Publish fixed calibration postfit implementation and pre-score review binding

This commit is contained in:
MeltAstra committed 2026-11-17 20:43:20 +08:00
1 parent d67da6382b
commit 1960762d2d
18 files changed
+70002

No files matched your search

@@ -380,3 +380,11 @@ Claude6421/7650均实读;唯一service `meltmind-gate-outcome-v2-20261117`,c
队列仅特征准备,完整后状态FEATURES_COMPLETE_PREDICTIONS_AND_COMPARISON_PENDING;无fit/预测/比较入口,0候选标签读取,42下载仍暂停。比较规则已固定,后续实现仍须Assist实际语义PASS才比较,所有后续成本从同72000账继续。实际行为看started/live/first-ten/complete,不能从本快照重复启动或改pins。B17已终态PRUNE并在原Slack共享F0CJ09KV407,不重启。参见docs/phase2-gate-outcome-v2-execution-2026-11-17.md、actual-start-audit.json。
## 最新下游实现固定,待一次Assist审阅(2026-11-17 20:42)
原v2特征666520首10成本门PASS:投影70291.80245297267/72000,含3600后处理reserve;最近68终态/12active,CPU12022.295320328,完整特征尚未终态。原source3443ae53及根不改,不重启。H2原队列不干扰。
独立postfit仅准备根`/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17`,implementation11af54fb5b1dc17f15b84e1c0076d13295611bf4,contract0ff33c5250f9f0a5e2067be9859d555d74d7f03cf1b91ed01955d62d56a90158,proof SHA4c87bcd8afda48c8245e66bf3a82f75b188e6a30fd62ba74f311548e454a8491;18合成PASS,源码/合同/expected-review-binding已固定。原493全部PASS+冻结+进程退出后才可接收输入;52原预测hash复用、493两冻结模型预测,7invalid留552分母。五折顺序CPU8,每折300CPU秒、比较600,总同72000且postfit自身≤3600;新增setup120,原账含原setup600只计一次。全部预测先冻结,score读取旧552目标前再次验真实语义PASS;0新fit/候选标签/解码。没有launch/worker/waiter,review.json尚未收到,不能把实现准备当成执行或独立审阅完成。参见docs/phase2-gate-outcome-v2-postfit-2026-11-17.md。下一产物是Assist对这一固定实现的实际PASS或具体缺陷;原特征仍持续运行,不等其终态才准备下游。
@@ -0,0 +1,15 @@
# Gate outcome v2:独立预测与比较实现
范围沿Claude1794916421/1794917650;原特征队列666520及其源码、合同、账保持不变。首10成本门已PASS:1475.164010345CPU秒,投影70291.80245297267/72000,含3600后处理储备。当前只准备下游实现和合成,不在原特征队列结束前执行预测/比较。
新入口tools/phase2_gate_outcome_postfit.py没有训练、VTU解码、候选标签或几何求解路径。运行前要求原493个不同任务全部PASS,完整features冻结,原协调器/worker均退出,原post-freeze-cost存在并核对;全部493结果/数组和52复用原文件按hash检查。原累计CPU(包括原setup600)仅计一次,新setup120、协调和全部子任务实际CPU共享72000总帽,同时后处理自身不超过3600;每次启动任务前保留120尾部余量。五个预测折按顺序在CPU8执行,每折300CPU秒、480墙钟秒、4GiB;比较600CPU秒、780墙钟秒、4GiB。失败即停,无自动重试、增额、拟合或替换模型。
预测固定原五折A模型19列、E1b G模型21列。493件使用新特征,float32输入,原t_set解码秒数,G无效原生NaN保持,负值不裁剪;原query rawIDs/计数/坐标Rg/near-wall逐项一致。52 exact seed件直接复用原预测文件hash,不重预测。7无效gate不生成预测。全部545预测文件hash进入manifest并Git冻结,之后才允许比较。
比较采用已固定7650口径:原552资格分母、7/552资格损失;两gate都有定义的成对件,组件内仅对这些件均值,组件等权,2000配对组件bootstrap、seed2026111408。不补0、不插补、不因单个缺失件删整组件;只有无任何有定义件的组件排除并报告。两模型field指标差CI各在±0.001、末点/Rg指标差CI各在±0.010内才接受。差值方向为label-gate误差减LOG-gate误差。边界含等号;零支持不通过,非有限值不被静默排除。未通过接受门不等于证明不等效,也不普遍否定所有gate-dependent方法。
原E1b逐件A/G指标按固定score hash复用;52 exact件的新指标即同一原指标,不读其目标。493件在审阅门后读取原已解码seconds缓存各一次,算原weighted relative-L2和原raw-ID破同值的末点/Rg;无候选标签或新解码。每次目标读取前记录case/hash。7无效件保留原对照指标但label-gate为null。
真正语义PASS要求review.json的verdict精确PASS,reviewed_commit、contract_sha256、implementation_proof_sha256、comparison_contract_sha256以及完整code_sha256逐项匹配expected-review-binding.json;实际源码/hash再核。review和binding及合同必须同时被implementation-review Git冻结引用。review门在运行入口和score worker目标读取前检查。普通hash冻结、FAIL、缺字段、其他commit/合同/比较规则、缺源码、未纳入冻结的review都不能放行。Assist审阅一次完整实现;此文不声称已获PASS。
当前实现准备根与正式接受回执记录在/home的独立postfit根;没有启动协调器、预测或比较。原特征终态/Assist实际PASS尚待,后续不重启666520,不把比较规则已冻结当成实现已审。
+6
View File
@@ -1,3 +1,9 @@
## 最新下游实现固定,待一次Assist审阅(2026-11-17 20:42)
原v2特征666520首10成本门PASS:投影70291.80245297267/72000,含3600后处理reserve;最近68终态/12active,CPU12022.295320328,完整特征尚未终态。原source3443ae53及根不改,不重启。H2原队列不干扰。
独立postfit仅准备根`/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17`,implementation11af54fb5b1dc17f15b84e1c0076d13295611bf4,contract0ff33c5250f9f0a5e2067be9859d555d74d7f03cf1b91ed01955d62d56a90158,proof SHA4c87bcd8afda48c8245e66bf3a82f75b188e6a30fd62ba74f311548e454a8491;18合成PASS,源码/合同/expected-review-binding已固定。原493全部PASS+冻结+进程退出后才可接收输入;52原预测hash复用、493两冻结模型预测,7invalid留552分母。五折顺序CPU8,每折300CPU秒、比较600,总同72000且postfit自身≤3600;新增setup120,原账含原setup600只计一次。全部预测先冻结,score读取旧552目标前再次验真实语义PASS;0新fit/候选标签/解码。没有launch/worker/waiter,review.json尚未收到,不能把实现准备当成执行或独立审阅完成。参见docs/phase2-gate-outcome-v2-postfit-2026-11-17.md。下一产物是Assist对这一固定实现的实际PASS或具体缺陷;原特征仍持续运行,不等其终态才准备下游。
## 最新实际启动:gate outcome v2唯一准备队列(2026-11-17 20:22)
Claude6421/7650均实读;唯一service `meltmind-gate-outcome-v2-20261117`,coordinator666520,根`/home/frank14f/MeltMind-runs/phase2-gate-outcome-calibration-v2-2026-11-17`;launch3443ae534e22de0bfdfa602355abf7f9a69591af,合同5ea5b5bd06f9c40e57bf4f992a805ba42980223961ac1adbc94814e8f6473c35,comparison合同6b79ace13af01f29c4291231d880ca6c1fa4c0f063e009c4498eafcb8d8df69b。真实4guard+4worker、首批Di/Df/moments已见,首件已PASS;原固定前10成本门尚待。10合成PASS;52hash复用/493重算/7无效资格分母,12核上限,CPU16–27避开B17保留核,原单件1800秒/4GiB,总72000CPU秒含准备600保守记账和3600postfit储备。原先60秒准备预留在真实启动前更正,初版固定对象与无科学执行的元数据KeyError均保留。
@@ -0,0 +1,9 @@
{
"first10_cpu_seconds": 1475.164010345,
"projected_total_cpu_seconds": 70291.80245297267,
"factor": 1.0,
"projection_kind": "node weighted mean, fixed total cap supplies margin",
"postfit_reserve": 3600,
"cap": 72000,
"passed": true
}
@@ -0,0 +1,12 @@
{
"status": "PASS",
"tests": 18,
"elapsed_seconds": 1.606,
"command": "python -m unittest tools.test_phase2_gate_outcome_postfit",
"scope": "8 metric/paired grouping tests,4 prediction/budget/terminal tests,6 actual file/Git semantic review tests; no validator or freeze mock, temporary path constants only",
"test_source_sha256": "ae3f0849b8cf0abba75d3a7fd7b14d756234a4c094c23e446c5aaa61b75d7c79",
"new_actual_predictions": 0,
"new_comparisons": 0,
"setup_cpu_charge": 120,
"unix": 1794919305.2485075
}
File diff suppressed because it is too large. Load diff
@@ -0,0 +1,45 @@
{
"reviewed_commit": "11af54fb5b1dc17f15b84e1c0076d13295611bf4",
"contract_sha256": "0ff33c5250f9f0a5e2067be9859d555d74d7f03cf1b91ed01955d62d56a90158",
"implementation_proof_sha256": "4c87bcd8afda48c8245e66bf3a82f75b188e6a30fd62ba74f311548e454a8491",
