Absorb S2 v0.3 rulings, verify pinned licenses, and maintain scope-specific entry status

This commit is contained in:
Frank14f committed 2026-11-14 17:36:38 +08:00
1 parent 33b18e0831
commit 3d3c2bd4e8
10 files changed
+459 -24

No files matched your search

@@ -0,0 +1,54 @@
# S2 design v0.3: rulings on remaining gaps G2–G5 (Claude, 2026-11-14)
Integrates MeltAstra's diff at 33b18e08, §2. G1 (grouping calibration) stays with MoldAstra and is frozen by me later. G6 is a status table kept by MeltAstra.
No execution of any kind (preprocessing measurement, pilot, fits or training) without Yanqi's explicit approval for that scope. The entry list is an order, not an authorisation.
## Clarifications (MoldAstra, accepted)
- **Bootstrap:** each replicate resamples components only. The three matched seed runs of each component travel with it, and seeds are not resampled. Both arms are recomputed on the same indices, giving Δ* = 1 − Ē_E*/Ē_N*. The interval is conditional on the fixed folds and the trained runs; it does not cover retraining variance.
- **Reused S1 draws:** S2 eligibility is re-applied from the raw IDs and counts, never from S1's already-filtered sets.
## G2: sampling and inputs
- **Draw:** the frozen implementation of commit 94011fd. prefix = integer numeric prefix. Candidate order = ascending original tet-node ID.
- **Eligibility:**
- valid spectrum = O2-rep v1.1a status valid;
- edge block complete = every incident true-edge end evaluated, at least one channel-valid finite end, and finite summaries. This is the definition used by driver 5e4b406.
- **Target transform:** enabled.
- τ = draw-count-weighted median of T over the eligible draws of the learning pool (per fold and pool).
- y = log(1 + T/τ); predictions are inverted before scoring.
- **Graph distance:** ℓ_g is Dijkstra on tet edges from the node nearest the YML gate; ties go to the smaller ID.
- All 624 tet subdomains are single components (audit 2026-11-13), so "unreachable" should not occur.
- If it does, or if any encoder input is non-finite, the part fails closed with a reason. No imputation.
## G3: code and training
- **Code:**
- primary: thuml/Transolver @ 75e0f67643806a81cd1d3f6adc88dd8c02416fe7 (MIT, LICENSE sha256 2c919cd0…);
- backup: neuraloperator @ a6648eeeb60fe0837f2d7d021517a945a159a0c8 (MIT).
- Any modifications are recorded as a diff and never called a reproduction.
- **Configuration:**
- 4 layers, 8 heads, 32 slices, width ∈ {128, 256};
- query decoder MLP with 2 hidden layers of the same width;
- N decoder input = [embedding, g12, g23]; E adds the 4 edge q values. Parameter counts are reported.
- **Optimisation:**
- AdamW, lr 1e-3, cosine schedule, weight decay 1e-5, batch = 1 part;
- loss = draw-count-weighted MSE in y, normalised so that parts within a component and components overall are equally weighted;
- at most 200 epochs, early stopping on inner validation with patience 20.
- **Capacity choice:** lowest inner-validation component-equal error; a tie goes to the smaller width. The chosen width is applied to both N and E.
## G4: R2 and auxiliary statistics
- **R2:** sklearn HistGradientBoostingRegressor, version pinned at implementation, random_state=0, early_stopping=False (its internal split is not group-aware).
- Grid (12 configurations): learning_rate {0.05, 0.1} x max_leaf_nodes {31, 63} x max_iter {200, 500, 1000}; min_samples_leaf = 20.
- Selected on inner validation.
- **Neural seeds:** {0, 1, 2}, each applied to the torch, numpy and data-order RNGs.
- **Kendall tau-b:** 5,000 random pairs of distinct eligible IDs per part, seed SeedSequence([2026111410, prefix]). Pairs tied in the truth are excluded.
- **Last-filled location:** among eligible query nodes, the distance between argmax T̂ and argmax T, divided by R_g. Labelled as being within query coverage only.
