Project 03 · Evidence evaluation

Evidence Utility Timeline

Which observation changed the conclusion — and why?

What the demo shows

Four reproducible scenarios reveal decisive corroboration, a late contradiction, redundant frames, and a missing modality. Each item receives an inspectable utility breakdown and leave-one-out comparison.

Implementation approach

A deterministic baseline combines uncertainty reduction, information gain, novelty, cross-modal corroboration, contradiction resolution, temporal coverage, and provenance quality, then subtracts redundancy, cost, and latency penalties. Weights are heuristic and explicit.

Public origins and related work

The prototype draws on information theory, value of information, Bayesian experimental design, active perception, multisensor fusion, provenance, influence analysis, data valuation, and calibration research. See the public origins note.

Limitations

Utility is not truth, legal weight, calibrated probability, Shapley value, or evidence validity. Scenario-authored weights and recommendations do not establish generalization, forensic validity, or operational reliability.

Provenance statement

All observations, source IDs, assets, annotations, and events are synthetic and independently created. This page contains no private source code or private repository link.

Screenshot gallery

Late synthetic inventory contradiction changes the hypothesis — A late contradiction prevents an overconfident conclusion.

Counterfactual removal of a synthetic sensor event — Removing the sensor increases uncertainty and reduces graph links.

Ranked evidence and next useful sensor recommendation — An inconclusive case recommends a bounded low-cost modality.

Try the static mini-demo

Step through compact synthetic arrivals, remove one item, and inspect the counterfactual—all in the browser.

Open the mini-demo

Compare notes

Working on value of information, evidence attribution, uncertainty, or active acquisition?

Share feedback