---
title: "Agent-readiness probe — live task success rates + circuit-breaker state"
method: GET
path: "/api/v1/health/agent-readiness"
tags: ["Discovery"]
---

# Agent-readiness probe — live task success rates + circuit-breaker state

`GET /api/v1/health/agent-readiness`

## Request

- Base URL: `https://www.apier.no`
- URL: `https://www.apier.no/api/v1/health/agent-readiness`
- Auth: none required

Zero-auth, per-IP rate-limited (Category A — 1000/min soft cap), and NEVER cached (`Cache-Control: no-store`). Call this BEFORE submitting a batch of compliance workflows to decide whether to proceed, throttle, or hand over to a human.

Returns two health signals. (1) `circuit_breaker` — the live Maskinporten circuit-breaker verdict: `CLOSED` = healthy, `OPEN` / `HALF_OPEN` = the upstream is failing or probing recovery, `UNKNOWN` = the breaker read itself failed. `status` is the boolean projection — `ready` only when the breaker is `CLOSED`, otherwise `degraded`. (2) `endpoints` — per-workflow-class live task success rate + p95 latency drawn from the `task_success_log` telemetry view.

Each `endpoints.*` entry carries a `source` flag: `live` when the value came from accumulated telemetry, `fallback` when no rows have been logged yet for that workflow class and a conservative static default (derived from load-test + staging) was substituted. Treat `fallback` rates as indicative, not measured. A `null` entry means the underlying read rejected unexpectedly — degrade gracefully rather than blocking.

This surface is distinct from the un-versioned infrastructure liveness check at `/api/health`: that one answers 'is the process up?', this one answers 'should an agent trust Apier with a workflow right now?'. The response always uses HTTP 200 — a degraded verdict is reported in the body, never via status code, so the probe always answers.

The envelope `_meta.source` is the honesty signal for the rates above: it reads `apier.no telemetry` when at least one endpoint's rate came from measured live telemetry, and the neutral `apier.no` when every endpoint is still serving its static fallback (the low-traffic default today, and the case when both reads reject) — so the envelope never advertises a measurement it doesn't have.

## Response `200`

Readiness payload. Always 200 — a degraded verdict is carried in `data.status`, not the HTTP status.