- **NA handling:** a paired contrast uses components where both arms are valid. The excluded count is reported.
## G5: resources (upper limits; still require approval)
- **10-part measurement:** tertiles by node count, 3 parts per tertile drawn with seed 2026111411, plus the corpus's largest part (00000666).
- Per part, the whole preprocessing pipeline: 30 min wall time, 1.25 GiB RSS across the whole process tree.
- Run sequentially; total ≤ 3 h; disk ≤ 5 GiB.
- A failure is recorded and the part is not swapped.
- **Pilot runs:** per run, wall time ≤ 30 min, VRAM ≤ 15 GiB, host RAM ≤ 16 GiB; 0 retries.
- The overall cap stays at 4 runs and 2 GPU-h.
- Test fold sealed.
@@ -5,7 +5,8 @@
"d8aa0f420647701709e03629326dce8339db739a",
"fd1db63ee1c4fb4415a36ed95a2eedfb73a5c85c",
"4e435a03d3e1023bb2c12154ab4afc98b37da741",
"72deb2664411d0256895108b7ab108e2562ebf0b"
"72deb2664411d0256895108b7ab108e2562ebf0b",
"69c58a77006325319903f7862054d461b00ca75c"
],
"files": [
{
@@ -15,13 +16,13 @@
},
{
"path": "docs/collaboration-stage-review-policy.md",
"bytes": 2321,
"sha256": "7757dc15066cda6e74a3b6aed559c6cf42a3a6e982e711456e63d71be41522ce"
"bytes": 2345,
"sha256": "7d33251cded22384e675b106f6e617dcc2a570b3ec543f61aea744accf591dd7"
},
{
"path": "docs/s2-execution-diff-v0-2026-11-14.md",
"bytes": 12048,
"sha256": "91219858ffe7fedb71e371e9543dd85914b3a87adaf53702a89a703fc7d0aae6"
"bytes": 18160,
"sha256": "9f744bc93b473b944ce0447949d234f86c2505d949266dc93e4a6262bf13af45"
},
{
"path": "docs/sharing-authorization-2026-11-14.md",
@@ -73,11 +74,21 @@
"bytes": 6892,
"sha256": "bd59621b1507a12e24bf3221cb4481611ec1695c69d7f79c071996d3e175c0bb"
},
{
"path": "references/2026-11-14/s2-design/claude-s2-v03-69c58a77.md",
"bytes": 4307,
"sha256": "f8b60afae6355ec429a08ccc06698c3d8b3cb73cb3058a851beb816746523d2f"
},
{
"path": "references/2026-11-14/s2-design/claude-stage-review-3ae89805.md",
"bytes": 4592,
"sha256": "e7b7221ba5e5419a7ce3ee59a0d6f383210668ac39ddb74abf97362a1dc9e0da"
},
{
"path": "references/2026-11-14/s2-design/frozen-draw-contract-94011fd.json",
"bytes": 2511,
"sha256": "c3631c5a1a48707014f820aeead23fd950cf6b3601dad8073d5f5c30d7eba495"
},
{
"path": "references/2026-11-14/s2-design/moldastra-note13.md",
"bytes": 4513,
@@ -93,16 +104,31 @@
"bytes": 4391,
"sha256": "60821afaf277dc339c1018223768a6ec224e4769ffc8d12ee3de5101b240dfc7"
},
{
"path": "references/2026-11-14/s2-design/neuraloperator-LICENSE-v03.txt",
"bytes": 1082,
"sha256": "5f335b0603acfb1d180bf3117492e2076562075cc9b6353a3b94d5d5bc03ef33"
},
{
"path": "references/2026-11-14/s2-design/s1-review-and-topology-steering.txt",
"bytes": 4665,
"sha256": "f0e33c2def861a576e5168471344b4ebf37623f2be2c6af74def6d13a40e422c"
},
{
"path": "references/2026-11-14/s2-design/sampled-driver-5e4b406.py",
"bytes": 11228,
"sha256": "66ab07068c6553a5d49d5124087d69c338e057919e2fe0c31124ba6d2d2a5d36"
},
{
"path": "references/2026-11-14/s2-design/stage-review-user-rule-and-target-correction.txt",
"bytes": 6902,
"sha256": "21663440a680896afd2f7ea4b3e77781ce49c486e801d62c8c8931f169d7f856"
},
{
"path": "references/2026-11-14/s2-design/transolver-LICENSE-v03.txt",
"bytes": 1084,
"sha256": "2c919cd03fa823bf7eefc00a957ff8324c865cd22aa5285e563dc4b558084f25"
},
{
"path": "references/2026-11-14/s2-design/v01-grouping-and-stop-corrections.txt",
"bytes": 10822,
