{"available":true,"floor":0.997,"coverage":{"overall":1,"routes":1,"jobs":1,"consumers":1,"governed":1594,"total":1594,"side_effecting_routes":1490,"governed_routes":1490,"jobs_total":91,"jobs_governed":91,"consumers_total":13,"consumers_governed":13,"read_only_routes":1694},"gaps":[],"disagreements":[],"jobs":[{"id":"engine/aleeth-brain-daily.startBrainDaily","control":"platform_internal","governed":true,"effects":"daily read-only diagnostic sweep; on issues writes a Brain finding and notifies"},{"id":"engine/aleeth-brain-proactive.startBrainProactive","control":"platform_internal","governed":true,"effects":"morning brief + proactive pushes to Shane (Slack/email); writes brain push log rows"},{"id":"engine/asap/jobs.startAsapCalibration","control":"gated","governed":true,"effects":"ASAP judge calibration (docs/asap/ASAP-v1-DESIGN.md sections 14 and 16). OFF unless ICA_ARM_ASAP_CALIBRATION=1; then every ASAP_CALIBRATION_HOURS (default 24), beating the asap-calibration daemon heartbeat. Runs the in-process reference suite with the live judges (one call per family, each pinned to its provider through engine/llm-provider callLLM) and appends one row to asap_calibration_runs, read back after the write. The public calibration endpoint reads the last row; no public request calls a model"},{"id":"engine/asap/jobs.startAsapContinuous","control":"gated","governed":true,"effects":"ASAP Continuous Mode (docs/asap/ASAP-v1-DESIGN.md section 22). OFF unless ICA_ARM_ASAP_CONTINUOUS=1; then every ASAP_CONTINUOUS_MINUTES (default 60), beating the asap-continuous daemon heartbeat. For each sealed engagement whose operator schedule is due it re-runs discovery, the arm-gated probes (ICA_ARM_ASAP_PROBES=1, scope-frozen target only, refused while the self-audit gate fails), the enrolled veracity claims and the economics, then drafts the next record in the same lineage: ledger entries, an asap_probe_batch artifact receipt sealed before state, and versioned work state in asap_work_state. It never seals a record: the customer co-signs every genesis in the portal and an operator seals it"},{"id":"engine/atc-engine.startATC","control":"platform_internal","governed":true,"effects":"correlates sensor signals into ica_atc_incidents and per-sensor coverage; executors (halt/quarantine/notify) run only when armed"},{"id":"engine/aws-alarm-intake.startAwsAlarmIntake","control":"platform_internal","governed":true,"effects":"on the runner-lease leader, every 60 s, reads the SQS queue ica-alarm-intake (EventBridge state changes of the ica-* CloudWatch alarms) with the task role and writes ALEETH platform rows of ica_alerts: an ALARM through captureAlert keyed by the alarm name, an OK as a cleared marker in the row metadata (no status change; a named person resolves), INSUFFICIENT_DATA dropped; a message is deleted only after its write is read back, else retried and kept by the dead-letter queue after 5 tries. Also takes the account's managed threat findings from the same queue as network detection alerts; when network detection is armed it seals them. Idle where the queue does not exist; no setting"},{"id":"engine/brain-autonomy-decide.startBrainAutonomySweep","control":"platform_internal","governed":true,"effects":"OFF by default; scheduled only under ICA_SHADOW_BRAIN_AUTONOMY=1 or ICA_ARM_BRAIN_AUTONOMY=1, every ICA_BRAIN_AUTONOMY_MINUTES (default 10), beating the brain-autonomy daemon heartbeat. Each tick reads the latest tsm Decision Advisor scan and the proposed cos_brain_actions, skips items already decided with the same evidence, and decides at most ICA_BRAIN_AUTONOMY_SWEEP_CAP (default 10): it asks the pre-action gate (POST /api/ingest/gate-request) under ALEETH Brain 01 and seals one brain_autonomy_decision receipt per item. Shadow acts on nothing and voids every execution token. Armed, an item that ICA allows and that sits inside the sealed Tier 3 charter is carried out through the native write (decision-execute RESOLVERS, agent-governance, the IAC scorer, the Brain action executors) and recorded as a governed agent step that spends the token; everything else stays on Shane's path. An error-budget breach, an anomalous decision rate or a failed decision seal writes a rail_guard halt on the Brain. Never writes ica_approval_requests and never sets a human claim"},{"id":"engine/ccm-scheduler.startCCMScheduler","control":"platform_internal","governed":true,"effects":"every 60s dispatches due ica_ccm_schedules per active customer and writes run rows + control results"},{"id":"engine/chain-anchor/job.startChainAnchorRelay","control":"platform_internal","governed":true,"effects":"OFF unless ICA_ARM_CHAIN_ANCHOR=1; then on the runner-lease holder, boot + 600 s and every ICA_CHAIN_ANCHOR_HOURS (default 6, 1 to 24), beating the chain-anchor-relay daemon heartbeat. Each tick runs engine/chain-anchor/relay.js relayOnce: reads the anchor checkpoint (engine/tlog-anchor-checkpoint.js latest), and for each chain in ICA_CHAIN_ANCHOR_CHAINS (at most 4) reads its sealed chain_anchor_tx history and the chain over JSON-RPC (https only, no redirects, answers capped at 2 MB), asks AWS KMS to sign one 32-byte digest with the relay key (ECDSA_SHA_256, MessageType DIGEST; the key never leaves KMS), seals a chain_anchor_tx \"signed\" receipt, then sends at most one zero-value EIP-1559 transaction from the relay address to itself whose calldata is ICA-ANCHORS-V1 || tree_size || root_hash, and seals broadcast, confirmed, failed or dropped receipts. Refuses before signing on a wrong chain id, a pending or unconfirmed transaction of its own, fees above max_fee_gwei, gas above gas_limit_max, the daily_tx_cap or daily_wei_cap, or a balance below the worst case. Sends only zero-value transactions and pays their gas from the relay address's own balance (funded by the owner, capped per chain per day); also reads the KMS public key (GetPublicKey). A tick fails its heartbeat on an error, an unfunded address, or no commitment for over 24 h while the tree grew. Writes nothing else in ICA, changes no gate outcome"},{"id":"engine/continuous-scoring.startCadence","control":"platform_internal","governed":true,"effects":"living-certification cadence: recomputes the 9-domain IAC score and writes score history"},{"id":"engine/duty-cadence-scheduler.startDutyCadence","control":"platform_internal","governed":true,"effects":"musters the Unit on an interval so each seat stands its governed watch; writes duty run receipts"},{"id":"engine/duty-executor-scheduler.startDutyExecutor","control":"platform_internal","governed":true,"effects":"ARM-GATED (ICA_ARM_DUTY_EXECUTOR) consumer of ica_agent_duties: each interval claims open duty orders and runs billet-routed governed work, writing ica_agent_steps and mutating duty status"},{"id":"engine/effect-relay.startEffectRelay","control":"platform_internal","governed":true,"effects":"transactional effect outbox relay (starts only when ICA_EFFECT_RELAY_INTERVAL_MIN is set): claims pending effects with SKIP LOCKED, seals effect_outbox_relay receipts, consumes once, settles, reclaims