diff --git a/docs/claude/phase1-synthesis-2026-11-17.md b/docs/claude/phase1-synthesis-2026-11-17.md index d61c55026..aca72d75d 100644 --- a/docs/claude/phase1-synthesis-2026-11-17.md +++ b/docs/claude/phase1-synthesis-2026-11-17.md @@ -9,24 +9,28 @@ Literature is marked as read in full by a tool (not page by page by a person); n ## What phase 1 established 1. **Mathematics.** Mid-surface or sheet extraction on real CAD is not reliable in general (MidSurfer: whole-part success 69 % on 213 CAD parts). The robust candidate is *paired-wall sheets with ports, with rib roots and thick bosses kept as raw 3D tets*; spectral or Reeb-graph structure is a supplement. -2. **Probe (synthetic plate, U-gap, T-rib, each meshed twice):** the paired-wall detection is exact and does not depend on the mesh; there are no false connections across gaps and no misclassified roots. Sheet coverage of the true volume is 100 / 80 / 69 %, the rest staying 3D. - - The current certificate only works for axis-aligned voxel geometry, so **0 Greif parts are eligible**. A general-domain certificate would be new engineering whose cost is unknown. +2. **Probe (synthetic plate, U-gap, T-rib; same exact shape and same anchors, two regular meshings each):** + - acceptance, thickness and covered volume are identical between the two meshings; + - thickness error median and P95 are 0, measured only at anchors with an analytic thickness; + - no false connections across gaps, no misclassified roots; + - sheet coverage of the true volume is 100 / 80 / 69.23 %, the rest staying 3D. + - **0 Greif parts were tried**, because the current certificate only supports axis-aligned regular voxel columns; eligibility of the whole library was not assessed part by part. A general-domain certificate would be new engineering whose cost is unknown. (Wording corrected per MoldAstra: not "mesh-independent", not "0 Greif parts eligible".) 3. **Physics.** Hele-Shaw / 2.5D holds in thin sheets and breaks at junctions, thick regions and the end of filling. Physical structure that can be embedded without new solvers: arrival-time causality along propagation paths, eikonal/level-set structure, and mass conservation on a fixed domain. None of these guarantees front accuracy. 4. **Machine learning.** Geometry-aware operators (GINO, Transolver-3 and others) show feasibility at about 400 to 500 geometries, but with random within-family splits and with **no fill-time evidence**. The recommended structure is true local connectivity, coarse global communication, and queries on the original domain. Geometry latents and physics latents should be kept separate. Many "small-data" claims count trajectories, not shapes. 5. **Our own facts.** - The faithful Gre24 GBRT (A) scores 7.08 %; the training-free EIK 9.59 %. - **EIK finds the last-filled location better than A.** - - The learning curve is flat. + - The learning curve keeps improving over the tested range, but slowly: A's nested pools give 7.703 / 7.308 / 7.083 %. D2 cannot decide whether longer training or more samples would help. - The S2 Transolver recipe scored 10.8 % (a different protocol, and the recipe may not have been adequate). - Greif v2 has 42 parts from 40 source keys not seen in v4 or v5: a candidate fresh hold-out set, not yet vetted. ## Proposed phase 2 (at most two pre-registered hypotheses; for Yanqi's choice) - **H1, physics prior plus learning (low cost, high prior):** learn a *correction* to the training-free eikonal map instead of t* from scratch. - - Arm: A's 19 features plus EIK's prediction as input, target t* - EIK. + - Arm (corrected per MoldAstra): EIK's dimensionless prediction *is* the strict volume quantile F, which is already one of A's 19 columns, so adding it as an input adds nothing. What changes is the **target**: F as a fixed skip connection, with the model learning t* - c F from the same 19 features. - Against A, with field L2, last-filled location and Kendall tau all pre-registered as co-primary or secondary. Optionally add a causality penalty along propagation paths. - Question: can one model keep EIK's end-of-fill accuracy and A's field accuracy? - Cost: hours of CPU. -- **H2, geometry-representation route (Yanqi's direction; higher cost):** a geometry-aware neural operator with the structure the ML map recommends, *using the H1 target (correction on EIK) as its physics prior*. +- **H2, geometry-representation route (after H1, if Yanqi chooses) (Yanqi's direction; higher cost):** a geometry-aware neural operator with the structure the ML map recommends, *using the H1 target (correction on EIK) as its physics prior*. - Against A and against H1, with the same metrics. - Question: does a learned geometry representation add information beyond hand-made distance and moment features, at about 400 training parts? - Cost: GPU days; the recipe needs care (S2's lesson).