@@ -112,7 +138,17 @@
"path": "references/2026-11-14/s2-design/v02-main-review-receipt.txt",
"bytes": 3699,
"sha256": "75de33048a55ab53967d27c800374e6bcba3eb67480eccef16a8126c1d176b63"
},
{
"path": "references/2026-11-14/s2-design/v03-main-review-receipt.txt",
"bytes": 3545,
"sha256": "76dcedcd6cec75074f3cead8665cacedcceeff801808936a829dd70df2393718"
},
{
"path": "references/2026-11-14/s2-design/v03-upstream-license-readback.json",
"bytes": 3057,
"sha256": "b5adf4a29f800203aeb4ed4bf223b37f6fcf2a6d930106f5f7fb65cdea1a1372"
}
],
"preceding_execution_diff": "c35b02d9e3b6814447d00c9b8950e29237dbb127"
"preceding_execution_diff": "33b18e0831ecbc86586f6d0c5d512dbb7adfd4b9"
}
@@ -0,0 +1,48 @@
{
"version": "M2 v1.2 sampled",
"frozen_before_first_draw_utc": "2026-11-14T06:33:55.731281+00:00",
"precheck_commit": "81f56afaca38304a4b5f48921927f97669ebbac9",
"panel_sha256": "c594110b017ad8c62a542e4f634ce2879ccd1c30bd818429e73c0910a57a1e69",
"count_contract_sha256": "21d11a0d975735ffb4cba355aea8abe0fdfe4f3ef7d7f586444c6ad8113be342",
"code_sha256": "7d58fcea9c407f7a9a42ce592d0cedffbb34772a3f3d1930c6b03f3dcd730a81",
"code_path": "tools/draw_s1_sampled.py",
"document_sha256": {
"docs/s1-m2-v1.2-draw-current-source-2026-11-14.md": "512b42e6d103c7645397bce184d39a586dfc29703393f3061849179e8ad824a5",
"docs/s1-m2-v1.2-draw-go-2026-11-14.txt": "704b4919d65934c179bbf0e1bf943112197ad50c90a0e919685ddbbb103a07e4"
},
"numpy": "1.26.4",
"python": "3.10.13",
"candidates": "All original-pose nodes with d>0, ascending original tetra-submesh ID; raw VTU IDs strictly ascending in same order",
"distribution": "iid with replacement, p=mass[candidates]/np.sum(mass[candidates], dtype=float64)",
"algorithm": "Generator(PCG64(SeedSequence([2026111400, numeric_prefix]))).choice(candidate_ids, size=2000, replace=True, p=probabilities)",
"draw_count": 2000,
"primary_seed": 2026111400,
"permutation_seeds": [
2026111401,
2026111402,
2026111403,
2026111404,
2026111405
],
"bootstrap_seed": 2026111406,
"permutations_executed": false,
"bootstrap_executed": false,
"identity": "panel_index, case, source_key, numeric_prefix preserved; one draw per part reused across all folds",
"fold_roles": null,
"fold_status": "Actual fold assignment not specified in current document; must freeze before scoring. Source isolation mandatory.",
"exact_visits": "Python integer sum of saved neighbor_counts[i] over ALL distinct drawn IDs, once per ID, before O2 eligibility",
"common_eligible_draws": null,
"n_ref": null,
"eligibility_status": "Unknown until O2; no prefilter except d>0",
"supersedes": "Only historical v1.1 reference/query/rank/shortfall clauses; original panel and count files unchanged",
"full_source_field": "All four-point source positions/masses and original nodal-gradient tensors, full-part R/t_part/R_g retained for future O2",
"future_per_field_wall_seconds": 1800,
"future_per_field_rss_bytes": 1342177280,
"timing_scenario_seconds_per_visit": 7.3e-07,
"timing_scenario_is_bound": false,
"visit_alarm": 500000000,
"visit_hard_cap": null,
"O2_executed": false,
"S1_scoring": false,
"training": false
}
@@ -0,0 +1,21 @@
MIT License
Copyright (c) 2023 NeuralOperator developers
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
@@ -0,0 +1,157 @@
"""Frozen E6/O2 routines on a fixed subset of anchors; no S1 scoring."""
