Decision record: naive-ordering result (Mold wording) and S3-B(a) well-posed definition
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
1 parent
0fca206ba9
commit
cc102bad23
1 file changed
+19
@@ -31,3 +31,22 @@ F. **O1** with a matched generic partition and a control cut, unchanged as the n
|
||||
|
||||
## 4. Order
|
||||
S3-A runs as approved. In parallel, the route-A falsification tests (O-vol oracle and reorder test) are cheap post-hoc CPU work on existing artifacts and are proposed next, together with the route-B distance definition, which must be fixed before any run. Route D needs Yanqi's input and should be asked now, because it decides what "useful" means for every later comparison.
|
||||
|
||||
## 5. Naive-ordering result (MeltMind 9b7121a4; exploration after S2 results were known). Wording adopted from Mold's review of the fixed report and raw JSON
|
||||
- **Given the true full-domain order, the linear volume map has small mean time error in this batch.** All rows use the same 617 parts / 209 components, scored at the original S2 query points with components weighted equally; only the domain used for the quantiles differs.
|
||||
- Full tet domain (d = 0 included), true order, training clock: 1.4782 %.
|
||||
- The same with the test part's true tet end time instead of the clock: 1.7611 %. The true end time constrains the last time, not the field L2, so this being slightly worse is not a contradiction.
|
||||
- Quantiles taken only on the eligible PPS domain, true order, training clock: 2.2723 %.
|
||||
- The 15 clocks were frozen on learning parts only. The true test order cannot be deployed.
|
||||
- **R2's ordering share cannot be obtained by subtracting 1.4782 % from 7.2993 %.** The two rows change the order source, the domain and the time scaling at once, and field errors have cross terms. 1.48 % is the relative L2 of query arrival times; the ~1 % solver step is a volume-fraction increment, so the two cannot be compared as a resolution. The oracle is a measured reference for one fixed construction, not a theoretical floor and not evidence of a learned order.
|
||||
- **The deployable conditional remap does not beat the original R2:** R2 remapped 7.3060 %, relative change -0.092 %, descriptive paired 95 % interval [-1.445 %, +1.239 %]. This post-processing has no net benefit. N/E/B0 remapped reach 9.6872/9.6893/8.7825 %, better than their own original time errors but still worse than R2. All 14,683 remapped rows keep tau and the last-filled location unchanged.
|
||||
- **Distance weighting:** D_w^1 gives 9.1370 % and D_w^2 gives 20.2380 % with this graph and clock. That supports keeping the first power as the reference here. Continuous eikonal, other conductances and real branch mechanisms remain open.
|
||||
- **LOG (a separate 624-part basis):** 411/624 within the fixed 1-percentage-point tolerance, maximum deviation about 34.01 pp; the flow-rate column of 611 parts has too little display precision to use. Label any side-by-side display as LOG volume deviation versus field time error over the 617 parts. A per-part join can show association, but start points, end points and volume definitions must be aligned before it explains a mechanism.
|
||||
|
||||
## 6. S3-B(a) ordering prior: well-posed definition (Mold; discussion only, not launched)
|
||||
- Single gate with a fixed Q > 0; no flux through solid walls; one gas reference pressure at the boundary between the wetted region and the unfilled capacity (this assumes venting); start from a finite wetted seed at the gate.
|
||||
- Solve the wetted-region pressure with symmetric positive edge conductances, q_ij = g_ij (p_i - p_j). Every wetted connected component that receives injection must reach a fixed-pressure front; otherwise positive net injection is incompatible with no-flux boundaries everywhere.
|
||||
- Each front control volume has a remaining capacity r_i, with inflows from several faces added. Freeze the fluxes for the event, take dt = min(r_i/q_i) over q_i > 0 only, update synchronously, merge simultaneous events, and stop when capacity is exhausted.
|
||||
- If positive unfilled volume remains with no reachable front, record it as blocked. Do not drop the d = 0 mass and do not declare the part filled.
|
||||
- This yields a conservative graph-ordering prior. To call it a Hele-Shaw discretisation, the edge conductance must state how it includes face scale, length, thickness and viscosity, and what it means under remeshing. A truly closed cavity with trapped air needs a gas-pressure or venting model.
|
||||
- With one gate, linear conductance and a uniform front pressure, scaling Q only changes the clock. The new, testable information is how flow splits between branches, and the resulting order.
|
||||
Reference in new issue
Block a user