Methodology

What we measure

Marine Sentinel is an intelligence layer over the connected maritime environmental system from catchment to coast to open ocean. Phase 1 focuses on pollution and release intelligence via the Nikoi historical demonstrator within the Singapore Maritime Environmental Intelligence Region (SMEIR).

Causal grammar

SOURCE → ACTIVITY/CONDITION → DISTURBANCE → RELEASE → PATHWAY
  → TRANSPORT → EXPOSURE → RECEPTOR → IMPACT/CONSEQUENCE

Epistemic separation

• WHAT WE OBSERVED
• WHAT WE INFER
• WHAT WE MODEL
• WHAT WE FORECAST
• WHAT WE DO NOT KNOW

Modes REPLAY / LIVE / SCENARIO are never conflated. Data labels are mandatory on every consequential datum.

Confidence

Confidence is multidimensional. Detection, classification, association, and causal attribution are separate scores. Every score exposes a WHY chain: inputs, weights, evidence, assumptions, missing information, contradictions, and model version.

We never present an unexplained black-box risk number as decision authority.

Attribution language

• associated
• potential source
• candidate source
• probable source
• high-confidence attribution
• confirmed by authority

These are not interchangeable. Phase 1 Nikoi vessel language remains at potential / candidate level.

What we never claim

• We never equate detection with attribution.
• We never equate association with causation.
• We never equate ecological connectivity with legal jurisdiction.
• We never label a vessel as a confirmed polluter without authority finding.
• We never use guilty / criminal / illegal polluter language as system status.
• We never hide contradictory source claims.
• We never disguise synthetic or historical data as live observation.

Candidate association (model v0.1.0)

Candidate Association Score ≈
    spatial relationship
  + temporal relationship
  + track compatibility
  + trajectory relationship
  + plume / transport compatibility
  + source characteristics
  + corroborating evidence
  − contradictory evidence

Formula is inspectable and versioned. Geographic proximity alone is never treated as causation.