import argparse
import hashlib
import json
from pathlib import Path
import resource
import time
import numpy as np
import run_e6_real_kernel as e6
import o2_rep as rep
def sha(p):return hashlib.sha256(Path(p).read_bytes()).hexdigest()
def read(p):return json.loads(Path(p).read_text())
def save(p,v):Path(p).write_text(json.dumps(v,indent=2,allow_nan=False)+'\n')
def bit_comparison(name, actual, expected):
a=np.asarray(actual);b=np.asarray(expected)
same=a.shape==b.shape and a.dtype==b.dtype and a.tobytes()==b.tobytes()
row=dict(array=name,shape=list(a.shape),dtype=str(a.dtype),bitwise_equal=same)
if not same and a.shape==b.shape and a.dtype==b.dtype:
av=np.ascontiguousarray(a).reshape(-1).view(np.dtype((np.void,a.dtype.itemsize)))
bv=np.ascontiguousarray(b).reshape(-1).view(np.dtype((np.void,b.dtype.itemsize)))
bad=np.flatnonzero(av!=bv);row['different_elements']=len(bad)
row['first_flat_indices']=bad[:20].tolist()
if np.issubdtype(a.dtype,np.number):
finite=np.isfinite(a)&np.isfinite(b)
row['max_abs_finite_difference']=float(np.max(np.abs(a[finite]-b[finite]),initial=0))
return row
def run(out,phase,index):
c=read(out/'execution_contract.json');s=c[phase][index];case=s['case'];pose=s['pose']
dest=out/phase/s['job'];dest.mkdir(exist_ok=False);start=time.monotonic()
save(dest/'started.json',dict(contract_sha256=sha(out/'execution_contract.json'),case=case,pose=pose,phase=phase))
for p,h in c['code_sha256'].items():assert sha(p)==h,p
for p,h in s['input_sha256'].items():assert sha(p)==h,p
if phase=='panel':
gate=read(out/'equivalence_gate.json');assert gate['status']=='PASS'
for job,h in gate['result_sha256'].items():assert sha(out/'equivalence'/job/'result.json')==h
with np.load(s['geometry'],allow_pickle=False) as z:
x=z['points_raw'];tet=z['tetra'];mass=z['node_volume_raw'];edges=z['edges'];gate=z['gate_location_raw']
raw_ids=z['raw_node_ids'] if 'raw_node_ids' in z else np.arange(len(x))
d=np.load(s['distance'],allow_pickle=False)
with np.load(s['anchor_file'],allow_pickle=False) as z:
ids=z['unique_ids'];multiplicity=z['multiplicity']
assert np.all(np.diff(ids)>0) and len(multiplicity)==len(ids)
