instruments / wink / EXP3-senses/predictions.json ยท raw
{
"schema_version": "v0",
"predictions": {
"P1": "vision recovers containment (parts within whole) but not sibling order; partial structure (WL vs truth 0.3-0.7)",
"P2": "nameless prose parses structure at least as well as named prose (EXP1-P3 lineage: the name imports priors)",
"P3": "vision channel lands spectrally closer to the deterministic graph than either language channel (pixels sit nearer geometry than words do)"
},
"frozen_at": "2026-08-21T13:21:13-0700"
}