@bli-cockpit/cli 0.2.45 → 0.2.47

package/dist/commands/local.js

@@ -9,6 +9,7 @@
  *   local-help.ts        what --help prints, and the recognised command names
  *   cli-io.ts            stdout/stderr/stdin plumbing and the production CliIo
  *   install-receipts.ts  the named step results this machine reports back
+ *   heartbeat.ts         the per-tick "I am alive, and these are my roots"
  *   local-auth.ts        which email, the OTP exchange, pairing with fallback
  *   collection-roots.ts  which folders may be collected, saved and read back
  *   local-discovery.ts   finding git worktrees inside those roots
@@ -36,6 +37,8 @@
  *   settings.ts          `cockpit settings` and `cockpit model`
  *   settings-render.ts   how a settings answer reads in a terminal
  *   team.ts              `cockpit team` — members, invite, role
+ *   autostart-heal.ts    the internal detached one-shot the macOS tick spawns
+ *                        to re-register launchd after the tick itself exits
  *
  * What stays here is orchestration: onboard, login/logout/start, the sync tick,
  * analyze, serve, autostart and agent-rules.
@@ -44,7 +47,8 @@ import path from "node:path";
 import { bufferedWritable, defaultExec, defaultIo, errorMessage, parseCapturedJson, replayCaptured, writeLine, } from "./cli-io.js";
 import { isLocalHelpRequest, localCommandHelp } from "./local-help.js";
 import { describeError, isMissingFileFailure } from "../health-detail.js";
-import { addInstallEvent, classifySyncHealthError, redactedSyncErrorDetail, reportInstallEventsBestEffort, } from "./install-receipts.js";
+import { sendCollectorHeartbeatBestEffort, } from "./heartbeat.js";
+import { addInstallEvent, classifySyncFailureRecords, classifySyncHealthError, redactedSyncErrorDetail, reportInstallEventsBestEffort, } from "./install-receipts.js";
 import { canReuseOnboardSession, pairLocalCollectorWithAuthFallback, readOnboardSessionReuseCandidate, requestPairingAccessToken, requestPairingAccessTokenDetailed, resolveInteractiveLoginEmail, resolveOnboardEmail, } from "./local-auth.js";
 import { collectionRootConsentAliases, persistOnboardingRootConfig, resolveOnboardingRootsForCommand, } from "./collection-roots.js";
 import { discoverCommandWorktrees, rememberDiscoveryLimits } from "./local-discovery.js";
@@ -75,16 +79,19 @@ import { autostartStatus, installAutostartAgent, uninstallAutostartAgent, } from
 import { DEFAULT_DASHBOARD_URL, ensureLocalCollectorConfig, getCollectorRuntimePaths, inspectLocalCollectorStatus, logoutLocalCollector, LOCAL_COLLECTOR_VERSION, readLocalCollectorConfig, readLocalCollectorSessionFile, startLocalWorkContext, } from "../local-state.js";
 import { acquireSyncLock } from "../sync-lock.js";
 import { runAttributedWorktreeSync, } from "./session-sync.js";
-import { COLLECTION_ROOT_REQUIRED, missingCollectionRootMessage, } from "../onboarding-roots.js";
+import { COLLECTION_ROOT_REQUIRED, CollectionRootRequiredError, missingCollectionRootMessage, } from "../onboarding-roots.js";
 import { rawEvidenceDedupSummary, rawEvidenceGcSummary, runRawEvidenceLocalGc, sweepDuplicateStagedRawEvidence, } from "../raw-evidence-gc.js";
 import { envWithNodeRuntimeOnPath, runScheduledSelfUpdate, } from "../scheduled-self-update.js";
 import { runAutostartSelfHeal, } from "../autostart-self-heal.js";
+import { rotateCollectorLogsBestEffort } from "../log-rotation.js";
+import { runAutostartHealDetached } from "./autostart-heal.js";
 import { normalizeCollectionRoots } from "../root-normalization.js";
 // `./local.js` is the published entry point for this command surface: the
 // public CLI's generated root, commands/root.ts, doctor.ts and the test suite
 // all import from here. Splitting the file must not move a name off it.
