Research approach
A fused display is not automatically fused evidence
MarineAware keeps source roles, time, collection failure, model output, and analytic judgement separate. The objective is not a more confident interface. It is a conclusion that an independent reader can inspect and challenge.
Evidence-led architecture
Preserve the build, state which sources succeeded, expose their coverage and limitations, reproduce the transformation, and abstain when the question cannot be answered.
The failure to avoid
Do not let missing collection, an analyst-set prior, or fluent model narration become an apparently current observation about the world.
The distinction, side by side
This is a methods comparison, not a vendor comparison or a claim of operational superiority.
| Evidence-led treatment | Unsupported shortcut | |
|---|---|---|
| Unit of analysis | A bounded question, time window, area, and preserved build | A general dashboard state |
| Source status | Retrieved, missing, stale, failed, or unsuitable by source | Configured sources presented as one fused feed |
| AIS absence | No position captured in the stated collection window | No vessel present |
| Imagery | Resolution and task suitability stated before inference | Any available image treated as a detector |
| AI role | Orchestration or labelled narration of computed evidence | Model language treated as observation or corroboration |
| Low evidence | Insufficient evidence or context only | A confident risk band |
| Correction | Original and revised conclusion, reason, date, and source | Silent overwrite |
Questions about the project
Is MarineAware an alternative maritime-intelligence platform?+
No. The public MarineAware project is an independent research demonstrator. It tests methods and failure states using bounded open data; it does not claim persistent coverage, operational deployment, or validated client outcomes.
Why compare source states rather than vendors?+
The research question is whether an institution can defend a conclusion from the evidence actually retrieved. Vendor capability and a specific build’s evidence are different claims.
What makes an output corroborated?+
Two suitably independent source types must support the same bounded observation, with compatible time and location. A model narration of one source is not a second source.
Challenge the evidence architecture.
MarineAware welcomes research critique, methods discussions, seminar invitations, and bounded collaboration proposals.