markerWait && markerToken && logs) logs = markerWait.stripPodResultMarker(logs, markerToken);
+        const deadlineMarker = podDeadlineResultMarker(logs, name, liveChildRunId);
+        if (deadlineMarker && state !== 'aborted' && state !== 'complete') {
+          state = 'failed';
+          failedReason = 'DeadlineExceeded';
+          debug.log('self-implement.pod', 'deadline-result-marker', { job: name, childRunId: liveChildRunId, ...deadlineMarker });
+          try { appendRunLedgerEntry({ runId: ledgerRunId, event: 'pod-deadline-result', data: { job: name, childRunId: liveChildRunId, stopCause: 'deadline', ...deadlineMarker } }, ledgerDir); }
+          catch { /* A full ledger must not hide the Pod's terminal marker. */ }
+        }
         let ledgerReason = '';
         let ledgerCollected = false;
         const collection = collectFullLogs('ledger-collect-incomplete', follower.owned ? new Set([liveChildRunId]) : undefined, (event, data) => {
@@ -3117,7 +3319,9 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
           continue;
         }
         // Pod 안 경로는 호스트에서 쓸 수 없다.
-        let disposition = parsed ? { ...parsed, worktreePath: undefined } : undefined;
+        let disposition: SelfImplementDisposition | undefined = parsed ? { ...parsed, worktreePath: undefined } : deadlineMarker
+          ? { stage: 'deadline', ok: false, ...(deadlineMarker.pr ? { prNumber: deadlineMarker.pr } : {}), ...(deadlineMarker.branch ? { branch: deadlineMarker.branch } : {}) }
+          : undefined;
         if (options.repairCommitOnly && state === 'complete' && !repairArtifactReady) {
           disposition = { ...disposition, worktreePath: undefined, ok: false, stage: 'repair-artifact-missing', detail: 'Pod repair result or bundle was not collected' };
         }
@@ -3261,8 +3465,8 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
         //   the terminal result line can be followed by the useful stderr that explains the actual exit.
         let noResultLogEvidence: Record<string, unknown> | null = null;
         const shouldCaptureFailureLog = state === 'failed' && !oomKilled && failedReason !== 'DeadlineExceeded';
-        if (shouldCaptureFailureLog || noResultDiagnostic?.includes('stage=unknown')) {
-          const missingStage = noResultDiagnostic?.includes('stage=unknown') ?? false;
+        const missingStage = noResultDiagnostic?.includes('stage=unknown') ?? false;
+        if (shouldCaptureFailureLog || missingStage) {
           const logRead = readPodNoResultLogTail({ run: kubectl, namespace, job: name, maxLines: POD_NO_RESULT_LOG_TAIL_LINES });
           if ('tail' in logRead) {
             noResultLogEvidence = missingStage
@@ -3272,11 +3476,13 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
               job: name, childRunId: liveChildRunId, diagnostic: noResultDiagnostic, lines: logRead.tail.length, logTail: logRead.tail,
             }, { compact: { stringMax: 500 } });
           } else {
+            const unavailable = `log-unavailable: ${logRead.unavailable}`;
             noResultLogEvidence = missingStage
-              ? { stderr: `log-unavailable: ${logRead.unavailable}` }
+              ? { stderr: unavailable }
               : { podLogUnavailable: logRead.unavailable };
             debug.log('self-implement.pod', missingStage ? 'no-result-log-tail' : 'failed-pod-log-tail', {
-              job: name, childRunId: liveChildRunId, diagnostic: noResultDiagnostic, logUnavailable: logRead.unavailable,
+              job: name, childRunId: liveChildRunId, diagnostic: noResultDiagnostic,
+              logUnavailable: missingStage ? unavailable : logRead.unavailable,
             }, { level: 'warn' });
           }
         }
@@ -3284,6 +3490,9 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
         if (noResultDiagnostic && !noResultDiagnostic.includes('stage=unknown')) {
           debug.log('self-implement.pod', 'no-result-ledger-diagnostic', { job: name, childRunId: liveChildRunId, diagnostic: noResultDiagnostic }, { compact: { stringMax: 500 } });
         }
+        // The marker can precede the container status by several seconds. Read the Pod after
+        // collecting the child logs, before the first terminal evidence row is emitted.
+        await refreshPodTermination();
         // POD-LAUNCH-PROVENANCE — append-only «finished» row paired with its started row on job·run.
         // Keyed on launchRunId like the started row; a reattached Job's inherited child run rides along as childRun.
         // exitCode stays null when the container's code was not read (never filled in as 0).
