P12/P13 boundary · Explainable Alerts

Surface attention. Preserve the evidence trail.

Explainable Alerts is the operational review surface for governed Health Intelligence findings. It does not create values, infer diagnoses or bypass the protocol boundary.

Health Intelligence
Fail-closed by design
A finding can only become an operational review state when reviewed evidence, data quality and context requirements are satisfied. Otherwise the surface remains informational, data-quality blocked or research-only.
P35.3 · Crew evidence review (non-diagnostic)

Read-only, membership-scoped evidence. Descriptive history is not an approved personal baseline. This panel never activates clinical or operational alerts.

Checking authenticated mission membership...

Current session

NO ACTIVE EVIDENCE

No alert is fabricated when mission evidence is absent. Findings appear only from real or explicitly synthetic evidence supplied to the governed intelligence engine.

Alert semantics

ATTENTION, NOT DIAGNOSIS

Alerts communicate what changed, why it matched, confidence, uncertainty, missing data and confounders.

Next governed step

PROTOCOL BOUNDARY

Any operational response is handled by the separate Action & Protocol Guidance layer.

Open guidance

Attention states

INFORMATION
Evidence is present, but the reviewed pattern is incomplete or does not justify an operational review state.
REVIEW
A reviewed multi-signal pattern matched with sufficient independent evidence and should be reviewed by the astronaut or authorized human operator.
DATA QUALITY
Poor, missing or conflicting evidence blocks health interpretation. The correct response is to inspect data quality or repeat measurement.
RESEARCH ONLY
The relationship is scientifically interesting but is explicitly prevented from becoming an operational alert.
Explainability contract
Every alert-facing finding preserves the same evidence, source IDs, confidence breakdown, missing roles, missing context, confounders, uncertainty and scientific guardrails produced by Health Intelligence.
Safety rails
No alert is a diagnosis or treatment recommendation.
Research-only findings cannot become operational alerts.
Poor-quality evidence fails closed to a data-quality state.
Confidence, uncertainty, confounders and missing context stay visible.
Medical escalation is separately governed by Action & Protocol Guidance.