- object
  - `success` true, required
  - `data` object, required
    - `status` 'ready' | 'degraded', required: `ready` only when the Maskinporten circuit breaker is CLOSED; otherwise `degraded`.
    - `circuit_breaker` 'CLOSED' | 'OPEN' | 'HALF_OPEN' | 'UNKNOWN', required: Live Maskinporten circuit-breaker state. UNKNOWN means the breaker state could not be confirmed — the read errored or fail-opened (e.g. a Supabase outage) — and is treated as degraded, NOT ready.
    - `endpoints` object, required: Per-workflow-class live success rate + p95 latency. A value of null means that read rejected unexpectedly.
      - `compliance_check` object, nullable, required: Live success rate + p95 latency for compliance-evaluation workflows, or null if the read rejected.
        - `success_rate` number, required: Fraction in [0,1] of compliance tasks that succeeded over the trailing 30 days.
        - `p95_latency_ms` number, required: 95th-percentile end-to-end latency in milliseconds.
        - `last_updated_at` string, date-time, nullable, required: Timestamp of the most recent logged task; null on the fallback path.
        - `source` 'live' | 'fallback', required: `live` = from telemetry; `fallback` = static default substituted.
        - `measured` boolean, required: Honesty flag. true ONLY when success_rate and p95_latency_ms came from measured live telemetry (source = "live"). false for the indicative static target served when no telemetry rows exist yet for this workflow class (source = "fallback") — so an agent never reads a hardcoded projection as a real measurement.
      - `altinn_delegation` object, nullable, required: Live success rate + p95 latency for Altinn acting-capacity workflows, or null if the read rejected.
        - `success_rate` number, required: Fraction in [0,1] of Altinn delegation tasks that succeeded over the trailing 30 days.
        - `p95_latency_ms` number, required: 95th-percentile end-to-end latency in milliseconds.
        - `last_updated_at` string, date-time, nullable, required: Timestamp of the most recent logged task; null on the fallback path.
        - `source` 'live' | 'fallback', required: `live` = from telemetry; `fallback` = static default substituted.
        - `measured` boolean, required: Honesty flag. true ONLY when success_rate and p95_latency_ms came from measured live telemetry (source = "live"). false for the indicative static target served when no telemetry rows exist yet for this workflow class (source = "fallback") — so an agent never reads a hardcoded projection as a real measurement.
  - `_meta` object, required: Metadata included in every response for trust and verifiability.
    - `rulebook_version` string, required: Current Rulebook version. Internal rule-version records carry structured origin metadata (`initial_seed` | `human_review` | `lovdata_drift_response` | `audit_finding` | `customer_question` | `legal_correction`) for operator forensics, but that attribution is deliberately NOT exposed on public responses to avoid leaking operational signal.
    - `data_freshness` string, date-time, required: When this data was last fetched (ISO 8601 UTC).
    - `last_verified` string, date-time, required: When this data was last verified against source (ISO 8601 UTC).
    - `source` string, required: Origin of this data. Defaults to `apier.no`. The agent-readiness probe (`GET /api/v1/health/agent-readiness`) is the one surface that may instead emit `apier.no telemetry` — it does so when at least one of its endpoint rates came from measured live telemetry, and falls back to `apier.no` when every rate is a static default; all other endpoints always emit `apier.no`.
    - `data_source` string: Authoritative origin of the underlying data (listed alongside rulebook_version, data_freshness, last_verified, legal_basis for Rulebook-influenced endpoints). Distinct from `source` (the response origin): `data_source` identifies the authoritative data publisher — e.g. 'Brønnøysund Enhetsregisteret + Apier Universal Rulebook'. Present on registry-backed endpoints.
    - `served_from` 'cache' | 'live' | 'static': Cache-aware endpoints only — 'cache' when the response was served from the local copy, 'live' when the gateway just refreshed from the upstream, 'static' when the payload was computed deterministically from pure code with no DB or upstream read (e.g. /v1/public/deadlines). Absent on endpoints that don't carry this semantic.
    - `stale` boolean: Cache-aware endpoints only — true when the upstream was unreachable and a within-limit cached row was served as a fallback. Absent on normal responses.
    - `schema_version` string, required: OpenAPI contract version at the time of the response; matches `info.version`. Always present on every `/v1/*` response so a future schema break can be traced to the exact rows / audit events that predate / postdate it. Agent-facing machine interfaces (v1/capabilities, workflows.json) may override with their own narrower surface version; all other endpoints inherit the manifest-level version.
    - `legal_basis` string: Legal basis under which the response data is reused. Present on Rulebook-influenced endpoints (v1/company/{org}/obligations, /v1/company/{org}/deadlines). Common value: 'NLOD — public registry reuse'.
    - `rule_count` integer: Rule-evaluation count — total rules evaluated for this request. Present on endpoints that invoke the Rulebook evaluator (v1/company/{org}/obligations).
    - `applicable_count` integer: Count of rules whose evaluation_result is 'applicable'. Present on evaluator endpoints.
    - `not_applicable_count` integer: Count of rules whose evaluation_result is 'not_applicable'. Present on evaluator endpoints.