stale claims"},{"id":"engine/evidence-mesh.startPeriodicEvidenceMesh","control":"platform_internal","governed":true,"effects":"pulls evidence from customer connectors on a cadence and seals evidence receipts (PANDORA machine 1)"},{"id":"engine/ica-api-catalog.startCatalogRefresh","control":"platform_internal","governed":true,"effects":"refreshes the governed external-API catalog scores/classifications in batches with evidence rows"},{"id":"engine/ica-drift-sensor.startDriftSensor","control":"platform_internal","governed":true,"effects":"latent-problem detection (index/row growth/timeouts); writes drift observations + incidents"},{"id":"engine/ica-provenance-server.startPeriodicSealing","control":"platform_internal","governed":true,"effects":"L5 provenance: OpenTimestamps upgrade poll + catch-up seals of unanchored ledger seals (ica_provenance_seals); no periodic mint"},{"id":"engine/ica-regulatory-analyst.startIcaExtractionAgent","control":"platform_internal","governed":true,"effects":"in-ICA extraction agent over ica_legislation_watch via engine/llm-provider; writes extraction proposals only"},{"id":"engine/ica-regulatory-analyst.startRegulatoryAnalyst","control":"platform_internal","governed":true,"effects":"HUMAN-GATED: proposes control updates from newly enacted laws into a proposals table; applies nothing"},{"id":"engine/ica-remediation-agent.startRemediationAgent","control":"platform_internal","governed":true,"effects":"LLM-in-the-loop remediation proposals, HUMAN-GATED; writes proposals only, never applies"},{"id":"engine/ica-self-heal-warden.startWarden","control":"platform_internal","governed":true,"effects":"reads FIRING sentinel incidents and runs idempotent reversible fixes from a static registry; writes warden actions"},{"id":"engine/ica-self-watch.startSelfWatch","control":"platform_internal","governed":true,"effects":"recomputes live posture per active cert, routes auto-remediable findings, writes ica_self_watch_runs (append-only)"},{"id":"engine/ica-sentinel.startSentinel","control":"platform_internal","governed":true,"effects":"self-monitoring probes every 5 min: ica_sentinel_probes state, ica_sentinel_incidents (append-only), heartbeat, Slack/SMTP alerts"},{"id":"engine/ledger-anchor-recount.startLedgerAnchorRecount","control":"platform_internal","governed":true,"effects":"hourly (ICA_ARM_LEDGER_ANCHOR): read-only recount of every ledger window anchored in the last 48h against the leaf_count its anchor signed; writes only its own ica_sentinel_heartbeats row (ledger_anchor_recount), which the Sentinel probe ledger.anchor.window_recount reads"},{"id":"engine/observability-slo.startCadence","control":"platform_internal","governed":true,"effects":"SLO measurement cadence over real telemetry; writes SLO snapshots + error-budget state"},{"id":"engine/pc-shadow-canary.startPcShadowCanary","control":"platform_internal","governed":true,"effects":"hourly per-tenant read of ica_agent_steps and ica_proof_carrying_log; writes only an ica_alerts row (ica_alert_upsert) when a tenant had steps but 0 log rows in 24h, or when a read failed"},{"id":"engine/push-alerts-job.startPushAlerts","control":"platform_internal","governed":true,"effects":"OFF unless ICA_PUSH_TRANSPORT is direct or broker; leader only. Every 60 s reads approvals waiting (ica_approval_requests pending) and open critical and high alerts (ica_alerts) from the last 15 minutes, routes each to the live phone subscriptions (ica_push_subscriptions) whose operator can read its tenant, claims (subscription, event) once in ica_push_deliveries, sends one Web Push (a fixed title and the severity word, no alert text) to a phone maker push host only, directly or through the Outbound Broker, seals push_sent or push_failed, and revokes a subscription the push service answers 404 or 410. At most 30 pushes a phone an hour"},{"id":"engine/receipt-signature-sweep.startReceiptSignatureSweep","control":"platform_internal","governed":true,"effects":"every 5 min (ICA_ARM_LEDGER_ANCHOR, off with ICA_RECEIPT_SWEEP_OFF=1): read-only re-hash and Ed25519 signature check of every receipt written since its last run, and of every receipt in each newly anchored ledger window; writes only its own ica_sentinel_heartbeats row (receipt_signature_sweep: cursors and open findings), which the Sentinel probe ledger.receipts.signature_sweep reads; changes no gate outcome"},{"id":"engine/reconciler-job.startReconciler","control":"platform_internal","governed":true,"effects":"OFF unless ICA_RECONCILER_JOB=shadow (unset, 0, off or any other value is off; no heartbeat is registered while off). On the runner-lease leader, hourly (boot + 720 s), beating the reconciler daemon heartbeat: reads the shadow-identity scan, the posture-drift snapshots, the newest monitor canary run and ica_compliance_findings (read only), decides each finding against engine/reconciler-rules.json (every row shadow; a table with any other mode stops the tick) and seals one reconciler_decision receipt per new acting, paging, cap-tripped or no-rule decision (deduped on finding and outcome for 24 hours; at most 50 a tick) plus one tick-summary receipt of counts for what was not measured or under a floor. A tick that could not seal, read, or measure any detector throws, so the heartbeat reads failing. It halts, revokes, holds and releases nothing and writes no other row. A seal refused because ica_receipts does not list the kind yet is counted and not retried"},{"id":"engine/security-closure-assignment-topoff.startSecurityClosureAssignmentTopOff","control":"platform_internal","governed":true,"effects":"keeps the critical closure lane owner-assigned by writing campaign receipts/provenance only"},{"id":"engine/sponsor-departure.startSponsorDepartureContain","control":"gated","governed":true,"effects":"Sponsor departure containment. OFF unless ICA_ARM_SPONSOR_DEPARTURE is shadow or enforce; then every ICA_SPONSOR_DEPARTURE_MINUTES (default 15), beating the sponsor-departure-contain daemon heartbeat. Reads the residual ICA authority of identities the IdP or ICA deactivated (engine/deauthorization-store.js). shadow writes one ica_operator_events row when the planned set changes and nothing else. enforce, on strong matches only and inside the per-identity and per-run caps: halts agent identities whose recorded sponsor left (engine/agent-authority.js setIdentityLifecycle, gate_decision receipt), halts roster agents whose business owner left (ica_rail_guard_halt_set, rail_guard_halt receipt), revokes live grants the person granted (ica_agent_authority_grants revoked_at, gate_decision receipt), reads each end state back and records each step in ica_operator_events"},{"id":"engine/uca/revocation-replica.startReplicaPull","control":"platform_internal","governed":true,"effects":"when a revocation store is configured, each replica pulls ica_identity_revocations on an interval into a per-tenant index; an unread or failed pull denies authority dispatch"},{"id":"engine/vuln-kev-watch.startKevWatch","control":"platform_internal","governed":true,"effects":"OFF