"comparison_contract_sha256": "6b79ace13af01f29c4291231d880ca6c1fa4c0f063e009c4498eafcb8d8df69b",
"code_sha256": {
"/home/frank14f/optane/MeltMind/tools/phase2_b13_e1b.py": "e0fb6c9ecb51fdae24706e792edccb5d2259082f06aa2e292e7a23772328330d",
"/home/frank14f/optane/MeltMind/tools/r4bc_common.py": "27543fff987975ece767167f9771e6621c825035a053805604e4a696794dfcbd",
"/home/frank14f/optane/MeltMind/tools/s2_recovery_supervisor.py": "f563ca3c2665f5baf0c04f296616e4640423a59b815fbb00479af0f29e57a919",
"/home/frank14f/optane/MeltMind/tools/s2_recovery_monitoring.py": "442c2a9c75e8c00b30a8f2e91cb52d87c466b19a80aea9128b10c021e61b922e",
"/home/frank14f/optane/MeltMind/tools/s2_models/features.py": "3b34ae61cf8012ebd58b7505a73d7aa43d8ee1890ac54f98d7917b67887a45a4",
"/home/frank14f/optane/MeltMind/tools/s2_custody_statistics.py": "188565801f6901d0192f79aec0889e632501e672a4149d76d1835efbb9b45447",
"/home/frank14f/optane/MeltMind/tools/s2_models/objectives.py": "5b6527a45f08690055d572335470ce5979e8bb86e1cab00de594f79ae21316b1",
"/home/frank14f/optane/MeltMind/tools/s2_custody_common.py": "588c646101be0d50f505f9a12a1c98361aa3437c8c594eecb7321831f717a741",
"/home/frank14f/optane/MeltMind/tools/freeze_g1_components.py": "d0cee60d95243ac2c59ef3f2244ba818315718c4e3d1895f2f9f064f1f7f5ecd",
"/home/frank14f/optane/MeltMind/tools/r4bc_guard.py": "437a756ddf10b125dce450f463e76e28df52724f8c5766e0a989dfa536805628",
"/home/frank14f/optane/MeltMind/tools/phase2_b1213_gate.py": "d5a023749e9d8bb6e04a2a2cbdbb8f0d9568924c534f36102d06707859853e10",
"/home/frank14f/optane/MeltMind/tools/gre24_models.py": "53b6fee8b700b9b982b06d29908ce2ce23e0c57512c46dbf74b2aade79e3ac3a",
"/home/frank14f/optane/MeltMind/tools/phase2_b1213_features.py": "d8e975b0a5362be7ff00fbf554f2b438b54af1b6b95f8993e7c2beae77f2c8f2",
"/home/frank14f/optane/MeltMind/tools/phase2_b12_math.py": "5465b34be5578cc62a7afc67e3cee3cd489ebb75ce6622a3fa320d1b81196c57",
"/home/frank14f/optane/MeltMind/tools/o1_v2.py": "7bfc81d820a1be7d40b97b2c0b65a616812cb7615b99d80c4ac589788282c291",
"/home/frank14f/optane/MeltMind/tools/o1_partition.py": "5e7dbf97604b9ea9646113a9847e7232020e5f9248fad388177c32ae2a7117da",
"/home/frank14f/optane/MeltMind/tools/o1_prepare.py": "1a695477ee6605d75e7e991205e3e6a49a0907a7733491af8d2771c7bc6a5c23",
"/home/frank14f/optane/MeltMind/tools/gre24_features.py": "92bca4d348f31246c2e931949e2ca38103881b1b784a621a42044db56a460026",
"/home/frank14f/optane/MeltMind/tools/boundary_distance.py": "a45c1cbce92521a96eb814913b4f465b926c0b3883932ab0b7c37d11db113ed7",
"/home/frank14f/optane/MeltMind/tools/audit_samples.py": "a3daf70ec3bcb77bf46d9dc4887d552cf7c422117880a169969780012278cd70",
"/home/frank14f/optane/MeltMind/tools/gre24_features_v2.py": "7dcd5ac218ffdc1ca149edcd87e8658fb3c59e18c0fb9ce47985c7f1357d2cdf",
"/home/frank14f/optane/MeltMind/tools/phase2_h1_worker_v2.py": "1543df85b615bb69b79bacd572074a7c77ae19dab110197601573a7e5805ed20",
"/home/frank14f/optane/MeltMind/tools/phase2_residual_math.py": "ffbd9624dec79bdaabf95244fe87209055908f31a977e73ea4377045d6f8fe62",
"/home/frank14f/optane/MeltMind/tools/phase2_fixed_io.py": "ea17fdfc7508f3f980e3cdf69534b1eb8ceb8a8e469c32451b2d5e16d0ada575",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_features.py": "87258094d0dde09cf2fa5fe362e38fc9ad348013a24d93038b6f947a97cf2397",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_queue.py": "e68704fc50512200d67f6dcd6cb55f406bd71f9d82bb9bafd7fd9086c3abaf68",
"/home/frank14f/optane/MeltMind/tools/prepare_phase2_gate_outcome.py": "f9652c41287d19d7387415e958ca5ca97d664f50420dd44fbfada77139cd0359",
"/home/frank14f/optane/MeltMind/tools/test_phase2_gate_outcome_features.py": "a465cdd76f12538f66ab9b517649cf5f0c5ae4ff24624a3302534f7cdf24351a",
"/home/frank14f/optane/MeltMind/tools/test_phase2_gate_outcome_queue.py": "ae6dfe50cf3f0759ebcea6af8619d26853cf6158e5e421f53b50b51d9a583916",
"/home/frank14f/optane/MeltMind/tools/phase2_home_freeze_io.py": "bf09204b85ed2a7038acdd202257f618474d1110a9d6d772f105a3ffb334d3ff",
"/home/frank14f/optane/MeltMind/tools/phase2_local_freeze_io.py": "60adf3382577613c5234f5cf30779336b9951b66e90c075cb7cd0adab9cc42dd",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_statistics.py": "8e9f781cc4975beff245d991a9b886e0d8e7bf6d66c0ff0337d76dccaddda8e1",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_review.py": "915617aeb8292d03c654e637bc34a014c254fd178a773f135d1806c4b41372a8",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_postfit.py": "b05ee93cfac1f032c84b5a0e33e6d052a4f3b507ad6c717aa6d42456ae4bc512",
"/home/frank14f/optane/MeltMind/tools/prepare_phase2_gate_outcome_postfit.py": "198877c113c10cbe05c43943f68411159566bc939d9e250a1c96838100aac8f1",
"/home/frank14f/optane/MeltMind/tools/test_phase2_gate_outcome_postfit.py": "ae3f0849b8cf0abba75d3a7fd7b14d756234a4c094c23e446c5aaa61b75d7c79",
"/home/frank14f/optane/MeltMind/docs/phase2-gate-outcome-v2-postfit-2026-11-17.md": "d36c860b1b5ecf96fe07fca18c906f5e35d559688355a8abbb7c3c9efa82e7c2"
}
}
@@ -0,0 +1,4 @@
{
"contract_sha256": "0ff33c5250f9f0a5e2067be9859d555d74d7f03cf1b91ed01955d62d56a90158",
"proof_sha256": "4c87bcd8afda48c8245e66bf3a82f75b188e6a30fd62ba74f311548e454a8491"
}
@@ -0,0 +1,91 @@
{
"commit": "11af54fb5b1dc17f15b84e1c0076d13295611bf4",
"files": {
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/contract.json": "0ff33c5250f9f0a5e2067be9859d555d74d7f03cf1b91ed01955d62d56a90158",
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/acceptance.json": "dbe0bd47eeb4b52407c57a2f1fe0e570f9570abed8aef4d16f9df1463a2591c6",
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/preparation-audit.json": "905a2a6bb20c08875b9dafce161817aa953ba804d97116047d1732da0802eadb",