 export { rootCommandNames, localCommandHelp } from "./local-help.js";
-export { classifySyncHealthError, redactedSyncErrorDetail, reportInstallEventsBestEffort, SYNC_ERROR_DETAIL_MAX_CHARS, } from "./install-receipts.js";
+export { buildCollectorHeartbeat, collectionRootLabel, collectionRootLabels, sendCollectorHeartbeatBestEffort, } from "./heartbeat.js";
+export { classifySyncFailureRecords, classifySyncHealthError, redactedSyncErrorDetail, reportInstallEventsBestEffort, SYNC_ERROR_DETAIL_MAX_CHARS, } from "./install-receipts.js";
 export { assertCollectionRootPersisted } from "./collection-roots.js";
 export { runSelfUpdate, SelfUpdateError, } from "./install-update.js";
 export async function runLocalCockpitCli(argv, io = defaultIo()) {
@@ -1091,6 +1098,11 @@ async function runStart(command, io) {
 }
 async function runSync(command, io) {
     const paths = getCollectorRuntimePaths(command.homeDir);
+    // BLI-3553, first thing in the tick: cap the scheduler's own logs. Nothing
+    // rotated them before, and two months of a 15-minute tick left 211 MB of
+    // sync.log on the reference Mac. Best-effort by construction — a rotation
+    // problem is its own log line, never a reason collection does not run.
+    await rotateCollectorLogsBestEffort(paths);
     const config = await readLocalCollectorConfig(paths).catch(() => null);
     const dashboardUrl = command.dashboardUrl ?? config?.dashboard_url ?? DEFAULT_DASHBOARD_URL;
     const minCliVersionAtStart = await reportInstallEventsBestEffort({
@@ -1103,6 +1115,10 @@ async function runSync(command, io) {
     });
     try {
         const result = await runSyncWithHealthReceipt(command, io);
+        // BLI-3551: every tick checks in, including one that collected nothing.
+        // This is the only writer of `last_seen_at` that does not need an envelope,
+        // so it is what separates a quiet machine from a dead one.
+        await sendSyncHeartbeat(command, io, dashboardUrl, result.heartbeat);
         const minCliVersion = await reportInstallEventsBestEffort({
             homeDir: command.homeDir,
             dashboardUrl,
@@ -1118,6 +1134,14 @@ async function runSync(command, io) {
         return result.exitCode;
     }
     catch (error) {
+        const errorCode = classifySyncHealthError(error);
+        // A machine whose sync THREW is still alive, and that is worth knowing —
+        // a device that stops checking in entirely is a different problem from one
+        // checking in with a failure every fifteen minutes.
+        await sendSyncHeartbeat(command, io, dashboardUrl, {
+            status: "fail",
+            reason: errorCode,
+        });
         const minCliVersion = await reportInstallEventsBestEffort({
             homeDir: command.homeDir,
             dashboardUrl,
@@ -1126,7 +1150,7 @@ async function runSync(command, io) {
                 {
                     step: "sync_complete",
                     status: "fail",
-                    error_code: classifySyncHealthError(error),
+                    error_code: errorCode,
                     error_detail: redactedSyncErrorDetail(error),
                 },
             ],
@@ -1138,6 +1162,28 @@ async function runSync(command, io) {
         throw error;
     }
 }
+/**
+ * The tick's check-in (BLI-3551).
+ *
+ * Resolving the roots is best-effort on purpose: a machine with NO approved
+ * root is exactly the machine whose silence needs explaining, so it still
+ * checks in — with an empty root list, which is itself the finding.
+ */
+async function sendSyncHeartbeat(command, io, dashboardUrl, facts) {
+    const roots = await resolveSyncCollectionRoots(command).catch(() => []);
+    await sendCollectorHeartbeatBestEffort({
+        homeDir: command.homeDir,
+        dashboardUrl,
+        roots,
+        facts,
+        io,
+    }).catch((error) => {
+        // The sender already swallows everything it knows about; this is the net
+        // for anything it does not, because a heartbeat must never fail a sync.