unless ICA_ARM_KEV_WATCH=1; then, on the runner-lease leader only, boot + 240 s and every ICA_KEV_WATCH_MINUTES (default 60, 15 to 1440), beating the kev-watch daemon heartbeat. Reads the public CISA Known Exploited Vulnerabilities feed (engine/vuln-exploit-feeds readKev, at most 55 minutes old; sends nothing about any tenant) and every open ica_vuln_findings row whose advisory id is a CVE, in one read of at most 5,000. Writes one ica_alerts row (ica_alert_upsert, deduped by the finding id) per open finding on KEV, critical and due 4 hours from detection for a critical finding, high and due 24 otherwise, and one more (deduped) when such a finding is past that deadline. Findings are read newest first. A resolved alert reopens on the next hour while its finding is still open and on KEV (fixing or accepting the finding stops it). A failed feed or findings read, a cut list or an alert that does not land fails the tick. Changes no finding, deadline or gate outcome and revokes nothing"},{"id":"engine/vuln-osv-import.startOsvImport","control":"platform_internal","governed":true,"effects":"OFF unless ICA_ARM_OSV_ADVISORIES=1; then boot + 600 s and every ICA_OSV_ADVISORIES_HOURS (default 24, 1 to 168), beating the osv-advisories daemon heartbeat. Reads every distinct (name, version, purl) in ica_sbom_components in one complete read; asks api.osv.dev (POST /v1/querybatch, GET /v1/vulns/<id>; read-only, no LLM) which advisories name each npm, PyPI and Go package and which affect each installed version, bounded at 10,000 packages, 20,000 versions and 5,000 records. Writes ica_cve_advisories rows with source osv only: new keys inserted with ignore-duplicates (an operator row under the same key is left as it is), changed and stale osv rows updated or deleted with source=eq.osv in the filter, then read back. Debian packages are reported, never written. Any failed read writes nothing; a failed read, write or read-back writes one ica_alerts row (ica_alert_upsert) and fails the tick. Changes no gate outcome and revokes nothing"},{"id":"engine/vuln-scan-job.startVulnScan","control":"platform_internal","governed":true,"effects":"OFF unless ICA_ARM_VULN_SCAN=1; then boot + 300 s and every ICA_VULN_SCAN_HOURS (default 6, 1 to 168), beating the vuln-scan daemon heartbeat. Lists every (tenant, tool_id) holding ica_sbom_components in one complete read and runs engine/vuln-dependency scanForVulns for each tool on its own, never for a whole tenant. Each scan inserts new ica_vuln_findings rows under a deterministic id (tenant, tool_id, component, version, CVE) with ignore-duplicates, so a re-scan or a second replica adds nothing, and reopens a finding marked patched while the scan still sees the affected version; accepted and waived findings are left as they are. Writes one ica_alerts row (ica_alert_upsert) per tool that opened or reopened findings, and one per failed scan, unread scope list or empty plane; a pass that did not check every tool fails the tick. Changes no gate outcome and revokes nothing"},{"id":"engine/vuln-self-sbom.startSelfSbom","control":"platform_internal","governed":true,"effects":"OFF unless ICA_SELF_SBOM is set (the AWS production stack sets it to the image and commit this task runs, the build-record bucket, its region and tenant tsm); then boot + 90 s and every ICA_SELF_SBOM_HOURS (default 1, 1 to 24), beating the self-sbom daemon heartbeat. Reads from S3, read only, the verified build record under image-builds/<its own commit>/ that names exactly this image, and the SBOM beside it, refusing an SBOM whose sha256 is not the recorded one (the task role reads that prefix and nothing else in the bucket). Reads the tool ica_sbom_components first and writes nothing when they are exactly this SBOM under sbom_id <tool>@<digest>; otherwise writes them through engine/vuln-dependency ingestSbom in parts of 500 under that one sbom_id, the last part with replace so components from earlier builds are removed, then reads them back and passes only on an exact match, and runs engine/vuln-dependency scanForVulns once for that one tool (inserting ica_vuln_findings under deterministic ids with ignore-duplicates, as the scheduled scan does), writing one ica_alerts row (source vuln-scan) when it opened or reopened findings and one when open findings name a package the image no longer holds. A failed read, write, read-back or scan writes one ica_alerts row (ica_alert_upsert) and fails the tick. Closes no finding, changes no gate outcome and revokes nothing"},{"id":"lib/decision-latency/job.startDecisionLatencyFlush","control":"platform_internal","governed":true,"effects":"every window, default 15 min (only when ICA_ARM_DECISION_LATENCY=1, which defaults off): ends the in-memory decision latency recorder window and seals one decision_latency_rollup receipt per replica per window, holding per decision path the sample count, p50, p95, p99, max and histogram bucket counts, the window bounds, the release commit and the replica id, then reads the receipt back; a failed flush holds the window for retry and beats the decision-latency-flush heartbeat overdue; writes nothing else in ICA and changes no gate outcome"},{"id":"lib/vault-token.startVaultTokenRenewal","control":"platform_internal","governed":true,"effects":"boot+30s, then inside the token TTL, only when ICA_SIGNER_KIND=vault and ICA_VAULT_ADDR and ICA_VAULT_TOKEN are set: renews the worker Vault token on auth/token/renew-self. A failed renewal marks daemon vault-token-renew overdue on that tick. It does not rotate the token and does not edit a Vault policy."},{"id":"middleware/shield-store.startSharedSync","control":"platform_internal","governed":true,"effects":"every 15 s (ICA_SHIELD_SYNC_MS), only when ICA_ARM_SHARED_SHIELD=1 and a service key is set: started by middleware/shield.js when it is loaded, not by server.js. Upserts the rate and auth IP blocks this replica decided into ica_shield_blocks (list, key, until_ms, replica, updated_at) over Supabase REST with the service key, then reads back every block still live and merges it into the local block lists, so a block decided on one replica applies on the others. Any error falls back to local-only. RLS is on with no policy, so only the service key reaches the table"},{"id":"routes/capture-governance-auto-tick.startCaptureGovernanceAutoTickRunner","control":"platform_internal","governed":true,"effects":"capture-governance auto-tick over the enrolled fleet: sweep + enforcement receipts per tenant"},{"id":"routes/security-closure-auto-wave.startSecurityClosureAutoWaveDagController","control":"platform_internal","governed":true,"effects":"security-closure auto-wave DAG controller: campaign wave receipts/provenance only, no native finding mutation"},{"id":"scheduleDaemon:aai-fairness-monitor","control":"platform_internal","governed":true,"effects":"boot + 480 s, then daily (off with ICA_FAIRNESS_MONITOR_OFF=1): for each tenant with industry-healthcare-provider-ai-2026 HC-AI-PA-001 bound, evaluates the four-fifths ratio and parity gap over the share of