"/home/frank14f/optane/MeltMind/tools/phase2_b13_e1b.py": "e0fb6c9ecb51fdae24706e792edccb5d2259082f06aa2e292e7a23772328330d",
"/home/frank14f/optane/MeltMind/tools/r4bc_common.py": "27543fff987975ece767167f9771e6621c825035a053805604e4a696794dfcbd",
"/home/frank14f/optane/MeltMind/tools/s2_recovery_supervisor.py": "f563ca3c2665f5baf0c04f296616e4640423a59b815fbb00479af0f29e57a919",
"/home/frank14f/optane/MeltMind/tools/s2_recovery_monitoring.py": "442c2a9c75e8c00b30a8f2e91cb52d87c466b19a80aea9128b10c021e61b922e",
"/home/frank14f/optane/MeltMind/tools/s2_models/features.py": "3b34ae61cf8012ebd58b7505a73d7aa43d8ee1890ac54f98d7917b67887a45a4",
"/home/frank14f/optane/MeltMind/tools/s2_custody_statistics.py": "188565801f6901d0192f79aec0889e632501e672a4149d76d1835efbb9b45447",
"/home/frank14f/optane/MeltMind/tools/s2_models/objectives.py": "5b6527a45f08690055d572335470ce5979e8bb86e1cab00de594f79ae21316b1",
"/home/frank14f/optane/MeltMind/tools/s2_custody_common.py": "588c646101be0d50f505f9a12a1c98361aa3437c8c594eecb7321831f717a741",
"/home/frank14f/optane/MeltMind/tools/freeze_g1_components.py": "d0cee60d95243ac2c59ef3f2244ba818315718c4e3d1895f2f9f064f1f7f5ecd",
"/home/frank14f/optane/MeltMind/tools/r4bc_guard.py": "437a756ddf10b125dce450f463e76e28df52724f8c5766e0a989dfa536805628",
"/home/frank14f/optane/MeltMind/tools/phase2_b1213_gate.py": "d5a023749e9d8bb6e04a2a2cbdbb8f0d9568924c534f36102d06707859853e10",
"/home/frank14f/optane/MeltMind/tools/gre24_models.py": "53b6fee8b700b9b982b06d29908ce2ce23e0c57512c46dbf74b2aade79e3ac3a",
"/home/frank14f/optane/MeltMind/tools/phase2_b1213_features.py": "d8e975b0a5362be7ff00fbf554f2b438b54af1b6b95f8993e7c2beae77f2c8f2",
"/home/frank14f/optane/MeltMind/tools/phase2_b12_math.py": "5465b34be5578cc62a7afc67e3cee3cd489ebb75ce6622a3fa320d1b81196c57",
"/home/frank14f/optane/MeltMind/tools/o1_v2.py": "7bfc81d820a1be7d40b97b2c0b65a616812cb7615b99d80c4ac589788282c291",
"/home/frank14f/optane/MeltMind/tools/o1_partition.py": "5e7dbf97604b9ea9646113a9847e7232020e5f9248fad388177c32ae2a7117da",
"/home/frank14f/optane/MeltMind/tools/o1_prepare.py": "1a695477ee6605d75e7e991205e3e6a49a0907a7733491af8d2771c7bc6a5c23",
"/home/frank14f/optane/MeltMind/tools/gre24_features.py": "92bca4d348f31246c2e931949e2ca38103881b1b784a621a42044db56a460026",
"/home/frank14f/optane/MeltMind/tools/boundary_distance.py": "a45c1cbce92521a96eb814913b4f465b926c0b3883932ab0b7c37d11db113ed7",
"/home/frank14f/optane/MeltMind/tools/audit_samples.py": "a3daf70ec3bcb77bf46d9dc4887d552cf7c422117880a169969780012278cd70",
"/home/frank14f/optane/MeltMind/tools/gre24_features_v2.py": "7dcd5ac218ffdc1ca149edcd87e8658fb3c59e18c0fb9ce47985c7f1357d2cdf",
"/home/frank14f/optane/MeltMind/tools/phase2_h1_worker_v2.py": "1543df85b615bb69b79bacd572074a7c77ae19dab110197601573a7e5805ed20",
"/home/frank14f/optane/MeltMind/tools/phase2_residual_math.py": "ffbd9624dec79bdaabf95244fe87209055908f31a977e73ea4377045d6f8fe62",
"/home/frank14f/optane/MeltMind/tools/phase2_fixed_io.py": "ea17fdfc7508f3f980e3cdf69534b1eb8ceb8a8e469c32451b2d5e16d0ada575",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_features.py": "87258094d0dde09cf2fa5fe362e38fc9ad348013a24d93038b6f947a97cf2397",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_queue.py": "e68704fc50512200d67f6dcd6cb55f406bd71f9d82bb9bafd7fd9086c3abaf68",
"/home/frank14f/optane/MeltMind/tools/prepare_phase2_gate_outcome.py": "f9652c41287d19d7387415e958ca5ca97d664f50420dd44fbfada77139cd0359",
"/home/frank14f/optane/MeltMind/tools/test_phase2_gate_outcome_features.py": "a465cdd76f12538f66ab9b517649cf5f0c5ae4ff24624a3302534f7cdf24351a",
"/home/frank14f/optane/MeltMind/tools/test_phase2_gate_outcome_queue.py": "ae6dfe50cf3f0759ebcea6af8619d26853cf6158e5e421f53b50b51d9a583916",
"/home/frank14f/optane/MeltMind/tools/phase2_home_freeze_io.py": "bf09204b85ed2a7038acdd202257f618474d1110a9d6d772f105a3ffb334d3ff",
"/home/frank14f/optane/MeltMind/tools/phase2_local_freeze_io.py": "60adf3382577613c5234f5cf30779336b9951b66e90c075cb7cd0adab9cc42dd",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_statistics.py": "8e9f781cc4975beff245d991a9b886e0d8e7bf6d66c0ff0337d76dccaddda8e1",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_review.py": "915617aeb8292d03c654e637bc34a014c254fd178a773f135d1806c4b41372a8",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_postfit.py": "b05ee93cfac1f032c84b5a0e33e6d052a4f3b507ad6c717aa6d42456ae4bc512",
"/home/frank14f/optane/MeltMind/tools/prepare_phase2_gate_outcome_postfit.py": "198877c113c10cbe05c43943f68411159566bc939d9e250a1c96838100aac8f1",
"/home/frank14f/optane/MeltMind/tools/test_phase2_gate_outcome_postfit.py": "ae3f0849b8cf0abba75d3a7fd7b14d756234a4c094c23e446c5aaa61b75d7c79",
"/home/frank14f/optane/MeltMind/docs/phase2-gate-outcome-v2-postfit-2026-11-17.md": "d36c860b1b5ecf96fe07fca18c906f5e35d559688355a8abbb7c3c9efa82e7c2",
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-calibration-v2-2026-11-17/comparison-contract.json": "6b79ace13af01f29c4291231d880ca6c1fa4c0f063e009c4498eafcb8d8df69b"
},
"git_paths": {
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/contract.json": "home-runs/phase2-gate-outcome-v2-postfit-2026-11-17/contract.json",
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/acceptance.json": "home-runs/phase2-gate-outcome-v2-postfit-2026-11-17/acceptance.json",
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/preparation-audit.json": "home-runs/phase2-gate-outcome-v2-postfit-2026-11-17/preparation-audit.json",
"/home/frank14f/optane/MeltMind/tools/phase2_b13_e1b.py": "tools/phase2_b13_e1b.py",
"/home/frank14f/optane/MeltMind/tools/r4bc_common.py": "tools/r4bc_common.py",
"/home/frank14f/optane/MeltMind/tools/s2_recovery_supervisor.py": "tools/s2_recovery_supervisor.py",