    - `insufficient_data_count` integer: Count of rules whose evaluation_result is 'insufficient_data' (Tier 2 gated). Present on evaluator endpoints. rule_count = applicable_count + not_applicable_count + insufficient_data_count.
    - `cache_age_ms` integer, nullable: Cache layer: milliseconds since the cached row was computed, or null on a cache miss (served_from === 'live'). Agents use it alongside data_freshness as a staleness signal. Present on endpoints wired to the cache (v1/company/{org}/obligations, /deadlines, /summary, /v1/public/obligations, /v1/public/deadlines).
    - `response_time_ms` integer: Speed signal: total server-side processing time in milliseconds, measured via performance.now() at the route boundary. Mirrored on the `X-Apier-Response-Time-Ms` header so agents can key on it without parsing the body. Present on every endpoint.
    - `response_timestamp` string, date-time: RFC 3339 / ISO 8601 UTC timestamp stamped by the response builder at wire-exit. Mirrored on the `provenance_log.response_timestamp` column by a fire-and-forget writer. Distinct from `data_freshness` (when upstream data was fetched) and `last_verified` (when the citation was verified); this is the moment the envelope left the server. Always populated at runtime — deliberately left out of the `required` list to avoid a cascading contract change that would invalidate every pre-existing example envelope in this spec (append-only response contracts). A future revision may promote it into the `required` list alongside a bulk example refresh.
    - `response_hash` string: SHA-256 hash of the canonical-JSON form of the envelope with `_meta` stripped before hashing, prefixed with `sha256:`. Mirrored on `provenance_log.response_hash` so a consumer can independently re-hash the bytes they received and compare against what we persisted (integrity receipt). Stripping `_meta` is load-bearing: volatile fields (response_time_ms, cache_age_ms, served_from) and the hash itself would otherwise feed into the digest and break the cache-hit-bytewise-identical invariant.
    - `source_snapshot_id` string: Optional free-text pointer into an upstream sync snapshot the response derives from — e.g. a Brreg sync's `brreg-<ISO>` snapshot id. Absent when the response is not tied to a specific upstream snapshot (capabilities, tools, static deadlines).
    - `is_sandbox` true: Present and `true` only when the response was generated from sandbox fixture data (currently `ALTINN_MODE=sandbox` on `/api/v1/altinn/list-acting-capacity`). Absent (NOT `false`) on every other response — the absent-vs-false distinction is deliberate so downstream agents branching on `if (response._meta.is_sandbox)` cannot be misled into treating non-sandbox responses as positively-confirmed-not-sandbox. Distinct from the `/v1/sandbox/*` `is_sandbox: true` literal in `SandboxMeta` (which carries the sandbox surface tag); this flag rides through the production `/api/v1/altinn/*` surface to forensically tag a sandbox-mode response that an OEM evaluator received.
    - `cost_nok` number: What this call debited from the caller's prepaid credit balance, in NOK, rendered from the integer-øre ledger (50 øre → 0.5; the ledger itself never leaves whole øre and no float arithmetic touches a balance). Present on the 2xx body of every metered company-data read — the same debit that sets the `X-Credits-Cost-Ore` header, so the two never disagree. `0` when nothing was charged: credit enforcement running dark (the response then also carries `X-Credit-Check: shadow`) or the sandbox surface. Absent on unmetered endpoints and on error envelopes — a refunded 5xx/403/404 claims no cost rather than a wrong one. Lives under `_meta`, which `response_hash` excludes, so the provenance receipt is unaffected; the route's ETag is computed before the credit wrapper stamps it, so a conditional request can still short-circuit to 304. Append-only addition.
    - `wallet_remaining_nok` number, nullable: The prepaid credit balance remaining AFTER this call's debit, in NOK (same rendering as `cost_nok`; mirrors the `X-Credits-Balance-Ore` header). `null` when no balance was read: credit enforcement running dark, or the sandbox surface. Present wherever `cost_nok` is. An agent planning a sequence of metered calls can divide this by the per-call price from GET /api/v1/pricing to know how many reads remain before a 402 INSUFFICIENT_CREDITS. Append-only addition.
    - `spend_cap_remaining_nok` number: What the calling key may still spend under its tighter spend cap (Europe/Oslo calendar day or month) AFTER this call's debit, in NOK (same rendering as `cost_nok`; never below 0). Present on the 2xx body of a metered read only when credit enforcement is live AND the key has a daily or monthly cap (set with PATCH /api/v1/account/keys/{id}); absent otherwise. When it reaches 0 the next metered call on this key returns 402 SPEND_CAP_REACHED until the cap resets. Append-only addition.

## Other responses

- `304`: Not Modified — the readiness payload is unchanged since the ETag supplied in `If-None-Match`. Empty body; `ETag` and `Cache-Control` headers are echoed.
- `429`: Per-minute rate limit reached for this (IP, endpoint) bucket on a zero-auth public endpoint. The `Retry-After` header indicates how many seconds to wait before the window resets. Public endpoints share a single soft-cap of 1000 requests per minute PER IP — there is no tier scaling, because callers aren't authenticated. Every allowed response also carries the standard `X-RateLimit-*` headers. On rare infrastructure-side failures the gateway fails OPEN and passes the request through with `X-RateLimit-Check: skipped`.

---

[API](https://skmtc.dev/apier/apis/apier-no-api.md) · [All operations](https://skmtc.dev/apier/apis/apier-no-api/llms.txt) · [OpenAPI document](https://skmtc.dev/apier/apis/apier-no-api/revisions/9ec19bc87311?raw)