Network, training record, contrasts and ablations

.gitattributes

@@ -33,3 +33,17 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+ablations/converged_posterior_12500/run/PLAN.json filter=lfs diff=lfs merge=lfs -text
+ablations/full_12500/run/PLAN.json filter=lfs diff=lfs merge=lfs -text
+ablations/no_posterior_12500/run/PLAN.json filter=lfs diff=lfs merge=lfs -text
+ablations/no_summaries_12500/run/PLAN.json filter=lfs diff=lfs merge=lfs -text
+ablations/summaries_only_12500/run/PLAN.json filter=lfs diff=lfs merge=lfs -text
+final/PLAN.json filter=lfs diff=lfs merge=lfs -text
+final/snapshots/epoch06/PLAN.json filter=lfs diff=lfs merge=lfs -text
+final/snapshots/epoch09/PLAN.json filter=lfs diff=lfs merge=lfs -text
+training_contrasts/errors_constant_maps_8k/run/PLAN.json filter=lfs diff=lfs merge=lfs -text
+training_contrasts/full_mix_8k/run/PLAN.json filter=lfs diff=lfs merge=lfs -text
+training_contrasts/no_errors_8k/run/PLAN.json filter=lfs diff=lfs merge=lfs -text
+training_contrasts/no_templates_8k/run/PLAN.json filter=lfs diff=lfs merge=lfs -text
+training_contrasts/site_encoder_only_8k/run/PLAN.json filter=lfs diff=lfs merge=lfs -text
+training_contrasts/summaries_only_88k/run/PLAN.json filter=lfs diff=lfs merge=lfs -text