appealed AI denials that held per opaque subgroup (engine/subgroup-outcomes.js, engine/aai-fairness.js); a measured run is sealed as one fairness_assessment receipt, a breach opens one advisory ica_compliance_findings row on carrier rule ICA-FAIRNESS-ADVISORY-001 and within-line closes it; unmeasured (no subgroup data, fewer than two measured subgroups, a cut-short read) writes nothing and closes nothing; changes no standing, score or halt"},{"id":"scheduleDaemon:approval-expiry-sweep","control":"platform_internal","governed":true,"effects":"boot + 150 s, then every 5 min (off with ICA_HOLD_EXPIRY_SWEEP_OFF=1): marks each customer ica_approval_requests row past its expiry as expired (conditional on still pending with no minted token; Company OS holds are left to engine/os-holds.js), seals one execution_grant receipt per expired hold with decision denied and approver system:expiry, records its hash on the row, and opens one approval_hold_expired self-audit finding per hold (dedupe key approval-hold-expired:<id>); never approves, never mints a token"},{"id":"scheduleDaemon:cert-drift","control":"platform_internal","governed":true,"effects":"boot + 120 s, then hourly (ICA_CERT_DRIFT_INTERVAL_MIN, 15 to 1440; it was 6 h until 2026-11-06): for each active or under_review cert, checks the worst agent score, open high/critical findings from the last 24h and a week-over-week drop in average agent score; inserts an ica_recertification_triggers row unless that cert already reported the same trigger type in the last 7 days; changes no cert status"},{"id":"scheduleDaemon:cert-lifecycle","control":"platform_internal","governed":true,"effects":"daily: moves active or under_review certs past their expiration_date to expired in ica_certification_register; inserts the expired and 90/60/30-day countdown ica_recertification_triggers rows and one ica_renewal_assessments row per cert inside 90 days, each unique per cert, so a re-run adds nothing"},{"id":"scheduleDaemon:clinical-fitness-attest","control":"platform_internal","governed":true,"effects":"boot + 510 s, then daily (off with ICA_CLINICAL_FITNESS_ATTEST_OFF=1): for each tenant with industry-healthcare-provider-ai-2026 HC-AI-FDA-003 bound, reads the newest fairness_assessment receipt of each clinical model its binding names (evaluator_config.clinical_systems), verified against ica_signing_keys, and attests clinical_models_with_fit_subgroup_verdict_pct through engine/kpi-evidence-attest.js (the evidence route path): one ica_pack_kpi_evidence row per tenant per UTC day; a miss emits one kpi_breach finding through ica_atc_compliance_emit_finding naming no agent; a met value closes this KPI breaches that name no agent. A binding that names no clinical models, or a cut-short read, writes nothing; refuses a KPI whose miss halts before any write, and never lifts a halt"},{"id":"scheduleDaemon:compliance-package-schedule","control":"platform_internal","governed":true,"effects":"only when ICA_ARM_COMPLIANCE_SCHEDULE=1: boot + 120 s, then every minute on the runner-lease holder; in the first 15 minutes of 06:00 and 18:00 CT mints that slot's compliance package once (a failed try is retried the next minute) (rpc ica_atc_compliance_mint_package: snapshots the active compliance rules into ica_compliance_packages, takes one L5 seal, stamps findings with no package yet); a slot already minted is refused by the database and reads already minted"},{"id":"scheduleDaemon:connected-app-audit","control":"platform_internal","governed":true,"effects":"boot + 1500 s, then every 15 min (off with ICA_CONNECTED_APP_WATCH_OFF=1; reads only with ICA_ARM_CONNECTED_APP_WATCH=1): for each active read connection in ica_connected_app_connections, under the scan lease, reads the customer directory's audit log and sign-in logs (GET only, with the connection's own read-only credential; app sign-ins from the beta endpoint); seals each finding (connected_app_finding) and each window (connected_app_scan, with a digest over its rows); writes ica_connected_app_audit_events, ica_connected_app_signins and ica_connected_app_events; moves the audit cursor only after the window is sealed; one alert per critical finding; never disables, deletes or writes anything in the directory; changes no gate outcome"},{"id":"scheduleDaemon:connected-app-watch","control":"platform_internal","governed":true,"effects":"boot + 1200 s, then every 6 h (off with ICA_CONNECTED_APP_WATCH_OFF=1; reads only with ICA_ARM_CONNECTED_APP_WATCH=1): for each active read connection in ica_connected_app_connections, reads the customer directory's apps, grants, directory roles and credential metadata through Microsoft Graph (GET only, with the connection's own read-only credential); seals each finding (connected_app_finding) and the scan (connected_app_scan); writes ica_connected_app_events and ica_connected_apps; one alert per critical finding; never disables, deletes or writes anything in the directory; changes no gate outcome"},{"id":"scheduleDaemon:economics-population","control":"platform_internal","governed":true,"effects":"boot + 150 s, then daily (ICA_ARM_ECONOMICS_POPULATION): for each customer with a registered system, inserts one ica_cost_records row, with a cost_record_ingested receipt, for each provider usage snapshot window of the last 30 days (ICA_ECONOMICS_POPULATION_WINDOW_DAYS) that has no record yet; a window already recorded is counted and skipped, and the ica_cost_records_snapshot_natural_key index refuses a second record once applied. Then seals one economic_ledger receipt and one ica_economic_ledgers row; changes no gate outcome"},{"id":"scheduleDaemon:forecast-quantum","control":"platform_internal","governed":true,"effects":"boot + 960 s, then daily (off with ICA_FORECAST_QUANTUM_OFF=1): reads ica_forecast_quantum_state (or the built-in defaults when none is seeded) and computes the safety-margin countdown as of now (engine/forecast-quantum.js computeHorizon); each fired trigger is posted once to the Operations Needs You inbox (source_id carries the trigger key and the state receipt). When ica_endurance_drill_schedule says the annual break drill is overdue or never ran, runs engine/forecast-endurance.js runAnnualDrill under engine:forecast-quantum-cadence: one ica_forecast_quantum_drills row and its quantum_horizon_drill receipt, one endurance_contract receipt and the schedule row moved a year on; a drill that does not pass is posted once to Needs You. Changes no trigger, trajectory or state row, re-signs no receipt"},{"id":"scheduleDaemon:framework-currency","control":"platform_internal","governed":true,"effects":"boot + 1500 s, then daily (off with ICA_FRAMEWORK_CURRENCY_OFF=1). Unarmed (ICA_ARM_RDE_LIVE_FETCH not 1): reads only, writes no row and seals nothing; posts one not-armed notice to the Operations Needs You inbox, repeated weekly (worker state framework-currency-alarm in ica_sentinel_heartbeats). Armed, at most one full run per 20 h (worker state framework-currency-run, marked running then done; a run left running past its 45 min budget plus 