Files
MeltMind/tools/build_s1_panel_review.py
T

87 lines
11 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""Publish complete panel count evidence; no kernel or S1 scoring."""
import ast
import csv
import hashlib
import json
from pathlib import Path
import shutil
import zipfile
def sha(p):return hashlib.sha256(Path(p).read_bytes()).hexdigest()
def read(p):return json.loads(Path(p).read_text())
def main():
out=Path('artifacts/s1-panel-precheck-2026-11-14');c=read(out/'count_contract.json');pc=read(out/'panel_contract.json');v=read(out/'verification.json')
assert v['selection_reproduced'] and len(v['parts'])==24
assert read(out/'three_part_scale_calibration.json')['status']=='PASS'
rows=[]
for s in c['cases']:
case=s['case'];g=read(out/(case+'.supervisor.json'));row=dict(case=case,tertile=s['tertile'],nodes=s['tet_nodes'],tets=s['tet_cells'],status=g['status'],wall_seconds=g['wall_seconds'],peak_RSS_MiB=g['peak_rss_bytes']/2**20)
if g['status']=='PASS':
r=read(out/case/'result.json');t=r['timings'];a=r['groups']['all'];row.update(radius=r['radius'],t_part=r['t_part'],four_point_sources=r['four_point_sources'],neighbor_visits=a['neighbor_visits'],neighbor_p50=a['quantiles']['p50'],neighbor_p99=a['quantiles']['p99'],neighbor_max=a['quantiles']['max'],alarm_5e8=r['neighbor_visit_alarm_exceeded'],geometry_load_seconds=t['load_geometry_and_topology_seconds'],distance_and_geometry_write_seconds=t['write_geometry_and_distance_seconds'],count_seconds=t['count_seconds'],future_kernel_memory_planning_MiB=r['future_kernel_memory_planning_bytes']/2**20,distance_brute_max_abs_error=r['distance_brute_max_abs_error'])
if g['status']=='PASS':
low=row['neighbor_visits']*.33e-6;high=row['neighbor_visits']*.73e-6
row.update(future_kernel_scenario_0_33us_seconds=low,future_kernel_scenario_0_73us_seconds=high,future_kernel_scenario_30min='both_scenarios_above' if low>1800 else 'straddles' if high>1800 else 'both_scenarios_below')
rows.append(row)
columns=list(dict.fromkeys(k for r in rows for k in r))
with (out/'summary.csv').open('w') as f:
w=csv.DictWriter(f,fieldnames=columns,lineterminator='\n');w.writeheader();w.writerows(rows)
passed=[r for r in rows if r['status']=='PASS'];alarms=[r for r in passed if r['alarm_5e8']]
lines=['# S1 panel:24件名单与仅邻居计数预检','',
f"24件中{len(passed)}件完成几何准备与全节点邻居计数;{len(rows)-len(passed)}件未完成/超限,逐件状态见表。没有计算核、原梯度张量、O2谱读出、S1分数或训练。计数预算通过不等于未来核计算或S1运行预算通过。",'',
'## 先公布、后准备几何与计数','',
'[Claude 13:37:37 GO](https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794634657902719);[缺缓存时几何预处理批准](https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794634967993989);[24件名单的预先公开消息](https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794635053208589)。全部名单公布后才开始第一件d计算。panel_contract保留名单生成时的未启动/壁距待裁量状态;后续count_contract引用13:42:47批准,解决该依赖,最终执行状态以监督回执与结果为准,未改写预先名单契约。','',