SHA256SUMS

@@ -0,0 +1,255 @@
+1bc3c0a7a998471e4746fbe49ee383807577a5dc85ff2419fffba63d191dc970  ablations/converged_posterior_12500/logs/attempt1_20260916_222119.log
+6604c0a860d4ffe2db1b70e25f6252e0abe14ab26f19cc79ef1eec9a4c94eb18  ablations/converged_posterior_12500/logs/command.txt
+342262ddc541845507c47b277d8a0dd1305bf2fb620da63062cd84f84161ef45  ablations/converged_posterior_12500/logs/harness_run.json
+eb712447e3696f38ed1119ce22687f19f46d77280c1cd19e54c7041b2b61b442  ablations/converged_posterior_12500/logs/memwatch.log
+f5ff6fb323ca321f6ce5ea2463ebaac8d2bd118ea847273918af60003151b8ba  ablations/converged_posterior_12500/logs/refit6_fit12500_e12.out
+c1a411ad2fbbb4f3f622e9eae9c64041a4b35c10231228206540733ae392a812  ablations/converged_posterior_12500/logs/refit6_fit12500_e12.supervisor.log
+1164d581e772b64f9f592f62e29d5e2dc4aa3f59e57f2193c8811cabb0d9bf48  ablations/converged_posterior_12500/run/DEVELOPMENT.json
+ab0d09d7217cdd34ee08c86e8bdfb32854fef90e150f0695669679a1672b5e9b  ablations/converged_posterior_12500/run/DONE.json
+8a14982cea247a926fbcaa569fc0f2bec9760670aee000ae156750524d99fb2f  ablations/converged_posterior_12500/run/PLAN.json
+0b0876dab6433080384933a3df9602762c5e7926817c8fa0b18ef46e8af1df86  ablations/converged_posterior_12500/run/all_inputs.final.pt
+57d9dcc0d8652b33b68176e6bd1f11dd6fb98f24c9bbf5370667326346a1e925  ablations/converged_posterior_12500/run/all_inputs/DONE.json
+10b35b0daa22087f88d21133ac9cf4ff1b26dec92afb0395f27eea0a20c78141  ablations/converged_posterior_12500/run/all_inputs/checkpoint.pt
+c89a32ad70612558a1816f0192666e8fb5e99150e309c8726e903203f95d906d  ablations/converged_posterior_12500/run/all_inputs/mechanism_evaluation.npz
+c84c0f7284797609b4780052fb7e84227a4e8cc6de5c2b280d8013155044e1f0  ablations/converged_posterior_12500/run/all_inputs/report.json
+5a70cb5d76d3991910571b968f00494e3c812ec5f39dac63fdea14db649eee74  ablations/full_12500/logs/attempt1_20260916_073515.log
+5329a53bd872343400c4a4385a26e6af6c7257b3bf6861cd0f1a2e45313fa2ba  ablations/full_12500/logs/command.txt
+df6248aeac3ec4c85fe22afa86a53d6418a9b891de2cf92a03ecf077cab5e07c  ablations/full_12500/logs/ctrl_fit12500_e12.out
+228d42c7c22d1d04d22888cc1d5402b85a48f3ad139bbeeb15e744c600c2a719  ablations/full_12500/logs/ctrl_fit12500_e12.supervisor.log
+ae052b34d2a1d646d97895650e86bb2311cf6b6408001eaf3f2bc9f6923600b4  ablations/full_12500/logs/harness_run.json
+0cb119a29f422bc222ffa7e2d75fa19fa24896aa8a3e6d10e2d5298b6721df7d  ablations/full_12500/logs/memwatch.log
+ed1bcd8d0e91c1ccd2ee968b9b55a832cfe6719118b936b232a336117e95b336  ablations/full_12500/run/DEVELOPMENT.json
+4d97adcb9e15fbef4901b0fcd4a15c5a994db9e0e43c9363466fc32657e9dd44  ablations/full_12500/run/DONE.json
+ccffa7a9e978c480998c8f0a73484c7e59fc891b9ce775b32208f8a6bec8a122  ablations/full_12500/run/PLAN.json
+0b76056a43d36029594d41e8a594801c06035b34b047d7d417c54df1c4e1126e  ablations/full_12500/run/all_inputs.final.pt
+d8f9ff89c798b042d6f1859ad1de5724a1b3a4f2e963bf721c333dff59ef94d9  ablations/full_12500/run/all_inputs/DONE.json
+925d3d3914ab3a140e84dd0931fc37a9de3e0bc829991ffb8728b9d51af0044c  ablations/full_12500/run/all_inputs/checkpoint.pt
+3394fd9dc43d7d05c607d6000a11e129b6a39033910b2677d6fc182cfcc9174c  ablations/full_12500/run/all_inputs/mechanism_evaluation.npz
+596c5ab57deedc774ade13a47b59d21cf664516fbbc7f09e4b02210fc1b88ffa  ablations/full_12500/run/all_inputs/report.json
+70ca82a10b387516f3812e1dccf0192f5fdd17d968e057fce7e57b2a0b74786c  ablations/no_posterior_12500/logs/attempt1_20260916_152815.log
+f8d543fbbb867ab6f2b4506423fc706292038f909c1e3ad245031c2c3f66add0  ablations/no_posterior_12500/logs/command.txt
+eb46ffd784576ff66a47c7f0b54f1971c73d2ee6e92f6325893955999c205592  ablations/no_posterior_12500/logs/drop_sidecar_fit12500_e12.out
+1d3448f1f6f61d9bf3b137fa46e660f1f7926e72bf7fdd3fa511b6dc2278781d  ablations/no_posterior_12500/logs/drop_sidecar_fit12500_e12.supervisor.log
+5798f63bf29ced24c6d206e3ff86e2d75903be774a868c450bf93bdb38d45336  ablations/no_posterior_12500/logs/harness_run.json
+25dc5d916a4b99b3bebe1ba778180eeb38dd9fdd68096cde2e08a25d85130b2f  ablations/no_posterior_12500/logs/memwatch.log
+f46be94331a2d116d889b3f59c03bb926396ef11237954206fe37614c48f3324  ablations/no_posterior_12500/run/DEVELOPMENT.json
+a57067d550f4b3756122c60bc1904520dabcae7e5d7f07ec6f441f37d07b7080  ablations/no_posterior_12500/run/DONE.json
+538b5152f1759146525ee192bddf45d2f5e253ef34d6541f5b88a9c6efc19317  ablations/no_posterior_12500/run/PLAN.json
+68b5d29fd594855b31b57b3b9e75a3eb1f51193016b95174e58e020085a37c81  ablations/no_posterior_12500/run/site_encoder_and_summaries.final.pt
+3e0222cb8e58404ccdd5978dc6f642237b5bd4944f329d94659d0ad1b7951959  ablations/no_posterior_12500/run/site_encoder_and_summaries/DONE.json
+58e35ac89a095b376bc35ceeab7b6fb0d00c11745b530c76e706f9eb5f4553b0  ablations/no_posterior_12500/run/site_encoder_and_summaries/checkpoint.pt
+545a7456adecf48557c274a4c0bfc984377617d577a593cd7c0f43b22f8c3268  ablations/no_posterior_12500/run/site_encoder_and_summaries/mechanism_evaluation.npz
+11294cc4afd8ea92c54287771cfe4416b5f65bccd4b1b17aa8970beff3736ee6  ablations/no_posterior_12500/run/site_encoder_and_summaries/report.json
+40d9fc1f20a777feebd315f58dae4a5069bfd6b3cd30db6da21b044b8b019271  ablations/no_summaries_12500/logs/attempt1_20260917_090003.log
+d8ea7e82c35b45108bfca826b4caf9946c3d76d32e3d50ee6dcef0515cb8d207  ablations/no_summaries_12500/logs/command.txt
+dc834afd996bc55de5e1c83e3e32870408a23bde9fcfe4de6de9747c2262b9de  ablations/no_summaries_12500/logs/harness_run.json
+ffe5c2ce7403d16e5a099b536a11f47fe09e42b13b2286609ffb614f2f0e5333  ablations/no_summaries_12500/logs/memwatch.log
+4f0def80326fe3e9dc744cef443e74ff99a4dc82d15173e61afcb72ef994bc23  ablations/no_summaries_12500/logs/rawsc_fit12500_e12.out
+c923770b25614f7657abb515b88afaea39900f31552a91ead4dca8409227ad56  ablations/no_summaries_12500/logs/rawsc_fit12500_e12.supervisor.log
+42c31717df529dbec28f6af012392f30b0404396d13fae83b543a475159aeef0  ablations/no_summaries_12500/run/DEVELOPMENT.json
+50109c17f69db404c78597fb846df44adc62c4cbfb376a353d1eebdbe402a630  ablations/no_summaries_12500/run/DONE.json
+e059e8bb12236337fccf7b6632c8bfcf7e6f3886826bfb101552baa1a654acf5  ablations/no_summaries_12500/run/PLAN.json
+be6515ed60d23f7ce0b75456f7882a66610d92e58c3d479c8265ed99bdf27b57  ablations/no_summaries_12500/run/site_encoder_and_posterior.final.pt
+fc06a133ae2c10baad42698e79862e82f61e428f08a525a544c6fa3a516b2282  ablations/no_summaries_12500/run/site_encoder_and_posterior/DONE.json
+99351656dad0cc8748c1130fe0070ea250d9d92e29b5908dc4b5aa933bb684a1  ablations/no_summaries_12500/run/site_encoder_and_posterior/checkpoint.pt
+a4f337764e7f77e3fc541ca8650e6ae94c481eb2d0228b24155e64fd75f69104  ablations/no_summaries_12500/run/site_encoder_and_posterior/mechanism_evaluation.npz
+6503331d8e2844765b54c5c69ce66db164cb421cf3e427841b65a005f90a596c  ablations/no_summaries_12500/run/site_encoder_and_posterior/report.json
+189a2b99b534b6d5e0f4d636a36fcace8dac624325d1e9a4388f17faea62024a  ablations/results/RESULTS_run.md
+2fe714c2e3725c018f23ac2e0d9c14fc3d8249d9a5b3bce1635288f8629c3ba8  ablations/results/ablation_paired.json
+394f7355b385def41f04feb6aee1b24e49330606ee72a307bd0374d521831d74  ablations/results/ablation_paired.py
+6eea57b39948d05a24d2f9ec79176f7be4bf6086f812e1334f620d787a97a22e  ablations/results/rawsc_paired.json
+6b349e76b35f1f4fec2651335bcd6320e4901bb18347df52f1cd0ce3933226c8  ablations/results/summary_only_paired.json
+6ec8570352e563bfa23f9877f01b28ca8eadbe3f5d61d8efb42a771c1af0369f  ablations/summaries_only_12500/logs/attempt1_20260917_083908.log
+d4ab30b73f872436e4878a4f25728a1c3df8bd57f4af81ab9da58c31b0f7fe08  ablations/summaries_only_12500/logs/command.txt
+7d261273e61c4534fe2178b71c436d9ab8d29f350a051891355f0d5d2ddb44ac  ablations/summaries_only_12500/logs/harness_run.json
+6f45f6374723f31d10809bf65bfa210d3acd948cddad47159ceb24af9955f737  ablations/summaries_only_12500/logs/memwatch.log
+2e9f219b33f73e14af8e13e38e9553af60b4c140be029b63a11d88fcf81d6622  ablations/summaries_only_12500/logs/summary_only_fit12500_e12.out
+4a294c27dde46d5fe90cb5ed9fd54db8f6ec78abcd5d091c35602ee15aaec4f9  ablations/summaries_only_12500/logs/summary_only_fit12500_e12.supervisor.log
+2510ecebef295ae4bb8e0c303bd6af2cbbdf819577142f12f5bf9090e20c27f5  ablations/summaries_only_12500/run/DEVELOPMENT.json
+3f23027a8c81d862a0b8fb92dadd146b386963bbf4233838e695f45ece1b887b  ablations/summaries_only_12500/run/DONE.json
+9c72e9b5844b598dd1a306d249f9f05f02574c9582ffc2a6f1d097232a96cd06  ablations/summaries_only_12500/run/PLAN.json
+f5d8b30b1d297929b731ea7a01e1ac6c71b1fbedf22a36ad7616dbf8d2a0ea09  ablations/summaries_only_12500/run/summaries_only.final.pt
+79c795535cc4b0813d472ae227ab185f00a7255a438531c6e05b6a592452d70e  ablations/summaries_only_12500/run/summaries_only/DONE.json
+c6fdaf5df9e564232d8b28929d1a749288d926b65dc45ee15015aaf5b1c472fa  ablations/summaries_only_12500/run/summaries_only/checkpoint.pt
+34f1a287906e64937cdf73fa481bb7164dbc26ac82beb592a23c8ff890713a9b  ablations/summaries_only_12500/run/summaries_only/mechanism_evaluation.npz
+fb5ec87b84a657081926b218cef0fa59455d3ef411c6ff913bd8e45448bdc82b  ablations/summaries_only_12500/run/summaries_only/report.json
+49a40ada70aaf3390cbc15945ee38c28edd1e122760f7851d0769c098263f6a8  final/DEVELOPMENT.json
+39caf385d0ffcf1c243a4c7369c849a8ec3fc65e643e6038606fac0497e49cc7  final/DONE.json
+ce8c2e983648b28bb997f8a7c4d26cc239c7379a0fbf81371d7becdac0114e76  final/INTAKE_VERIFY.json
+b51974c00e406115d94d53b5f0a6d4cf485b6fd94f2242b9e31e5bf688234a5d  final/PLAN.json
+66439a336847122deb2f13fb96e407aeff4fddd593a2168d1ca580e1a0c28ab5  final/checkpoint.pt
+2c1ccc1f3b0b76ea670ecfb7ce0ae2d93494159389c41f22847cf4364e789e4b  final/compshare_sha256.txt
+df7693f3c69d287de01e6ad717ad23a0813e56facae8757b9e78f092c655b9e7  final/report.json
+d7ef435c1701d0dee7db3e4f29e7bafce95dfaeba727d1d74197f5c0c7207815  final/run_DONE.json
+c64a86c1ae4bc69574d8ef00375e008e1cd8090409539108c0df8bfad263b100  final/snapshots/epoch06/INTAKE.json
+ec6a80536fc5d533a901d9176dec88921e3fd37da39848b5a68d479bbc36fefb  final/snapshots/epoch06/PLAN.json
+85aed5c2e3ef067ae69199c37c398b03d5e895b369a98c74c2e84f524ee42e57  final/snapshots/epoch06/epoch6_weights_as_best.resume.pt
+0e53c837302cf2e99f5bd074a8b8f840fac4d9f03be2c3a9111b0d16db84a8c1  final/snapshots/epoch06/mechanism_evaluation.npz
+a833df391d4ab4d114c54357edbd5bc3e277b2c18d19b0a89b23f26b1f593cc5  final/snapshots/epoch06/plan_gates.json
+99b54c3a26635cc3a8adbfa56cf3387db4eae780067ee8c303aa7656ce6d8602  final/snapshots/epoch06/resume_summary.json
+ec6a80536fc5d533a901d9176dec88921e3fd37da39848b5a68d479bbc36fefb  final/snapshots/epoch09/PLAN.json
+537be7ae9a69abbb3a49b6d435e4f37c75a84c12432bdef341a885ed3cfd5ba9  final/snapshots/epoch09/epoch9_weights_as_best.resume.pt
+b6607fa0ee68fa887a7bc049d2397bab0f43290e8443c0bbf42308d92e0f4bec  final/snapshots/epoch09/mechanism_evaluation.npz
+e1b6fb0397de740c75e5396415ed3fc4f97346158b1c80872f286a1ab606b329  final/test_predictions.npz
+ad9198184f8b753b4aa24f9a40dc9dc5fd58a65688b8139c4f7d774a623a7b0a  final/training_log/attempt1_20260913_212030.log
+35b95f59fbfa1396ee81b28d80092b2afc29ff1550e889970dbeceed9a0de16c  final/training_log/command.txt
+00b6355cb7ef88451d8f3430b751e2c17d41efcb4a0cba5f116f04a79030f3ff  final/training_log/launcher.out
+47907998380b9605b9220f3d0d2e05ad64a9fe73d0afb6714f46fdbc5cb9e438  final/training_log/memwatch.log
+49e2f4f788553c300f4113e95be5745446196e748f47d57f41f5bc838f7af55b  final/training_state.pt
+af91cdcd650597aace6f6c709b7a59a7b31fd1774b825a78483a2dcfef85d069  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260920_125005.log
+15986710d06148674cc68700a982b01588353c5cbf17da3643c84ddb6173024b  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260920_175936.log
+a07e5509873ce572a3522608b5a0c003e0187482f5b21e4be99b55543a768f5f  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260920_230027.log
+118908608321f6fb871be52a53c1eb61e93e9800d585c3b1c2c6f05ded071bdd  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260921_165520.log
+4e3e6862f90f852497e53d77f2775603363a03f220c5659c7787dac481a154e5  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260921_203711.log
+4f1af56834fdb838f8df56c5bca70f8d7f6b1d14a84a01ab69b0ddd976830f8a  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260922_002806.log
+af67d276e44c0e0756b4a9f6d9361eec27dd8ddf09e808047b29aaca023287bf  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260922_034821.log
+7e7ca7b0aecddb0c2981110c87f875cfc1377af735d530bd421431b206fdbd55  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260922_084956.log
+aa3939551ee494e3188666e1377528883f5cfc29da16b8540e5cceee59c5de7e  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260922_122810.log
+8e65be5b97a63acfe486ba14e7995b74e5aa1ebfd7da769f4ed576fd6f35774b  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260922_163200.log
+5752b7b93f1217ee7e78d40b104e1f3c5b40c18271af1f8309c87da82ade128b  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260922_194414.log
+283c7b7bfef0e884021384a3bf3a0f67bfab95ba54284e1d1724162e11d1d570  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260922_234824.log
+ca0d2b5948bcb86036b3537f37ad821836d725b7ee682a55717ee496cb84e37e  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260923_042104.log
+2658bd505b740604bdd4f4b4aca335697dc00beed8c4f4606fc579e62978c729  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260923_082320.log
+62af17ad63b784760ecc094c08d9ab563bfeeaa6a7b4e715c433e5cf4fee1d78  training_contrasts/errors_constant_maps_8k/logs/attempt1_20260923_132425.log
+e0b734a2c480e1070949506ce211004ac4f3fbd29b23fa224a822744e3e7128c  training_contrasts/errors_constant_maps_8k/logs/command.txt
+eae72e48831bf626fcc1e40485a98d4cd2f4b3707397a5d9ef321ec58a9be650  training_contrasts/errors_constant_maps_8k/logs/harness_run.json
+4ea542c1b31a8e79b70d3bf9379e1b9adf2490af5880dec15f2a5aa423df1a69  training_contrasts/errors_constant_maps_8k/logs/memwatch.log
+b6ef09bf9bbf320ef116bef472ba96f012d2906f535f9034f3a114b4f72a9219  training_contrasts/errors_constant_maps_8k/run/DEVELOPMENT.json
+0c45eef3badf102958c0e8f227966ffcd63cd847e204c24582700b9c1f9f994d  training_contrasts/errors_constant_maps_8k/run/DONE.json
+b0d521c29e01bbba62b8d166b738f5452751261f7405f51fbe340520716e28eb  training_contrasts/errors_constant_maps_8k/run/PLAN.json
+ce0158e6cb7e53c298b62c6f32b6d8c4d552abf2429e16b21b50873afe58a7c0  training_contrasts/errors_constant_maps_8k/run/all_inputs.final.pt
+6e82731f6cdb9a024fedb372eba103e352f2a6ff8a1022005a799826b54eb6a2  training_contrasts/errors_constant_maps_8k/run/all_inputs/DONE.json
+00c0afe5fbeb25961e8a23344e68d590f6d1ebc3a7ff2bf7f95e8cecdea98fe6  training_contrasts/errors_constant_maps_8k/run/all_inputs/checkpoint.pt
+bd5057bbb7435f706cfb77479c866f10a71a7fa895b22a51c271537a6340cf70  training_contrasts/errors_constant_maps_8k/run/all_inputs/mechanism_evaluation.npz
+c79bcdc49c8a8d574e3867b7425c8426ef6e090ca4da041268e851b2d9354cef  training_contrasts/errors_constant_maps_8k/run/all_inputs/report.json
+fa644202d008c783eed5059001171a131cca4093f269d1bee3f25a7a85c4b6b8  training_contrasts/full_mix_8k/logs/attempt1_20260920_125015.log
+116bcd76762e23031e22fc53909fbbc69fed67b89fdeca2106637f9281ec0a82  training_contrasts/full_mix_8k/logs/attempt1_20260920_175047.log
+a1fcb948a2ea13780f41bfecfb12d90425dcbd67a338701ea12e6eea19b4487c  training_contrasts/full_mix_8k/logs/attempt1_20260920_225136.log
+b686578a73130b2a5777ee832c7d6a35e5a7379fd421cd55657708d0d6efd03a  training_contrasts/full_mix_8k/logs/attempt1_20260921_035149.log
+585460a4d5766c21716bee979c9d5e4bd462850339000d9401452512019eab73  training_contrasts/full_mix_8k/logs/attempt1_20260921_085213.log
+af4a8488dc8f76ec5ccc6f6e0b87bc7ae8744ca405dc4fb60614666d6b7a107d  training_contrasts/full_mix_8k/logs/attempt1_20260921_135357.log
+585441a5d848cf1f8bbf6251d8b6159ae8b720aa6ab873a89c51d5e563cd6adc  training_contrasts/full_mix_8k/logs/attempt1_20260921_185427.log
