diff --git a/stage7/stage7b_v2_source_anchor_repeatability_diagnostic_v1/README.md b/stage7/stage7b_v2_source_anchor_repeatability_diagnostic_v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..3cc6924b466597498bca32a02fe0074f08e0cd38 --- /dev/null +++ b/stage7/stage7b_v2_source_anchor_repeatability_diagnostic_v1/README.md @@ -0,0 +1,107 @@ +# Stage 7b v2 source-anchor diagnosis + +Status: the formal v2a run is **indeterminate** and immutable. It completed all +four source evaluations, failed the preregistered attacked-anchor gate, and +stopped before any specialist or shared treatment training. The frozen +`1e-9` tolerance remains unchanged. + +## Formal boundary + +| Field | Frozen value | +|---|---| +| Git commit | `369d609` | +| Protocol SHA-256 | `3720fb0b19bfc240a195f3ae4e3bf4e54d098e42070bf18ee5e31b3e0da6e3e7` | +| L40S output | `/root/data/runs/stage7b_per_vehicle_vs_shared_continuation_l40s_v2a` | +| Decision | `source_anchor_indeterminate_no_treatment_training_started` | +| Optimization updates | `0` | +| Completed source vehicles | `4/4` | +| Specialist/shared arms started | no | + +All clean source anchors passed exactly. All four source textures matched their +frozen tensor hashes, all recalls matched exactly, the detector remained frozen, +and all prediction/texture values were finite. The attacked AP anchors did not +meet the preregistered absolute tolerance: + +| Vehicle | Expected attacked AP50:95 | Observed attacked AP50:95 | Absolute difference | AP50 absolute difference | Recall difference | +|---|---:|---:|---:|---:|---:| +| Audi E-Tron | 0.230838334468 | 0.230855441453 | 1.71069849e-5 | 1.76543594e-6 | 0 | +| Tesla Model 3 | 0.208439495225 | 0.208427824712 | 1.16705126e-5 | 7.28626890e-5 | 0 | +| Ford Mustang | 0.268039408449 | 0.268035550513 | 3.85793510e-6 | 5.03712717e-6 | 0 | +| Jeep Wrangler Rubicon | 0.017725985776 | 0.017704725282 | 2.12604948e-5 | 1.39422143e-5 | 0 | + +These are small numerical differences, but their size does not override the +frozen validity rule. No Stage 7b capacity, oracle-recovery, or specialization +outcome exists. + +## Current diagnosis + +The formal Stage 7a source endpoint ran at commit +`4ec8e18888d6130edafdcff8d497953814d8d7d6` on the same named L40S host runtime +with Python 3.10.20 and PyTorch 2.4.0. A repository diff from that commit to the +v2a commit is empty for the endpoint's core evaluation, renderer, detector, +dataset, and detection-metric code: + +```text +method/evaluationCamouflage.py +renderLib/ +detectors/ +dataset/ +utils/detection_metrics.py +``` + +The exact clean endpoint and exact generated-texture hashes narrow the observed +drift to the rendered attacked path, or to detector sensitivity to a minute +change produced by that path. They do not identify a unique CUDA, rasterizer, +interpolation, conversion, or detector kernel. The frozen numerical environment +also records `deterministic_algorithms=false` and `cudnn_deterministic=false`, so +bitwise endpoint reproduction cannot be assumed merely from matching package and +GPU identities. + +## Frozen source-only diagnostic + +Before observing any repeat result, the next diagnostic is fixed as exactly two +additional evaluations of all four source seed-42 endpoints. Together with v2a, +this gives three current-runtime observations per vehicle. The diagnostic: + +- uses the exact v2a source runtime configs; +- changes only `evaluation.results_folder_name` to fresh outputs; +- uses the same checkpoint, four manifests, scene statistics, seed, renderer, + detector, thresholds, and AP code; +- performs no optimizer construction or update; +- cannot call specialist/shared training; +- cannot relax `1e-9`, replace the Stage 7a anchor, or authorize a formal result; +- cannot access or use Nissan for any selection. + +Interpretation is frozen before the repeats: + +1. If attacked prediction-row hashes vary while clean rows and texture hashes + remain exact, the evidence establishes current-runtime non-repeatability in + the rendered attacked endpoint pipeline. +2. If all current attacked rows repeat exactly but remain different from the + historical anchor, the evidence establishes a stable current-versus-historical + endpoint difference, without uniquely locating its cause. +3. Any diagnostic failure remains diagnostic; it cannot be converted into a + training authorization. + +The diagnostic result will determine what protocol redesign should be sent for +a narrow external review. It will not retroactively validate the stopped v2a +run. + +The first diagnostic launch attempt at commit `d08b0ab` stopped before creating +its output directory or loading a detector/dataset. Its input validator assumed +that JSON object keys retained protocol order, while the canonical JSON writer +sorts them alphabetically. The nuisance-only repair validates the exact vehicle +set and then reconstructs protocol order; repeat count, inputs, endpoint code, +interpretation, and all prohibitions remain unchanged. + +## Evidence + +The complete immutable v2a source artifacts, including all four prediction-row +tables, evaluation summaries, configs, decision, preflight, and runtime-access +record, are archived under: + +`experiments/provenance/stage7_20260804/stage7b_v2_source_anchor_indeterminate_v2a/` + +This document follows the paper claim boundary: observations are separated from +causal interpretations, and no Stage 7b treatment claim is made from a failed +validity gate. diff --git a/stage7/stage7b_v2_source_anchor_repeatability_diagnostic_v1/SHA256SUMS b/stage7/stage7b_v2_source_anchor_repeatability_diagnostic_v1/SHA256SUMS new file mode 100644 index 0000000000000000000000000000000000000000..6495ad1826d6d226aa5ac0aa10066583e3436619 --- /dev/null +++ b/stage7/stage7b_v2_source_anchor_repeatability_diagnostic_v1/SHA256SUMS @@ -0,0 +1,95 @@ +aed3401575a97cd76fa044ec37aef5d636627da83017e2c0d881dfe3aa62b990 README.md +868a7544070df88cd7d1fdb195ca71d3df044b12eb9a9af56d3be2a30d40e982 diagnostic_v1/console.log +c8cf030abdb9c86163faf205febb352a3c009591b90614fe83106ad85828dbff diagnostic_v1/decision.json +a2e8ba81b6e86477646f30f083618721b50e5117fb9ebc5d9af374166dfa33bf diagnostic_v1/partial_repeat_results.json +30615ea9612959a70d0346c29cf6251c433ea4b17d9f062df572ca9586ba355f diagnostic_v1/preflight.json +7d51dc85dac9c39ae999c7df25f2a091e95be9299a08a322a16d5bab3a47de86 diagnostic_v1/repeat_01/source/audi_etron/evaluation/per_seed/42/samples.csv +2f7761e54e9d3179578dcda6e121857bea9b144e702f31c493468ad82651d1f1 diagnostic_v1/repeat_01/source/audi_etron/evaluation/per_seed/42/summary.json +7d51dc85dac9c39ae999c7df25f2a091e95be9299a08a322a16d5bab3a47de86 diagnostic_v1/repeat_01/source/audi_etron/evaluation/samples.csv +73ee5fae14ad22b0c56da225287b6aaa96edc464cd875a6c84b26003f421b3dc diagnostic_v1/repeat_01/source/audi_etron/evaluation/summary.json +e13549317dab7fe3621ec902f7b19f77550c3c6ed4dc890c9796c8ba18ffd57f diagnostic_v1/repeat_01/source/audi_etron/source_evaluation_result.json +5a89ddb6ed48519af627411eee16cc6a482684990a4b7a7a79da825736a3339b diagnostic_v1/repeat_01/source/audi_etron/source_evaluation_runtime.json +30000468f1f8834d577cf50f9b2ae8ad287fb6f34e7f61a0a6cb035e28eb1660 diagnostic_v1/repeat_01/source/audi_etron/source_prediction_rows.json +69caed9ece06a08a5f55a188151ebae876555be4dc72ad9d50d7e284a8a5c72d diagnostic_v1/repeat_01/source/ford_mustang/evaluation/per_seed/42/samples.csv +d36e5d4f63eac600118d5a352e3057bb9e8f4f0afeab7017600a3d492be6330c diagnostic_v1/repeat_01/source/ford_mustang/evaluation/per_seed/42/summary.json +69caed9ece06a08a5f55a188151ebae876555be4dc72ad9d50d7e284a8a5c72d diagnostic_v1/repeat_01/source/ford_mustang/evaluation/samples.csv +28745f4f3f069afcb913075caeb2aa69c4a7e46147388a5c7def7fef51be4a18 diagnostic_v1/repeat_01/source/ford_mustang/evaluation/summary.json +22c093fdc5ad59dce2ffc85daef39034f18a30bc947c567ac409faa9e23021e0 diagnostic_v1/repeat_01/source/ford_mustang/source_evaluation_result.json +33dcd20534b1bdbf4f24d029ee0fb08374a7c62026ba9b46c3f5901d2324f716 diagnostic_v1/repeat_01/source/ford_mustang/source_evaluation_runtime.json +c44ca97bce835c544f99c4b30825f65df737eed34aab3acf24bac3393012f743 diagnostic_v1/repeat_01/source/ford_mustang/source_prediction_rows.json +038fe76b3d7738b2bf7a4c31c5cf1c03517006419e4148614c18bf6c643beb93 diagnostic_v1/repeat_01/source/jeep_wrangler_rubicon/evaluation/per_seed/42/samples.csv +0ace7d0648f3dabaf4ad145a509b91d33a56869447ea3753fb58414e6abc939f diagnostic_v1/repeat_01/source/jeep_wrangler_rubicon/evaluation/per_seed/42/summary.json +038fe76b3d7738b2bf7a4c31c5cf1c03517006419e4148614c18bf6c643beb93 diagnostic_v1/repeat_01/source/jeep_wrangler_rubicon/evaluation/samples.csv +4d56dbc035b310862b2ce63f8c4ecdc776a56af9316f350e48d1f1dce3524da3 diagnostic_v1/repeat_01/source/jeep_wrangler_rubicon/evaluation/summary.json +4320224d46929ff50e79749b5e646fc3bac46f645342108062f61ffae68c7435 diagnostic_v1/repeat_01/source/jeep_wrangler_rubicon/source_evaluation_result.json +62b1c5c1422d9883fca85585551c5e22649f7099383818a21327a33e7a60ad45 diagnostic_v1/repeat_01/source/jeep_wrangler_rubicon/source_evaluation_runtime.json +2b61134b4de633934af5a53380d70fe2006c59a03caad7185bab842050503ada diagnostic_v1/repeat_01/source/jeep_wrangler_rubicon/source_prediction_rows.json +3bc78832c244c766598554028e06ec7b03cd6f4fb49855378b5b80902a1a63bd diagnostic_v1/repeat_01/source/tesla_model3/evaluation/per_seed/42/samples.csv +5a727924ca7cb162c69d497470177ee432a43f16dcac4b21d5c598068c6d5236 diagnostic_v1/repeat_01/source/tesla_model3/evaluation/per_seed/42/summary.json +3bc78832c244c766598554028e06ec7b03cd6f4fb49855378b5b80902a1a63bd diagnostic_v1/repeat_01/source/tesla_model3/evaluation/samples.csv +ed56831e7481e46ba98283db0f034673659be76cf5ea6d01540e1a35143758ef diagnostic_v1/repeat_01/source/tesla_model3/evaluation/summary.json +1ec6251de55528d9d8d90aeb18727a3b52804d9be446b84a07741cc52d3862e6 diagnostic_v1/repeat_01/source/tesla_model3/source_evaluation_result.json +7561a94a0e1acf5c4deefd50650b8c3029db85aa4341114d3738e305c218f2a7 diagnostic_v1/repeat_01/source/tesla_model3/source_evaluation_runtime.json +6982361f14b9ba7b14c92cea2caf96f1962e210cb1b72ce5432bfff96dadef17 diagnostic_v1/repeat_01/source/tesla_model3/source_prediction_rows.json +763a2cb74d87b9e278356632265278344391906b68670144ac9383535c04fee2 diagnostic_v1/repeat_02/source/audi_etron/evaluation/per_seed/42/samples.csv +f3be54757e2f088e3ef8eb3f5c410d01607220d19e8ff826997923fe986666ca diagnostic_v1/repeat_02/source/audi_etron/evaluation/per_seed/42/summary.json +763a2cb74d87b9e278356632265278344391906b68670144ac9383535c04fee2 diagnostic_v1/repeat_02/source/audi_etron/evaluation/samples.csv +d14c2b47fedf2c0ce0ec9e4b8f0e8685a15a07e25b4ee733e62188060fcea822 diagnostic_v1/repeat_02/source/audi_etron/evaluation/summary.json +eb4520eaa2b0049273bd1a30c308fb0720044369838b0c41f1ae34d42d0beeb2 diagnostic_v1/repeat_02/source/audi_etron/source_evaluation_result.json +70a4f7a8897a4678dcd1ebe2e4bc37147adf458f97137f21c8227e5f903f9c37 diagnostic_v1/repeat_02/source/audi_etron/source_evaluation_runtime.json +0ccd0ebcf541149842b1d23ca5be8d2cdcd9407e2739f6e672bba953c9447b53 diagnostic_v1/repeat_02/source/audi_etron/source_prediction_rows.json +9e365989fd1451489375b6487f7b12c1ad827abe1a631a3de2620d517f38b323 diagnostic_v1/repeat_02/source/ford_mustang/evaluation/per_seed/42/samples.csv +6817416b718cc7e4d1c0d82d337924ece49a5efebda5a391be6ab1efb2816d91 diagnostic_v1/repeat_02/source/ford_mustang/evaluation/per_seed/42/summary.json +9e365989fd1451489375b6487f7b12c1ad827abe1a631a3de2620d517f38b323 diagnostic_v1/repeat_02/source/ford_mustang/evaluation/samples.csv +561a0a25ace20bac8acdf7443da687b9771db9c29b53f538e2245a81fd8ed432 diagnostic_v1/repeat_02/source/ford_mustang/evaluation/summary.json +a0b8f30ffa7cc99f4e17aed44c552c20cbd5c9abe15208b25b93540d117d25a6 diagnostic_v1/repeat_02/source/ford_mustang/source_evaluation_result.json +99892a5adb25a5ef62bbdab4d1dcf6f33b731e75e2d1877373e5418576142547 diagnostic_v1/repeat_02/source/ford_mustang/source_evaluation_runtime.json +be058b645bc3da1c3aba2f165b985afa3acd54967975d920fa03f3190f738e84 diagnostic_v1/repeat_02/source/ford_mustang/source_prediction_rows.json +33265fd82e3bd66397c2dbd3575ad94534b5d71ffd815cb82cc18f30fd2454e0 diagnostic_v1/repeat_02/source/jeep_wrangler_rubicon/evaluation/per_seed/42/samples.csv +6bf40d6dc1d244293767e504b15ddf633da926cd8fb500501cf74527f05a54d5 diagnostic_v1/repeat_02/source/jeep_wrangler_rubicon/evaluation/per_seed/42/summary.json +33265fd82e3bd66397c2dbd3575ad94534b5d71ffd815cb82cc18f30fd2454e0 diagnostic_v1/repeat_02/source/jeep_wrangler_rubicon/evaluation/samples.csv +e962be926ded91b84dd59f53d5e269e516e34aafb8200ec5e5a73ed7b3e57b13 diagnostic_v1/repeat_02/source/jeep_wrangler_rubicon/evaluation/summary.json +e9c22f3b98e6f5f879aee48a65b541d56e7bcc91e1a4c39a088acfe2803297c2 diagnostic_v1/repeat_02/source/jeep_wrangler_rubicon/source_evaluation_result.json +62eb0a52459acaea46108bfa23cc12f22d093c80c38bd27b458ea3ee987ed39c diagnostic_v1/repeat_02/source/jeep_wrangler_rubicon/source_evaluation_runtime.json +4c930baa5a9a46f99121b6915f92692844e68cc1f7178953ed83002654e7f532 diagnostic_v1/repeat_02/source/jeep_wrangler_rubicon/source_prediction_rows.json +2d13213f912c8754d2dd0e6098039413669ceefb4c00cf7fc76da0f92821ffaf diagnostic_v1/repeat_02/source/tesla_model3/evaluation/per_seed/42/samples.csv +4a61da0cbd3a521c138862a8d88d9be2eb78aabf6d3e4121a55492fc42862d5a diagnostic_v1/repeat_02/source/tesla_model3/evaluation/per_seed/42/summary.json +2d13213f912c8754d2dd0e6098039413669ceefb4c00cf7fc76da0f92821ffaf diagnostic_v1/repeat_02/source/tesla_model3/evaluation/samples.csv +00b6ca72fbf2e51057195c1c9f2199c698fe995e041150be294d7f80aa2579e2 diagnostic_v1/repeat_02/source/tesla_model3/evaluation/summary.json +5408874bd36b2dbab2d639e5fa739e8262e48df964140708a77706c841a30117 diagnostic_v1/repeat_02/source/tesla_model3/source_evaluation_result.json +650e05d4f7b6391529994ef3dbd83ce85adc3f87bce65ebe2ebcfe04d7cfa7e7 diagnostic_v1/repeat_02/source/tesla_model3/source_evaluation_runtime.json +6e7142cc0f45869a9a869ad54d7be1466cd3fd3992fa4cafa938ea9d897d4293 diagnostic_v1/repeat_02/source/tesla_model3/source_prediction_rows.json +3cbebbd0a7bb29cba70134ab80ecf954cb6ce9a458905bb69697265201c95f23 diagnostic_v1/runtime_file_access.json +9dc7b3aa3a211603e990824a3262a3395ae74e52267a4e1de3421e003a21a0c5 formal_v2a/decision.json +79114918e6cc9e803073633250f7c304f4354d10f3c9e95898e45563adb6e283 formal_v2a/partial_source_results.json +4c629bce6baae9ab2fc5df6a022d293073a889599d1eea8305f33c2b7106b9d0 formal_v2a/preflight.json +3a0cf254fcd37ccdfbe3f5422ae6ec971ecaa6d80f7adb1e5ce6897d01451613 formal_v2a/runtime_file_access.json +d26f947413dd88a565fec87abe06e90450916f3dce930201f0c69b9d13a5f708 formal_v2a/source/audi_etron/evaluation/per_seed/42/samples.csv +967ac6ea99a9aa8d817a69eb988feb5d2c9110bae488b1d82ad789e05d4075c1 formal_v2a/source/audi_etron/evaluation/per_seed/42/summary.json +d26f947413dd88a565fec87abe06e90450916f3dce930201f0c69b9d13a5f708 formal_v2a/source/audi_etron/evaluation/samples.csv +25b61966392b0a7bec1cb29bb91f42eeffcad17cbbfc1e3fa2cfb9f38079d3c2 formal_v2a/source/audi_etron/evaluation/summary.json +aa88070151c3db93f410da45c84f6b66f2ce9bd857f2c0b630c340f70a822b47 formal_v2a/source/audi_etron/source_evaluation_result.json +032cfbf46dd494f3508357940c61af980ef28fb513adb36a59264394fbe9d39e formal_v2a/source/audi_etron/source_evaluation_runtime.json +e025b4a564684cecc832bb76515ceac58ee5665c4db4f8a46d971d1f48908553 formal_v2a/source/audi_etron/source_prediction_rows.json +ab106ff362726d8edd680f0b2c86499a0abdd040c63abf715f722e60fec4e5e0 formal_v2a/source/ford_mustang/evaluation/per_seed/42/samples.csv +006744fbe6f714de4e7229069ec1da0d5bb724fa84c1ed18a0123c22a842b83c formal_v2a/source/ford_mustang/evaluation/per_seed/42/summary.json +ab106ff362726d8edd680f0b2c86499a0abdd040c63abf715f722e60fec4e5e0 formal_v2a/source/ford_mustang/evaluation/samples.csv +b4c19b750650ddbe4034dff6a77e063536f79d2d8eb352972a55a871be399c48 formal_v2a/source/ford_mustang/evaluation/summary.json +467105e0270573e96a070861b4097a3197f6ac47a9fd736f1cd00cc1e528f714 formal_v2a/source/ford_mustang/source_evaluation_result.json +cf341b312005aee6fe6e11bc98c5a5974cdabd42b4f498e44a8c18ec53ee48f8 formal_v2a/source/ford_mustang/source_evaluation_runtime.json +ac106c57c98f5f689bc9af5f4458af65325f73fea3735b3a5df0737c5b450e7c formal_v2a/source/ford_mustang/source_prediction_rows.json +a40e769e81c74ac4527a1b3f539c24eb880ef932ec5e07eca24c6688bf71575c formal_v2a/source/jeep_wrangler_rubicon/evaluation/per_seed/42/samples.csv +cbb75f6fcb46b372e2eb15228b5594cec043bab58fdb211372cf1422a0c8811b formal_v2a/source/jeep_wrangler_rubicon/evaluation/per_seed/42/summary.json +a40e769e81c74ac4527a1b3f539c24eb880ef932ec5e07eca24c6688bf71575c formal_v2a/source/jeep_wrangler_rubicon/evaluation/samples.csv +9cbc53353eea3f213a1f71fce2b0eb64848b956a2488f606c514e2ce2df91893 formal_v2a/source/jeep_wrangler_rubicon/evaluation/summary.json +504881ce9ed2ebb5df4246b1eed5cd65ccbc7dd0e618e06853fc441c142122b7 