15 min no longer blocks): links each active ica_policy_frameworks row to one official source (its ica_legislation_watch row source_url by the analyst slug rule, else data/frameworks/framework-sources.json) and inserts any missing reg_discovery_sources row (source_id fw:<slug>; never edits or disables an existing source; never for a non-https, explicit-port, IP-literal or internal URL). For each such source due (not fetched in the last day) whose host the egress allowlist names, healthy sources oldest first and then at most 20 sources failing 3+ times in a row (a window that rotates by day, so each is retried), at most 200 per run and within 45 min, runs engine/regulatory-discovery-engine.js runFetch (GET of the official source; a non-2xx answer, a redirect off the source host (a leading www. aside) or a non-text body is a failed fetch; one reg_discovery_fetch_jobs row; a reg_discovery_snapshots row only when the text changed; a detected change adds one reg_discovery_changes row, one pending reg_discovery_review_queue item and one evidence_bundle receipt). Seals one evidence_bundle receipt per completed run (kind framework_currency_audit, tenant tsm); a seal that fails is reported in the run log and fails its daemon heartbeat. Posts up to 10 detected changes one by one and the rest in one roll-up, and, when the kinds of finding change or weekly, failing, stale, unchecked, egress-blocked, unsafe or unwatched sources, to Needs You. Never writes a framework, control, crosswalk or pack, and approves nothing"},{"id":"scheduleDaemon:halt-handoff","control":"platform_internal","governed":true,"effects":"boot + 75 s, then every 60 s (off with ICA_HALT_HANDOFF_OFF=1): for each agent under an active halt (lib/agent-halt-state.js getActiveHalts), seals one work_handoff receipt and then moves its open ica_agent_duties (ASSIGNED or IN_PROGRESS, conditional on still open) to HANDED_OFF with handed_to the charter accountable owner (or tenant operators, named), handed_off_at and handoff_receipt_hash. A drill halt moves only a duty whose task starts drill:handoff-proof. A hand-off not sealed moves nothing. Does nothing until the 2026-11-04 duty hand-off migration is applied. Writes nothing else"},{"id":"scheduleDaemon:handoff-overdue-sweep","control":"platform_internal","governed":true,"effects":"boot + 660 s, then hourly (off with ICA_HANDOFF_OVERDUE_OFF=1): reads ica_handoffs across tenants for offers still in state offered with no decision past the expiry in their signed statement (created on or after 2026-11-05, ICA_HANDOFF_OVERDUE_SINCE moves it) and opens one handoff_offer_overdue self-audit finding per offer (dedupe key handoff-offer-overdue:<customer>:<handoff id>; an offer that ever had a finding is not raised again), at most 25 per tick; the finding carries ids, party ids, times and the statement digest, never the statement; changes no hand-off row, seals no decision"},{"id":"scheduleDaemon:held-key-outside-use","control":"platform_internal","governed":true,"effects":"boot + 1800 s, then every 6 h (off with ICA_HELD_KEY_OUTSIDE_USE_OFF=1): reads active ica_provider_connections, the newest ica_provider_admin_snapshots and ica_gateway_transactions outcome rows; with each tenant's admin connection reads the provider's daily usage report per held key (Anthropic and OpenAI; GET only); inserts one critical ica_ai_incidents row (failure mode held_key_used_outside_ica, its id derived from the tenant, key and day) and one critical alert per held key and day whose provider usage exceeds what ICA sent with it; a high alert instead when some of ICA's own calls that day could not be counted; one medium alert per held key it cannot match to a provider key id, and per tenant and day whose check cannot run; one medium alert a day per held key that got no verdict on either day (the watch is blind to it), and per tenant and provider whose usage it cannot read (unmeasured); never disables, rotates or writes a key; changes no gate outcome"},{"id":"scheduleDaemon:incident-overdue-sweep","control":"platform_internal","governed":true,"effects":"boot + 210 s, then every 5 min, only when ICA_ARM_INCIDENT_OVERDUE_SWEEP=1: runs routes/incident-lifecycle.js sweepOverdue (the same code as POST /api/cert/incidents/sweep-overdue) under engine:incident-overdue-sweep; reads open ica_ai_incidents past their phase, DFARS 72-hour, VA one-hour, EU AI Act Article 73, GDPR Article 33 or developer safety report or imminent-disclosure clock and seals one incident_timeline receipt plus one ica_operator_events row per breach not already escalated (deduped per phase entry or due time); moves no incident row, notifies no one; a seal or read that did not land fails the tick"},{"id":"scheduleDaemon:monitor-canaries","control":"platform_internal","governed":true,"effects":"boot + 600 s, then daily (off with ICA_MONITOR_CANARIES_OFF=1): runs the built-in bank of planted items through ICA's per-action monitors (engine/monitor-canaries.js runAndSeal) and seals a governance_action receipt; a monitor below standard opens a draft ica_ai_incidents row"},{"id":"scheduleDaemon:opfor-recheck","control":"platform_internal","governed":true,"effects":"boot + 420 s, then hourly (ICA_OPFOR_RECHECK_MIN, 15 to 1440; off with ICA_OPFOR_RECHECK_OFF=1): for every academy-enrolled agent that already has a receipt-backed opfor:battery-v1 exam, and every certified one that has none (the first-exam duty excludes certified agents; it is re-confirmed certified just before sealing), re-runs the static battery plus failure memory against its current posture (engine/opfor/recheck.js; no model call) and compares trials run, passed and critical escapes with the latest sealed exam. An unchanged verdict writes nothing. A changed verdict is sealed as one opfor-engagement artifact receipt and one ica_proving_ground_trials row (engine/opfor/seal.js), at most 50 per pass (ICA_OPFOR_RECHECK_MAX_SEALS), certified agents included unless ICA_OPFOR_RECHECK_CERTIFIED=0. Each agent's posture is read strictly (engine/opfor/posture.js resolveTarget strict): a failed tenant, pack, manifest, skill or tool-grant read, or an engagement not graded against the agent's own layer, seals nothing for that agent; a failed roster, exam or certified-list read seals nothing at all. A changed verdict that is now clean answers that agent's earlier escape in the same battery (lib/proving-ground-tally.js), closing the episode, which can restore its standing; releases no halt"},{"id":"scheduleDaemon:pq-leg-completer","control":"platform_internal","governed":true,"effects":"only when ICA_ARM_PQ_HYBRID=1 and ICA_ARM_PQ_ALL_KINDS=1: boot + 120 s, then every 60 s, reads ica_receipts created at or after ICA_PQ_ALL_KINDS_FROM (at most the last 24 hours, older than one minute) with no inline ML-DSA leg and not already in ica_pq_backfill (at most 1,000 read and 300 signed a tick), signs each over its own canonical bytes with lib/receipt-signer.js pqSign (AWS KMS in production) and inserts one ica_pq_backfill row per receipt (ignore-duplicates); never updates ica_receipts, never signs a body that does not recompute to its hash; logs \"pq-leg MISSING\" and fails the tick when a receipt older than 15 minutes still has no leg (alarm ica-prod-pq-leg-missing). Without a valid ICA_PQ_ALL_KINDS_FROM it fails the tick and reads nothing (engine/pq-leg-completer.js; Shane 2026-11-13)."