+6861da29a623412fac29be677ea7ed8a7a1670155065cc1dd6a84c54cb485775  training_contrasts/full_mix_8k/logs/attempt1_20260921_220426.log
+e3a07dca8573e11bfddbcee8123b1f7f94ad6cf50e720321d622447ad4ba3179  training_contrasts/full_mix_8k/logs/attempt1_20260922_011121.log
+6a23006de1fa6a693bb6535d4a77ff3699ad0a89ee869638d03deb5964eaa28b  training_contrasts/full_mix_8k/logs/attempt1_20260922_042005.log
+83926d454bedf0f6b9b191b18be70c64aa73b48ae1976d12560beb42de0a38d1  training_contrasts/full_mix_8k/logs/attempt1_20260922_081316.log
+e854e60943e0dfc9bcce28a65321e4fc6dbe1312ad807f4e9b4760ab046dde75  training_contrasts/full_mix_8k/logs/attempt1_20260922_111937.log
+8910d5462d14bf1b23c432fb92bab256029868cea3114a4d45f022c95dfed8f8  training_contrasts/full_mix_8k/logs/attempt1_20260922_144257.log
+41b1f55466f1da822dbeff75a2fe1d14e4cb9e6cbf7414b722b6e38f1d13a1a4  training_contrasts/full_mix_8k/logs/command.txt
+14804db6404c2dd10a0d9d6e55e70f0d5e468d638957b4f2940c65aec0049a78  training_contrasts/full_mix_8k/logs/harness_run.json
+8e867c27d4079dbe7fa782bf8a46ee33690a45901a8842fa102b451be978dfd7  training_contrasts/full_mix_8k/logs/memwatch.log
+4fd3a3e58689b50b9266e14de8b118737315b8d9ed70f06e3165085f4ff28015  training_contrasts/full_mix_8k/run/DEVELOPMENT.json
+fb9deae8014b397c4b44a78366f1af8fcd709d0bb6045cd364f4bbe5dc111b72  training_contrasts/full_mix_8k/run/DONE.json
+bf474e85120d99ba0d6c4bd292cd25f7217f49d898a444f688c0bb3312a5c883  training_contrasts/full_mix_8k/run/PLAN.json
+b1020f35e937622b9138eb85ae24bd68174dbda517df310db7be9dc01eae0276  training_contrasts/full_mix_8k/run/all_inputs.final.pt
+64769e53b5fb33a4b6870f1c6b959148da4950c800846ad8ed8f2364aa5bcdb1  training_contrasts/full_mix_8k/run/all_inputs/DONE.json
+8e514fe804063357691d86b2d5b72728ded7161a0b73353278f1cc9165236534  training_contrasts/full_mix_8k/run/all_inputs/checkpoint.pt
+0fb9baac8ad667755d7274a43fada75c430535b691d2b5ff4bf2e7e77b11ecd6  training_contrasts/full_mix_8k/run/all_inputs/mechanism_evaluation.npz
+c207db0b3172d5f843fc3911d3dae213517c4eb05699a49387df1844ac1148b8  training_contrasts/full_mix_8k/run/all_inputs/report.json
+e5ca921507358368868b36f89bf31f23a68012b5d21d637276f4ad92d9ea223e  training_contrasts/no_errors_8k/logs/attempt1_20260920_125005.log
+225582954db34ddad484c4e08237e9607baeaf187706c0dee13f5644c6bc6a07  training_contrasts/no_errors_8k/logs/attempt1_20260920_175934.log
+09648d7ee5ff5c17da312968ea94fd828dbfad68b9d155df01560ca6d4e9dec2  training_contrasts/no_errors_8k/logs/attempt1_20260920_230032.log
+09648d7ee5ff5c17da312968ea94fd828dbfad68b9d155df01560ca6d4e9dec2  training_contrasts/no_errors_8k/logs/attempt1_20260921_165043.log
+c787f3971f90835981bc3ea6c291b282394afeaae140ece0ebc4caf9a8604f53  training_contrasts/no_errors_8k/logs/attempt1_20260921_215233.log
+f815b2856b830e85cfe0a83df8a045afe8524f2e6c5d2d446af7ad5a0b1f1293  training_contrasts/no_errors_8k/logs/attempt1_20260922_010455.log
+baabf29fbf476f156981189700f289e1c6c7b6e31608abc6efc1bf1a0ddf8a86  training_contrasts/no_errors_8k/logs/attempt1_20260922_060523.log
+039cb6a0df86664bd242ae387d547b820a2266f5b39f9534b70f51eb4c23bb3a  training_contrasts/no_errors_8k/logs/attempt1_20260922_092936.log
+cafab51f6f60256957dbc362f26d7a52ea6e61dda96ecceef30b93592013c3e4  training_contrasts/no_errors_8k/logs/attempt1_20260922_142624.log
+ec8052f3e0fd21a2fa832db3a2f909d5625c17601d42736a9e3e0d8d842b3528  training_contrasts/no_errors_8k/logs/attempt1_20260922_191845.log
+3bf859624d54d8b402c49de59f3897067750c51e4a8b99ee0941fb44d284e8af  training_contrasts/no_errors_8k/logs/attempt1_20260922_223543.log
+3bf859624d54d8b402c49de59f3897067750c51e4a8b99ee0941fb44d284e8af  training_contrasts/no_errors_8k/logs/attempt1_20260923_033802.log
+3bf859624d54d8b402c49de59f3897067750c51e4a8b99ee0941fb44d284e8af  training_contrasts/no_errors_8k/logs/attempt1_20260923_083825.log
+2bd22944f34c9480fd7f960085549314721e1a6782be8161669fce73621e2392  training_contrasts/no_errors_8k/logs/attempt1_20260923_133848.log
+f5f0f09a342cc289a3dc8871aef4abf485be31a5247dd22dd5f2a090489982c9  training_contrasts/no_errors_8k/logs/attempt1_20260923_174053.log
+72a0a3fb2baefe1b9dc23bc8999946c073e5c38ca2d6fcc9bcae1555b42040fa  training_contrasts/no_errors_8k/logs/attempt1_20260923_211429.log
+d855bf07e9dc9d94f63277a0a1de66af6cebdfc0ea636e4c967be4b1b6fc07a3  training_contrasts/no_errors_8k/logs/attempt1_20260924_012905.log
+193951b80da56f871899fb496827c0971f1aa937a91e4fe71d2c3d7e73dcbcf8  training_contrasts/no_errors_8k/logs/attempt1_20260924_043403.log
+0c18501ab920877084c7d8a5a29b7640432f5ca1f0ab781962c2df24c412da9d  training_contrasts/no_errors_8k/logs/command.txt
+433828fd3ab403f73c0d160f8a73c095fc19c5aa549fe854caccf3389f47d2f0  training_contrasts/no_errors_8k/logs/harness_run.json
+fb641b02db922aec5b7514d028fd63b5bb13f1063b07830bc9c1f66da563d092  training_contrasts/no_errors_8k/logs/memwatch.log
+f8d81b4dd7f8ce447406cc92b1b1255ab9a0e59e91ca674cead3b73de8d1a655  training_contrasts/no_errors_8k/run/DEVELOPMENT.json
+4f2382d6f6d9c15dbb187b2bc3278ca537b6438a0d28a57c407d70792a187325  training_contrasts/no_errors_8k/run/DONE.json
+48705fb3c8344acfbc3fbe569f881ca510822ba5ae7d92dbf80fa917c08a652f  training_contrasts/no_errors_8k/run/PLAN.json
+9611c72d56f9de258a1444525c58947e6a9a33fd46082ca221aca860a44412e9  training_contrasts/no_errors_8k/run/all_inputs.final.pt
+0c03d42dbbc975c02300b6500f197eb786f9186269158703e64bbd1b08e56d7a  training_contrasts/no_errors_8k/run/all_inputs/DONE.json
+f2e493033471daeb4d365197e2cb43e1552872a359c7fdacfab18c2a96473993  training_contrasts/no_errors_8k/run/all_inputs/checkpoint.pt
+74453e0afae280e75f801a3a137eb915919c268c95ac47fcc9b99a0536a80e05  training_contrasts/no_errors_8k/run/all_inputs/mechanism_evaluation.npz
+2604d72d08b1d27621b141182028db2b7d8df62046768661e71355826500f755  training_contrasts/no_errors_8k/run/all_inputs/report.json
+632c6ddeb7ac30fe54684185862f4e9450683c4df4cc16b542f0b33f4af47815  training_contrasts/no_templates_8k/logs/attempt1_20260920_125007.log
+101383d258d9af1a674fde44ad203ad632f58ed8f418a9531d5f48d560ef3845  training_contrasts/no_templates_8k/logs/attempt1_20260920_175934.log
+f10e837f1f1f6fcfea4ae63690d72e4fe6fb73c0a4f3e95bce21c39b895fdf40  training_contrasts/no_templates_8k/logs/attempt1_20260920_230028.log
+80b8b53c467969c529aa831385e381a23312becd99744afb13386199eb545040  training_contrasts/no_templates_8k/logs/attempt1_20260921_040052.log
+b10e3878f74441bb452e88a43d945a3d77a37f3de3dfea9ce32f5489b99b0e25  training_contrasts/no_templates_8k/logs/attempt1_20260921_090136.log
+741809d5642701fefcfb77d4b6cfc068be93d8c493b41a3c38745381742d4990  training_contrasts/no_templates_8k/logs/attempt1_20260921_140655.log
+af87d64c0c1d7d436ad3ab0ed9210bdb61afbdb287cd34972f2b580a17bd34e7  training_contrasts/no_templates_8k/logs/attempt1_20260921_190654.log
+bf1addbe0a3fbd668f85eb561173880a8b5d7bb60edaac2ad1444f071dea9259  training_contrasts/no_templates_8k/logs/attempt1_20260921_220831.log
+74471dc773c28a2e279145903f15b190cbe841666059e37a0708eaf88547d9b4  training_contrasts/no_templates_8k/logs/attempt1_20260922_013722.log
+9cfc85079c216e6165d8ff7ea419e9f54dd3cdd628ee4051800516e216076fac  training_contrasts/no_templates_8k/logs/attempt1_20260922_060531.log
+6d4ed382912d1315ab9deccf6143312763a0f8e51559710adc5f69fc395f7b9f  training_contrasts/no_templates_8k/logs/attempt1_20260922_090304.log
+6af877eeb68ae6dfa4d643e292071063660b08b113d7f8595b91da0694800480  training_contrasts/no_templates_8k/logs/attempt1_20260922_120644.log
+3bc5aaeccc24aa9b0dfb388ff40da0ba90ca6ea33b84f68057e0fa0efbd540fe  training_contrasts/no_templates_8k/logs/attempt1_20260922_152925.log
+0fb705fdbc60696bea8962b32c92486c3cda67d43efb54ceb952ac08c93203e9  training_contrasts/no_templates_8k/logs/attempt1_20260922_194157.log
+c6191782411849b232b801098e26a5bc26d071fe0a63f785822eebdafe821b86  training_contrasts/no_templates_8k/logs/command.txt
+cd56037ed1da46b8d73dcde16de7d6eab41794910e797840711940a7a11e367a  training_contrasts/no_templates_8k/logs/harness_run.json
+4db0f6ec4935e0375ba40c34c00f51f27e0dad1f68c2eab7d2b2807ae528590e  training_contrasts/no_templates_8k/logs/memwatch.log
+775588451d1ddce1e9245fa1261460597728682cbd9ac2c154e4ace85a9950ce  training_contrasts/no_templates_8k/run/DEVELOPMENT.json
+0ca39427bb831f58a86555539e0a9c7f9e74ef3393c57e181e08eb859c50ce1e  training_contrasts/no_templates_8k/run/DONE.json
+47c290772b1708d00f2382135e9b50a92f1bb17690401ddc028de23137336c26  training_contrasts/no_templates_8k/run/PLAN.json
+36506072f1e03a569afda45fe1916028ada429a1af8c9c0331226d894dcc7642  training_contrasts/no_templates_8k/run/all_inputs.final.pt
+cd8065f3c71290c3582987891fd6f32bfee558d92912a8bc8228cca16657e0a1  training_contrasts/no_templates_8k/run/all_inputs/DONE.json
+6c32947acf8625a0291d24ae8bae7da260d6aff7d6e67e08bffd2343506ee175  training_contrasts/no_templates_8k/run/all_inputs/checkpoint.pt
+1d7b325725fe36101057d4d00ca6b765f686fd9d159c9450c0054c6fc05cf0ad  training_contrasts/no_templates_8k/run/all_inputs/mechanism_evaluation.npz
+6d660005b2e3a84a0c775eee3fb01a2e753b4563c64ab85a70d44a6e3658b6a8  training_contrasts/no_templates_8k/run/all_inputs/report.json
+4f4f2108f661437508c89b5f5ef1add00d363aad6dbc0e032ddd84a7fa4efc36  training_contrasts/results/RESULTS.md
+10eeb8dfe7e5a6a6742383378652a5b38fd82efa6425f0bc3e68bc2eee12a3cc  training_contrasts/results/RUNBOOK.md
+857d2cfa759211d0f04874f52f57147902d3cc3cd0e6f336abb11c43d56b1067  training_contrasts/results/eval_all_20260925.npz
+3377c212687416d72e74f58cb66de435afd38c155e467164825bff3982aa2b38  training_contrasts/results/paired_20260925.json
+5ccd519a102d6d999be000825d7f91581cf82c7262e5336687009fc4983aa993  training_contrasts/site_encoder_only_8k/logs/attempt1_20260921_171408.log
+efe4c9349479b616a591851dd24b3cca7d9c9499d9369c53583e9cf1d2cb8536  training_contrasts/site_encoder_only_8k/logs/attempt1_20260921_205024.log
+fc612315922585fe22933e7031f2673ced6df039829fea7396225ed61ed93bbe  training_contrasts/site_encoder_only_8k/logs/attempt1_20260922_015155.log
+a36c3f0a0db78ef5809e81770daf58da3e68fd64f63c5df326314c49beb1f829  training_contrasts/site_encoder_only_8k/logs/attempt1_20260922_050515.log
+1541030a363f55c92a730642a6a2ec958d3647eee67ceb3c6a2dab2fdd912f5c  training_contrasts/site_encoder_only_8k/logs/attempt1_20260922_084956.log
+fc57f1574bba776ba55a70472f5954ff352509fcd5135a8679ccded2a1fef366  training_contrasts/site_encoder_only_8k/logs/attempt1_20260922_132701.log
+a1f9631f031238e6913ecfdd3d259e830705209243daa1d7814aa8524989fb35  training_contrasts/site_encoder_only_8k/logs/attempt1_20260922_165326.log
+a71644af72b508a7e06cbc287ea6d41b78b5f75bac2cb7ecf64d3ebf2476c3b3  training_contrasts/site_encoder_only_8k/logs/attempt1_20260922_200503.log
+e5e11012b2816324c8fab6605b0c7450a7686329d2ba3cf7736acabf27c92180  training_contrasts/site_encoder_only_8k/logs/attempt1_20260922_233512.log
+2a6ad1405b6c7d72aa7e0d13390bfce8757c651f9a02c2e519ddac8128566b31  training_contrasts/site_encoder_only_8k/logs/attempt1_20260923_023808.log
+e44ec5bc3fd0b11446887645e1d4b0b92b7d0ebe8a34f20aa709ac2fca46c850  training_contrasts/site_encoder_only_8k/logs/attempt1_20260923_073843.log
+5389547e0e88e9db0dfed992925818cc9a5b80ce4104ef782e6a30481448c0f4  training_contrasts/site_encoder_only_8k/logs/attempt1_20260923_105531.log
+5ea8fa95962193f3830e18a6b655cafaa900bd15ab5bcd02f155265fd11ce51c  training_contrasts/site_encoder_only_8k/logs/attempt1_20260923_141109.log
+91e34bd88a9c505885b77aceff8b84394c41b31c4efe49a950b121f6d43b387f  training_contrasts/site_encoder_only_8k/logs/attempt1_20260923_171515.log
+ff688bc0bc791d414d9c0c9850baf6fd9dd4766ace1bdbd5f454bc6b5d67a394  training_contrasts/site_encoder_only_8k/logs/command.txt
+4728ef5c0db2dff85741b6f7122a6e0ec8e3e53f966f80c7dcd0916a6daf6a02  training_contrasts/site_encoder_only_8k/logs/harness_run.json
+503060d38435ee04734dc0b79195afed8ef2fdba33e687144313e28ca7e87489  training_contrasts/site_encoder_only_8k/logs/memwatch.log
+3c186cea121c3b3f945124a0e52b96297dc15e6d5f1831d167a56014ee4af56f  training_contrasts/site_encoder_only_8k/run/DEVELOPMENT.json
+74d48ba6aa13bc9eb8a8338b1695499979e1e6c4b476431278aad3c0ba642fdf  training_contrasts/site_encoder_only_8k/run/DONE.json
+23f1705596aa35b32a98f7a5010f58b137cee486ec1ef382bf3d6c6533ad03ca  training_contrasts/site_encoder_only_8k/run/PLAN.json
+a0f9f7bc610e6955c478d16189bdd4b967ee59dff6f3955f237030483583b8cc  training_contrasts/site_encoder_only_8k/run/site_encoder_only.final.pt
+104e589636352314cf0742d1466dccf365cff2f1d38eabbf349ed3e9e5c4c735  training_contrasts/site_encoder_only_8k/run/site_encoder_only/DONE.json
+db51c28a1a2597baaf22340a822b165a1d7b141d9bc7bb377f64dcd8ddc8d890  training_contrasts/site_encoder_only_8k/run/site_encoder_only/checkpoint.pt
+80cddec711ffe5820a30f79babb4f7df0267674527c34553fc5581f4d35e3722  training_contrasts/site_encoder_only_8k/run/site_encoder_only/mechanism_evaluation.npz
+bf15a1f00b945818d2223b348441518264c1c1b2cca93b9c12c44e00d8f6487c  training_contrasts/site_encoder_only_8k/run/site_encoder_only/report.json
+c481f3fa5bd7bfa34149805326ca6f05afc296a2fa21577bd97f75c02abb4567  training_contrasts/summaries_only_88k/logs/attempt1_20260921_175608.log
+c27e850c4cd243d67c4da5dddc2a021b4a597e845a6fb27e58c4bef9e70a43c6  training_contrasts/summaries_only_88k/logs/command.txt
+dbdcce0f090a2b8eaf5d89cb230ff08407da4d27d10d19422b14393a8d795581  training_contrasts/summaries_only_88k/logs/harness_run.json
+601ca5ec94f4fd3ff63621e20174ca1e2c252196768217852108ded5e006bcb9  training_contrasts/summaries_only_88k/logs/memwatch.log
+a24ba29a050bd189d6d88ba194e4c37484afc30b00ff7a573307660a8d9eaeab  training_contrasts/summaries_only_88k/run/DEVELOPMENT.json
+a51222c0deb50e7089b6f7bacbcd3d6518e2499abe0e16da2d03a9c824897531  training_contrasts/summaries_only_88k/run/DONE.json
+9c708ae9b152ca2dcbb0031a78efbc744b039a8b3b3a8632032ccbbbda355140  training_contrasts/summaries_only_88k/run/PLAN.json
+cbcc74b53c7fcbb3f4eda0cc2614b2ff71a665ab4fc59223d4694539d3a22acc  training_contrasts/summaries_only_88k/run/summaries_only.final.pt
+ed579f52f5f6bbe6046bf8d62faadfe56ccb4488b80129fa15a7f4f0af69fd3e  training_contrasts/summaries_only_88k/run/summaries_only/DONE.json
+28fa4bc629958b4497babf9dc438902155c864b202cf0e1eb9b2ffd62c7810a7  training_contrasts/summaries_only_88k/run/summaries_only/checkpoint.pt
+390258c8bd82e94223a400465a434c0e5aba43c141d03439bc66e52f7093d737  training_contrasts/summaries_only_88k/run/summaries_only/mechanism_evaluation.npz
+5dd0b966d1f5e559c043a6f9a3839db27a5b925b166a22151fe214b0ced5da26  training_contrasts/summaries_only_88k/run/summaries_only/report.json
+81e696d8884babe20d3ad2fbf8aa02df295c37de377ccde1c4d28dfea46dff6d  training_contrasts/summaries_only_8k/logs/attempt1_20260921_171408.log
+f6f15da576d61f5ac1e30845150720b612a684f4894419f7496f117b03e1eb39  training_contrasts/summaries_only_8k/logs/command.txt
+bb6eed22493ea94f28b84af4b0ecad2040f4f1cd15ef4b04d2e593ee7a758586  training_contrasts/summaries_only_8k/logs/harness_run.json
+84c18c4cefa8c74ffba8f69649dc4e42f264b33ae0debeabb32d153769c1e46e  training_contrasts/summaries_only_8k/logs/memwatch.log
+e09ee3496c911fce8ad54cc02b33106374857c83b35b184216af0d8eb9a3789b  training_contrasts/summaries_only_8k/run/DEVELOPMENT.json
+3a2b982cc11c1403bba50470ac9cbe7d05fc978142fccb09bc34d22990efa6f6  training_contrasts/summaries_only_8k/run/DONE.json
+1762b29f7d5fa0c41a3f42b0e373f4e154611a2a56b6eb3bb70d5f325e928478  training_contrasts/summaries_only_8k/run/PLAN.json
+e1e8b7dc85095fa2cedfbfd037a3451d3608955ce097c1058d90fa7c95de7b32  training_contrasts/summaries_only_8k/run/summaries_only.final.pt
+1faa161d4d963200606f236ec77f09f744a72b402bdd724b2dafeb8c27e4980b  training_contrasts/summaries_only_8k/run/summaries_only/DONE.json
+62d214164ddf826a9aa5c36472b8e77df4eb8ce0a08c4d441a977cdb7dc12955  training_contrasts/summaries_only_8k/run/summaries_only/checkpoint.pt
+0b5fd85d829cab037c726286a4bc106c425f2939a7a63dc17c075e5c0ec778fe  training_contrasts/summaries_only_8k/run/summaries_only/mechanism_evaluation.npz
+740aed007b3f82c787187f03f10d202ccab6c51205bf6aa5c9cc6cd23bdb52d2  training_contrasts/summaries_only_8k/run/summaries_only/report.json