'由既有624件静态卡和249来源组表,每来源取最低数字前缀;按(tet_nodes,数字前缀,case)稳定排序,83/83/83等数三段。PCG64 seed=20261114依低/中/高各无放回抽8,名单按稳定顺序发布。只使用case/source/tet_nodes选件,不按标签、O2、成本、缓存或质量旗标重抽。本轮选中24件均为纯tet。source_key是既有来源分组,不声称已经排尽跨source近重复。','',
f"名单SHA256 `{pc['panel_sha256']}`。panel.csv为24件,representatives.csv保留全部249代表件及层/排名/入选标志。缺缓存没有换件;没有进行替补抽样。",'',
'## 固定口径与执行','',
'仅从既有ZIP读取坐标、tet连接和gate来源;没有解码Fill_time或动态标签数组。利用已验证SurfaceDistance,对tet子域边界三角形计算节点欧氏距离,边界节点d=0。整件tet集总质量m_i下的d中位数(包括d=0)给t_part=4 median,R=10 median=2.5t_part。每件保留30个固定节点的直接三角形穷举对照及误差;不以Rg或包围盒替代R。','',
'源点为冻结E6四面体四点坐标/质量,不构造方向张量。原始全节点做return_length邻居计数,查询批128、workers=1,不存全邻接列表。记录全节点、正壁距、边界三类计数与集总质量。查询计数包含距离等于R的源点,与原E6查询口径一致;后续核权重为零的边界支撑点并未在本轮按权重过滤。','',
'每件几何准备与计数共用1800秒/1.25GiB,外部20ms墙钟/VmRSS/VmHWM监督,退出后再核生产进程峰值。顺序执行,排他锁,开始标记/日志不覆盖;超限停止该件且保留证据,不自动替换。5e8访问仅预警,无通用访问硬上限。','',
'## 实测摘要','',
f"已完成件总节点{sum(r['nodes'] for r in passed):,};总邻居访问{sum(r['neighbor_visits'] for r in passed):,};{len(alarms)}件超过5e8预警。各件墙钟范围{min(r['wall_seconds'] for r in passed):.2f}–{max(r['wall_seconds'] for r in passed):.2f}秒,最大实测RSS {max(r['peak_RSS_MiB'] for r in passed):.2f} MiB。",'',
'| 件 | 层 | 节点 | 访问总数 | R原单位 | 墙钟秒 | 峰值MiB | 5e8预警 |','|---|---:|---:|---:|---:|---:|---:|---|']
for r in rows:lines.append(f"| {r['case'][:8]} | {r['tertile']} | {r['nodes']:,} | {r.get('neighbor_visits','NA'):,} | {r.get('radius',float('nan')):.6g} | {r['wall_seconds']:.2f} | {r['peak_RSS_MiB']:.2f} | {r.get('alarm_5e8','NA')} |" if r['status']=='PASS' else f"| {r['case'][:8]} | {r['tertile']} | {r['nodes']:,} | {r['status']} | NA | {r['wall_seconds']:.2f} | {r['peak_RSS_MiB']:.2f} | NA |")
lines+=['','按[Claude 13:48:33指出的0.33–0.73微秒/访问](https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794635313907359),summary.csv附两种未来核时间粗略情景及其是否跨30分钟。这不是本轮实测、置信区间或保证界,未纳入完整O2读出,也不是自动替换/放行规则;统一交领队裁量。','',
'summary.csv另外给出几何装载、壁距/几何写盘、计数耗时、p50/p99/max邻居数与未来核内存规划值。该规划值沿用E6的保守经验模型:计数峰值+72S源张量+72N原张量+120N输出+256×最大邻居数临时空间+128MiB;不是实测核RSS或保证界,且未纳入O2节点/边读出。没有用本轮计数时间替代核耗时,也未据此宣布未来O2通过。','',
'## 实施表与验证','',
'运行前panel_contract.json已声明面板/参照/五个置换种子、稳定排序、严格零gap模式及FP诊断abs=1e-10,并同时保留原始超量与精确正超量计数。不截断或重抽。六臂中归一化分层shuffle仅辅助敏感性,主比较是B对同维A+S和C对同维普通edge块shuffle。此表不构成S1执行批准。','',
'three_part_scale_calibration.json如实记录补做时序:panel队列之后才完成850/778/666回归;原ZIP重新读取的节点/tet/质量/边界与冻结缓存逐项一致,各30点距离对照、全部缓存d/m的median/t_part/R复算通过。没有声称panel前已做,也没有重复完整距离场或核。其复跑需要先前O2固定包的三件缓存与原始ZIP。','',
'verification.json独立复现24件抽选,核对全部结果文件SHA、全部保存计数摘要、半径/壁距边界约定;每件最多3个锚点直接与全部四点源坐标比较距离,复核计数完全一致。它没有重复壁距或计算核。','',
'ZIP包括全部24件几何缓存、d、全节点计数、统计/监督日志、预先名单及契约、来源/裁量快照、复现源码和来源组/静态卡。原data.zip不重复打包,原档案SHA和成员CRC/大小随几何记录保存。附带静态卡的标签/审计列未用于选件或本次计数。','',