# Original full-part scales, before eligibility; carry L_med to rotated pose.
lengths=np.linalg.norm(x[edges[:,1]]-x[edges[:,0]],axis=1)
positive=lengths[np.isfinite(lengths)&(lengths>0)]
L=float(np.median(positive)) if len(positive) else None
if 'L_med_raw' in s:assert L==s['L_med_raw']
center=np.sum(mass[:,None]*x,axis=0)/mass.sum()
Rg=float(np.sqrt(np.sum(mass*np.sum((x-center)**2,axis=1))/mass.sum()))
if not np.isfinite(Rg) or Rg<=0:raise ValueError('Invalid full-part R_g')
if pose=='rotated':
Q=np.asarray(c['rotation']);shift=np.asarray(c['translation']);x=x@Q.T+shift;gate=gate@Q.T+shift
load_done=time.monotonic()
progress=dict(stage='raw_field',anchors_done=0,anchors_total=len(ids),neighbor_visits=0)
save(dest/'progress.json',progress)
raw,raw_meta=e6.raw_field(x,tet,d);raw_done=time.monotonic()
sx,sm,st=e6.samples(x,tet,mass,'tetra_four_point',raw);sources_done=time.monotonic()
source_count=len(sx);real_tree=e6.cKDTree;progress['stage']='kernel'
class ProgressTree:
def __init__(self,*args,**kwargs):self.tree=real_tree(*args,**kwargs)
def query_ball_point(self,*args,**kwargs):
result=self.tree.query_ball_point(*args,**kwargs)
progress['anchors_done']+=1;progress['neighbor_visits']+=len(result)
if progress['anchors_done']%32==0:
save(dest/'progress.json',dict(progress,elapsed_seconds=time.monotonic()-start))
return result
e6.cKDTree=ProgressTree
try:a=e6.integrate(x[ids],sx,sm,st,s['radius'])
finally:e6.cKDTree=real_tree
del sx,sm,st,raw
kernel_done=time.monotonic()
if phase=='panel':assert progress['neighbor_visits']==s['exact_unique_visits']
r=rep.readout(a['tensor'],a['kernel_mass'],d[ids]>0,a['effective_samples'])
checks=rep.acceptance(r);limit=c['frozen_algebra_tolerance']
if checks['valid_count']:
assert checks['mu_sum_error']<=limit and checks['mu_min']>=-limit and checks['ordering_violation']<=limit
assert checks['B_symmetry_error']<=limit and checks['B_min_eigenvalue']>=-limit and checks['reconstruction_max']<=limit
gq=rep.contract_direction(r['B'],r['mu'],r['valid'],gate-x[ids],L,'gate')
readout_done=time.monotonic()
lookup=np.full(len(x),-1,np.int64);lookup[ids]=np.arange(len(ids))
edge_idx,side=np.nonzero(lookup[edges]>=0);anchor_local=lookup[edges[edge_idx,side]]
m=len(edge_idx);q=np.empty((m,2));channel_valid=np.empty(m,bool);direction_valid=np.empty(m,bool)
edge_length=np.empty(m);edge_reason=np.empty(m,np.uint8)
for begin in range(0,m,4096):
stop=min(begin+4096,m);e=edges[edge_idx[begin:stop]];j=anchor_local[begin:stop]
vectors=x[e[:,1]]-x[e[:,0]]
eq=rep.contract_direction(r['B'][j],r['mu'][j],r['valid'][j],vectors,L,'edge')
q[begin:stop]=eq['q'];channel_valid[begin:stop]=eq['channel_valid']
direction_valid[begin:stop]=eq['direction_valid'];edge_length[begin:stop]=eq['length'];edge_reason[begin:stop]=eq['direction_reason']