ablations/converged_posterior_12500/logs/attempt1_20260916_222119.log

@@ -0,0 +1,27 @@
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/sidecars/refit6_dropin/chain0_level32.npz"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/sidecars/refit6_dropin/chain0_level32.npz"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/sidecars/refit6_dropin/chain0_level32.npz"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/sidecars/refit6_dropin/chain0_level32.npz"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+/cloud/cloud-ssd1/dlcoal/venv/lib/python3.10/site-packages/torch/distributed/c10d_logger.py:83: UserWarning: barrier(): using the device under current context. You can specify `device_id` in `init_process_group` to mute this warning.
+  return func(*args, **kwargs)
+[rank0]:[W916 22:34:34.047336109 ProcessGroupNCCL.cpp:5310] Guessing device ID based on global rank. This can cause a hang if rank to GPU mapping is heterogeneous. You can specify device_id in init_process_group()
+{"arm": "raw_plus_all_sidecar", "epoch": 1, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3874870885334078, "fit_seconds": 2099.781686428003, "inner_raw_sites_consumed": 206175097, "inner_seconds": 43.7657260739943, "inner_validation_mae": 0.24260066779330372, "inner_validation_rmse": 0.36720536727036024, "learning_rate": 0.0002951444302461922, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 2, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.35157089066779285, "fit_seconds": 2020.885373572004, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.76424893899821, "inner_validation_mae": 0.23346009119972586, "inner_validation_rmse": 0.36140650023232634, "learning_rate": 0.0002809086200392825, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 3, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3405153757004071, "fit_seconds": 1983.8647596539813, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.52227248996496, "inner_validation_mae": 0.22293252874538302, "inner_validation_rmse": 0.34500043445112416, "learning_rate": 0.000258262716319083, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 4, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3326181621285621, "fit_seconds": 1937.9455639480148, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.63561340601882, "inner_validation_mae": 0.21684347880259155, "inner_validation_rmse": 0.33961601418682646, "learning_rate": 0.00022875, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 5, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3242470915037239, "fit_seconds": 1917.5532338590128, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.86571745702531, "inner_validation_mae": 0.21324309011921286, "inner_validation_rmse": 0.3313257635639196, "learning_rate": 0.00019438171392710925, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 6, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31854615230379535, "fit_seconds": 1909.9167725280276, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.43009391799569, "inner_validation_mae": 0.22143272884190082, "inner_validation_rmse": 0.33546912790565625, "learning_rate": 0.00015750000000000003, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 7, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31350130430094064, "fit_seconds": 1911.4870361229987, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.303684029961005, "inner_validation_mae": 0.20916711370274424, "inner_validation_rmse": 0.3273764426346878, "learning_rate": 0.00012061828607289084, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 8, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3084819702345976, "fit_seconds": 1906.5582509580418, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.00038634199882, "inner_validation_mae": 0.2094024477340281, "inner_validation_rmse": 0.3275464521364425, "learning_rate": 8.625000000000005e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 9, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.304159666418543, "fit_seconds": 1894.5831168919685, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.70847871003207, "inner_validation_mae": 0.20332511628791688, "inner_validation_rmse": 0.323939514300953, "learning_rate": 5.6737283680917e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 10, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3004740411733424, "fit_seconds": 1867.3456651599845, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.34333942498779, "inner_validation_mae": 0.20176592399552465, "inner_validation_rmse": 0.3227285966947815, "learning_rate": 3.4091379960717494e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 11, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.29723618097072063, "fit_seconds": 1890.3383352709934, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.875337061006576, "inner_validation_mae": 0.19974477234110236, "inner_validation_rmse": 0.32016222555350266, "learning_rate": 1.9855569753807785e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 12, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.2952669044399332, "fit_seconds": 1862.5043647200218, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.08403986803023, "inner_validation_mae": 0.19794908697158098, "inner_validation_rmse": 0.31870359428220024, "learning_rate": 1.4999999999999999e-05, "seed": 20260728, "world_size": 4}

ablations/converged_posterior_12500/logs/command.txt

@@ -0,0 +1,3 @@
+cwd /cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32
+env DLCOAL_TRAIN_DEPLOY=/cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32 DLCOAL_COHORT_SPLIT=99996,5000,4248 DLCOAL_FORMAL_EPOCHS=12 OMP_NUM_THREADS=8 PYTHONDONTWRITEBYTECODE=1 CUBLAS_WORKSPACE_CONFIG=:4096:8 MASTER_ADDR=127.0.0.1 MASTER_PORT=29711
+/cloud/cloud-ssd1/dlcoal/venv/bin/torchrun --rdzv-backend=c10d --rdzv-endpoint=127.0.0.1:29711 --nnodes=1 --nproc_per_node=4 -m training.carrier_raw_summary_fusion_runner --registry-dir /cloud/cloud-ssd1/dlcoal/registry109k --unphased-cache /cloud/cloud-ssd1/dlcoal/level_caches109k/chain0_level32/unphased_evidence_cache.npz --sidecar-npz /cloud/cloud-ssd1/dlcoal/ablation_20260916/sidecars/refit6_dropin/chain0_level32.npz --expected-selection-rows 4248 --out-dir /cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/refit6_fit12500_e12 --stage pilot --development-single-arm raw_plus_all_sidecar --seed 20260728 --epochs 12 --formal-epochs 12 --pilot-rows 13500 --pilot-fit-rows 12500 --pilot-inner-rows 500 --pilot-evaluation-rows 500 --development-gpu-class 4090 --packed-hap-bytes 16 --d-model 128 --host-cache-gib 48 --host-budget-fraction transient=0.27 --host-budget-fraction artifact=0.16 --host-budget-fraction segment_plan=0.08 --host-budget-fraction map_exposure=0.015 --host-budget-fraction map=0.02 --host-budget-fraction shift=0.005 --host-budget-fraction prefetch=0.35 --prefetch-rows 10 --stage-workers 4 --host-cache-sites 20000000000 --accept-cache-refill --resume 

