Skip to content

Hawaii Signal Gap Analysis (Cromwell's Beach 2026)

This document reviews why the strict pre-incident Cromwell's Beach replay remained low, and what Hawaii-focused signal coverage improvements are most likely to improve operational surveillance awareness.

Scope limits:

  • no scoring-weight changes
  • no attack-probability claims
  • no shark-intent claims
  • no speculative signals treated as historical fact

Case Scope

  • Case: Cromwell's Beach near Diamond Head, Honolulu, Oahu
  • Date: 2026-05-30
  • Pre-incident sensitivity window: approximately 06:15-06:30 HST
  • Replay variants:
  • swimming_across_channel
  • surfing_or_paddling_out
  • Timeline guardrail:
  • post-incident Cromwell's confirmation and later Ala Moana Bowls reports are excluded from pre-incident runs

Why Pre-Incident Stayed Low

The strict pre-incident runs had very limited live/verified early-morning signals. Habitat/activity context was present, but high-value short-latency observational layers were mostly missing. In this configuration, AI1SAD remained in a bounded low surveillance band before incident-time reports were ingested.

Pre-Incident Input Audit

Status keys used below:

  • available
  • unavailable
  • stale
  • not_applicable
  • low_confidence
  • missing_provider
  • future_provider_candidate
Signal Layer Status Notes
Rainfall / runoff available + low_confidence Engine supports weather/rainfall context, but this replay run had no strong, high-confidence local weather observations in-window.
SST / anomaly available + unavailable Adapter exists, but replay input was missing SST/anomaly observations for this pre-incident window.
NOAA/NWS alerts available + unavailable Provider exists for U.S. alerts; no decisive pre-incident alert signal was active in this replay input.
Surf conditions missing_provider No dedicated live surf-state ingestion adapter yet (period/height/energy by spot/time).
Channel / reef-edge habitat available Baseline habitat context (reef/dropoff/channel) supported and used.
Tidal state static_baseline_support Phase 23 adds static/offline tide-window context; no live tide-state ingestion is enabled yet.
Current direction static_baseline_support Phase 23 adds static/offline nearshore current-direction/speed and channel-flow context; no live current feed is enabled yet.
Water clarity / turbidity static_baseline_support Phase 24 adds static/offline Oahu visibility baselines; no live clarity/turbidity ingestion is enabled yet.
Human exposure available + low_confidence Static exposure profiles exist, but no high-resolution time-of-day local occupancy feed in this run.
Early-morning surf/swim overlap available + low_confidence Activity variant is modeled, but overlap intensity by minute/spot is not live-ingested.
Shark sightings available + unavailable Sighting field supported, but strict pre-incident run intentionally had no verified pre-06:30 ingestion.
Baitfish presence available + unavailable + low_confidence Biological adapter exists, but no verified in-window baitfish event signal in this case input.
Turtle migration / nesting context available + low_confidence Supported as bounded biological context when present; not a strong active driver in this replay input.
Biological events (general) available + unavailable Provider exists (static/manual), but no active verified event supplied for this pre-incident window.
Vessel/fishing context available + unavailable Provider exists (static/manual), but no active nearby vessel/fishing signal in this pre-incident input.
Hawaii seasonal species profiles available Regional pack context exists and contributed bounded contextual points only.