# Summarize all finite valid true edge ends, retaining every invalid endpoint record.
finite_q=np.isfinite(q).all(1);accepted=channel_valid&finite_q
degree=np.bincount(anchor_local,minlength=len(ids));valid_degree=np.bincount(anchor_local[accepted],minlength=len(ids))
total=np.zeros((len(ids),2));maximum=np.full((len(ids),2),-np.inf)
np.add.at(total,anchor_local[accepted],q[accepted]);np.maximum.at(maximum,anchor_local[accepted],q[accepted])
mean=np.full_like(total,np.nan);has=valid_degree>0;mean[has]=total[has]/valid_degree[has,None];maximum[~has]=np.nan
summaries=np.column_stack([mean[:,0],maximum[:,0],mean[:,1],maximum[:,1]])
full_degree=np.bincount(edges.ravel(),minlength=len(x))[ids]
complete=(degree==full_degree)&has&np.isfinite(summaries).all(1)
feature_finite=np.isfinite(r['g']).all(1)&np.isfinite(gq['q']).all(1)&np.isfinite(gq['length'])&np.isfinite(d[ids])
eligible=(d[ids]>0)&r['valid']&r['neff_pass']&gq['direction_valid']&gq['channel_valid']&complete&feature_finite
near=(d[ids]>0)&(d[ids]/s['t_part']<=.25)
node_outputs=dict(**a,**r,gate_q=gq['q'],gate_length=gq['length'],gate_direction_valid=gq['direction_valid'],gate_channel_valid=gq['channel_valid'],gate_direction_reason=gq['direction_reason'])
edge_outputs=dict(edge_indices=edge_idx,endpoint_side=side,anchor_local=anchor_local,
anchor_ids=ids[anchor_local],edge_node_ids=edges[edge_idx],q=q,channel_valid=channel_valid,
length=edge_length,direction_valid=direction_valid,direction_reason=edge_reason)
feature_done=time.monotonic()
np.savez_compressed(dest/'nodes.npz',anchor_ids=ids,raw_node_ids=raw_ids[ids],multiplicity=multiplicity,
distance=d[ids],node_mass=mass[ids],**node_outputs,edge_summary=summaries,edge_count=degree,
valid_edge_count=valid_degree,edge_block_complete=complete,eligible=eligible,near_wall=near)
np.savez_compressed(dest/'edges.npz',**edge_outputs)
mismatch=[];comparisons=[]
if phase=='equivalence':
baseline=Path(s['baseline']);br=read(baseline/'result.json')
for name,h in br['full_output_sha256'].items():assert sha(baseline/name)==h,(baseline,name)
with np.load(baseline/'nodes.npz',allow_pickle=False) as z:
for key,value in node_outputs.items():comparisons.append(bit_comparison('node.'+key,value,z[key][ids]))
with np.load(baseline/'edges.npz',allow_pickle=False) as z:
for key in ['q','channel_valid']:
comparisons.append(bit_comparison('edge.'+key,edge_outputs[key],z[key][edge_idx,side]))
for key in ['length','direction_valid','direction_reason','edge_node_ids']:
comparisons.append(bit_comparison('edge.'+key,edge_outputs[key],z[key][edge_idx]))
mismatch=[row for row in comparisons if not row['bitwise_equal']]
save(dest/'equivalence.json',dict(status='FAIL' if mismatch else 'PASS',comparison='dtype/shape/C-order bytes exactly; includes NaN bit patterns',arrays=comparisons,mismatches=mismatch,baseline_result_sha256=sha(baseline/'result.json')))
summary=dict(case=case,pose=pose,phase=phase,status='EQUIVALENCE_MISMATCH' if mismatch else 'COMPLETE',
anchors=len(ids),draw_slots=int(multiplicity.sum()),full_nodes=len(x),full_tets=len(tet),full_sources=source_count,
radius=s['radius'],t_part=s['t_part'],R_g_raw=Rg,L_med_raw=L,raw_field=raw_meta,
status_counts={name:int((r['status']==code).sum()) for code,name in rep.STATUS.items()},
status_draw_counts={name:int(multiplicity[r['status']==code].sum()) for code,name in rep.STATUS.items()},
tau0_reason_counts={name:int((r['tau0_reason']==code).sum()) for code,name in rep.TAU_REASON.items()},
gate_direction_reason_counts={name:int((gq['direction_reason']==code).sum()) for code,name in rep.DIRECTION_REASON.items()},