"/home/frank14f/optane/MeltMind/tools/s2_recovery_monitoring.py": "tools/s2_recovery_monitoring.py",
"/home/frank14f/optane/MeltMind/tools/s2_models/features.py": "tools/s2_models/features.py",
"/home/frank14f/optane/MeltMind/tools/s2_custody_statistics.py": "tools/s2_custody_statistics.py",
"/home/frank14f/optane/MeltMind/tools/s2_models/objectives.py": "tools/s2_models/objectives.py",
"/home/frank14f/optane/MeltMind/tools/s2_custody_common.py": "tools/s2_custody_common.py",
"/home/frank14f/optane/MeltMind/tools/freeze_g1_components.py": "tools/freeze_g1_components.py",
"/home/frank14f/optane/MeltMind/tools/r4bc_guard.py": "tools/r4bc_guard.py",
"/home/frank14f/optane/MeltMind/tools/phase2_b1213_gate.py": "tools/phase2_b1213_gate.py",
"/home/frank14f/optane/MeltMind/tools/gre24_models.py": "tools/gre24_models.py",
"/home/frank14f/optane/MeltMind/tools/phase2_b1213_features.py": "tools/phase2_b1213_features.py",
"/home/frank14f/optane/MeltMind/tools/phase2_b12_math.py": "tools/phase2_b12_math.py",
"/home/frank14f/optane/MeltMind/tools/o1_v2.py": "tools/o1_v2.py",
"/home/frank14f/optane/MeltMind/tools/o1_partition.py": "tools/o1_partition.py",
"/home/frank14f/optane/MeltMind/tools/o1_prepare.py": "tools/o1_prepare.py",
"/home/frank14f/optane/MeltMind/tools/gre24_features.py": "tools/gre24_features.py",
"/home/frank14f/optane/MeltMind/tools/boundary_distance.py": "tools/boundary_distance.py",
"/home/frank14f/optane/MeltMind/tools/audit_samples.py": "tools/audit_samples.py",
"/home/frank14f/optane/MeltMind/tools/gre24_features_v2.py": "tools/gre24_features_v2.py",
"/home/frank14f/optane/MeltMind/tools/phase2_h1_worker_v2.py": "tools/phase2_h1_worker_v2.py",
"/home/frank14f/optane/MeltMind/tools/phase2_residual_math.py": "tools/phase2_residual_math.py",
"/home/frank14f/optane/MeltMind/tools/phase2_fixed_io.py": "tools/phase2_fixed_io.py",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_features.py": "tools/phase2_gate_outcome_features.py",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_queue.py": "tools/phase2_gate_outcome_queue.py",
"/home/frank14f/optane/MeltMind/tools/prepare_phase2_gate_outcome.py": "tools/prepare_phase2_gate_outcome.py",
"/home/frank14f/optane/MeltMind/tools/test_phase2_gate_outcome_features.py": "tools/test_phase2_gate_outcome_features.py",
"/home/frank14f/optane/MeltMind/tools/test_phase2_gate_outcome_queue.py": "tools/test_phase2_gate_outcome_queue.py",
"/home/frank14f/optane/MeltMind/tools/phase2_home_freeze_io.py": "tools/phase2_home_freeze_io.py",
"/home/frank14f/optane/MeltMind/tools/phase2_local_freeze_io.py": "tools/phase2_local_freeze_io.py",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_statistics.py": "tools/phase2_gate_outcome_statistics.py",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_review.py": "tools/phase2_gate_outcome_review.py",
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_postfit.py": "tools/phase2_gate_outcome_postfit.py",
"/home/frank14f/optane/MeltMind/tools/prepare_phase2_gate_outcome_postfit.py": "tools/prepare_phase2_gate_outcome_postfit.py",
"/home/frank14f/optane/MeltMind/tools/test_phase2_gate_outcome_postfit.py": "tools/test_phase2_gate_outcome_postfit.py",
"/home/frank14f/optane/MeltMind/docs/phase2-gate-outcome-v2-postfit-2026-11-17.md": "docs/phase2-gate-outcome-v2-postfit-2026-11-17.md",
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-calibration-v2-2026-11-17/comparison-contract.json": "home-runs/phase2-gate-outcome-calibration-v2-2026-11-17/comparison-contract.json"
},
"remote_published": false,
"local_repository": "/tmp/meltmind-phase2-local-fixed-20261117"
}
@@ -0,0 +1,11 @@
{
"status": "PREPARED_NOT_LAUNCHED",
"original_feature_pid": 666520,
"original_feature_pins_verified": true,
"comparison_rule_fixed": true,
"models_hash_checked": 10,
"no_labels_opened": true,
"no_prediction_or_comparison": true,
"no_new_waiter": true,
"unix": 1794919305.2486906
}
@@ -0,0 +1,4 @@
{
"contract_sha256": "0ff33c5250f9f0a5e2067be9859d555d74d7f03cf1b91ed01955d62d56a90158",
"proof_sha256": "413cddcfd93a0ac32d676601bc31061f89d4692f5151aa83d416eefae6c408ff"
}
@@ -0,0 +1,17 @@
{
"commit": "3d5a69dc14704153e3c265cf24703d397490a5b2",
"files": {
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/contract.json": "0ff33c5250f9f0a5e2067be9859d555d74d7f03cf1b91ed01955d62d56a90158",
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/expected-review-binding.json": "c47166bda1ba4060298a9d7f547c02fdf3bd2b2472f402acf1b73e3fa4d77228",
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/implementation-git-freeze.json": "4c87bcd8afda48c8245e66bf3a82f75b188e6a30fd62ba74f311548e454a8491",
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/implementation-authorization.json": "3e2754925021a04c321fefb33709f98efd7c7ca089e3855f93882d23d0de20ec"
},
"git_paths": {
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/contract.json": "home-runs/phase2-gate-outcome-v2-postfit-2026-11-17/contract.json",
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/expected-review-binding.json": "home-runs/phase2-gate-outcome-v2-postfit-2026-11-17/expected-review-binding.json",
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/implementation-git-freeze.json": "home-runs/phase2-gate-outcome-v2-postfit-2026-11-17/implementation-git-freeze.json",
"/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17/implementation-authorization.json": "home-runs/phase2-gate-outcome-v2-postfit-2026-11-17/implementation-authorization.json"
},
"remote_published": false,
"local_repository": "/tmp/meltmind-phase2-local-fixed-20261117"
}
+162
View File
@@ -0,0 +1,162 @@
"""Single continuation of fixed calibration: reuse models, predict, freeze, compare."""
