05 · EXPLAINABILITY & AUDIT TRAIL
Why the system recommended what it did
Every contributing factor, every source, every timestamp. Cryptographically signed and immutable.
X-01
RECOMMENDATION LOGIC
- +18Vessel course deviationANOMALY SCOREDetected via AIS + radar fusion.
- +12AIS interruption (4 min)UNCERTAINTYPossible spoofing or equipment failure.
- +20UAV proximity (7 km)RISK IMPACTOutside notified flight corridors.
- −15 minWeather degradationRESPONSE WINDOWReduces COA-2 viability after 38 min.
- HOLDNo hostile confirmationHUMAN VERIFICATION REQUIREDEngagement decisions reserved for human commanders.
HUMAN-IN-THE-LOOP · SentinelOps does not recommend or authorise use-of-force actions. It supports human-led assessment, planning, and escalation workflows only — all consequential decisions require explicit human commander approval and two-person confirmation.
M-01
MODEL CARDS
- sentinel-anom-3Anomaly detectionv3.1.7Performance · 0.84 F1
- sentinel-coa-2COA generationv2.4.0Performance · 0.79 BLEU-eval
- sentinel-llm-7bBrief synthesisv7.0Performance · 0.81 conf
L-01
IMMUTABLE AUDIT LOG
14:32:04INGESTAIS · Course delta detected (MMSI 273-XXXXXX)via ais-feed-01
14:32:07FUSIONCross-checked with radar track R-2241via fusion-engine
14:32:09ANOMALYA-118 raised · score 0.87via model:sentinel-anom-3
14:36:01INGESTAIS dropout (4 min) — uncertainty +12via ais-feed-01
14:41:18INGESTUAV detection 7 km NE perimetervia radar-grid-04
14:41:22ANOMALYA-119 raised · score 0.81via model:sentinel-anom-3
14:51:00OPSBRAVO-2 availability confirmedvia ops-net
14:58:11BRIEFSituation brief generated · conf 0.81via model:sentinel-llm-7b
14:58:11COAThree COAs proposed · COA-2 recommendedvia model:sentinel-coa-2
14:58:14HUMANAwaiting commander approvalvia user:lindqvist.a
SHA-256 · 9f2c…b41aCHAIN VALID ✓