+        console.error("[heartbeat] the check-in threw and was dropped", JSON.stringify({ reason: "heartbeat_threw", ...describeError(error) }));
+        return false;
+    });
+}
 /**
  * BLI-2721: after the tick's collection and self-update are done and
  * reported, repair a broken/legacy autostart registration in place (Windows
@@ -1199,7 +1245,10 @@ async function reportAutostartSelfHealOutcome(command, io, dashboardUrl, result)
         command: "sync",
         events: [
             {
-                step: "autostart_repair",
+                // Windows repairs in place and keeps the name already in the receipts
+                // and the runbook; the macOS path only SCHEDULES a detached repair, so
+                // it reports under its own step (BLI-3553).
+                step: result.step ?? "autostart_repair",
                 status: result.status,
                 ...(result.status === "ok" ? {} : { error_code: result.reason }),
                 ...(result.detail ? { error_detail: result.detail } : {}),
@@ -1298,10 +1347,24 @@ function scheduledSelfUpdateInstallEvent(result) {
         ? `forced_min_version ${result.min_version}`
         : null;
     if (result.status === "ok") {
+        // BLI-3551: this used to be `update ok` with an empty detail unless the
+        // floor forced it. One machine posted that receipt daily for nine releases
+        // while sitting on 0.2.37, and nobody could tell "already current" from
+        // "installed something" from "npm answered nothing" — three different
+        // situations wearing one word. The success branch names itself now.
+        const okDetail = [
+            forcedDetail,
+            result.reason === "updated" && result.previous_version && result.installed_version
+                ? `installed ${result.previous_version}→${result.installed_version}`
+                : result.reason,
+            result.target_version ? `target ${result.target_version}` : null,
+        ]
+            .filter((part) => Boolean(part))
+            .join("; ");
         return {
             step: "update",
             status: "ok",
-            ...(forcedDetail ? { error_detail: forcedDetail } : {}),
+            ...(okDetail ? { error_detail: okDetail } : {}),
         };
     }
     const detail = [
@@ -1341,6 +1404,10 @@ async function runSyncWithHealthReceipt(command, io) {
                 status: "skipped",
                 error_code: "live_sync_paused_during_backfill",
             },
+            heartbeat: {
+                status: "skipped",
+                reason: "live_sync_paused_during_backfill",
+            },
         };
     }
     // Single-flight: a launchd timer and a manual sync must not interleave the
@@ -1367,14 +1434,28 @@ async function runSyncWithHealthReceipt(command, io) {
                 status: "skipped",
                 error_code: "sync_already_running",
             },
+            heartbeat: { status: "skipped", reason: "sync_already_running" },
         };
     }
     try {
-        const { exitCode, failureReasons } = await runSyncLocked(command, io);
+        const { exitCode, failureReasons, failureRecords, notice, sessionsObserved, sessionsOutsideRoot, } = await runSyncLocked(command, io);
+        const counts = {
+            sessionsObserved,
+            sessionsOutsideRoot,
+        };
         if (exitCode === 0) {
+            // BLI-3551: an `ok` tick can still have something to say. `nothing_in_root`
+            // is the receipt that separates "this machine is alive and its operator
+            // works outside the approved roots" from "this machine is dead", which
+            // until now looked identical from the dashboard.
             return {
                 exitCode,
-                completion: { step: "sync_complete", status: "ok" },
+                completion: {
+                    step: "sync_complete",
+                    status: "ok",
+                    ...(notice ? { error_detail: notice } : {}),
+                },
+                heartbeat: { status: "ok", reason: notice, ...counts },
             };
         }
         // A sync that fails by exit code says exactly as much as one that throws.
@@ -1382,14 +1463,18 @@ async function runSyncWithHealthReceipt(command, io) {
         // failure rows carried a null detail and the real reason was reachable only
         // by running `cockpit status` on the machine itself (BLI-2526).
         const reasonText = failureReasons.join("; ");
+        // The bucket comes from the records the deciding branches wrote, not from
+        // this sentence (BLI-3551). The sentence is still the detail.