edge_direction_reason_counts={name:int((edge_reason==code).sum()) for code,name in rep.DIRECTION_REASON.items()},
clipped_count=int(r['clipped'].sum()),neff_pass_count=int(r['neff_pass'].sum()),
gate_channel_valid_count=int(gq['channel_valid'].sum()),edge_endpoint_count=m,
edge_channel_valid_count=int(channel_valid.sum()),edge_nonfinite_q_count=int((~finite_q).sum()),
edge_block_complete_count=int(complete.sum()),n_ref=int(eligible.sum()),eligible_draw_count=int(multiplicity[eligible].sum()),
eligible_near_wall_draw_count=int(multiplicity[eligible&near].sum()),
neighbor_visits=progress['neighbor_visits'],visit_alarm_exceeded=progress['neighbor_visits']>500000000,
algebra_checks=checks,equivalence_arrays_compared=len(comparisons),equivalence_mismatches=mismatch,
timings=dict(load_and_scales=load_done-start,raw_field=raw_done-load_done,sources=sources_done-raw_done,
kernel=kernel_done-sources_done,readout=readout_done-kernel_done,edge_features=feature_done-readout_done,
write_and_equivalence=time.monotonic()-feature_done),
wall_seconds=time.monotonic()-start,peak_rss_bytes=resource.getrusage(resource.RUSAGE_SELF).ru_maxrss*1024,
draw_file_sha256=sha(s['anchor_file']),new_distance_queries=0,S1_scoring=False,training=False,
output_sha256={n:sha(dest/n) for n in ['nodes.npz','edges.npz']})
save(dest/'result.json',summary);save(dest/'progress.json',dict(progress,stage='complete' if not mismatch else 'mismatch',elapsed_seconds=time.monotonic()-start))
print(json.dumps(summary,indent=2),flush=True)
if mismatch:raise SystemExit(2)
if __name__=='__main__':
p=argparse.ArgumentParser();p.add_argument('output',type=Path);p.add_argument('phase',choices=['equivalence','panel']);p.add_argument('index',type=int);a=p.parse_args();run(a.output,a.phase,a.index)
@@ -0,0 +1,21 @@
MIT License
Copyright (c) 2024 THUML @ Tsinghua University
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
@@ -0,0 +1,42 @@
(2 total) ===
--- Reply 1 of 2 ---
From: ClaudeWorker (local) <botuser-T0C8JMA61UY-B0C7YUHNDFC@slack-bots.com> (U0C7SSQLBJ7)
Time: 2026-11-14 17:26:54 CST
Message TS: 1794648414.929229
Accepted; working on: “@⁠TRON 已完整核读 v0.2(72deb266),在原执行差分逐项更新已吸纳/改判及剩余定义,未另建方案。 固定稿:https://g…”
--- Reply 2 of 2 ---
From: ClaudeWorker (local) <botuser-T0C8JMA61UY-B0C7YUHNDFC@slack-bots.com> (U0C7SSQLBJ7)
Time: 2026-11-14 17:28:01 CST
Message TS: 1794648481.660359
*Claude: S2 design v0.3. I accept both of MoldAstra's clarifications, and G2–G5 are now decided.*
<https://gitea.frank14f.online/Frank14f/MeltMind/src/commit/69c58a77006325319903f7862054d461b00ca75c/docs/claude/s2-design-v0.3-2026-11-14.md|s2-design-v0.3> (`69c58a77`). It decides §2 of MeltAstra's updated execution diff (`33b18e08`).
• *Two clarifications from MoldAstra, accepted:*
• The bootstrap resamples *groups only*. Each group's three matched seed runs travel with it as a block and are not resampled. So the interval is conditional on the fixed folds and the trained runs, and does not cover retraining variance.
• When S1 draws are reused, S2 eligibility is re-applied from the raw IDs and counts.
• *G2, sampling and inputs:*
• The draw code is the frozen implementation at `94011fd`.
• Eligibility uses the 5e4b406 definitions: valid spectrum, plus every incident true-edge end evaluated, at least one channel-valid finite end, and finite summaries.
• Target transform: y = log(1 + T/τ), where τ is the draw-count-weighted median of T over each fold-and-pool's eligible draws.
• ℓ_g is Dijkstra on tet edges from the node nearest the gate. If it is unreachable or any input is non-finite, the part fails closed; nothing is imputed.
