Public API, v1

This document

This document

Public estate topology

The canonical public estate map. Rich manifest components remain authoritative for runtime roles, layers, dependencies, routes, and health metadata. Public, non-archived Atlas repositories missing from the manifest are added as source-only repository components. Explicit exclusions such as simple-proxy are never exposed.

Worker registry

Every Worker in the estate with its self-declared /_meta document, rebuilt hourly by atlas-api-index and reshaped here into a stable public form.

RAG search over the estate corpus

Semantic search across the estate's own documentation (decisions, READMEs, case studies). Each request costs a real embedding on local hardware, so the anonymous limit is ten per minute per IP. Client IPs stop at the edge; the corpus never sees them.

q required, 500 chars max  ·  top_k optional, min 1, max 10, default 5

Estate statistics

Repository totals, nineteen component verdicts, and uptime measured from the ten-minute estate probe. The additive component_details map preserves healthy, degraded, down, and unknown with evidence source, measurement time, detail, and latency; the original boolean components map remains for compatibility. Uptime accrues from first deploy inside a rolling window rather than inventing history.

Per-day probe counters for error budget maths

The raw per-day ok/total counters behind the uptime numbers, one probe pass every ten minutes, pruned past the rolling window. Successful probes also carry round-trip milliseconds (ms_sum, ms_count) so consumers can show a measured average latency. Targets are deliberately not here; they live with the status page, so tuning a target never touches this Worker.

Infra health state

The current verdict from the specular-sentinel pipeline: Ollama reachability, corpus health, the RAG search path end to end, and WSL2 IP drift. Staleness is recomputed at read time; a silent sentinel reads as down, not as its last good report.

Sentinel ingest (internal)

Accepts observation reports from specular-sentinel. Bearer-authenticated; documented for completeness, not for public use.

RAG query statistics

Counts only, by design: aggregate numbers are safe to publish, fragments of visitor queries are not, and client IPs never entered the pipeline at all.

Corpus summary ingest (internal)

Accepts hourly query summaries from atlas-corpus. Bearer-authenticated; documented for completeness, not for public use.

Public assurance evidence index

Latest conformance and chaos evidence summaries, each tied to a producer timestamp, source commit, and content fingerprint.

Latest estate conformance report

Latest estate conformance report

history optional

Latest chaos assurance report

Latest chaos assurance report

history optional

Conformance evidence ingest

Conformance evidence ingest

Chaos evidence ingest

Chaos evidence ingest

Derived reliability results for every measured service

Error budgets, day-granular burn rates, coverage, and explicit states derived deterministically from the /v1/slo counters and the published policy. Services without approved objectives appear in the unmeasured list; missing, stale, malformed, or insufficient evidence is a stated condition, never health. The report is served with an explicit stale marker once past its freshness bound.

One service's derived reliability result

One service's derived reliability result

service_id required, 64 chars max

The active published reliability policy

Targets, windows, and measurement sources from the canonical atlas-infra policy, plus the explicit unmeasured list. Targets are public presentation data already; no credential or evidence payload appears here.

Release baseline for atlas-journey-watch verification

An atlas-journey-watch/release-baseline/v1 document comparing the recent measurement window against the objective window, using average latency because per-day aggregates cannot support percentiles. Serves an honest 503 when evidence is insufficient, stale, or unavailable.

service_id required, 64 chars max

Reliability policy ingest

Reliability policy ingest

SVG status badge

A shields-flat badge reading N/M operational across the estate's nineteen probed components. Sixty second cache; embeds cleanly in GitHub READMEs.