+        const errorCode = classifySyncFailureRecords(failureRecords);
         return {
             exitCode,
             completion: {
                 step: "sync_complete",
                 status: "fail",
-                error_code: classifySyncHealthError(reasonText),
+                error_code: errorCode,
                 error_detail: redactedSyncErrorDetail(reasonText),
             },
+            heartbeat: { status: "fail", reason: errorCode, ...counts },
         };
     }
     finally {
@@ -1407,6 +1492,10 @@ function syncResult(run) {
     return {
         exitCode: run.ok ? 0 : 1,
         failureReasons: run.ok ? [] : run.failure_reasons,
+        failureRecords: run.ok ? [] : run.failure_records,
+        notice: run.notice,
+        sessionsObserved: run.sessions_observed,
+        sessionsOutsideRoot: run.sessions_outside_root,
     };
 }
 /**
@@ -1492,6 +1581,7 @@ async function reportNoWorktreeSync(command, io, run, dedup) {
             mode: "no_worktrees",
             status: collectionRunStatus,
             collection_complete: run.ok,
+            ...(run.notice ? { notice: run.notice } : {}),
             codex_sessions: run.summary,
             raw_evidence_gc: gc,
             raw_evidence_dedup: dedup,
@@ -1499,6 +1589,11 @@ async function reportNoWorktreeSync(command, io, run, dedup) {
         return syncResult(run);
     }
     writeLine(run.ok ? io.stdout : io.stderr, `Tower sync ${collectionRunStatus}: no git worktrees under this root; session scan ran.`);
+    if (run.notice) {
+        // Says out loud what the receipt now says to the dashboard: the sessions
+        // this machine ran were all outside the folders it is allowed to look at.
+        writeLine(io.stdout, `Every session seen this run was outside your approved folders (${run.notice}). Nothing was collected, and nothing is broken.`);
+    }
     writeAgentSessionSummary(io, run.summary);
     if (gc && !gc.skipped)
         writeLine(io.stdout, rawEvidenceGcSummary(gc));
@@ -1558,7 +1653,7 @@ async function resolveSyncCollectionRoots(command) {
     if (savedRoots.length > 0) {
         return collectionRootConsentAliases(savedRoots);
     }
-    throw new Error(`${COLLECTION_ROOT_REQUIRED}: no explicit or saved collection root is available.`);
+    throw new CollectionRootRequiredError(`no explicit or saved collection root is available.`);
 }
 async function runAnalyze(command, io) {
     const syncStdout = [];
@@ -1701,6 +1796,17 @@ async function runServe(command, io) {
     return 0;
 }
 async function runAutostart(command, io) {
+    if (command.action === "heal-detached") {
+        // BLI-3553: the internal one-shot the scheduled macOS tick spawns. It has
+        // its own receipt and its own exit code; nothing else in this function
+        // applies to it.
+        return runAutostartHealDetached({
+            homeDir: command.homeDir,
+            dashboardUrl: command.dashboardUrl,
+            parentPid: command.parentPid ?? null,
+            json: command.json,
+        }, io);
+    }
     const exec = io.exec ?? defaultExec();
     const repoRoots = command.action === "install" || command.action === "status"
         ? await resolveAutostartRoots(command.homeDir, command.repoRoot)

package/dist/commands/ops-render.js

@@ -80,6 +80,8 @@ export function renderOpsStatus(payload, dim) {
                 lines.push(dim(`       scheduled by ${row.configFile} (${row.cron ?? "?"})`));
         }
     }
+    if (payload.fleet)
+        lines.push(...renderFleet(payload.fleet, dim));
     const slack = payload.skips?.slack;
     const external = payload.skips?.external;
     if (slack || external) {
@@ -93,6 +95,33 @@ export function renderOpsStatus(payload, dim) {
     }
     return lines;
 }
+/**
+ * The laptops. Red first, because a person reading this at 9am should not have
+ * to scroll past nine healthy machines to find the dead one.