Signal-Gap Matrix

Signal Current AI1SAD support Pre-incident availability Confidence impact Surveillance usefulness Recommended next action Provider/dataset candidate Priority
Live sightings ingestion Structured replay/provider fields exist Low (none ingested pre-window) High High Add low-latency verified sightings adapter with strict timestamps/source attribution Honolulu Ocean Safety / lifeguard dispatch-compatible feed (if policy/permitted) high
Surf-line / lifeguard observations Partial manual/context only Low High High Add structured observation ingest for patrol/lifeguard reports Lifeguard ops log ingestion adapter (manual/API bridge) high
Tide state Static/offline baseline adapter Partial baseline only Medium High Add live or pre-fetched tide-phase/height ingestion after static validation NOAA CO-OPS, PacIOOS South Shore Oahu ROMS high
Current direction/speed Static/offline baseline adapter Partial baseline only Medium High Add live or pre-fetched nearshore current context ingestion PacIOOS South Shore Oahu ROMS, PacIOOS Oahu ROMS, PacIOOS Main Hawaiian Islands ROMS, NOAA CO-OPS currents high
Turbidity / water clarity Static/offline baseline adapter Partial baseline only Medium High Add source-timestamped prefetch after static validation NOAA CoastWatch, PacIOOS water quality, Hawaii beach water-quality datasets high
Reef-channel habitat mapping Baseline habitat flags Partial Medium High Increase granularity from boolean habitat to mapped reef-channel zones Static Hawaii habitat map pack high
Hawaii human exposure timing Static profiles Partial Medium High Add time-sliced exposure curves (weekday/weekend/time-of-day) Local beach-use schedule/profile dataset high
Biological prey context Static/manual biology adapter Low Medium Medium Expand verification workflow and freshness decay for prey-context signals Curated biological event registry with expiry medium
SST/anomaly freshness Adapter exists, often offline/mock Low to partial Medium Medium Improve ingest freshness and coverage checks NOAA CoastWatch operational ingestion path medium
NOAA/NWS hazard linkage Provider exists Partial Low to medium Medium Keep as supporting context; improve alert-to-zone relevance mapping NOAA/NWS alerts + localized spatial mapping medium
Vessel/fishing context Static/manual adapter Low Low to medium Medium Expand nearshore effort proxies where policy-safe Public harbor/launch schedules or curated local logs low
Turtle seasonal context Biological context support Partial Low Low to medium Keep bounded, background-only unless stacked Hawaii seasonal ecology calendar low

Strongest Likely Gaps (Ranked)

  1. Live sightings ingestion
  2. Surf-line / lifeguard observation ingestion
  3. Tide/current context: static/offline baseline support added in Phase 23; live PacIOOS/NOAA ingestion remains future work.
  4. Live water clarity / turbidity context beyond static baselines
  5. Reef-channel habitat mapping granularity
  6. Hawaii-specific human exposure timing
  7. Biological prey-context freshness/verification

Do Not Tune Yet

This single incident should not be used to retune scoring weights. The disciplined sequence is:

  1. improve signal coverage and freshness first
  2. run strict timeline-separated replays across a wider Hawaii cohort
  3. evaluate calibration only after broader dataset expansion

Calibration should follow improved data coverage, not precede it.

Future Hawaii Replay Cohort Plan

Target cohort: 10-20 Hawaii replay cases with:

  • mixed fatal and non-fatal incidents
  • mixed islands and coast types
  • mixed activity contexts (swim/surf/dive/fishing/paddle)
  • mixed seasons and environmental states
  • matched quiet-day controls
  • strict hindsight separation in every case

Evaluation outputs for cohort review:

  • surveillance band movement timing
  • confidence deltas by missing-source class
  • false-elevation and missed-elevation patterns
  • provider freshness impact on operational interpretation

A. Hawaii habitat mapping adapter
B. Tide/current adapter: Phase 23 static/offline baseline complete; future work should add source-timestamped ingestion.
C. Water clarity/turbidity adapter
D. Live sightings ingestion adapter
E. Hawaii replay cohort buildout
F. Calibration review (post-cohort only)

Phase 22 Status

Phase 22 adds a static/offline Hawaii habitat mapping adapter and baseline Oahu demo profiles for:

  • Cromwell's Beach / Diamond Head
  • Kaikoo / Hale Mano channel context
  • Waikiki / Ala Moana south-shore context
  • reef-channel demo baseline
  • shallow-reef edge demo baseline
  • sandy-bottom quiet-day baseline

These are historic/static habitat baselines with source-date metadata, not live habitat observations.

Operational Bottom Line

For Cromwell's Beach pre-incident conditions, AI1SAD behaved as a bounded low-confidence, low-surveillance system because key short-latency Hawaii observational layers were unavailable. The highest-value improvement path is better early local signal ingestion and habitat/ocean-state coverage, not weight tuning to one event.