import argparse,fcntl,json,os,resource,signal,sys,time
from pathlib import Path
import numpy as np
from phase2_fixed_io import BASE
from phase2_h1_worker_v2 import checked,sha,save
from phase2_home_freeze_io import verify,freeze
from phase2_gate_outcome_review import validate_review
from phase2_gate_outcome_statistics import two_metrics,summarize
from phase2_b1213_features import cached
from s2_custody_common import prediction_inputs
from r4bc_guard import supervise
from s2_recovery_monitoring import disk_usage
def read(p):return json.loads(Path(p).read_text())
def config(root):
c=read(root/'contract.json');verify(root,'implementation')
if c['total_cpu_seconds']!=72000 or c['postfit_cpu_seconds']!=3600 or c['new_fits']!=0 or c['candidate_label_reads']!=0:raise PermissionError('Fixed scope')
for p,h in {**c['source_sha256'],**c['code_sha256']}.items():checked(p,h)
prep=read(c['preparation_contract']);comparison=read(c['comparison_contract'])
if comparison['rule']!=c['comparison_rule'] or comparison['bands']!=c['bands']:raise ValueError('Fixed7650 statistics changed')
if len(c['parts'])!=552 or len({r['case'] for r in c['parts']})!=552 or {r['case'] for r in c['parts']}!=set(prep['original_cases']):raise ValueError('Original552 roster')
old=read(checked(comparison['original_prediction_manifest'],comparison['original_prediction_manifest_sha256']))['entries'];by={r['case']:r for r in c['parts']}
if len(old)!=552 or set(by)!={e['case'] for e in old}:raise ValueError('Original prediction roster')
for e in old:
row=by[e['case']];test=[a['fold'] for a in row['assignments'] if a['role']=='test']
if test!=[e['fold']] or row['component_id']!=e['component_id']:raise ValueError('Frozen fold/component identity')
return c,prep,comparison,{e['case']:e for e in old}
def terminal_inputs(root,c,prep):
p=Path(c['preparation_root']);verify(p,'features');done=read(p/'complete.json');cost=read(p/'post-freeze-cost.json')
if done['status']!='FEATURES_COMPLETE_PREDICTIONS_AND_COMPARISON_PENDING' or not done['accounting_complete'] or len(done['records'])!=493 or any(r['guard']['status']!='PASS' for r in done['records']):raise PermissionError('All493 original PASS required')
if {r['index'] for r in done['records']}!=set(range(493)):raise ValueError('Distinct493 terminal jobs')
if not cost['within_cap'] or cost['cumulative_cpu_seconds']<done['cumulative_cpu_seconds']:raise ValueError('Complete original CPU ledger required')
import psutil
for process in psutil.process_iter(['pid','cmdline']):
args=process.info['cmdline'] or []
if any(x.endswith('phase2_gate_outcome_queue.py') for x in args):raise PermissionError('Original feature processes still active')
m=read(p/'features-manifest.json');new=m['recomputed'];exact=m['exact_reuse'];invalid=m['invalid_cases']
if len(new)!=493 or exact!=prep['exact_reuse'] or invalid!=prep['invalid_cases']:raise ValueError('Fixed493/52/7 partition')
expected={item['row']['case']:i for i,item in enumerate(prep['parts'])}
if len({e['case'] for e in new})!=493 or set(expected)!={e['case'] for e in new}:raise ValueError('Recomputation roster')
for e in new:
d=Path(e['path']);r=read(checked(d/'result.json',e['result_sha256']))
if e['index']!=expected[e['case']] or r['case']!=e['case'] or r['output_sha256']!=e['output_sha256']:raise ValueError('Feature result identity')
for n,h in e['output_sha256'].items():checked(d/n,h)
for e in exact:
for path,h in e['source_sha256'].items():checked(path,h)
return dict(prior_cpu_seconds=cost['cumulative_cpu_seconds'],original_complete_sha256=sha(p/'complete.json'),original_cost_sha256=sha(p/'post-freeze-cost.json'),features_manifest_sha256=sha(p/'features-manifest.json'),recomputed=new,exact_reuse=exact,invalid_cases=invalid,terminal_pid_absent=True)
def adopted(root,c):
verify(root,'inputs');a=read(root/'adopted-inputs.json');p=Path(c['preparation_root'])
checked(p/'complete.json',a['original_complete_sha256']);checked(p/'post-freeze-cost.json',a['original_cost_sha256']);checked(p/'features-manifest.json',a['features_manifest_sha256'])
if a['prior_cpu_seconds']!=read(p/'post-freeze-cost.json')['cumulative_cpu_seconds']:raise ValueError('Original cost changed')
return a
def query_identity(q,z):
for name in ['raw_ids','counts','coordinates_over_Rg','near_wall']:
if not np.array_equal(q[name],z[name]):raise ValueError('Frozen query identity: '+name)
def predict_arrays(A,G,features,extra,raw,q,t_set):
ix=np.searchsorted(raw,q['raw_ids'])
if np.any(ix>=len(raw)) or not np.array_equal(raw[ix],q['raw_ids']):raise ValueError('Query IDs outside full-node array')
if features.shape!=(len(raw),19) or extra.shape!=(len(raw),2) or not np.isfinite(features).all() or np.isinf(extra).any():raise ValueError('Invalid19+2 features')
if not np.isfinite(t_set) or t_set<=0:raise ValueError('Original clock')
x=features[ix].astype(np.float32);g=np.column_stack([features[ix],extra[ix]]).astype(np.float32)
out=dict(A=np.asarray(A.predict(x),np.float64)*t_set,G=np.asarray(G.predict(g),np.float64)*t_set)
if any(v.shape!=(len(ix),) or not np.isfinite(v).all() for v in out.values()):raise ValueError('Nonfinite prediction; no auto exclusion')
return out
def predict(root,fold):
import joblib,xgboost as xgb
from threadpoolctl import threadpool_limits
c,prep,cmp,old=config(root);a=adopted(root,c);by={e['case']:e for e in a['recomputed']};spec=cmp['models'][fold]
if spec['fold']!=fold or xgb.__version__!='2.1.4' or np.__version__!='1.26.4':raise ValueError('Model/library identity')
A=joblib.load(checked(spec['A']['path'],spec['A']['sha256']));G=xgb.XGBRegressor();G.load_model(checked(spec['G']['path'],spec['G']['sha256']));A.set_params(n_jobs=1);G.set_params(n_jobs=1)
if A.get_booster().num_features()!=19 or G.get_booster().num_features()!=21:raise ValueError('Model feature widths')
dest=root/'predict'/str(fold);dest.mkdir(parents=True,exist_ok=False);entries=[]
with threadpool_limits(1):
for row in c['parts']:
case=row['case'];base=old[case]
if base['fold']!=fold or case not in by:continue
e=by[case];path=Path(e['path']);q=prediction_inputs(row);raw=cached(row,'raw_ids')
with np.load(checked(base['path'],base['sha256']),allow_pickle=False) as z:query_identity(q,z)
f=np.load(checked(path/'features.npy',e['output_sha256']['features.npy']),allow_pickle=False);g=np.load(checked(path/'G_extra.npy',e['output_sha256']['G_extra.npy']),allow_pickle=False)
values=predict_arrays(A,G,f,g,raw,q,row['preparation']['t_set']);out=dest/(case+'.npz');np.savez_compressed(out,**values,**{n:q[n] for n in ['raw_ids','counts','coordinates_over_Rg','near_wall']})
entries.append(dict(case=case,component_id=row['component_id'],fold=fold,path=str(out),sha256=sha(out),original=base,exact_reuse=False,negative_predictions={k:int(np.sum(v<0)) for k,v in values.items()}));save(dest/'progress.json',dict(parts=len(entries),labels_read=False))
save(dest/'manifest.json',dict(entries=entries,labels_read=False,new_fits=0))
def freeze_predictions(root,c,cmp,old,a):
entries=[]
for f in range(5):entries.extend(read(root/'predict'/str(f)/'manifest.json')['entries'])
if len(entries)!=493 or {e['case'] for e in entries}!={e['case'] for e in a['recomputed']}:raise ValueError('All493 new predictions required')
for e in a['exact_reuse']:
base=old[e['case']]
if e['prediction']!=base:raise ValueError('Exact52 original prediction identity')
entries.append(dict(case=e['case'],component_id=base['component_id'],fold=base['fold'],path=base['path'],sha256=base['sha256'],original=base,exact_reuse=True))
if len({e['case'] for e in entries})!=545:raise ValueError('Distinct545 predictions')
for e in entries:checked(e['path'],e['sha256'])
save(root/'predictions-manifest.json',dict(entries=entries,invalid_cases=a['invalid_cases'],eligible_denominator=552,new_parts=493,reused_parts=52,labels_read=False))
freeze(root,'predictions',[root/'contract.json',root/'adopted-inputs.json',root/'predictions-manifest.json',*[Path(p) for p in c['code_sha256']]])
def score(root):
c,prep,cmp,old=config(root);adopted(root,c);verify(root,'predictions');validate_review(root) # Before ANY target values or comparison.