},{"id":"scheduleDaemon:pq-readiness-rescan","control":"platform_internal","governed":true,"effects":"only when ICA_ARM_PQ_RESCAN=1: boot + 900 s, then daily, for each tenant that already has an ica_pq_readiness_inventories row, re-runs engine/pq-readiness-scan.js runInventory (outbound TLS handshakes, no request sent; operator-named endpoints of the prior inventory carried over), inserts one inventory row and its CycloneDX CBOM, and seals one aibom_document receipt per run, as POST /api/ica/pq-readiness/scan does"},{"id":"scheduleDaemon:receipt-anomaly-scan","control":"platform_internal","governed":true,"effects":"boot + 600 s, then hourly (off with ICA_RECEIPT_ANOMALY_SCAN_OFF=1): for each tenant with a seal chain, runs the same scan as GET /api/ica/anomaly/scan (reads ica_receipts and ica_agent_steps through narrow aggregates) and seals each new finding as one anomaly_recommendation receipt (advisory; a finding already sealed in the last 24 h by this process is not sealed again); a tenant whose receipts cannot be read fails the tick; changes no gate outcome and halts nothing"},{"id":"scheduleDaemon:review-sample-draws","control":"platform_internal","governed":true,"effects":"boot + 480 s, then hourly (off with ICA_REVIEW_DRAWS_OFF=1): for each active ica_review_policies row, every closed period not yet drawn, seeded by the first transparency-log anchor after it closed (engine/review-schedule.js tick); seals a governance_action receipt per draw before writing ica_review_draws. Skips quietly until migration 2026-11-05-standing-review.sql is applied"},{"id":"scheduleDaemon:revocation-drill","control":"platform_internal","governed":true,"effects":"boot + 780 s, then every ICA_REVOCATION_DRILL_INTERVAL_H hours (1 to 168, default 24; off with ICA_REVOCATION_DRILL_OFF=1; nothing runs until ICA_REVOCATION_DRILL_AGENTS names a tenant:agent test agent): per test agent, when the tenant passport gate enforces, issues up to two one-minute ica_authority_passports rows for that agent (each sealing an authority_passport receipt) and revokes every one it issued, makes up to four draft-class Tool Gateway calls (tool ica_drill:draft, sealed tool_gateway_decision receipts, counted in the test agent's breakers, usage windows and suspicion budget like any call; no capability is redeemed), seals one revocation_effect proof (engine/revocation-proof.js) and one governance_action receipt with PASS or FAIL (engine/revocation-drill.js). Grants no authority beyond the test agent, never approves anything for a person"},{"id":"scheduleDaemon:self-audit-beacon","control":"platform_internal","governed":true,"effects":"boot + 20 s, then every 60 s (off with ICA_SELF_AUDIT_BEACON_OFF=1): GETs the latest drand quicknet round and NIST beacon 2.0 pulse, verifies each signature, and holds the epoch in memory; seals one beacon_degraded or beacon_restored receipt when consensus is lost or returns; writes nothing else and changes no gate outcome"},{"id":"scheduleDaemon:self-audit-formal","control":"platform_internal","governed":true,"effects":"boot + 900 s, then daily (off with ICA_SELF_AUDIT_FORMAL_OFF=1): where Java and tla2tools.jar are present, model-checks every entry of the formal suite and seals one formal_verification_run receipt; otherwise reads formal_verification_run receipts; opens one formal_divergence finding per green that fails, red that passes, or running model with no passing run; writes nothing else and changes no gate outcome"},{"id":"scheduleDaemon:self-audit-governance","control":"platform_internal","governed":true,"effects":"boot + 300 s, then hourly (off with ICA_SELF_AUDIT_GOVERNANCE_OFF=1): reads ica_governance_actions, ica_approval_requests, ica_agent_steps, ica_provenance_registry and ica_self_audit_findings; seals one self_audit_finding receipt and writes one open ica_self_audit_findings row per new anomaly (pair concentration, fast approval, approval burst, post-finding spike), per break-glass past its post-incident review, per agent-hour of the last complete hour with a tool_call step and no gateway decision (bypass_detected), and per tenant with an unsettled provenance replacement; writes nothing else and changes no gate outcome"},{"id":"scheduleDaemon:self-audit-temporal","control":"platform_internal","governed":true,"effects":"boot + 360 s, then every 10 min (off with ICA_SELF_AUDIT_TEMPORAL_OFF=1): reads ledger_merkle_root and root_temporal_proof receipts and ica_proof_archive; seals one root_temporal_proof receipt (a Wesolowski VDF over the root, its anchor receipt and beacon epoch) per anchored root not yet proven, at most three a run; inserts the deterministic archive bundle of each anchored root that has none into append-only ica_proof_archive, oldest first, at most 25 a run; writes nothing else and changes no gate outcome"},{"id":"scheduleDaemon:self-audit-windows","control":"platform_internal","governed":true,"effects":"boot + 240 s, then every 6 h (off with ICA_SELF_AUDIT_WINDOWS_OFF=1): reads ica_seal_chain, ica_receipts, ica_proof_archive and the newest worm_archive_sync receipt; seals one self_audit_observation_window receipt per new period package per tenant and one self_audit_tier_completeness receipt per run; writes nothing else and changes no gate outcome"},{"id":"scheduleDaemon:stale-gate-close","control":"platform_internal","governed":true,"effects":"boot + 690 s, then hourly (shadow, no switch): reads every ica_agent_steps row at gate_state pending (id, tenant, session, tool name, time only) and, for the sessions owning a step older than 24 h, whether that session has any step in the last 24 h; logs one [STALE-GATE-CLOSE] count line per tenant and a total; writes no row, seals nothing, approves nothing; a failed or cut-short read fails the tick"},{"id":"scheduleDaemon:subgroup-outcomes-attest","control":"platform_internal","governed":true,"effects":"boot + 420 s, then daily (off with ICA_SUBGROUP_ATTEST_OFF=1): for each tenant with industry-healthcare-provider-ai-2026 HC-AI-PA-001 bound, computes the appeal overturn rate per opaque subgroup from ica_outcome_reconciliations (engine/subgroup-outcomes.js) and, when measured, attests ai_pa_denial_appeals_overturn_rate_window_pct through engine/kpi-evidence-attest.js (the evidence route path): one ica_pack_kpi_evidence row per tenant per UTC day; a miss emits one kpi_breach finding through ica_atc_compliance_emit_finding naming no agent; a met value closes this KPI breaches that name no agent. Never attests an unmeasured figure, refuses a KPI whose miss halts before any write, and never lifts a halt"},{"id":"scheduleDaemon:suspicion-calibrate","control":"platform_internal","governed":true,"effects":"boot + 540 s, then daily (off with ICA_SUSPICION_CALIBRATE_OFF=1): for