+ */
+export function renderFleet(fleet, dim) {
+    const lines = ["", `FLEET  ${fleet.summary ?? "(no summary)"}`];
+    if (fleet.readError) {
+        lines.push(`  the fleet could not be read (${fleet.readError}); nothing is known about any machine`);
+        return lines;
+    }
+    const devices = fleet.devices ?? [];
+    const rank = (colour) => colour === "red" ? 0 : colour === "amber" ? 1 : 2;
+    const ordered = [...devices].sort((left, right) => rank(left.colour) - rank(right.colour));
+    if (ordered.length === 0) {
+        lines.push(dim("  no live collector device is registered at all"));
+    }
+    for (const device of ordered) {
+        const text = `  ${device.line ?? device.deviceId ?? "(device)"}`;
+        // A healthy machine is dimmed, never dropped: "which laptops are fine" is
+        // the other half of the question, and a list that only shows failures
+        // cannot answer "is everybody else collecting?".
+        lines.push(device.colour === "red" || device.colour === "amber" ? text : dim(text));
+    }
+    if (fleet.noDeviceLine)
+        lines.push(dim(`  ${fleet.noDeviceLine}`));
+    return lines;
+}
 function renderSkipLedger(ledger, dim) {
     const lines = [];
     const name = ledger.relation ?? "skips";

package/dist/commands/ops.js

@@ -78,6 +78,12 @@ async function runOpsStatus(command, io, tower) {
         unhealthy: unhealthy.length,
         unhealthy_ids: unhealthy.map((row) => row.id ?? "?"),
         with_skips: Boolean(command.skips),
+        // The laptops (BLI-3550), counted on the same line: a run that shows a
+        // green board and says nothing about the fleet cannot answer "was
+        // anybody's machine dead this morning?".
+        fleet_devices: payload.fleet?.counts?.devices ?? null,
+        fleet_red: payload.fleet?.counts?.red ?? null,
+        fleet_amber: payload.fleet?.counts?.amber ?? null,
     })}`);
     return unhealthy.length > 0 ? 1 : 0;
 }

package/dist/commands/public-root.js

@@ -15,7 +15,7 @@ export async function runCockpitCli(argv, io) {
   }
 
   if (command === "--version" || command === "-V" || command === "version") {
-    writeLine(io?.stdout ?? process.stdout, "0.2.45");
+    writeLine(io?.stdout ?? process.stdout, "0.2.47");
     return 0;
   }
 

package/dist/commands/session-sync.js

@@ -143,17 +143,34 @@ function claudeAttributionReadFailureCount(scan) {
 }
 export function sourceScanRetryReason(source, scan) {
     const reasons = new Set();
-    const readFailureCount = source === "codex"
-        ? codexAttributionReadFailureCount(scan)
-        : claudeAttributionReadFailureCount(scan);
-    if (readFailureCount > 0) {
-        reasons.add(`${source}_session_store_read_failed`);
-    }
+    const failure = sourceScanFailureReason(source, scan);
+    if (failure)
+        reasons.add(failure);
+    // Kept HERE and nowhere else (BLI-3551): a repo that is not on disk is a
+    // reason to widen the next scan window, because the transcript fallback can
+    // still attribute it. It is not a reason to call this sync failed — see
+    // `sourceScanFailureReason`.
     if (scan.results.some((result) => result.reason === "repo_not_on_disk")) {
         reasons.add("repo_not_on_disk");
     }
     return reasons.size > 0 ? [...reasons].sort().join(",") : null;
 }
+/**
+ * The part of the scan outcome that is a genuine FAILURE: the session store
+ * itself could not be read, so sessions that exist were not seen.
+ *
+ * Split from {@link sourceScanRetryReason} in BLI-3551. The two used to be one
+ * function, so `repo_not_on_disk` — a label the attribution umbrella finding
+ * already established is not a defect (nothing was deleted; the transcript
+ * names a path git no longer tracks) — failed the sync on every tick for three
+ * operators. A retry hint and a failure are different claims.
+ */
+export function sourceScanFailureReason(source, scan) {
+    const readFailureCount = source === "codex"
+        ? codexAttributionReadFailureCount(scan)
+        : claudeAttributionReadFailureCount(scan);
+    return readFailureCount > 0 ? `${source}_session_store_read_failed` : null;
+}
 async function reconcileSourceScanRetry(options) {
     if (options.reason) {
         await recordSourceRetryFailure(options.paths, {
@@ -447,25 +464,47 @@ export async function runAttributedWorktreeSync(options) {
     });
     // Same conditions as before, one per line, each writing down its own reason.
