Skip to content
NO ACCOUNT · REFERENCE SPACECRAFT · AUTHORITY LEVEL A0

Try the Interlock

A root cause is easy to show. The harder question is what happens when the evidence for it is missing. Here is the same anomaly investigated twice on JANTAR-REF-1, the synthetic reference spacecraft: once with the failure mode's documentation in the knowledge graph, and once with it removed. The detector, the reasoning and the telemetry are identical. Watch what changes — and what refuses to.

THE ANOMALY

FM-EPS-002: Solar Array String Interconnection Loss

A seeded deviation on the EPS subsystem of JANTAR-REF-1 (synthetic reference spacecraft), on the channels solar_array_current. The real detector forms the event; the real investigation graph runs it; the citation checker judges the result.

RUN ONE · FULL KNOWLEDGE GRAPH

Investigated with its documentation

Verdict: FM-EPS-002: Solar Array String Interconnection Loss

The deviation on solar_array_current matches the documented signature of FM-EPS-002 (Solar Array String Interconnection Loss): solar_array_current drops by a quantized ~33% step during a sunlit period. Observed telemetry confirms the signature: solar_array_current min 5.544, max 7.7, mean 6.622 A over the window. The recommended mitigation is SOP-EPS-102.

Cited causeHeld for a human — nothing releasedCitation integrity holdsNo undocumented cause claimed
Hypotheses weighed
  • FM-EPS-002 — supported, support 1
Citations on the report
  • eps_manual:solar-arch#2.1
  • kg:FailureMode/fm-string-loss
Drafted recovery procedure
  1. Execute SOP-EPS-102 (Solar Array String Reconfiguration) as the primary corrective action for FM-EPS-002. cited
  2. Monitor solar_array_current until values return within limits and confirm recovery. cited
  3. If the condition persists, escalate to the on-call systems engineer for a contingency safe-mode decision. ungrounded — blocks release
Held for approval on JANTAR-REF-1 (synthetic reference spacecraft). Release is blocked while a step is ungrounded; nobody approves anything here.
RUN TWO · GROUNDING REMOVED

Investigated without it

Verdict: INSUFFICIENT: no failure mode confirmed against telemetry

Available evidence did not confirm a single failure mode; escalate for manual review.

Declined — no documented cause fitsNothing stagedCitation integrity holdsNo undocumented cause claimed
Citations on the reportnone — a refusal cites nothing
Drafted recovery procedurenothing staged
The failure mode FM-EPS-002 and its 1 documentation section(s) were removed from the knowledge graph. The anomaly, the detector, the reasoning model and the telemetry are identical across the pair.

From this release's pair, commit 2bc4474, read from the same API image production serves. Reasoning: deterministic, mode airgap. Synthetic throughout: the craft is the reference spacecraft, the telemetry is a seeded synthetic deviation, and nothing here describes a real mission. Advisory only: a grounded draft is held for a human; nothing is approved, released or transmitted. The moat benchmark runs this pair for every documented failure mode and publishes the rates on the scorecard.

Production is scaled to zero when idle; this click wakes it, which can take a minute.
WHAT THIS PAIR IS ONE CASE OF

The moat benchmark runs this for every documented failure mode

On a component with a documented sibling the second run degrades to that sibling instead of declining — honest either way, because the removed mode is never claimed. The rates below are that benchmark's, read from the scorecard, and the section they come from is on the home page.

1.000grounded honesty rate
1.000grounding integrity rate
1.000ablation effect rate

Measured on synthetic · JANTAR-REF-1. mode=airgap · model=deterministic · grounded_faults=16 · moat_holds=True. Source: make bench-moat. Measured at commit 2bc4474. From this release’s scorecard. Published at /api/system/scorecard.

THE WHOLE CONSOLE, SAME SYNTHETIC CRAFT

Walk the operator console without an account

Fleet health, casefiles with their citations, the approval queue and the ground and risk views, all on JANTAR-REF-1's seeded replay. It runs in your browser on bundled data: it calls no Jantar service, and nothing you approve or file leaves the tab.