each ica_suspicion_budgets row, reads the tenant's sealed tool_gateway_decision suspicion scores of the last 7 days and a count of its non-read decisions, sets audit_threshold and defer_threshold for the budget (engine/suspicion-gate.js calibrateTick), sealing a governance_action receipt first. Skips quietly until migration 2026-11-05-suspicion-budgets.sql is applied"},{"id":"setInterval:academyTick","control":"platform_internal","governed":true,"effects":"30 min Academy sweep (write:true): certifies/conditional/insufficient verdicts per agent"},{"id":"setInterval:agentAssessmentTick","control":"platform_internal","governed":true,"effects":"hourly per-agent continuous assessment batch over the last 24h; writes assessment runs"},{"id":"setInterval:certEnforceTick","control":"platform_internal","governed":true,"effects":"6h: advisory cert enforcement (under_review / renewal recommendation); never auto-suspends or revokes"},{"id":"setInterval:constraintSurvivalAttestTick","control":"platform_internal","governed":true,"effects":"hourly (Constraint Survival Attestation): for each open constraint scope observed since its last attestation (at most once per 24 hours per scope), reads that scope's observations and seals one constraint_survival_attestation receipt; updates attested_at on ica_constraint_commitments; decides nothing"},{"id":"setInterval:detectionTuningTick","control":"platform_internal","governed":true,"effects":"6h when due (OMB M-26-14 App. C Collection Operations): trains the detection tuning model on per-finding human dispositions from ica_detection_label_daily, backtests it out of time, writes one ica_detection_tuning_runs row and seals a detection_tuning_run receipt; proposals are recorded, never applied"},{"id":"setInterval:egressLockTick","control":"platform_internal","governed":true,"effects":"5 min, only where ICA_EGRESS_LOCK is set: reads the egress lock's four CloudWatch log groups and its live configuration with the task role (read only), seals one egress_lock_window receipt per closed window (the cursor), one egress_lock_refusal receipt and one ica_atc_signals row (sensor egress-lock) per refused or would-refuse destination, and egress_lock_allowlist / egress_lock_coverage receipts when the verified allowlist or the coverage changes; changes no network setting"},{"id":"setInterval:exceptionExpiryTick","control":"platform_internal","governed":true,"effects":"daily sweep: patches exception rows past their expiry to the expired status (idempotent)"},{"id":"setInterval:iacCertEnforceTick","control":"platform_internal","governed":true,"effects":"6h: IAC cert standing enforcement; seals an iac_cert_standing_change receipt FIRST, never moves without it"},{"id":"setInterval:iacTick","control":"gated","governed":true,"effects":"6h: fleet IAC rescoring per customer (scores, layer telemetry, posture snapshots) + IMPERIUM self-watch steps. Since 2026-10-30 each customer's recompute is decided first by the Tool Gateway under the self-watch agent (ICA_SELF_WATCH_AGENT_HANDLE) and runs only on an allow; with the handle unset it runs ungated as before and says so in the log"},{"id":"setInterval:ledgerAnchorTick","control":"platform_internal","governed":true,"effects":"6h (ICA_ARM_LEDGER_ANCHOR): seals one chained ledger_merkle_root receipt over ica_receipts"},{"id":"setInterval:mcpStandingTick","control":"platform_internal","governed":true,"effects":"6h: re-probes certified MCP servers; can only move register status to under_review"},{"id":"setInterval:networkDetectionTick","control":"platform_internal","governed":true,"effects":"5 min: network detection over the egress lock's logs and other listed VPC flow logs, read only; when armed it seals its alerts; changes no network setting"},{"id":"setInterval:reconcileAndReportTick","control":"platform_internal","governed":true,"effects":"hourly; at 03:00 UTC daily legislation reconciliation + daily report render; at 06:00 UTC the device posture drift check (off unless ICA_ARM_DEVICE_POSTURE_DRIFT=1; writes posture attestations, shadow findings and one sealed run receipt per target)"},{"id":"setInterval:replayTick","control":"platform_internal","governed":true,"effects":"hourly (ICA_ARM_REPLAY_EXECUTOR): drains queued forensic replays; each terminates completed/failed"},{"id":"setInterval:siemForwardTick","control":"platform_internal","governed":true,"effects":"10 min: forwards every baseline log source (per-source inclusive cursors) to each active SIEM destination with a stored secret; writes delivery receipts and forward cursors"},{"id":"setInterval:toolInventoryTick","control":"platform_internal","governed":true,"effects":"daily (OMB M-26-14 inventory): captures tools observed in agent steps into ica_tool_catalog as observed_unattested, never overwriting an attested row; writes the tool_inventory_sync heartbeat"},{"id":"setInterval:tsaAnchorTick","control":"platform_internal","governed":true,"effects":"daily (ICA_ARM_TSA_ANCHOR): RFC 3161 timestamp over the anchored Merkle root; seals a tsa receipt. When ICA_QTSA_URL is set, the same tick also asks that qualified TSA for a token and seals an rfc3161_qualified_timestamp receipt only when the token verifies as qualified against the pinned EU trusted-list snapshot; a refusal seals nothing"},{"id":"setInterval:watcherTick","control":"platform_internal","governed":true,"effects":"4h: legislation watcher pulls sources and creates/updates ica_legislation_watch rows"},{"id":"setInterval:wormArchiveSyncTick","control":"platform_internal","governed":true,"effects":"boot + 180 s, then every 6 h (only when AWS_PROOF_ARCHIVE_BUCKET is set and credentials resolve: AWS_PROOF_ARCHIVE_KEY and _SECRET, or the ECS task role when neither is set): reads ica_proof_archive bundles from an hour before the last synced checkpoint and PUTs each to s3://<bucket>/checkpoints/ with If-None-Match: *, so an archived bundle answers 412 and is never rewritten; seals one worm_archive_sync receipt per run that lands, none for a run that stops; writes nothing else in ICA"},{"id":"worker:reg-analyst-worker.mjs","control":"gated","governed":true,"effects":"external launchd worker: pulls the reg-worker queue over a bearer token, runs claude -p, posts raw proposals back; the server decides everything"}],"consumers":[{"file":"engine/break-glass-orchestrator.js","role":"observer","control":"platform_internal","governed":true},{"file":"engine/effect-relay.js","role":"consumer","control":"platform_internal","governed":true},{"file":"engine/event-projector.js","role":"consumer","control":"platform_internal","governed":true},{"file":"engine/event-spine.js","role":"observer","control":"platform_internal","governed":true},{"file":"engine/failure-semantics.js","role":"both","control":"platform_internal","governed":true},{"file":"engine/horizon-projector.js","role":"observer","control":"platform_internal","governed":true},{"file":"engine/observability-slo.js","role":"observer","control":"platform_internal","governed":true},{"file":"lib/event-outbox.js","role":"emitter","control":"platform_internal","governed":true},{"file":"lib/reliability.js","role":"observer","control":"platform_internal","governed":true},{"file":"lib/transactional-outbox.js","role":"both","control":"platform_internal","governed":true},{"file":"routes/observability-slo.js","role":"observer","control":"platform_internal","governed":true},{"file":"scripts/jurisdiction-shadow-lab-packet.js","role":"observer","control":"platform_internal","governed":true},{"file":"scripts/prove-build-order.js","role":"observer","control":"platform_internal","governed":true}],"inputs_digest":"9072875c2cf9235c0ea0e2086f2d314b74656cfd66979a7cdf5258a5ddb792fc","basis":"static