formal_v2a/source/jeep_wrangler_rubicon/source_evaluation_result.json +97a5ece811b497c4e45c013e7062b9f358a51fc565ee11dd8f43809460a3caeb formal_v2a/source/jeep_wrangler_rubicon/source_evaluation_runtime.json +cdf77dbc2fdda0977fdf388e88f313a41e93d342cbb523916c1f6a3d40b18470 formal_v2a/source/jeep_wrangler_rubicon/source_prediction_rows.json +a09891fd63a44e327c28b25187a8f5f1c579eeb3fa8805c745688920a853a734 formal_v2a/source/tesla_model3/evaluation/per_seed/42/samples.csv +2fea7273a05e541f1f50f8ec4994b3254035248a119f3967dcf69da0d4e3347c formal_v2a/source/tesla_model3/evaluation/per_seed/42/summary.json +a09891fd63a44e327c28b25187a8f5f1c579eeb3fa8805c745688920a853a734 formal_v2a/source/tesla_model3/evaluation/samples.csv +08b51b1293b4707bdf6ff697cbfbf23a03762eb4c726a8115c5899efe1ab70be formal_v2a/source/tesla_model3/evaluation/summary.json +9325965ea3c9febcfa4d40740cd42dc07fc525e1ec023db2fb8a3b5763a9c41c formal_v2a/source/tesla_model3/source_evaluation_result.json +a2be396185e621c45467e29078d2929acae3caea91796ec15a011740853a0c45 formal_v2a/source/tesla_model3/source_evaluation_runtime.json +db3ff99694f773e79cebc4895285755237cc2a546887dc46910f960922c39cdd formal_v2a/source/tesla_model3/source_prediction_rows.json +398fca7ea0d66797c937c28bf3748af24ef0a658f0a70335a78d23bfc4c2ebba pre_outcome_failed_attempt/pre_outcome_failed_attempt_d08b0ab.txt diff --git a/stage7/stage7b_v2_source_anchor_repeatability_diagnostic_v1/diagnostic_v1/console.log b/stage7/stage7b_v2_source_anchor_repeatability_diagnostic_v1/diagnostic_v1/console.log new file mode 100644 index 0000000000000000000000000000000000000000..92f9f11c8836ed1ad5c0a6850b34323822d6a102 --- /dev/null +++ b/stage7/stage7b_v2_source_anchor_repeatability_diagnostic_v1/diagnostic_v1/console.log @@ -0,0 +1,6059 @@ +2026-08-04 17:45:50,111 - INFO - enable console log +2026-08-04 17:45:50,111 - INFO - enable file log,the log file is in the log +2026-08-04 17:45:50,416 - INFO - Try to loading data from /root/data/hf-releases/drc-data-0664d82e/raw/stage2_town10_v1. +2026-08-04 17:45:50,473 - INFO - Detector: YoloV5 is selected,trying to load the detector. +2026-08-04 17:45:51,817 - INFO - Try to loading weight named yolov5x6.pt +Fusing layers... +YOLOv5x6 summary: 574 layers, 140730220 parameters, 0 gradients, 209.6 GFLOPs +2026-08-04 17:45:54,360 - INFO - Detector frozen for input-gradient optimization: modules=1, parameters=140730220 +2026-08-04 17:45:56,299 - INFO - Geometry loaded from mask=asset/audi_etron_carla_attack/audi_etron_carla_attack_mask.png, position=True, normal=True +2026-08-04 17:45:56,299 - INFO - Universal camouflage initialized: style=forest, latent=8x32x64, texture=2048x2048, stochastic_train=True, highres_residual=False, spatial_geometry_decoder=True, canonical_surface_decoder=False +2026-08-04 17:45:57,106 - INFO - Loaded evaluation generator checkpoint from /root/data/releases/model-d38b274e/stage2_town10_universal_latent32x64_v1/checkpoints/epoch_0006.pt + Evaluating train seed=42: 0%| | 0/144 [00:00 + raise SystemExit(main()) + File "/root/data/drc/tools/diagnose_stage7b_source_anchor_repeatability.py", line 250, in main + decision, original, frozen = _validate_formal_boundary( + File "/root/data/drc/tools/diagnose_stage7b_source_anchor_repeatability.py", line 173, in _validate_formal_boundary + raise ValueError("Formal source results are incomplete or reordered") +ValueError: Formal source results are incomplete or reordered