dest=root/'score';dest.mkdir(exist_ok=False);manifest=read(root/'predictions-manifest.json');entries={e['case']:e for e in manifest['entries']};invalid=set(manifest['invalid_cases']);prior={r['case']:r for r in read(checked(c['original_score'],c['source_sha256'][c['original_score']]))['rows']}
if len(entries)!=545 or invalid!=set(prep['invalid_cases']) or set(entries)&invalid or set(entries)|invalid!={r['case'] for r in c['parts']}:raise ValueError('Complete eligibility roster')
rows=[];access=[];save(dest/'review-consumed.json',dict(verdict='PASS',review_sha256=sha(root/'review.json'),unix=time.time(),before_target_access=True))
for row in c['parts']:
case=row['case'];base=old[case];prior_row=prior[case]
if prior_row['fold']!=base['fold'] or prior_row['component_id']!=row['component_id']:raise ValueError('Original score roles')
reference=dict(A={k:prior_row['A'][k] for k in ['relative_l2','last_location']},G={k:prior_row['new'][k] for k in ['relative_l2','last_location']})
out=dict(case=case,component_id=row['component_id'],fold=base['fold'],LOG_gate=reference,label_gate=None,eligibility='INVALID_GATE' if case in invalid else 'VALID')
if case not in invalid:
e=entries[case]
if e['original']!=base or e['component_id']!=row['component_id'] or e['fold']!=base['fold']:raise ValueError('Prediction fold/component identity')
if e['exact_reuse']:
checked(e['path'],e['sha256']);out['label_gate']=reference;out['reuse']='Exact seed set, original prediction and metrics by hash'
else:
access.append(dict(case=case,target_sha256=row['v5_cache']['output_sha256']['seconds.npy'],existing_old552_cache=True));save(dest/'target-access.json',access)
truth=cached(row,'seconds');raw=cached(row,'raw_ids')
with np.load(checked(e['path'],e['sha256']),allow_pickle=False) as z,np.load(checked(base['path'],base['sha256']),allow_pickle=False) as zz:
query_identity(z,zz);ix=np.searchsorted(raw,z['raw_ids'])
if np.any(ix>=len(raw)) or not np.array_equal(raw[ix],z['raw_ids']):raise ValueError('Score query IDs')
out['label_gate']={arm:two_metrics(z[arm],truth[ix],z) for arm in ['A','G']}
rows.append(out);save(dest/'progress.json',dict(parts=len(rows)))
result=summarize(rows,{r['component_id'] for r in c['parts']});save(dest/'result.json',dict(status='COMPLETE',**result,rows=rows,eligibility_denominator=552,invalid_gate_parts=sorted(invalid),eligibility_loss=7/552,exact_seed_prediction_reuse=52,new_prediction_parts=493,original_exact_set_FAIL_preserved=True,candidate_label_reads=0,new_decodes=0,new_fits=0,comparison_rule=cmp['rule'],review_sha256=sha(root/'review.json')))
def remaining_cap(prior,post_spent,requested):
remaining=min(3600-post_spent,72000-prior-post_spent)-120
if requested>remaining:raise RuntimeError('Insufficient shared budget for fixed job; no retry/extension')
return requested
def run(root):
start=time.process_time();u0=resource.getrusage(resource.RUSAGE_CHILDREN);c,prep,cmp,old=config(root);validate_review(root);lock=(root/'run.lock').open('a');fcntl.flock(lock,fcntl.LOCK_EX|fcntl.LOCK_NB)
if (root/'started.json').exists():raise FileExistsError('Single postfit launch; no restart')
os.sched_setaffinity(0,[8]);save(root/'started.json',dict(pid=os.getpid(),unix=time.time(),phase='VERIFY_ORIGINAL_TERMINAL',new_fits=0,candidate_label_reads=0));prior=None;records=[];charged=0.;requested=[]
def post_cost():
u=resource.getrusage(resource.RUSAGE_CHILDREN);return c['setup_cpu_seconds']+time.process_time()-start+max(charged,u.ru_utime+u.ru_stime-u0.ru_utime-u0.ru_stime)
def job(mode,index,cap):
nonlocal charged
if requested or disk_usage(root)>c['disk_bytes']:raise RuntimeError('Signal/disk stop')
cap=remaining_cap(prior,post_cost(),cap);d=root/'guards'/f'{mode}-{index}'
g=supervise([sys.executable,'tools/phase2_gate_outcome_postfit.py',mode,str(root),str(index)],d,dict(cpu_seconds=cap,wall_seconds=cap+180,rss_bytes=4*2**30,disk_bytes=c['disk_bytes']))
charged+=g['cpu_seconds'];records.append(dict(mode=mode,index=index,guard=g));save(root/'live-progress.json',dict(stage=mode,records=records,prior_cpu_once=prior,postfit_cpu_seconds=post_cost(),cumulative_cpu_seconds=prior+post_cost()))
if g['status']!='PASS' or post_cost()>3600 or prior+post_cost()>72000:raise RuntimeError('Postfit guard/budget stop')
old_handlers={s:signal.signal(s,lambda signum,frame:requested.append(signum)) for s in [signal.SIGTERM,signal.SIGINT]}
try:
a=terminal_inputs(root,c,prep);prior=a['prior_cpu_seconds'];save(root/'adopted-inputs.json',a);freeze(root,'inputs',[root/'contract.json',root/'adopted-inputs.json',Path(c['preparation_root'])/'complete.json',Path(c['preparation_root'])/'post-freeze-cost.json',Path(c['preparation_root'])/'features-manifest.json'])
for f in range(5):job('predict',f,300)
freeze_predictions(root,c,cmp,old,a);job('score',0,600)
save(root/'complete.json',dict(status='COMPLETE',decision=read(root/'score/result.json')['decision'],records=records,prior_cpu_once=prior,postfit_cpu_seconds=post_cost(),cumulative_cpu_seconds=prior+post_cost(),accounting_complete=True,new_fits=0,new_decodes=0,retry=False))
freeze(root,'terminal',[root/'contract.json',root/'complete.json',root/'score/result.json',root/'review.json',root/'predictions-manifest.json'])
save(root/'post-terminal-cost.json',dict(cumulative_cpu_seconds=prior+post_cost(),postfit_cpu_seconds=post_cost(),within_cap=prior+post_cost()<=72000 and post_cost()<=3600))
except BaseException as exc:
save(root/'complete.json',dict(status='STOPPED',reason=type(exc).__name__+': '+str(exc),records=records,prior_cpu_once=prior,postfit_cpu_seconds=post_cost(),cumulative_cpu_seconds=prior+post_cost() if prior is not None else None,accounting_complete=prior is not None,retry=False));raise
finally:
for s,h in old_handlers.items():signal.signal(s,h)
if __name__=='__main__':
p=argparse.ArgumentParser();p.add_argument('action',choices=['run','predict','score']);p.add_argument('root',type=Path);p.add_argument('index',nargs='?',type=int,default=0);a=p.parse_args()
if a.action=='run':run(a.root)
elif a.action=='predict':predict(a.root,a.index)
else:score(a.root)
+20
View File
@@ -0,0 +1,20 @@
"""Exact semantic PASS and immutable source/contract identity before custody."""
import json
from pathlib import Path
from phase2_h1_worker_v2 import sha,checked
from phase2_home_freeze_io import verify
def read(p):return json.loads(Path(p).read_text())
def validate_review(root):
root=Path(root);c=read(root/'contract.json');proof=verify(root,'implementation');binding=read(root/'expected-review-binding.json');review=read(root/'review.json')
expected=dict(reviewed_commit=proof['commit'],contract_sha256=sha(root/'contract.json'),implementation_proof_sha256=sha(root/'implementation-git-freeze.json'),comparison_contract_sha256=c['comparison_contract_sha256'],code_sha256=c['code_sha256'])
if binding!=expected:raise PermissionError('Review binding differs from implementation')
if review.get('verdict')!='PASS':raise PermissionError('Actual exact PASS required')
for name,value in expected.items():
if review.get(name)!=value:raise PermissionError('Review identity mismatch: '+name)
checked(c['comparison_contract'],expected['comparison_contract_sha256'])
for p,h in expected['code_sha256'].items():checked(p,h)
frozen=verify(root,'implementation-review')
for p in [root/'review.json',root/'expected-review-binding.json',root/'contract.json']:
if frozen['files'].get(str(p.resolve()))!=sha(p):raise PermissionError('Review freeze does not include exact required file')
return review
+34
View File
@@ -0,0 +1,34 @@
"""Fixed7650 paired component equivalence; array/record only, no data I/O."""
from collections import defaultdict
import numpy as np
from r4bc_common import paired_difference
from s2_models.objectives import relative_l2
from s2_custody_statistics import last_location
METRICS={'relative_l2':.001,'last_location':.010}
def two_metrics(pred,truth,query):
p=np.asarray(pred);t=np.asarray(truth)
if p.shape!=t.shape or p.ndim!=1 or not len(p) or not np.isfinite(p).all() or not np.isfinite(t).all():raise ValueError('Invalid metric arrays')
return dict(relative_l2=relative_l2(p,t,query['counts'],np.ones(len(t),bool)),last_location=dict(value=last_location(t,p,query['raw_ids'],query['coordinates_over_Rg']),reason=None))
def summarize(rows,original_components):
if len({r['case'] for r in rows})!=len(rows):raise ValueError('Duplicate part')
result={};all_components=set(original_components)
for arm in ['A','G']:
result[arm]={}
for metric,band in METRICS.items():
paired=defaultdict(list);excluded=[]
for r in rows:
if r['component_id'] not in all_components:raise ValueError('Unknown component')
new=None if r['label_gate'] is None else r['label_gate'][arm][metric]['value']
old=None if r['LOG_gate'] is None else r['LOG_gate'][arm][metric]['value']
if new is None or old is None:excluded.append(r['case']);continue
if not np.isfinite([new,old]).all():raise ValueError('Nonfinite metric must not be silently dropped')
paired[r['component_id']].append((float(new),float(old)))
means={g:np.mean(v,axis=0) for g,v in paired.items()}
new={g:float(v[0]) for g,v in means.items()};old={g:float(v[1]) for g,v in means.items()}
ci=paired_difference(new,old,seed=2026111408,replicates=2000)
passes=ci['low'] is not None and ci['high'] is not None and ci['low']>=-band and ci['high']<=band
result[arm][metric]=dict(label_minus_LOG=ci,label_mean=float(np.mean(list(new.values()))) if new else None,LOG_mean=float(np.mean(list(old.values()))) if old else None,band=[-band,band],passed=passes,paired_parts=sum(map(len,paired.values())),excluded_parts=excluded,component_part_counts={g:len(v) for g,v in paired.items()},excluded_components=sorted(all_components-set(paired)),excluded_component_count=len(all_components-set(paired)))
passed=all(v['passed'] for arm in result.values() for v in arm.values())
return dict(decision='PASS_OUTCOME_EQUIVALENCE' if passed else 'FAIL_OUTCOME_EQUIVALENCE',accepted_for_v2=passed,comparisons=result,interpretation='Acceptance within fixed bands; a failed gate is not proof of non-equivalence or failure of every gate-dependent method')
+29
View File
@@ -0,0 +1,29 @@
"""Local immutable barriers for explicitly approved /home run roots."""