ablations/converged_posterior_12500/logs/harness_run.json

@@ -0,0 +1,29 @@
+{
+  "run": "refit6_fit12500_e12",
+  "started_utc": "2026-09-16T22:21:19Z",
+  "launcher": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/bin/train_ablation.sh",
+  "launcher_sha256": "31689adea48e27b63cda2cad9c0b1ae5a34cb42f5d3285d32834ce578e5398e6",
+  "deploy": "/cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32",
+  "deploy_manifest_sha256": "659f0427dc079a0bdb2ad3ae8ff8a328e98df6eeeb623532c34cb9a2bcd0b502",
+  "arm": "raw_plus_all_sidecar",
+  "fit_rows": 12500,
+  "inner_rows": 500,
+  "evaluation_rows": 500,
+  "epochs": 12,
+  "seed": 20260728,
+  "d_model": 128,
+  "world_size": 4,
+  "gpu_class": "4090",
+  "registry": "/cloud/cloud-ssd1/dlcoal/registry109k",
+  "unphased_cache": "/cloud/cloud-ssd1/dlcoal/level_caches109k/chain0_level32/unphased_evidence_cache.npz",
+  "sidecar_npz": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/sidecars/refit6_dropin/chain0_level32.npz",
+  "sidecar_bytes": 107621430,
+  "host_cache_gib": 48,
+  "extra_args": "",
+  "strict_budgets": true,
+  "master_port": 29711,
+  "smoke": false,
+  "out_dir": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/refit6_fit12500_e12",
+  "log_dir": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/refit6_fit12500_e12",
+  "free_gb_at_start": 157
+}

ablations/converged_posterior_12500/logs/refit6_fit12500_e12.out

@@ -0,0 +1,43 @@
+=== data disk free 157 GB (floor 30 GB)
+=== deploy-hash guard: the three runner files match /cloud/cloud-ssd1/dlcoal/prep_fixed32/EXPECTED_SHA256.txt
+=== deploy verified: manifest 659f0427dc079a0bdb2ad3ae8ff8a328e98df6eeeb623532c34cb9a2bcd0b502 (34 artifacts) at /cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32
+=== row-count/rank deadlock pre-check (the deploy's rank guard)
+=== FIT=12500 at 4 ranks: 391 optimizer steps per epoch on every rank, world 1's
+=== memwatch every 60s -> /cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/refit6_fit12500_e12/memwatch.log (cgroup memory.max 398358216704)
+=== RUN=refit6_fit12500_e12 ARM=raw_plus_all_sidecar FIT=12500 INNER=500 EVAL=500 SEED=20260728 EPOCHS=12 D_MODEL=128 SMOKE=0
+=== 4 x 4090  host cache 48 GiB  OMP=8  stage workers 4  prefetch 10
+=== cache=/cloud/cloud-ssd1/dlcoal/level_caches109k/chain0_level32/unphased_evidence_cache.npz
+=== sidecar=/cloud/cloud-ssd1/dlcoal/ablation_20260916/sidecars/refit6_dropin/chain0_level32.npz
+=== OUT=/cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/refit6_fit12500_e12
+=== logs=/cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/refit6_fit12500_e12  retries=2  MASTER_PORT=29711
+=== 09-16 22:21:19 attempt 1/3, log /cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/refit6_fit12500_e12/attempt1_20260916_222119.log
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/sidecars/refit6_dropin/chain0_level32.npz"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/sidecars/refit6_dropin/chain0_level32.npz"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/sidecars/refit6_dropin/chain0_level32.npz"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/sidecars/refit6_dropin/chain0_level32.npz"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+/cloud/cloud-ssd1/dlcoal/venv/lib/python3.10/site-packages/torch/distributed/c10d_logger.py:83: UserWarning: barrier(): using the device under current context. You can specify `device_id` in `init_process_group` to mute this warning.
+  return func(*args, **kwargs)
+[rank0]:[W916 22:34:34.047336109 ProcessGroupNCCL.cpp:5310] Guessing device ID based on global rank. This can cause a hang if rank to GPU mapping is heterogeneous. You can specify device_id in init_process_group()
+{"arm": "raw_plus_all_sidecar", "epoch": 1, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3874870885334078, "fit_seconds": 2099.781686428003, "inner_raw_sites_consumed": 206175097, "inner_seconds": 43.7657260739943, "inner_validation_mae": 0.24260066779330372, "inner_validation_rmse": 0.36720536727036024, "learning_rate": 0.0002951444302461922, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 2, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.35157089066779285, "fit_seconds": 2020.885373572004, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.76424893899821, "inner_validation_mae": 0.23346009119972586, "inner_validation_rmse": 0.36140650023232634, "learning_rate": 0.0002809086200392825, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 3, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3405153757004071, "fit_seconds": 1983.8647596539813, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.52227248996496, "inner_validation_mae": 0.22293252874538302, "inner_validation_rmse": 0.34500043445112416, "learning_rate": 0.000258262716319083, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 4, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3326181621285621, "fit_seconds": 1937.9455639480148, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.63561340601882, "inner_validation_mae": 0.21684347880259155, "inner_validation_rmse": 0.33961601418682646, "learning_rate": 0.00022875, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 5, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3242470915037239, "fit_seconds": 1917.5532338590128, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.86571745702531, "inner_validation_mae": 0.21324309011921286, "inner_validation_rmse": 0.3313257635639196, "learning_rate": 0.00019438171392710925, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 6, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31854615230379535, "fit_seconds": 1909.9167725280276, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.43009391799569, "inner_validation_mae": 0.22143272884190082, "inner_validation_rmse": 0.33546912790565625, "learning_rate": 0.00015750000000000003, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 7, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31350130430094064, "fit_seconds": 1911.4870361229987, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.303684029961005, "inner_validation_mae": 0.20916711370274424, "inner_validation_rmse": 0.3273764426346878, "learning_rate": 0.00012061828607289084, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 8, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3084819702345976, "fit_seconds": 1906.5582509580418, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.00038634199882, "inner_validation_mae": 0.2094024477340281, "inner_validation_rmse": 0.3275464521364425, "learning_rate": 8.625000000000005e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 9, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.304159666418543, "fit_seconds": 1894.5831168919685, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.70847871003207, "inner_validation_mae": 0.20332511628791688, "inner_validation_rmse": 0.323939514300953, "learning_rate": 5.6737283680917e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 10, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3004740411733424, "fit_seconds": 1867.3456651599845, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.34333942498779, "inner_validation_mae": 0.20176592399552465, "inner_validation_rmse": 0.3227285966947815, "learning_rate": 3.4091379960717494e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 11, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.29723618097072063, "fit_seconds": 1890.3383352709934, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.875337061006576, "inner_validation_mae": 0.19974477234110236, "inner_validation_rmse": 0.32016222555350266, "learning_rate": 1.9855569753807785e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 12, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.2952669044399332, "fit_seconds": 1862.5043647200218, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.08403986803023, "inner_validation_mae": 0.19794908697158098, "inner_validation_rmse": 0.31870359428220024, "learning_rate": 1.4999999999999999e-05, "seed": 20260728, "world_size": 4}
+runner exit=0 (attempt 1)
+=== launcher wall time 24775 s
+launcher exit=0

ablations/converged_posterior_12500/logs/refit6_fit12500_e12.supervisor.log

@@ -0,0 +1,4 @@
+09-16 22:21:16 supervisor start pid 628915 for RUN=refit6_fit12500_e12 (max relaunch 4, unknown 1, interval 120s)
+09-16 22:21:16 OUT=/cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/refit6_fit12500_e12  FIT=12500 EPOCHS=12 SEED=20260728 ARM=raw_plus_all_sidecar SIDECAR=/cloud/cloud-ssd1/dlcoal/ablation_20260916/sidecars/refit6_dropin/chain0_level32.npz
+09-16 22:21:16 first launch -> refit6_fit12500_e12.out (RESUME_OK=0)
+09-17 05:15:34 DONE.json present; supervisor exits