     // The old version was a single boolean chain: correct, and completely mute.
-    const failureReasons = new Set();
-    const fail = (condition, reason) => {
+    //
+    // Since BLI-3551 each condition also writes down the LABEL it is classified
+    // by, beside the rendered string a person reads. The health receipt reads the
+    // label; nothing parses the sentence back apart.
+    const failureRecords = new Map();
+    const add = (record) => {
+        if (!failureRecords.has(record.rendered)) {
+            failureRecords.set(record.rendered, record);
+        }
+    };
+    const fail = (condition, label, rendered = label) => {
         if (condition)
-            failureReasons.add(reason);
+            add({ label, rendered });
     };
     for (const { worktree, sync } of outcomes) {
         if (sync.status !== "uploaded") {
             // The spooled reason is the most specific thing anyone has, so lead with
             // it and name the worktree it belongs to — a fleet failure is usually one
             // repo, and "which one" is the first question asked.
-            failureReasons.add(sync.status === "spooled" && sync.failure_reason
-                ? `${worktree.worktree_label}:${sync.failure_reason}`
-                : `${worktree.worktree_label}:upload_${sync.status}`);
+            add(sync.status === "spooled" && sync.failure_reason
+                ? {
+                    label: sync.failure_class,
+                    rendered: `${worktree.worktree_label}:${sync.failure_reason}`,
+                    http_status: sync.failure_http_status,
+                }
+                : {
+                    label: "upload_not_completed",
+                    rendered: `${worktree.worktree_label}:upload_${sync.status}`,
+                });
         }
         for (const reason of sync.raw_evidence_failure_reasons ?? []) {
-            failureReasons.add(`raw_evidence:${reason}`);
+            add({
+                label: "raw_evidence_upload_failed",
+                rendered: `raw_evidence:${reason}`,
+            });
         }
         for (const reason of sync.raw_evidence_retry_reasons ?? []) {
-            failureReasons.add(`raw_evidence_retry:${reason}`);
+            add({
+                label: "raw_evidence_retry_required",
+                rendered: `raw_evidence_retry:${reason}`,
+            });
         }
         fail(sync.raw_evidence_deferred_byte_budget > 0, "deferred_byte_budget");
         fail(sync.raw_evidence_deferred_object_budget > 0, "deferred_object_budget");
@@ -474,28 +513,110 @@ export async function runAttributedWorktreeSync(options) {
     fail(claudeAttribution.session_limit_applied, "claude_session_limit_applied");
     fail(codexAttributionReadFailureCount(codexAttribution) > 0, "codex_session_read_failed");
     fail(claudeAttributionReadFailureCount(claudeAttribution) > 0, "claude_session_read_failed");
-    const codexScanRetry = sourceScanRetryReason("codex", codexAttribution);
-    if (codexScanRetry)
-        failureReasons.add(`codex_scan:${codexScanRetry}`);
-    const claudeScanRetry = sourceScanRetryReason("claude_code", claudeAttribution);
-    if (claudeScanRetry)
-        failureReasons.add(`claude_scan:${claudeScanRetry}`);
-    fail(reportRequired && !report.posted, `session_report_unposted:${report.reason ?? "unknown"}`);
+    // BLI-3551: the scan's RETRY reason and the scan's FAILURE reason are two
+    // different questions, and answering both with one function is what put
+    // `claude_scan:repo_not_on_disk` on every tick of three machines. A repo that
+    // is not on disk is a label on the session (the attribution umbrella finding:
+    // nothing was deleted, the transcript simply names a path git no longer
+    // knows). It still widens the next scan window; it is not a failed sync.
+    const codexScanFailure = sourceScanFailureReason("codex", codexAttribution);
+    if (codexScanFailure) {
+        add({ label: "codex_scan_read_failed", rendered: `codex_scan:${codexScanFailure}` });
+    }
+    const claudeScanFailure = sourceScanFailureReason("claude_code", claudeAttribution);
+    if (claudeScanFailure) {
+        add({
+            label: "claude_scan_read_failed",
+            rendered: `claude_scan:${claudeScanFailure}`,
+        });
+    }
+    // BLI-3551: a tick that observed only sessions from outside the operator's
+    // approved roots has nothing to post, and that is the consent boundary
+    // working — not a failure. It used to fail as
+    // `session_report_unposted:no_successful_sync`, whose word "session" then
+    // classified as `auth_failed`; one machine reported a broken credential 377
+    // times in 38 hours while its token had eleven weeks left. The withhold
+    // decision itself is untouched (adapters/attribution-core.ts) — only what it
+    // is CALLED.