import fcntl,json,shutil,subprocess,hashlib
from pathlib import Path
from phase2_h1_worker_v2 import sha,checked,save
from phase2_local_freeze_io import STORE,LOCK,BRANCH,BASE
HOME_RUNS=Path('/home/frank14f/MeltMind-runs')
def key(path):
p=Path(path).resolve()
if p.is_relative_to(HOME_RUNS):return 'home-runs/'+str(p.relative_to(HOME_RUNS))
return str(p.relative_to(BASE))
def freeze(root,name,files):
with LOCK.open('a') as lock:
fcntl.flock(lock,fcntl.LOCK_EX)
def git(*a):return subprocess.check_output(['git','-C',str(STORE),'-c','gc.auto=0',*a],text=True).strip()
if git('branch','--show-current')!=BRANCH or git('status','--porcelain'):raise RuntimeError('Local writer not clean')
hashes={};paths={}
for f in files:
f=Path(f).resolve();k=key(f);p=STORE/k;p.parent.mkdir(parents=True,exist_ok=True);shutil.copyfile(f,p);hashes[str(f)]=sha(f);paths[str(f)]=k
subprocess.run(['git','-C',str(STORE),'add','--',*paths.values()],check=True)
if git('diff','--cached','--name-only'):subprocess.run(['git','-C',str(STORE),'-c','user.name=MeltAstra','-c','user.email=meltrastra@local','commit','-q','-m','Immutable home run '+root.name+' '+name],check=True)
commit=git('rev-parse','HEAD')
for p,h in hashes.items():
if hashlib.sha256(subprocess.check_output(['git','-C',str(STORE),'show',commit+':'+paths[p]])).hexdigest()!=h:raise ValueError('Git content mismatch')
save(root/(name+'-git-freeze.json'),dict(commit=commit,files=hashes,git_paths=paths,remote_published=False,local_repository=str(STORE)))
save(root/(name+'-authorization.json'),dict(contract_sha256=sha(root/'contract.json'),proof_sha256=sha(root/(name+'-git-freeze.json'))))
def verify(root,name):
a=json.loads((root/(name+'-authorization.json')).read_text());checked(root/'contract.json',a['contract_sha256']);p=checked(root/(name+'-git-freeze.json'),a['proof_sha256']);d=json.loads(p.read_text())
for f,h in d['files'].items():checked(f,h)
return d
@@ -0,0 +1,30 @@
"""Prepare only; never launches model prediction, score or a waiter."""
import json,time
from pathlib import Path
from phase2_fixed_io import BASE
from phase2_h1_worker_v2 import checked,sha,save
from phase2_home_freeze_io import freeze,verify
from phase2_gate_outcome_postfit import config
def read(p):return json.loads(Path(p).read_text())
def main():
root=Path('/home/frank14f/MeltMind-runs/phase2-gate-outcome-v2-postfit-2026-11-17')
if root.exists():raise FileExistsError('Existing postfit root: inspect, do not duplicate')
prep_root=Path('/home/frank14f/MeltMind-runs/phase2-gate-outcome-calibration-v2-2026-11-17');prep=read(prep_root/'contract.json');verify(prep_root,'launch');verify(prep_root,'comparison-contract');cmp=read(prep_root/'comparison-contract.json');groot=BASE/'artifacts/phase2-b13-e1b-2026-11-17';science=read(groot/'contract.json')
assert sha(prep_root/'contract.json')==cmp['preparation_contract_sha256'];assert prep['total_cpu_seconds']==72000 and cmp['rule']['approval']=='1794917650.044649'
sources={str(p):sha(p) for p in [prep_root/'contract.json',prep_root/'comparison-contract.json',groot/'contract.json',groot/'predictions-manifest.json',groot/'score/result.json']}
for m in cmp['models']:
for arm in ['A','G']:sources[m[arm]['path']]=m[arm]['sha256']
for p,h in sources.items():checked(p,h)
codes={str(BASE/p):h for p,h in prep['code_sha256'].items()}
for p in ['tools/phase2_gate_outcome_statistics.py','tools/phase2_gate_outcome_review.py','tools/phase2_gate_outcome_postfit.py','tools/prepare_phase2_gate_outcome_postfit.py','tools/test_phase2_gate_outcome_postfit.py','docs/phase2-gate-outcome-v2-postfit-2026-11-17.md']:
codes[str(BASE/p)]=sha(BASE/p)
c=dict(version='gate-outcome-v2-postfit-v1',approval='1794916421.489089',comparison_approval='1794917650.044649',total_cpu_seconds=72000,postfit_cpu_seconds=3600,setup_cpu_seconds=120,worker_cpu=8,disk_bytes=3*2**30,new_fits=0,candidate_label_reads=0,preparation_root=str(prep_root),preparation_contract=str(prep_root/'contract.json'),comparison_contract=str(prep_root/'comparison-contract.json'),comparison_contract_sha256=sha(prep_root/'comparison-contract.json'),comparison_rule=cmp['rule'],bands=cmp['bands'],parts=science['parts'],original_score=str(groot/'score/result.json'),source_sha256=sources,code_sha256=codes,new_prediction_parts=493,exact_reuse_parts=52,invalid_gate_parts=7,full_original_terminal_required=True,predictions_freeze_before_score=True,actual_semantic_PASS_required=True,no_retry=True)
root.mkdir();save(root/'contract.json',c)
save(root/'acceptance.json',dict(status='PASS',tests=18,elapsed_seconds=1.606,command='python -m unittest tools.test_phase2_gate_outcome_postfit',scope='8 metric/paired grouping tests,4 prediction/budget/terminal tests,6 actual file/Git semantic review tests; no validator or freeze mock, temporary path constants only',test_source_sha256=sha(BASE/'tools/test_phase2_gate_outcome_postfit.py'),new_actual_predictions=0,new_comparisons=0,setup_cpu_charge=120,unix=time.time()))
save(root/'preparation-audit.json',dict(status='PREPARED_NOT_LAUNCHED',original_feature_pid=666520,original_feature_pins_verified=True,comparison_rule_fixed=True,models_hash_checked=10,no_labels_opened=True,no_prediction_or_comparison=True,no_new_waiter=True,unix=time.time()))
freeze(root,'implementation',[root/'contract.json',root/'acceptance.json',root/'preparation-audit.json',*[Path(p) for p in codes],prep_root/'comparison-contract.json'])
proof=verify(root,'implementation');binding=dict(reviewed_commit=proof['commit'],contract_sha256=sha(root/'contract.json'),implementation_proof_sha256=sha(root/'implementation-git-freeze.json'),comparison_contract_sha256=c['comparison_contract_sha256'],code_sha256=codes)
save(root/'expected-review-binding.json',binding);freeze(root,'review-binding',[root/'contract.json',root/'expected-review-binding.json',root/'implementation-git-freeze.json',root/'implementation-authorization.json']);config(root)
print(json.dumps(dict(root=str(root),implementation_commit=proof['commit'],contract_sha256=sha(root/'contract.json'),implementation_proof_sha256=sha(root/'implementation-git-freeze.json'),binding_sha256=sha(root/'expected-review-binding.json'),actual_PASS_pending=True,no_launch=True)))
if __name__=='__main__':main()
+94
View File
@@ -0,0 +1,94 @@
import copy,json,tempfile,unittest,subprocess
from pathlib import Path
from unittest.mock import patch
import numpy as np
import phase2_home_freeze_io as fio
from phase2_h1_worker_v2 import sha
from phase2_gate_outcome_review import validate_review
from phase2_gate_outcome_statistics import summarize,two_metrics
from phase2_gate_outcome_postfit import predict_arrays,query_identity,remaining_cap,terminal_inputs
from s2_custody_statistics import part_metrics
def metric(v):return dict(relative_l2=dict(value=v),last_location=dict(value=v))
