When no signal is not calm: a Bab-el-Mandeb evidence audit
Public MarineAware demonstrator audit
Observed audit values
Challenge
The page displayed a risk score of 41 and the band “calm”. The audit asked what a reader could legitimately infer from the sources actually present in that build.
Approach
The public page was captured on 17 August 2026. The disclosed formula, score inputs, source states, timestamps, imagery limits, and narration type were recorded and the score was recomputed independently.
Outcome
The score reproduced exactly, but all 41 points came from 55 percent of an analyst-set structural prior. No contemporaneous reporting, thermal value, or AIS position was present. Weather and broad imagery supplied context only.
Finding
The captured score was arithmetically correct. It was not an observation that Bab-el-Mandeb was calm.
The page contained no retrieved dated reporting, no thermal count, and no captured AIS position. Its NASA true-colour image was explicitly too coarse to resolve vessels. Weather described mild conditions at the area’s centroid, but weather did not enter the risk formula.
The score was:
round(0.55 × 74 + 0 × 42 + 0 + 0) = 41Every point came from the structural prior. The defensible conclusion was: weather and broad visual context were available, but activity evidence was insufficient to assess contemporary maritime-security risk.
What the empty sources meant
“No AIS positions captured” describes the collection result. It does not show that no vessel was present. An empty sample can reflect receiver coverage, the collection window, rate limits, connection failure, legitimate switch-off, deliberate non-broadcast, or true absence.
No retrieved reporting item means only that the configured retrieval returned no item. It does not establish that no relevant event occurred. The absent thermal field supported no conclusion. The broad true-colour image supplied context, not vessel detection.
Why freshness mattered
The page was ten calendar days older than the capture date although the site described a daily cadence. A build can be internally reproducible and still mislead a reader who assumes current collection. A public page should expose the build time, source-level last success, collection failures, and a clear stale state.
The design change
A more faithful output would separate structural exposure, observed activity,
physical conditions, evidence completeness, and analytic judgement. It would
show insufficient evidence when the sources required for the intended claim
are absent.
The point is institutional as much as technical. Information power includes the ability to preserve evidence, distinguish observation from prior belief, expose collection failure, and stop an interface from converting “unknown” into a decision.