ablations/converged_posterior_12500/run/DEVELOPMENT.json

@@ -0,0 +1,580 @@
+{
+  "accumulate_rows": 32,
+  "arm_reports": [
+    {
+      "all_raw_sites_consumed_when_raw_enabled": true,
+      "arm": "raw_plus_all_sidecar",
+      "best_epoch": 12,
+      "best_inner_validation_rmse": 0.31870359428220024,
+      "checkpoint_artifact": {
+        "path": "checkpoint.pt",
+        "sha256": "10b35b0daa22087f88d21133ac9cf4ff1b26dec92afb0395f27eea0a20c78141"
+      },
+      "curve": [
+        {
+          "epoch": 1,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3874870885334078,
+          "fit_seconds": 2099.781686428003,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 43.7657260739943,
+          "inner_validation_mae": 0.24260066779330372,
+          "inner_validation_rmse": 0.36720536727036024,
+          "learning_rate": 0.0002951444302461922,
+          "world_size": 4
+        },
+        {
+          "epoch": 2,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.35157089066779285,
+          "fit_seconds": 2020.885373572004,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.76424893899821,
+          "inner_validation_mae": 0.23346009119972586,
+          "inner_validation_rmse": 0.36140650023232634,
+          "learning_rate": 0.0002809086200392825,
+          "world_size": 4
+        },
+        {
+          "epoch": 3,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3405153757004071,
+          "fit_seconds": 1983.8647596539813,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.52227248996496,
+          "inner_validation_mae": 0.22293252874538302,
+          "inner_validation_rmse": 0.34500043445112416,
+          "learning_rate": 0.000258262716319083,
+          "world_size": 4
+        },
+        {
+          "epoch": 4,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3326181621285621,
+          "fit_seconds": 1937.9455639480148,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.63561340601882,
+          "inner_validation_mae": 0.21684347880259155,
+          "inner_validation_rmse": 0.33961601418682646,
+          "learning_rate": 0.00022875,
+          "world_size": 4
+        },
+        {
+          "epoch": 5,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3242470915037239,
+          "fit_seconds": 1917.5532338590128,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.86571745702531,
+          "inner_validation_mae": 0.21324309011921286,
+          "inner_validation_rmse": 0.3313257635639196,
+          "learning_rate": 0.00019438171392710925,
+          "world_size": 4
+        },
+        {
+          "epoch": 6,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.31854615230379535,
+          "fit_seconds": 1909.9167725280276,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.43009391799569,
+          "inner_validation_mae": 0.22143272884190082,
+          "inner_validation_rmse": 0.33546912790565625,
+          "learning_rate": 0.00015750000000000003,
+          "world_size": 4
+        },
+        {
+          "epoch": 7,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.31350130430094064,
+          "fit_seconds": 1911.4870361229987,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.303684029961005,
+          "inner_validation_mae": 0.20916711370274424,
+          "inner_validation_rmse": 0.3273764426346878,
+          "learning_rate": 0.00012061828607289084,
+          "world_size": 4
+        },
+        {
+          "epoch": 8,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3084819702345976,
+          "fit_seconds": 1906.5582509580418,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.00038634199882,
+          "inner_validation_mae": 0.2094024477340281,
+          "inner_validation_rmse": 0.3275464521364425,
+          "learning_rate": 8.625000000000005e-05,
+          "world_size": 4
+        },
+        {
+          "epoch": 9,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.304159666418543,
+          "fit_seconds": 1894.5831168919685,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.70847871003207,
+          "inner_validation_mae": 0.20332511628791688,
+          "inner_validation_rmse": 0.323939514300953,
+          "learning_rate": 5.6737283680917e-05,
+          "world_size": 4
+        },
+        {
+          "epoch": 10,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3004740411733424,
+          "fit_seconds": 1867.3456651599845,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.34333942498779,
+          "inner_validation_mae": 0.20176592399552465,
+          "inner_validation_rmse": 0.3227285966947815,
+          "learning_rate": 3.4091379960717494e-05,
+          "world_size": 4
+        },
+        {
+          "epoch": 11,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.29723618097072063,
+          "fit_seconds": 1890.3383352709934,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.875337061006576,
+          "inner_validation_mae": 0.19974477234110236,
+          "inner_validation_rmse": 0.32016222555350266,
+          "learning_rate": 1.9855569753807785e-05,
+          "world_size": 4
+        },
+        {
+          "epoch": 12,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.2952669044399332,
+          "fit_seconds": 1862.5043647200218,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.08403986803023,
+          "inner_validation_mae": 0.19794908697158098,
+          "inner_validation_rmse": 0.31870359428220024,
+          "learning_rate": 1.4999999999999999e-05,
+          "world_size": 4
+        }
+      ],
+      "device": "cuda",
+      "direct_64_bin_output": true,
+      "epochs_ran": 12,
+      "evaluation_raw_sites_consumed": 216060598,
+      "evaluation_seconds": 193.62224856001558,
+      "experiment_signature": "3c52f767133c97ef0391e1f85c26fd55fc3fd8eae0b62af74d509e2faedd378d",
+      "fit_rows": 12500,
+      "fit_seconds": 23202.764159113052,
+      "fixed_epoch_budget": 12,
+      "gpu_name": "NVIDIA GeForce RTX 4090",
+      "hazard_tail_survival_dual_or_class_head": false,
+      "initial_state_sha256": "7ecf840102b35595f25ab8b6f6d6fbe36ad9fed367b184e33444a2f76b85b100",
+      "inner_rows": 500,
+      "inner_validation_seconds": 451.2989377200138,
+      "mechanism_evaluation_artifact": {
+        "path": "mechanism_evaluation.npz",
+        "sha256": "c89a32ad70612558a1816f0192666e8fb5e99150e309c8726e903203f95d906d"
+      },
+      "mechanism_evaluation_metrics": {
+        "constant_predictor_rmse": 0.402485317024815,
+        "label_spread_strata": {
+          "flat": {
+            "constant_predictor_rmse": 0.15479404880509584,
+            "label_spread_range": [
+              0.0,
+              0.24
+            ],
+            "mae": 0.11638418236762528,
+            "n_samples": 123,
+            "rmse": 0.1666460079969525
+          },
+          "moderate": {
+            "constant_predictor_rmse": 0.31611850295071287,
+            "label_spread_range": [
+              0.24,
+              0.4
+            ],
+            "mae": 0.17969141369589517,
+            "n_samples": 194,
+            "rmse": 0.25780015022998143
+          },
+          "structured": {
+            "constant_predictor_rmse": 0.5661845833240783,
+            "label_spread_range": [
+              0.4,
+              null
+            ],
+            "mae": 0.2753556079226113,
+            "n_samples": 183,
+            "rmse": 0.4279684648403553
+          }
+        },
+        "mae": 0.1991309298761189,
+        "n_samples": 500,
+        "per_bin_mae": [
+          0.24541986513137817,
+          0.24275072979927062,
+          0.2361834740638733,
+          0.22871926307678223,
+          0.21860558199882507,
+          0.2107172451019287,
+          0.20524972558021545,
+          0.2006639165878296,
+          0.19632365822792053,
+          0.20003465056419373,
+          0.19908141922950745,
+          0.19513570833206176,
+          0.1975396203994751,
+          0.19236923027038574,
+          0.18944601798057556,
+          0.1762158832550049,
+          0.1797312889099121,
+          0.17341783261299135,
+          0.17471158719062804,
+          0.17241287088394164,
+          0.17414124393463135,
+          0.17465080666542054,
+          0.17217784476280212,
+          0.1664047372341156,
+          0.1636147611141205,
+          0.16704572343826293,
+          0.16207540321350097,
+          0.16396289110183715,
+          0.1616450309753418,
+          0.15843560600280762,
+          0.15533437490463256,
+          0.1543867049217224,
+          0.1500585241317749,
+          0.15423594236373903,
+          0.15794178581237792,
+          0.16371115636825562,
+          0.16855746603012084,
+          0.17061732721328735,
+          0.19002954244613648,
+          0.19865961551666259,
+          0.20483355951309204,
+          0.20369744729995728,
+          0.2009083924293518,
+          0.19554463720321655,
+          0.18747902011871337,
+          0.19448473501205443,
+          0.19685277557373046,
+          0.20001736974716186,
+          0.20405663537979127,
+          0.2141436767578125,
+          0.214137704372406,
+          0.21937893915176393,
+          0.22215965843200683,
+          0.22793112421035766,
+          0.232182879447937,
+          0.23632795763015746,
+          0.23525073957443238,
+          0.2353964352607727,
+          0.23886531734466554,
+          0.24779381036758422,
+          0.25894999980926514,
+          0.25870959854125974,
+          0.2666563906669617,
+          0.28620465087890623
+        ],
+        "per_bin_rmse": [
+          0.32389942541581696,
+          0.3221050596170423,
+          0.31547731993166983,
+          0.3085287089747273,
+          0.29654394675620016,
+          0.2868082326356518,
+          0.2826639791828525,
+          0.27855689560038577,
+          0.27563319517660295,
+          0.28917464301230367,
+          0.29132474834869276,
+          0.28294229651123515,
+          0.29396326000817713,
+          0.2899133402273961,
+          0.28902025737165143,
+          0.2626366945341214,
+          0.27792930151651957,
+          0.2655847500881156,
+          0.2653641756809541,
+          0.26178172722476284,
+          0.2577647923958844,
+          0.2557381725169564,
+          0.25850987397502895,
+          0.23835091515404838,
+          0.24720428245005976,
+          0.26378937004988584,
+          0.25246553298877694,
+          0.24396350187800078,
+          0.2371889523905255,
+          0.23113402850054382,
+          0.22573255281408236,
+          0.22636431967295242,
+          0.22312301899911913,
+          0.24099931135469332,
+          0.25037903901054653,
+          0.2604456970960359,
+          0.2698299655162242,
+          0.2739312065935742,
+          0.312408364845823,
+          0.33982437251592745,
+          0.3509942049613907,
+          0.3485459183554334,
+          0.3412307920446075,
+          0.3303430885999946,
+          0.31690196754189076,
+          0.33891677716679264,
+          0.34304589894691867,
+          0.3417638096252389,
+          0.35209537511151395,
+          0.3735255952647202,
+          0.37070427840789777,
+          0.3735245450460494,
+          0.380085053737494,
+          0.38439452139715974,
+          0.38562083221577537,
+          0.39368245684457076,
+          0.3912439947511535,
+          0.389836819084466,
+          0.3934634475971372,
+          0.3978874729131184,
+          0.4148704753583624,
+          0.4122615862266757,
+          0.4211285521440208,
+          0.4449343652233928
+        ],
+        "rmse": 0.31568012279194996
+      },
+      "mechanism_evaluation_rows": 500,
+      "mechanism_evaluation_rows_requested": 500,
+      "model_name": "carrier_raw_summary_capacity_matched_direct64_nw_residual_v1",
+      "model_parameters": 1026409,
+      "optimizer_boundaries_sha256": "ce18181bfb956b04c297d2cb57aa005b40b1e38d281c91b7bb45b430faf64adf",
+      "peak_cuda_allocated_bytes": 12191421440,
+      "peak_cuda_reserved_bytes": 13323206656,
+      "publication_benchmark": false,
+      "row_order_sha256": "ff48ba7533087a7f438b18488c5d13f2260e3734220b11666507796956c2689f",
+      "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+      "seed": 20260728,
+      "segment_reduction_backend": "padded",
+      "selection_access_count": 0,
+      "selection_rows_read": 0,
+      "stage": "pilot",
+      "status": "complete",
+      "summary_only_skips_raw_io": false,
+      "trainable_parameters": 1026409,
+      "training_objective": {
+        "mode_weight_alpha": 0.0,
+        "principal_basis_rows": 0,
+        "spectral_decoupling": 0.0
+      },
+      "training_raw_sites_consumed": 18535440261,
+      "training_seconds": 23202.764159113052
+    }
+  ],
+  "development_single_arm": "raw_plus_all_sidecar",
+  "development_state_control": "carrier_correct",
+  "epochs": 12,
+  "evidence_description": {
+    "sidecar": {
+      "diploid_set_binding": "filename_only: the sidecar records no diploid identity; its chain-0 level-32 set is asserted by its basename, and checked only offline by verify_cohort_artifacts [6] against the PSMC+ store",
+      "members": [
+        "sidecar",
+        "sidecar_present"
+      ],
+      "present_flag": "declared",
+      "row_binding": "positional: sidecar row i describes cache row i"
+    },
+    "view": {
+      "column_order": "ladder_columns exactly as the level-cache builder stored them (permutation order); never sorted, never re-derived",
+      "columns_member": "ladder_columns",
+      "every_row_carries_every_stream": true,
+      "nw_log10_one_step_rows": 0,
+      "nw_log10_rule": "per cache row: nw_log10 == float32(log10(max(nw_proxy, tiny))) to within one float32 step",
+      "registry_context_log2_haplotypes": 7.0,
+      "site_count_oracle": "per row: |nw_proxy * 4 * a_64 * mutation_exposure - view sites| < 0.5",
+      "site_rule": "a site is staged iff 0 < derived < called over the view's 64 haplotypes; a half-called diploid contributes its one observed allele"
+    }
+  },
+  "evidence_identity": {
+    "sidecar": {
+      "array_sha256": "a4c09638f3bac6e84a7ad6b2800606ed29106cd325dead2397ca519b4cafb397",
+      "basename": "chain0_level32.npz",
+      "file_sha256": "3b44482f739e0da4fc46c702eb22b0c97fccb80325d555f50748222ccbddb63a",
+      "present_rows": 104996,
+      "present_sha256": "c4e31d9b42a839fef1f9f158fd70ef3079e33fcb6dfc5bb38c43c54ec97e14cf",
+      "rows": 104996,
+      "width": 256
+    },
+    "view": {
+      "chain": 0,
+      "columns_sha256": "17ea45c1d3bf97d51c78698020497c8aeb1460160e7f0a5325ec7041010b0791",
+      "columns_shape": [
+        104996,
+        32
+      ],
+      "context_log2_haplotypes": 6.0,
+      "diploids": 32,
+      "nw_proxy_sha256": "81a36cf53d0ca56d92ffa52285085c39a3206d044fa418a8a5f692b1514b24d3",
+      "row_diploids": 64,
+      "schema": "dlcoalsim-fixed-diploid-view-v1",
+      "stream_presence_sha256": "821039ed142a7cc57bfad84a169bc6f5619100d27504d6b948f9c12d74c2c608"
+    }
+  },
+  "experiment_signature": "3c52f767133c97ef0391e1f85c26fd55fc3fd8eae0b62af74d509e2faedd378d",
+  "formal_authorization": false,
+  "gate_eligible": false,
+  "ibs_length_bins": [],
+  "loader_performance": {
+    "artifact_cache": {
+      "bytes": 1952639996,
+      "entries": 33,
+      "evictions": 26531,
+      "hits": 40,
+      "label": "packed_development_rows",
+      "max_bytes": 2061584302,
+      "max_sites": 5000000000,
+      "misses": 26564,
+      "rejected_oversize": 0,
+      "sites": 76208682
+    },
+    "artifact_opens": 26564,
+    "cached_row_keyspace": "registry development indices only",
+    "cpu_dense_state_bytes_avoided": 683211142272,
+    "cpu_dense_state_h2d_bytes": 0,
+    "cpu_dense_states_cached": 0,
+    "gpu_state_controls_staged": 0,
+    "map_cache": {
+      "bytes": 257330048,
+      "entries": 1126,
+      "evictions": 23818,
+      "hits": 26264,
+      "label": "observed_or_deranged_maps",
+      "max_bytes": 257698037,
+      "max_sites": 10000000000,
+      "misses": 24944,
+      "rejected_oversize": 0,
+      "sites": 16082002
+    },
+    "map_exposure_cache": {
+      "bytes": 193220608,
+      "entries": 2576,
+      "evictions": 17805,
+      "hits": 5223,
+      "label": "map_exposure",
+      "max_bytes": 193273528,
+      "max_sites": 10000000000,
+      "misses": 20381,
+      "rejected_oversize": 0,
+      "sites": 6038144
+    },
+    "map_exposure_requests": 25604,
+    "map_requests": 51208,
+    "packed_call_bytes_staged": 402354176828,
+    "pinned_bytes_staged": 481690117144,
+    "prefetch": {
+      "max_bytes": 4509715660,
+      "max_sites": 400000000,
+      "peak_bytes": 957711124,
+      "peak_sites": 38104715,
+      "queue_rows": 10,
+      "stage_workers": 4
+    },
+    "row_sites_staged": 21350348196,
+    "rows_loaded": 26604,
+    "rows_staged": 25604,
+    "segment_plan_cache": {
+      "bytes": 112849120,
+      "entries": 10838,
+      "evictions": 0,
+      "hits": 14766,
+      "label": "carrier_segment_plans",
+      "max_bytes": 1030792151,
+      "max_sites": 5000000000,
+      "misses": 10838,
+      "rejected_oversize": 0,
+      "sites": 4348553669
+    },
+    "selection_rows_cached": 0,
+    "selection_rows_read": 0,
+    "shift_cache": {
+      "bytes": 0,
+      "entries": 0,
+      "evictions": 0,
+      "hits": 0,
+      "label": "carrier_control_shifts",
+      "max_bytes": 64424509,
+      "max_sites": 5000000000,
+      "misses": 0,
+      "rejected_oversize": 0,
+      "sites": 0
+    },
+    "sites_staged": 19266247123,
+    "transient_artifact": {
+      "max_bytes": 3478923509,
+      "max_sites": 1050000000,
+      "peak_bytes": 886371872,
+      "peak_sites": 33217818
+    },
+    "view_site_count_checks": 25604
+  },
+  "model_config": {
+    "bin_attention_temperature": false,
+    "carrier_pool_seeds": 0,
+    "carrier_region_chunk": 512,
+    "carrier_scale_levels": [],
+    "carrier_tree_levels": [],
+    "carrier_width": 32,
+    "chunk_sites": 65536,
+    "coalescence_auxiliary": false,
+    "context_dim": 2,
+    "d_model": 128,
+    "dropout": 0.0,
+    "extreme_tap_levels": [],
+    "individual_width": 32,
+    "leaf_width_bp": 100,
+    "long_block_dilation_cap": 0,
+    "macro_levels": 7,
+    "n_heads": 4,
+    "n_long_blocks": 11,
+    "n_region_scales": 4,
+    "n_regional_blocks": 3,
+    "ordered_leaf_scan_width": 0,
+    "overlapping_regions": false,
+    "pair_concordance_first_order": false,
+    "pair_concordance_scales_bp": [],
+    "pair_concordance_width": 8,
+    "region_downsample_mode": "conv",
+    "region_sfs_bins": 0,
+    "scale_film": false,
+    "segment_reduction_backend": "padded",
+    "sidecar_dim": 256,
+    "site_genetic_span_features": false,
+    "site_lag_span_features": false,
+    "site_ld_lags": [],
+    "site_pair_identity_lags": [],
+    "site_state_width": 0,
+    "site_width": 8,
+    "smooth_time_query_modes": 0,
+    "tied_row_attention_heads": 0,
+    "tree_tap_levels": [],
+    "use_checkpoint": false,
+    "use_site_frequency": false
+  },
+  "objective": {
+    "mode_weight_alpha": 0.0,
+    "spectral_decoupling": 0.0
+  },
+  "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+  "score_resume_from": "",
+  "seed": 20260728,
+  "selection_access_count": 0,
+  "selection_rows_read": 0,
+  "stage": "development_single_arm",
+  "stage_rows": {
+    "census": false,
+    "evaluation": 500,
+    "fit": 12500,
+    "inner": 500
+  },
+  "status": "complete"
+}