AST scan + live Express router + declared manifest (governance/action-manifest.yaml), reconciled by scripts/enforcement-coverage.js in CI; the floor only ratchets up","coverage_basis":"identity: a unit counts when it passes any observable control (auth, tool gateway, shared secret, credential, front-door policy decision)","coverage_basis_by_kind":{"routes":"a side-effecting HTTP route counts when a control is observable on its request path (auth, tool gateway, shared secret, a named credential verifier the handler calls, or the front-door policy decision the route carries, which always enforces)","jobs":"a background job is discovered from source and counts when governance/action-manifest.yaml declares it platform_internal or gated and states its effects; no request-path control is observed for a job","consumers":"a queue consumer is discovered from source and counts when governance/action-manifest.yaml declares it platform_internal or gated; no request-path control is observed for a consumer"},"policy_enforced":{"available":true,"definition":"side-effecting HTTP routes whose effect passes a policy decision point that can refuse it: the Tool Gateway (capability burn), the UCA Action PEP while armed, or the front-door PEP (always enforced), out of all side-effecting HTTP routes","enforced":1477,"tool_gateway_enforced":9,"front_door_enforced":6,"action_pep_enforced":1462,"total":1490,"ratio":0.9913,"policy_evaluated_shadow":0,"uca_pep_armed":true,"identity_only":13,"identity_only_by_design":[{"unit":"POST /api/auth/login","reason":"refuses every request (403) and issues no session; its only write records the refused attempt. Operator sign-in runs at POST /api/auth/operator/login behind the front-door policy point"},{"unit":"POST /api/auth/logout","reason":"signs the caller out; it ends only the caller's own session, which no policy should be able to refuse"},{"unit":"POST /api/auth/operator/logout","reason":"signs the caller out; it ends only the caller's own session, which no policy should be able to refuse"},{"unit":"POST /api/ica/capture","reason":"takes in sensor evidence for the caller's own organization; ICA's verdict on that activity is made from this evidence"},{"unit":"POST /api/ica/discovery","reason":"the same evidence intake as POST /api/ica/capture, for activity found in the caller's own organization"},{"unit":"POST /api/ica/egress-redeem/observe","reason":"records a refusal that a layer outside ICA has already made; it authorizes nothing and changes no decision"}],"identity_only_by_design_withheld":{"count":7,"reason":"counted here and not named, because naming them on a public page would help an attacker; each is declared with its reason in ICA's governance manifest and checked by the coverage scan"},"identity_only_open":0,"ungoverned":0,"note":"Distinct from coverage.overall, which counts identity controls. Routes behind the UCA Action PEP in shadow are evaluated and recorded but not counted as enforced.","policy_version":"ICA-UCA-PEP-1","policy_basis":{"available":true,"tool_gateway":"the Tool Gateway decides the call and burns a single-use capability","action_pep":{"policy_version":"ICA-UCA-PEP-1","state":"armed","decides_on":["action","actor_kind","channel","property"],"rule_sets":{"command":{"deny":4,"allow":10},"accession":{"deny":2,"allow":3},"brain_sender":{"deny":4,"allow":16},"credential":{"deny":1,"allow":1}},"note":"Armed: any outcome but allow refuses the request, and a decision that cannot be recorded refuses it."},"front_door":{"policy_version":"ICA-FRONT-DOOR-1","state":"always_enforced","actions":["auth.operator.login","auth.operator.mfa","public.readiness.submit","public.asap.survey","auth.saml.acs","public.sales.unlock"],"decides_on":["action","actor_kind","channel","property"],"conditions":["payload_malformed","tenant_scoped","challenge_stale","pair_locked","account_locked","address_rate","account_rate"],"rule_set":{"deny":5,"allow":1},"fail_modes":{"auth.operator.login":"open","auth.operator.mfa":"open","public.readiness.submit":"closed","public.asap.survey":"closed","auth.saml.acs":"open","public.sales.unlock":"closed"},"note":"Always enforced: any outcome but allow refuses the request. When the decision cannot be made or sealed, a sign-in step goes on to its unchanged password and code checks and the questionnaire refuses (503)."}}},"point_of_use":{"available":true,"sensor":"browser (Chrome and Edge, Manifest V3), sensors/browser/","ai_sites":42,"surfaces":["submit","paste","file"],"detector_categories":[{"category":"private_key","severity":"critical"},{"category":"aws_access_key_id","severity":"critical"},{"category":"openai_key","severity":"critical"},{"category":"anthropic_key","severity":"critical"},{"category":"openrouter_key","severity":"critical"},{"category":"xai_key","severity":"critical"},{"category":"groq_key","severity":"critical"},{"category":"huggingface_token","severity":"critical"},{"category":"perplexity_key","severity":"critical"},{"category":"github_token","severity":"critical"},{"category":"slack_token","severity":"critical"},{"category":"google_api_key","severity":"high"},{"category":"jwt","severity":"high"},{"category":"bearer_token","severity":"high"},{"category":"keyed_secret","severity":"high"},{"category":"ssn","severity":"high"},{"category":"credit_card","severity":"high"},{"category":"email","severity":"medium"},{"category":"phone","severity":"low"}],"text_leaves_the_machine":false,"decisions":["allow","coach","redact","block"],"policy_version":"ica-point-of-use-v1","default_posture":"observe","default_failure_mode":"fail_closed","census_version":"ica-ai-tool-census-v2","gaps":[{"id":"desktop_ai_apps","what":"Desktop AI apps on staff laptops (outside a browser) have no point-of-use sensor yet; coding agents are covered by the Claude Code and Copilot hooks (sensors/)."},{"id":"binary_files","what":"Binary files (PDF, images, office documents) and files over 1 MB are not read by the browser sensor; in enforce the failure mode decides them (by default fail_closed: held back)."},{"id":"other_browsers","what":"The browser sensor is a Chrome and Edge (Manifest V3) extension; other browsers have no point-of-use sensor yet, so organizations block unmanaged browsers by policy."},{"id":"in_app_navigation","what":"A page reached by in-app navigation onto an AI path of a site the browser does not load fresh is not screened until a full load."}]}}