def row(case,component,new,old):
return dict(case=case,component_id=component,label_gate=None if new is None else {a:metric(new) for a in ['A','G']},LOG_gate=None if old is None else {a:metric(old) for a in ['A','G']})
class Statistics(unittest.TestCase):
def test_partial_component_kept_only_empty_component_drops(self):
d=summarize([row('a','x',1,1),row('b','x',None,1),row('c','y',None,1)],['x','y']);v=d['comparisons']['A']['relative_l2']
self.assertEqual(v['component_part_counts'],{'x':1});self.assertEqual(v['excluded_components'],['y']);self.assertEqual(v['paired_parts'],1);self.assertEqual(d['decision'],'PASS_OUTCOME_EQUIVALENCE')
def test_paired_parts_not_independent_arm_averages(self):
d=summarize([row('a','x',1,1),row('b','x',10,None),row('c','x',None,30)],['x']);v=d['comparisons']['A']['relative_l2'];self.assertEqual(v['label_mean'],1);self.assertEqual(v['LOG_mean'],1)
def test_equal_component_not_equal_part_weighting(self):
d=summarize([row('a','x',2,1),row('b','x',2,1),row('c','y',1,1)],['x','y']);self.assertEqual(d['comparisons']['G']['relative_l2']['label_minus_LOG']['value'],.5)
def test_both_models_both_metrics_required(self):
r=row('a','x',0,0);r['label_gate']['G']['last_location']['value']=.011;d=summarize([r],['x']);self.assertFalse(d['accepted_for_v2']);self.assertTrue(d['comparisons']['A']['last_location']['passed'])
def test_closed_band_boundaries_and_zero_support(self):
r=row('a','x',.001,0);d=summarize([r],['x']);self.assertTrue(d['accepted_for_v2']);self.assertFalse(summarize([row('a','x',None,1)],['x'])['accepted_for_v2'])
def test_duplicates_and_nonfinite_rejected(self):
r=row('a','x',0,0)
with self.assertRaises(ValueError):summarize([r,r],['x'])
with self.assertRaises(ValueError):summarize([row('a','x',float('nan'),0)],['x'])
def test_metrics_match_original_with_weighted_counts_and_raw_ties(self):
p=np.array([2.,2.,1.]);t=np.array([1.,3.,3.]);q=dict(counts=np.array([1,3,2]),raw_ids=np.array([5,2,8]),coordinates_over_Rg=np.eye(3),near_wall=np.array([True,False,True]));a=two_metrics(p,t,q);b=part_metrics(p,t,q['counts'],q['near_wall'],q['raw_ids'],q['coordinates_over_Rg'],1)
self.assertEqual(a['relative_l2'],b['relative_l2']);self.assertEqual(a['last_location'],b['last_location'])
def test_zero_truth_denominator_retains_undefined(self):
q=dict(counts=np.ones(2),raw_ids=np.arange(2),coordinates_over_Rg=np.zeros((2,3)));d=two_metrics(np.ones(2),np.zeros(2),q);self.assertIsNone(d['relative_l2']['value'])
class PredictAndBudget(unittest.TestCase):
def test_native_nan_query_order_and_seconds(self):
class Model:
def __init__(self,n):self.n=n
def predict(self,x):
assert x.dtype==np.float32 and x.shape[1]==self.n
if self.n==21:assert np.isnan(x[:,-2:]).all()
return x[:,0]+self.n
f=np.arange(57,dtype=float).reshape(3,19);q={'raw_ids':np.array([2,9])};out=predict_arrays(Model(19),Model(21),f,np.full((3,2),np.nan),np.array([2,5,9]),q,.4)
np.testing.assert_allclose(out['A'],np.array([19,57])*.4);np.testing.assert_allclose(out['G'],np.array([21,59])*.4)
def test_query_mismatch_fails_before_predict(self):
with self.assertRaises(ValueError):predict_arrays(None,None,np.zeros((2,19)),np.zeros((2,2)),np.array([1,2]),{'raw_ids':np.array([3])},1)
a=dict(raw_ids=np.arange(2),counts=np.ones(2),coordinates_over_Rg=np.zeros((2,3)),near_wall=np.zeros(2));b=dict(a,counts=np.array([1,2]))
with self.assertRaises(ValueError):query_identity(a,b)
def test_shared_cost_charged_once_and_both_caps_enforced(self):
self.assertEqual(remaining_cap(68000,120,600),600)
with self.assertRaises(RuntimeError):remaining_cap(71600,120,300)
with self.assertRaises(RuntimeError):remaining_cap(10000,3400,300)
def test_missing_feature_freeze_rejects_adoption(self):
with tempfile.TemporaryDirectory() as p:
with self.assertRaises(FileNotFoundError):terminal_inputs(Path(p),{'preparation_root':p},{})
class SemanticReview(unittest.TestCase):
def setUp(self):
self.tmp=tempfile.TemporaryDirectory();p=Path(self.tmp.name);self.base=p/'base';self.home=p/'runs';self.store=p/'store'
for d in [self.base,self.home,self.store]:d.mkdir()
subprocess.run(['git','init','-q','-b','codex/test',str(self.store)],check=True)
self.patches=[patch.object(fio,k,v) for k,v in dict(BASE=self.base,HOME_RUNS=self.home,STORE=self.store,LOCK=p/'lock',BRANCH='codex/test').items()]
for x in self.patches:x.start()
self.root=self.home/'postfit';self.root.mkdir();self.code=self.base/'code.py';self.code.write_text('fixed code');self.comparison=self.base/'comparison.json';self.comparison.write_text('{"band":0.001}')
self.c=dict(comparison_contract=str(self.comparison),comparison_contract_sha256=sha(self.comparison),code_sha256={str(self.code):sha(self.code)})
(self.root/'contract.json').write_text(json.dumps(self.c));fio.freeze(self.root,'implementation',[self.root/'contract.json',self.code,self.comparison]);proof=fio.verify(self.root,'implementation')
self.binding=dict(reviewed_commit=proof['commit'],contract_sha256=sha(self.root/'contract.json'),implementation_proof_sha256=sha(self.root/'implementation-git-freeze.json'),comparison_contract_sha256=sha(self.comparison),code_sha256=self.c['code_sha256'])
(self.root/'expected-review-binding.json').write_text(json.dumps(self.binding));self.good=dict(verdict='PASS',**self.binding)
def tearDown(self):
for x in self.patches:x.stop()
self.tmp.cleanup()
def receipt(self,d,include=True):
(self.root/'review.json').write_text(json.dumps(d));files=[self.root/'contract.json']
if include:files.extend([self.root/'review.json',self.root/'expected-review-binding.json'])
fio.freeze(self.root,'implementation-review',files)
def test_actual_frozen_pass(self):self.receipt(self.good);self.assertEqual(validate_review(self.root)['verdict'],'PASS')
def test_fail_missing_and_unrelated_receipts(self):
for v in [dict(self.good,verdict='FAIL'),self.binding,{'unrelated':'review'}]:
self.receipt(v)
with self.assertRaises(PermissionError):validate_review(self.root)
def test_wrong_exact_identities(self):
for k in ['reviewed_commit','contract_sha256','implementation_proof_sha256','comparison_contract_sha256','code_sha256']:
v=dict(self.good);v[k]={} if k=='code_sha256' else 'wrong';self.receipt(v)
with self.assertRaises(PermissionError):validate_review(self.root)
def test_review_omitted_from_freeze(self):
self.receipt(self.good,False)
with self.assertRaises(PermissionError):validate_review(self.root)
def test_changed_source_rejected(self):
self.receipt(self.good);self.code.write_text('changed')
with self.assertRaises(ValueError):validate_review(self.root)
def test_changed_comparison_rejected(self):
self.receipt(self.good);self.comparison.write_text('changed')
with self.assertRaises(ValueError):validate_review(self.root)
if __name__=='__main__':unittest.main()