'解压后的保存结果复核:','```sh','python tools/verify_s1_panel_precheck.py artifacts/s1-panel-precheck-2026-11-14','```','需要NumPy/SciPy;原始ZIP仅在重新生成几何时需要。代码/元数据路径以解压根为当前目录。所有来源文档保留其历史阶段状态,以最后带时间戳的Slack裁量为准。','']
(out/'README.md').write_text('\n'.join(lines))
dest=Path('artifacts/s1-panel-review-delivery-2026-11-14');dest.mkdir(exist_ok=False)
include={p for p in out.rglob('*') if p.is_file() and p.name!='execution.lock'}
include.update(Path(p) for p in c['code_sha256']);include.update(Path(p) for p in c['metadata_sha256'])
include.update([Path('tools/select_s1_panel.py'),Path('tools/verify_s1_panel_precheck.py'),Path('tools/check_s1_scale_calibration.py'),Path(__file__),Path('artifacts/intake-2026-11-13/source_groups.json'),Path('artifacts/routes-2026-11-13/static_data_card.csv'),Path('artifacts/o2-rep-v1.1a-2026-11-14/execution_contract.json')])
include.update(Path('docs').glob('s1-*.md'));include.update(Path('docs').glob('s1-*.txt'));include.add(Path('docs/sharing-authorization-2026-11-14.md'))
pending=[p for p in include if p.suffix=='.py'];scanned=set()
while pending:
p=pending.pop()
if p in scanned:continue
scanned.add(p)
for node in ast.walk(ast.parse(p.read_text())):
names=[node.module.split('.')[-1]] if isinstance(node,ast.ImportFrom) and node.module else [x.name.split('.')[-1] for x in node.names] if isinstance(node,(ast.Import,ast.ImportFrom)) else []
for name in names:
q=Path('tools')/(name+'.py')
if q.exists() and q not in scanned:include.add(q);pending.append(q)
# __file__ can be absolute; archive paths are always repository-relative.
include={p.relative_to(Path.cwd()) if p.is_absolute() else p for p in include}
manifest=[dict(path=str(p),bytes=p.stat().st_size,sha256=sha(p)) for p in sorted(include)]
archive=dest/'MeltMind-S1-panel-counts-20261114.zip'
with zipfile.ZipFile(archive,'x',zipfile.ZIP_DEFLATED,compresslevel=6) as z:
for row in manifest:z.write(row['path'],row['path'])
z.writestr('PACKAGE_MANIFEST.json',json.dumps(manifest,indent=2)+'\n')
(dest/'PACKAGE_MANIFEST.json').write_text(json.dumps(manifest,indent=2)+'\n')
for name in ['README.md','panel.csv','representatives.csv','panel_contract.json','count_contract.json','summary.csv','verification.json','three_part_scale_calibration.json']:shutil.copy2(out/name,dest/name)
for s in c['cases']:
case=s['case'];shutil.copy2(out/(case+'.supervisor.json'),dest/(case+'.supervisor.json'))
if (out/case/'result.json').exists():shutil.copy2(out/case/'result.json',dest/(case+'.result.json'))
hashes={p.name:sha(p) for p in sorted(dest.iterdir()) if p.is_file()};(dest/'SHA256SUMS').write_text(''.join(f'{h} {n}\n' for n,h in hashes.items()))
print(json.dumps(dict(archive_bytes=archive.stat().st_size,archive_sha256=hashes[archive.name],files=len(manifest),destination=str(dest)),indent=2))
if __name__=='__main__':main()