@@ -3359,7 +3568,10 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
             if (unfitRetry) return unfitRetry;
             if (!await acquireAdmission()) return admissionFailure();
             member = await acquireSlot();
-            if (options.pool && !member) return { exitCode: null, output: '', error: { code: 'aborted', message: 'aborted before a pool slot opened' } };
+            if (options.pool && !member) {
+              if (lastSlotAttempt) recordSlotObservation(lastSlotAttempt, 'rejected', lastSlotReasonCode ?? 'aborted');
+              return { exitCode: null, output: '', error: { code: 'aborted', message: 'aborted before a pool slot opened', ...(lastSlotReasonCode ? { reasonCode: lastSlotReasonCode } : {}) } };
+            }
             if (member) context = member.context;
             if (input.signal?.aborted) return { exitCode: null, output: '', error: { code: 'aborted', message: 'aborted before OOM retry' } };
             continue;
@@ -3428,6 +3640,17 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
         const deadlineExceeded = state === 'failed' && failedReason === 'DeadlineExceeded';
         const deadlineSeconds = jobDeadlineSeconds;
         if (deadlineExceeded) {
+          const jobCreatedAt = deadlineJobCreatedAt;
+          const timing = podJobFinishedTiming({ jobCreatedAt, jobCompletedAt: deadlineJobCompletedAt,
+            podFinishedAt: typeof podTermination?.finishedAt === 'string' && Number.isFinite(Date.parse(podTermination.finishedAt))
+              ? podTermination.finishedAt : null,
+            observedAt: new Date((options.now ?? Date.now)()).toISOString() });
+          const lastStage = deadlineMarker?.stage ?? disposition?.stage ?? null;
+          debug.log('pod.job', 'finished', {
+            job: name, jobCreatedAt, finishedAt: timing.finishedAt, finishedAtSource: timing.finishedAtSource,
+            elapsedMs: timing.elapsedMs,
+            deadlineSec: deadlineSeconds, tier: oomRetried ? retryTier : memoryTier, lastStage,
+          });
           debug.log('self-implement.pod', 'deadline-exceeded', {
             job: name,
             deadlineSeconds,
@@ -3455,6 +3678,12 @@ export function podSelfImplementSpawn(options: PodSpawnOptions = {}): SelfImplem
           const salvageOutcome = /^ELANOUS_POD_SALVAGE \S+ \S+$/m.test(logs) ? 'pushed' : logs.includes('ELANOUS_POD_SALVAGE_NONE') ? 'none' : 'unknown';
           passJourneyNode('salvage', { provenance: 'self-implement-pod', outcome: salvageOutcome, data: { job: name } });
         }
+        const repairGateReceipt = disposition?.stage === 'gate-failed' ? gateFailureReceipt(disposition.gate) : null;
+        if (repairGateReceipt) {
+          for (const line of gateFailureReceiptLines(repairGateReceipt)) console.log(line);
+          debug.log('self-implement.pod', 'gate-failure-receipt', { job: name, childRunId: liveChildRunId,
+            signature: repairGateReceipt.signature, tests: repairGateReceipt.tests });
+        }
         return {
           exitCode: podExitCode,
           output: tail,
@@ -3610,7 +3839,9 @@ export function podNoResultLogTailLines(source: string, maxLines: number = POD_N
   let chars = 0;
   for (const line of lines.slice(-Math.max(0, maxLines)).reverse()) {
     if (chars >= POD_NO_RESULT_LOG_TAIL_MAX_CHARS) break;
-    const bounded = boundedPodLogText(line, Math.min(POD_NO_RESULT_LOG_TAIL_LINE_CHARS, POD_NO_RESULT_LOG_TAIL_MAX_CHARS - chars));
+    const redacted = redactSecretText(line);
+    const limit = Math.min(POD_NO_RESULT_LOG_TAIL_LINE_CHARS, POD_NO_RESULT_LOG_TAIL_MAX_CHARS - chars);
+    const bounded = redacted.length > limit ? `${redacted.slice(0, limit)}…[truncated]` : redacted;
     if (!bounded) continue;
     result.push(bounded);
     chars += bounded.length;
@@ -3632,7 +3863,7 @@ export function readPodNoResultLogTail(args: {
 }): PodNoResultLogRead {
   const maxLines = args.maxLines ?? POD_NO_RESULT_LOG_TAIL_LINES;
   try {
-    const result = args.run(['-n', args.namespace, 'logs', `job/${args.job}`, `--tail=${maxLines}`]);
+    const result = args.run(['-n', args.namespace, 'logs', `job/${args.job}`, '-c', 'child', `--tail=${maxLines}`]);
     if (result.status !== 0) {
       const reason = boundedPodLogText(result.stderr || result.stdout || '', POD_NO_RESULT_LOG_REASON_MAX_CHARS);
       return { unavailable: reason || `kubectl logs exit ${result.status}` };
@@ -3928,7 +4159,7 @@ export async function rebuildLocalPodImageOnce(deps: {
         });
       });
     });
-    const result = await build('docker/harness/build.sh');
+    const result = await build(['docker', 'harness', 'build.sh'].join(sep));
     if (result.status !== 0) {
       return { image, action: 'failed', error: `--substrate pod: 이미지 굽기 실패 rc=${result.status}${result.error ? ` (${(result.error as NodeJS.ErrnoException).code ?? result.error.message})` : ''}: ${(result.stdout + result.stderr).slice(-400)}` };
     }