ablations/converged_posterior_12500/run/DONE.json

@@ -0,0 +1,11 @@
+{
+  "experiment_signature": "3c52f767133c97ef0391e1f85c26fd55fc3fd8eae0b62af74d509e2faedd378d",
+  "gate_eligible": false,
+  "report_sha256": "1164d581e772b64f9f592f62e29d5e2dc4aa3f59e57f2193c8811cabb0d9bf48",
+  "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+  "seed": 20260728,
+  "selection_access_count": 0,
+  "selection_rows_read": 0,
+  "stage": "development_single_arm",
+  "status": "complete"
+}

ablations/converged_posterior_12500/run/PLAN.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8a14982cea247a926fbcaa569fc0f2bec9760670aee000ae156750524d99fb2f
+size 29598005

ablations/converged_posterior_12500/run/all_inputs.final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b0876dab6433080384933a3df9602762c5e7926817c8fa0b18ef46e8af1df86
+size 16735186

ablations/converged_posterior_12500/run/all_inputs/DONE.json

@@ -0,0 +1,19 @@
+{
+  "arm": "raw_plus_all_sidecar",
+  "checkpoint_artifact": {
+    "path": "checkpoint.pt",
+    "sha256": "10b35b0daa22087f88d21133ac9cf4ff1b26dec92afb0395f27eea0a20c78141"
+  },
+  "experiment_signature": "3c52f767133c97ef0391e1f85c26fd55fc3fd8eae0b62af74d509e2faedd378d",
+  "mechanism_evaluation_artifact": {
+    "path": "mechanism_evaluation.npz",
+    "sha256": "c89a32ad70612558a1816f0192666e8fb5e99150e309c8726e903203f95d906d"
+  },
+  "report_sha256": "c84c0f7284797609b4780052fb7e84227a4e8cc6de5c2b280d8013155044e1f0",
+  "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+  "seed": 20260728,
+  "selection_access_count": 0,
+  "selection_rows_read": 0,
+  "stage": "pilot",
+  "status": "complete"
+}

ablations/converged_posterior_12500/run/all_inputs/checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10b35b0daa22087f88d21133ac9cf4ff1b26dec92afb0395f27eea0a20c78141
+size 4169376

ablations/converged_posterior_12500/run/all_inputs/mechanism_evaluation.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c89a32ad70612558a1816f0192666e8fb5e99150e309c8726e903203f95d906d
+size 581024

ablations/converged_posterior_12500/run/all_inputs/report.json

@@ -0,0 +1,369 @@
+{
+  "all_raw_sites_consumed_when_raw_enabled": true,
+  "arm": "raw_plus_all_sidecar",
+  "best_epoch": 12,
+  "best_inner_validation_rmse": 0.31870359428220024,
+  "checkpoint_artifact": {
+    "path": "checkpoint.pt",
+    "sha256": "10b35b0daa22087f88d21133ac9cf4ff1b26dec92afb0395f27eea0a20c78141"
+  },
+  "curve": [
+    {
+      "epoch": 1,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3874870885334078,
+      "fit_seconds": 2099.781686428003,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 43.7657260739943,
+      "inner_validation_mae": 0.24260066779330372,
+      "inner_validation_rmse": 0.36720536727036024,
+      "learning_rate": 0.0002951444302461922,
+      "world_size": 4
+    },
+    {
+      "epoch": 2,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.35157089066779285,
+      "fit_seconds": 2020.885373572004,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.76424893899821,
+      "inner_validation_mae": 0.23346009119972586,
+      "inner_validation_rmse": 0.36140650023232634,
+      "learning_rate": 0.0002809086200392825,
+      "world_size": 4
+    },
+    {
+      "epoch": 3,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3405153757004071,
+      "fit_seconds": 1983.8647596539813,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.52227248996496,
+      "inner_validation_mae": 0.22293252874538302,
+      "inner_validation_rmse": 0.34500043445112416,
+      "learning_rate": 0.000258262716319083,
+      "world_size": 4
+    },
+    {
+      "epoch": 4,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3326181621285621,
+      "fit_seconds": 1937.9455639480148,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.63561340601882,
+      "inner_validation_mae": 0.21684347880259155,
+      "inner_validation_rmse": 0.33961601418682646,
+      "learning_rate": 0.00022875,
+      "world_size": 4
+    },
+    {
+      "epoch": 5,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3242470915037239,
+      "fit_seconds": 1917.5532338590128,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.86571745702531,
+      "inner_validation_mae": 0.21324309011921286,
+      "inner_validation_rmse": 0.3313257635639196,
+      "learning_rate": 0.00019438171392710925,
+      "world_size": 4
+    },
+    {
+      "epoch": 6,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.31854615230379535,
+      "fit_seconds": 1909.9167725280276,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.43009391799569,
+      "inner_validation_mae": 0.22143272884190082,
+      "inner_validation_rmse": 0.33546912790565625,
+      "learning_rate": 0.00015750000000000003,
+      "world_size": 4
+    },
+    {
+      "epoch": 7,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.31350130430094064,
+      "fit_seconds": 1911.4870361229987,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.303684029961005,
+      "inner_validation_mae": 0.20916711370274424,
+      "inner_validation_rmse": 0.3273764426346878,
+      "learning_rate": 0.00012061828607289084,
+      "world_size": 4
+    },
+    {
+      "epoch": 8,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3084819702345976,
+      "fit_seconds": 1906.5582509580418,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.00038634199882,
+      "inner_validation_mae": 0.2094024477340281,
+      "inner_validation_rmse": 0.3275464521364425,
+      "learning_rate": 8.625000000000005e-05,
+      "world_size": 4
+    },
+    {
+      "epoch": 9,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.304159666418543,
+      "fit_seconds": 1894.5831168919685,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.70847871003207,
+      "inner_validation_mae": 0.20332511628791688,
+      "inner_validation_rmse": 0.323939514300953,
+      "learning_rate": 5.6737283680917e-05,
+      "world_size": 4
+    },
+    {
+      "epoch": 10,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3004740411733424,
+      "fit_seconds": 1867.3456651599845,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.34333942498779,
+      "inner_validation_mae": 0.20176592399552465,
+      "inner_validation_rmse": 0.3227285966947815,
+      "learning_rate": 3.4091379960717494e-05,
+      "world_size": 4
+    },
+    {
+      "epoch": 11,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.29723618097072063,
+      "fit_seconds": 1890.3383352709934,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.875337061006576,
+      "inner_validation_mae": 0.19974477234110236,
+      "inner_validation_rmse": 0.32016222555350266,
+      "learning_rate": 1.9855569753807785e-05,
+      "world_size": 4
+    },
+    {
+      "epoch": 12,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.2952669044399332,
+      "fit_seconds": 1862.5043647200218,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.08403986803023,
+      "inner_validation_mae": 0.19794908697158098,
+      "inner_validation_rmse": 0.31870359428220024,
+      "learning_rate": 1.4999999999999999e-05,
+      "world_size": 4
+    }
+  ],
+  "device": "cuda",
+  "direct_64_bin_output": true,
+  "epochs_ran": 12,
+  "evaluation_raw_sites_consumed": 216060598,
+  "evaluation_seconds": 193.62224856001558,
+  "experiment_signature": "3c52f767133c97ef0391e1f85c26fd55fc3fd8eae0b62af74d509e2faedd378d",
+  "fit_rows": 12500,
+  "fit_seconds": 23202.764159113052,
+  "fixed_epoch_budget": 12,
+  "gpu_name": "NVIDIA GeForce RTX 4090",
+  "hazard_tail_survival_dual_or_class_head": false,
+  "initial_state_sha256": "7ecf840102b35595f25ab8b6f6d6fbe36ad9fed367b184e33444a2f76b85b100",
+  "inner_rows": 500,
+  "inner_validation_seconds": 451.2989377200138,
+  "mechanism_evaluation_artifact": {
+    "path": "mechanism_evaluation.npz",
+    "sha256": "c89a32ad70612558a1816f0192666e8fb5e99150e309c8726e903203f95d906d"
+  },
+  "mechanism_evaluation_metrics": {
+    "constant_predictor_rmse": 0.402485317024815,
+    "label_spread_strata": {
+      "flat": {
+        "constant_predictor_rmse": 0.15479404880509584,
+        "label_spread_range": [
+          0.0,
+          0.24
+        ],
+        "mae": 0.11638418236762528,
+        "n_samples": 123,
+        "rmse": 0.1666460079969525
+      },
+      "moderate": {
+        "constant_predictor_rmse": 0.31611850295071287,
+        "label_spread_range": [
+          0.24,
+          0.4
+        ],
+        "mae": 0.17969141369589517,
+        "n_samples": 194,
+        "rmse": 0.25780015022998143
+      },
+      "structured": {
+        "constant_predictor_rmse": 0.5661845833240783,
+        "label_spread_range": [
+          0.4,
+          null
+        ],
+        "mae": 0.2753556079226113,
+        "n_samples": 183,
+        "rmse": 0.4279684648403553
+      }
+    },
+    "mae": 0.1991309298761189,
+    "n_samples": 500,
+    "per_bin_mae": [
+      0.24541986513137817,
+      0.24275072979927062,
+      0.2361834740638733,
+      0.22871926307678223,
+      0.21860558199882507,
+      0.2107172451019287,
+      0.20524972558021545,
+      0.2006639165878296,
+      0.19632365822792053,
+      0.20003465056419373,
+      0.19908141922950745,
+      0.19513570833206176,
+      0.1975396203994751,
+      0.19236923027038574,
+      0.18944601798057556,
+      0.1762158832550049,
+      0.1797312889099121,
+      0.17341783261299135,
+      0.17471158719062804,
+      0.17241287088394164,
+      0.17414124393463135,
+      0.17465080666542054,
+      0.17217784476280212,
+      0.1664047372341156,
+      0.1636147611141205,
+      0.16704572343826293,
+      0.16207540321350097,
+      0.16396289110183715,
+      0.1616450309753418,
+      0.15843560600280762,
+      0.15533437490463256,
+      0.1543867049217224,
+      0.1500585241317749,
+      0.15423594236373903,
+      0.15794178581237792,
+      0.16371115636825562,
+      0.16855746603012084,
+      0.17061732721328735,
+      0.19002954244613648,
+      0.19865961551666259,
+      0.20483355951309204,
+      0.20369744729995728,
+      0.2009083924293518,
+      0.19554463720321655,
+      0.18747902011871337,
+      0.19448473501205443,
+      0.19685277557373046,
+      0.20001736974716186,
+      0.20405663537979127,
+      0.2141436767578125,
+      0.214137704372406,
+      0.21937893915176393,
+      0.22215965843200683,
+      0.22793112421035766,
+      0.232182879447937,
+      0.23632795763015746,
+      0.23525073957443238,
+      0.2353964352607727,
+      0.23886531734466554,
+      0.24779381036758422,
+      0.25894999980926514,
+      0.25870959854125974,
+      0.2666563906669617,
+      0.28620465087890623
+    ],
+    "per_bin_rmse": [
+      0.32389942541581696,
+      0.3221050596170423,
+      0.31547731993166983,
+      0.3085287089747273,
+      0.29654394675620016,
+      0.2868082326356518,
+      0.2826639791828525,
+      0.27855689560038577,
+      0.27563319517660295,
+      0.28917464301230367,
+      0.29132474834869276,
+      0.28294229651123515,
+      0.29396326000817713,
+      0.2899133402273961,
+      0.28902025737165143,
+      0.2626366945341214,
+      0.27792930151651957,
+      0.2655847500881156,
+      0.2653641756809541,
+      0.26178172722476284,
+      0.2577647923958844,
+      0.2557381725169564,
+      0.25850987397502895,
+      0.23835091515404838,
+      0.24720428245005976,
+      0.26378937004988584,
+      0.25246553298877694,
+      0.24396350187800078,
+      0.2371889523905255,
+      0.23113402850054382,
+      0.22573255281408236,
+      0.22636431967295242,
+      0.22312301899911913,
+      0.24099931135469332,
+      0.25037903901054653,
+      0.2604456970960359,
+      0.2698299655162242,
+      0.2739312065935742,
+      0.312408364845823,
+      0.33982437251592745,
+      0.3509942049613907,
+      0.3485459183554334,
+      0.3412307920446075,
+      0.3303430885999946,
+      0.31690196754189076,
+      0.33891677716679264,
+      0.34304589894691867,
+      0.3417638096252389,
+      0.35209537511151395,
+      0.3735255952647202,
+      0.37070427840789777,
+      0.3735245450460494,
+      0.380085053737494,
+      0.38439452139715974,
+      0.38562083221577537,
+      0.39368245684457076,
+      0.3912439947511535,
+      0.389836819084466,
+      0.3934634475971372,
+      0.3978874729131184,
+      0.4148704753583624,
+      0.4122615862266757,
+      0.4211285521440208,
+      0.4449343652233928
+    ],
+    "rmse": 0.31568012279194996
+  },
+  "mechanism_evaluation_rows": 500,
+  "mechanism_evaluation_rows_requested": 500,
+  "model_name": "carrier_raw_summary_capacity_matched_direct64_nw_residual_v1",
+  "model_parameters": 1026409,
+  "optimizer_boundaries_sha256": "ce18181bfb956b04c297d2cb57aa005b40b1e38d281c91b7bb45b430faf64adf",
+  "peak_cuda_allocated_bytes": 12191421440,
+  "peak_cuda_reserved_bytes": 13323206656,
+  "publication_benchmark": false,
+  "row_order_sha256": "ff48ba7533087a7f438b18488c5d13f2260e3734220b11666507796956c2689f",
+  "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+  "seed": 20260728,
+  "segment_reduction_backend": "padded",
+  "selection_access_count": 0,
+  "selection_rows_read": 0,
+  "stage": "pilot",
+  "status": "complete",
+  "summary_only_skips_raw_io": false,
+  "trainable_parameters": 1026409,
+  "training_objective": {
+    "mode_weight_alpha": 0.0,
+    "principal_basis_rows": 0,
+    "spectral_decoupling": 0.0
+  },
+  "training_raw_sites_consumed": 18535440261,
+  "training_seconds": 23202.764159113052
+}