+    const sessionsOutsideRoot = sessions.filter((session) => OUTSIDE_APPROVED_ROOT_REASONS.has(session.attribution_reason)).length;
+    const nothingInRoot = nothingInRootCount({
+        sessionCount: sessions.length,
+        outsideRootCount: sessionsOutsideRoot,
+        outcomes,
+        reportPosted: report.posted,
+        reportReason: report.reason,
+    });
+    fail(reportRequired && !report.posted && nothingInRoot === null, "session_report_unposted", `session_report_unposted:${report.reason ?? "unknown"}`);
     // `ok` may already be false from the per-worktree loop above; the outcome
     // scan re-derives that, so the two agree by construction.
-    ok = ok && failureReasons.size === 0;
-    if (!ok && failureReasons.size === 0) {
-        failureReasons.add(SYNC_FAILED_WITHOUT_REASON);
+    ok = ok && failureRecords.size === 0;
+    if (!ok && failureRecords.size === 0) {
+        add({
+            label: SYNC_FAILED_WITHOUT_REASON,
+            rendered: SYNC_FAILED_WITHOUT_REASON,
+        });
+    }
+    const notice = ok && nothingInRoot !== null ? `nothing_in_root:${nothingInRoot}` : null;
+    if (notice) {
+        // The success branch says something too: this is the receipt that proves a
+        // quiet machine is a working machine, and the count is what tells a coach
+        // that someone is working entirely outside the approved boundary.
+        console.error("[session-sync] nothing to collect inside the approved roots", JSON.stringify({
+            reason: "nothing_in_root",
+            sessions_outside_root: nothingInRoot,
+            collection_root_count: collectionRoots.length,
+            next_action: "widen the approved roots (an operator decision) if this machine should be collecting here",
+        }));
     }
+    const records = [...failureRecords.values()].sort((a, b) => a.rendered.localeCompare(b.rendered));
     return {
         ok,
-        failure_reasons: [...failureReasons].sort(),
+        failure_reasons: records.map((record) => record.rendered),
+        failure_records: records,
+        notice,
+        sessions_observed: sessions.length,
+        sessions_outside_root: sessionsOutsideRoot,
         outcomes,
         codexAttribution,
         claudeAttribution,
         summary,
     };
 }
+/**
+ * Reasons attribution gives when a session's working directory is not inside
+ * any approved collection root.
+ *
+ * Exact labels, not a pattern — the same discipline the classifier now follows.
+ * `attribution-core.ts` writes both of these and nothing else means
+ * "outside the boundary".
+ */
+const OUTSIDE_APPROVED_ROOT_REASONS = new Set([
+    "cwd_outside_scanned_worktrees",
+    "no_matching_worktree_signals",
+]);
+/**
+ * How many observed sessions were outside the approved roots, when that
+ * accounts for ALL of them and nothing else went wrong — otherwise `null`.
+ *
+ * Deliberately narrow. It requires that no worktree was synced at all (so no
+ * upload could have succeeded or failed), that every session observed this tick
+ * names an outside-the-root reason, and that the unposted report is the
+ * `no_successful_sync` shape rather than a spooled report that failed to flush.
+ * Anything else keeps its failure.
+ */
+export function nothingInRootCount(options) {
+    if (options.reportPosted)
+        return null;
+    if (options.reportReason !== "no_successful_sync")
+        return null;
+    if (options.outcomes.length > 0)
+        return null;
+    if (options.sessionCount === 0)
+        return null;
+    return options.outsideRootCount === options.sessionCount
+        ? options.outsideRootCount
+        : null;
+}
 export const ATTRIBUTION_STATE_RANK = CODEX_SESSION_ATTRIBUTION_STATE_RANK;
 function normalizeCodexResult(result) {
     return {