An active set of stimuli provides FinalImageId — the image key in semantics and in the understanding model.
getDiffImportance). Over time it accumulates a semantic set in SemanticObjectImportance: for each triple of conditions, meanImportance and samplesCount are recalculated.
EmotionId from the combination of base contexts) + image (finalImageId). The index SemanticMemoryIndex[well][emotionId][finalImageId] stores a summary for OR and for the understanding model.
| Condition (training) | meanImportance | samplesCount |
|---|---|---|
| Normal, emotion A | −0,3 | 4 |
| Well, emotion A | +0,8 | 2 |
| Poor, emotion B | +1,1 | 7 |
samplesCount ≥ OR_well_known (default threshold 10 in code). On the stimulus buttons you see the remaining amount to the threshold. In Poor, the threshold for starting OR does not block the same way as in Normal/Well; for Food/Water special rules apply.
ImageUnderstandingModelArr[finalImageId] is built: a digest of (wellNormaBad, emotionId) combinations — average significance, number of samples, and the latest z. This is what the operator opens via the i icon on the stimulus button.
samplesCount growssamplesCount) and stabilizing meanImportance across conditions. Additionally: with “quiet” semantics (|mean| ≤ 0.5) and calm vitals, the significance lens triggers — the averaged signal mixes an amplified component from the raw diff so weak differences do not disappear into zero (when vitals are under stress, the lens turns off).