{
  "version": 1,
  "scope": "Distilled local observations, not calibrated physics or general painting-quality claims. Raw scenes, images, rejected variants and reports remain local.",
  "studies": {
    "v41": {"topic":"branch structure","renders":14,"result":"Authored branch hierarchy and lighter contacts improved one tree locally; jitter did not repair topology.","limit":"One reference; no tree-growth model or novice transfer proof."},
    "v42": {"topic":"stroke planning","renders":17,"result":"Intermediate observation improved local contacts; the final canopy did not beat the retained v41 painting.","limit":"A local improvement is not a whole-painting improvement."},
    "v43": {"topic":"fresh-agent summer painting","renders":11,"profile":"watercolor-standard-v1","model":"paper-brush-v3","novice":"One no-history agent used the public guide, prewet/reserve/edges/hold and trial API without painting coaching.","result":"The same initial-background marks reordered water-first removed unwanted seams. One detail-stage wait changed from 18 to 90 seconds preserved details. Longer hold was rejected because reserved white shrank.","verification":"Final PNG, canonical JSON and independently replayed pixels agreed.","limit":"One agent and subject; grain, simple masses and muddy overlaps remain. Historical card examples were not newly visually validated. Water-first and 90 seconds are not universal prescriptions."},
    "v44": {"topic":"cottage and cloud reference study","renders":15,"selected":"11-connected-volume-v3","result":"Replanned brush turnarounds preserved cloud white; a faint continuous underwash connected detached shadow masses. Fine added edge strokes and the lifting approach were rejected for the final silhouette.","materialComparison":"Changing only both model/profile tags to v4 improved paper whiteness but blurred shapes. Reducing water in 24 strokes to one quarter did not resolve it. A dry-paper variant produced flat bands.","verification":"Selected PNG and fractional replay pixels/material state agreed.","limit":"Partial likeness only; cloud contour, paper texture and warm paper color differ. Existing brush.lift is not a tissue contact/capacity model; no validated tissue recipe."}
  },
  "unresolved": ["Dedicated tissue contact, finite absorption and wetness timing matrix", "Transfer of successful swatches to large compositions", "Paper color, texture and edge transport evaluated separately", "Repeatable quality across fresh agents and different subjects"],
  "promotion": "Use observations as conditional hypotheses with keep/reject criteria. Retain standard v3 by default; do not change the fixed sheet to hide a brushwork failure."
}