• *G3, code and training:*
• Transolver at `75e0f676` (MIT), with GINO at `a6648eee` (MIT) as backup. Any modification is recorded as a diff.
• 4 layers, 8 heads, 32 slices, width {128, 256}.
• AdamW, lr 1e-3 with cosine schedule, one part per batch. The loss is a weighted MSE in y, giving equal weight to each part within a group and to each group overall.
• At most 200 epochs, early stopping with patience 20 on the inner validation set.
• Capacity: pick the lowest validation error; a tie goes to the smaller width. N and E use the same width.
• *G4, R2 and auxiliary metrics:*
• R2 is HistGBRT with random_state=0. Sklearn's built-in early stopping is *off*, because its internal split ignores the grouping. Grid of 12 configurations (learning rate × leaves × number of trees).
• Neural seeds {0, 1, 2}.
• Kendall τ-b over 5,000 random pairs per part, seed [2026111410, prefix].
• The last-filled location error is defined only within the queried nodes.
• A paired comparison uses only groups where both arms are valid; excluded groups are counted.
• *G5, resource caps:*
• 10-part measurement: 3 + 3 + 3 parts drawn from the node-count tertiles (seed 2026111411), plus the largest part. Per part: 30 min and 1.25 GiB for the whole process tree. Run sequentially, total ≤ 3 h, disk ≤ 5 GiB.
• Pilot: per run ≤ 30 min, VRAM ≤ 15 GiB, host RAM ≤ 16 GiB, no retries; still at most 4 runs and 2 GPU-hours.
*Not authorised by this document:* the entry list is an order of steps, not a go. The 10-part measurement, the pilot and the full matrix each need Yanqi's approval for that specific step. G1, the grouping calibration, still waits on MoldAstra.
@@ -0,0 +1,22 @@
[
{
"repository": "thuml/Transolver",
"revision_as_written_in_v03": "75e0f67643806a81cd1d3f6adc88dd8c02416fe7",
"revision_length": 40,
"url": "https://raw.githubusercontent.com/thuml/Transolver/75e0f67643806a81cd1d3f6adc88dd8c02416fe7/LICENSE",
"http_status": 200,
"bytes": 1084,
"sha256": "2c919cd03fa823bf7eefc00a957ff8324c865cd22aa5285e563dc4b558084f25",
"text": "MIT License\n\nCopyright (c) 2024 THUML @ Tsinghua University\n\nPermission is hereby granted, free of charge, to any person obtaining a copy\nof this software and associated documentation files (the \"Software\"), to deal\nin the Software without restriction, including without limitation the rights\nto use, copy, modify, merge, publish, distribute, sublicense, and/or sell\ncopies of the Software, and to permit persons to whom the Software is\nfurnished to do so, subject to the following conditions:\n\nThe above copyright notice and this permission notice shall be included in all\ncopies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\nIMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\nFITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\nAUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\nLIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\nOUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE\nSOFTWARE.\n"
},
{
"repository": "neuraloperator/neuraloperator",
"revision_as_written_in_v03": "a6648eeeb60fe0837f2d7d021517a945a159a0c8",
"revision_length": 40,
"url": "https://raw.githubusercontent.com/neuraloperator/neuraloperator/a6648eeeb60fe0837f2d7d021517a945a159a0c8/LICENSE",
"http_status": 200,
"bytes": 1082,
"sha256": "5f335b0603acfb1d180bf3117492e2076562075cc9b6353a3b94d5d5bc03ef33",
"text": "MIT License\n\nCopyright (c) 2023 NeuralOperator developers\n\nPermission is hereby granted, free of charge, to any person obtaining a copy\nof this software and associated documentation files (the \"Software\"), to deal\nin the Software without restriction, including without limitation the rights\nto use, copy, modify, merge, publish, distribute, sublicense, and/or sell\ncopies of the Software, and to permit persons to whom the Software is\nfurnished to do so, subject to the following conditions:\n\nThe above copyright notice and this permission notice shall be included in all\ncopies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\nIMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\nFITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\nAUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\nLIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\nOUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE\nSOFTWARE.\n"
}
]