ablations/full_12500/logs/attempt1_20260916_073515.log

@@ -0,0 +1,27 @@
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+/cloud/cloud-ssd1/dlcoal/venv/lib/python3.10/site-packages/torch/distributed/c10d_logger.py:83: UserWarning: barrier(): using the device under current context. You can specify `device_id` in `init_process_group` to mute this warning.
+  return func(*args, **kwargs)
+[rank0]:[W916 07:48:12.382149932 ProcessGroupNCCL.cpp:5310] Guessing device ID based on global rank. This can cause a hang if rank to GPU mapping is heterogeneous. You can specify device_id in init_process_group()
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"arm": "raw_plus_all_sidecar", "epoch": 1, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3877548377247502, "fit_seconds": 2098.6562211629935, "inner_raw_sites_consumed": 206175097, "inner_seconds": 43.76968279702123, "inner_validation_mae": 0.24319069107994437, "inner_validation_rmse": 0.36749338920035934, "learning_rate": 0.0002951444302461922, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 2, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3521013502409691, "fit_seconds": 2030.155105318001, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.58497067500139, "inner_validation_mae": 0.23384933852776885, "inner_validation_rmse": 0.361742992787689, "learning_rate": 0.0002809086200392825, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 3, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.34141303742492995, "fit_seconds": 1986.695635454962, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.980467153014615, "inner_validation_mae": 0.22482555044442415, "inner_validation_rmse": 0.34681167203216906, "learning_rate": 0.000258262716319083, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 4, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.33391291771515436, "fit_seconds": 1940.0182267469936, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.16284603095846, "inner_validation_mae": 0.21823913822695612, "inner_validation_rmse": 0.34150910726154, "learning_rate": 0.00022875, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 5, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3259226433894777, "fit_seconds": 1916.2042815189925, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.927870412997436, "inner_validation_mae": 0.21493795347958802, "inner_validation_rmse": 0.33303814109650054, "learning_rate": 0.00019438171392710925, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 6, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3203502120929844, "fit_seconds": 1901.2896965210093, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.509612301015295, "inner_validation_mae": 0.22188886200636626, "inner_validation_rmse": 0.3363407414213797, "learning_rate": 0.00015750000000000003, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 7, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31557447811582906, "fit_seconds": 1899.350969287043, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.88181267399341, "inner_validation_mae": 0.21017889048904181, "inner_validation_rmse": 0.3286568574059646, "learning_rate": 0.00012061828607289084, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 8, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31071123407939055, "fit_seconds": 1901.4595357929938, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.862473885994405, "inner_validation_mae": 0.21104949691519143, "inner_validation_rmse": 0.3292946363575192, "learning_rate": 8.625000000000005e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 9, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3066627180539208, "fit_seconds": 1898.602330002992, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.6989472709829, "inner_validation_mae": 0.20480307349190116, "inner_validation_rmse": 0.3253764053289252, "learning_rate": 5.6737283680917e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 10, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.30293540239452327, "fit_seconds": 1883.31971246202, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.698872285021935, "inner_validation_mae": 0.20199428956583143, "inner_validation_rmse": 0.3230924535828584, "learning_rate": 3.4091379960717494e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 11, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.29976934745606265, "fit_seconds": 1890.5131154309493, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.774635092995595, "inner_validation_mae": 0.2002225075364113, "inner_validation_rmse": 0.320565721433626, "learning_rate": 1.9855569753807785e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 12, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.2978171607796869, "fit_seconds": 1865.1841960400343, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.16861139499815, "inner_validation_mae": 0.19898510994389654, "inner_validation_rmse": 0.3195314575125945, "learning_rate": 1.4999999999999999e-05, "seed": 20260728, "world_size": 4}

ablations/full_12500/logs/command.txt

@@ -0,0 +1,3 @@
+cwd /cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32
+env DLCOAL_TRAIN_DEPLOY=/cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32 DLCOAL_COHORT_SPLIT=99996,5000,4248 DLCOAL_FORMAL_EPOCHS=12 OMP_NUM_THREADS=8 PYTHONDONTWRITEBYTECODE=1 CUBLAS_WORKSPACE_CONFIG=:4096:8 MASTER_ADDR=127.0.0.1 MASTER_PORT=29711
+/cloud/cloud-ssd1/dlcoal/venv/bin/torchrun --rdzv-backend=c10d --rdzv-endpoint=127.0.0.1:29711 --nnodes=1 --nproc_per_node=4 -m training.carrier_raw_summary_fusion_runner --registry-dir /cloud/cloud-ssd1/dlcoal/registry109k --unphased-cache /cloud/cloud-ssd1/dlcoal/level_caches109k/chain0_level32/unphased_evidence_cache.npz --sidecar-npz /cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz --expected-selection-rows 4248 --out-dir /cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/ctrl_fit12500_e12 --stage pilot --development-single-arm raw_plus_all_sidecar --seed 20260728 --epochs 12 --formal-epochs 12 --pilot-rows 13500 --pilot-fit-rows 12500 --pilot-inner-rows 500 --pilot-evaluation-rows 500 --development-gpu-class 4090 --packed-hap-bytes 16 --d-model 128 --host-cache-gib 48 --host-budget-fraction transient=0.27 --host-budget-fraction artifact=0.16 --host-budget-fraction segment_plan=0.08 --host-budget-fraction map_exposure=0.015 --host-budget-fraction map=0.02 --host-budget-fraction shift=0.005 --host-budget-fraction prefetch=0.35 --prefetch-rows 10 --stage-workers 4 --host-cache-sites 20000000000 --accept-cache-refill --resume 

ablations/full_12500/logs/ctrl_fit12500_e12.out

@@ -0,0 +1,43 @@
+=== data disk free 157 GB (floor 30 GB)
+=== deploy-hash guard: the three runner files match /cloud/cloud-ssd1/dlcoal/prep_fixed32/EXPECTED_SHA256.txt
+=== deploy verified: manifest 659f0427dc079a0bdb2ad3ae8ff8a328e98df6eeeb623532c34cb9a2bcd0b502 (34 artifacts) at /cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32
+=== row-count/rank deadlock pre-check (the deploy's rank guard)
+=== FIT=12500 at 4 ranks: 391 optimizer steps per epoch on every rank, world 1's
+=== memwatch every 60s -> /cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/ctrl_fit12500_e12/memwatch.log (cgroup memory.max 398358216704)
+=== RUN=ctrl_fit12500_e12 ARM=raw_plus_all_sidecar FIT=12500 INNER=500 EVAL=500 SEED=20260728 EPOCHS=12 D_MODEL=128 SMOKE=0
+=== 4 x 4090  host cache 48 GiB  OMP=8  stage workers 4  prefetch 10
+=== cache=/cloud/cloud-ssd1/dlcoal/level_caches109k/chain0_level32/unphased_evidence_cache.npz
+=== sidecar=/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz
+=== OUT=/cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/ctrl_fit12500_e12
+=== logs=/cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/ctrl_fit12500_e12  retries=2  MASTER_PORT=29711
+=== 09-16 07:35:15 attempt 1/3, log /cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/ctrl_fit12500_e12/attempt1_20260916_073515.log
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+/cloud/cloud-ssd1/dlcoal/venv/lib/python3.10/site-packages/torch/distributed/c10d_logger.py:83: UserWarning: barrier(): using the device under current context. You can specify `device_id` in `init_process_group` to mute this warning.
+  return func(*args, **kwargs)
+[rank0]:[W916 07:48:12.382149932 ProcessGroupNCCL.cpp:5310] Guessing device ID based on global rank. This can cause a hang if rank to GPU mapping is heterogeneous. You can specify device_id in init_process_group()
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"arm": "raw_plus_all_sidecar", "epoch": 1, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3877548377247502, "fit_seconds": 2098.6562211629935, "inner_raw_sites_consumed": 206175097, "inner_seconds": 43.76968279702123, "inner_validation_mae": 0.24319069107994437, "inner_validation_rmse": 0.36749338920035934, "learning_rate": 0.0002951444302461922, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 2, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3521013502409691, "fit_seconds": 2030.155105318001, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.58497067500139, "inner_validation_mae": 0.23384933852776885, "inner_validation_rmse": 0.361742992787689, "learning_rate": 0.0002809086200392825, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 3, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.34141303742492995, "fit_seconds": 1986.695635454962, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.980467153014615, "inner_validation_mae": 0.22482555044442415, "inner_validation_rmse": 0.34681167203216906, "learning_rate": 0.000258262716319083, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 4, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.33391291771515436, "fit_seconds": 1940.0182267469936, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.16284603095846, "inner_validation_mae": 0.21823913822695612, "inner_validation_rmse": 0.34150910726154, "learning_rate": 0.00022875, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 5, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3259226433894777, "fit_seconds": 1916.2042815189925, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.927870412997436, "inner_validation_mae": 0.21493795347958802, "inner_validation_rmse": 0.33303814109650054, "learning_rate": 0.00019438171392710925, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 6, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3203502120929844, "fit_seconds": 1901.2896965210093, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.509612301015295, "inner_validation_mae": 0.22188886200636626, "inner_validation_rmse": 0.3363407414213797, "learning_rate": 0.00015750000000000003, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 7, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31557447811582906, "fit_seconds": 1899.350969287043, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.88181267399341, "inner_validation_mae": 0.21017889048904181, "inner_validation_rmse": 0.3286568574059646, "learning_rate": 0.00012061828607289084, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 8, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31071123407939055, "fit_seconds": 1901.4595357929938, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.862473885994405, "inner_validation_mae": 0.21104949691519143, "inner_validation_rmse": 0.3292946363575192, "learning_rate": 8.625000000000005e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 9, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3066627180539208, "fit_seconds": 1898.602330002992, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.6989472709829, "inner_validation_mae": 0.20480307349190116, "inner_validation_rmse": 0.3253764053289252, "learning_rate": 5.6737283680917e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 10, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.30293540239452327, "fit_seconds": 1883.31971246202, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.698872285021935, "inner_validation_mae": 0.20199428956583143, "inner_validation_rmse": 0.3230924535828584, "learning_rate": 3.4091379960717494e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 11, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.29976934745606265, "fit_seconds": 1890.5131154309493, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.774635092995595, "inner_validation_mae": 0.2002225075364113, "inner_validation_rmse": 0.320565721433626, "learning_rate": 1.9855569753807785e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all_sidecar", "epoch": 12, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.2978171607796869, "fit_seconds": 1865.1841960400343, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.16861139499815, "inner_validation_mae": 0.19898510994389654, "inner_validation_rmse": 0.3195314575125945, "learning_rate": 1.4999999999999999e-05, "seed": 20260728, "world_size": 4}
+runner exit=0 (attempt 1)
+=== launcher wall time 24780 s
+launcher exit=0

ablations/full_12500/logs/ctrl_fit12500_e12.supervisor.log

@@ -0,0 +1,4 @@
+09-16 07:35:13 supervisor start pid 525532 for RUN=ctrl_fit12500_e12 (max relaunch 4, unknown 1, interval 120s)
+09-16 07:35:13 OUT=/cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/ctrl_fit12500_e12  FIT=12500 EPOCHS=12 SEED=20260728 ARM=raw_plus_all_sidecar SIDECAR=<default>
+09-16 07:35:13 first launch -> ctrl_fit12500_e12.out (RESUME_OK=0)
+09-16 14:29:30 DONE.json present; supervisor exits

ablations/full_12500/logs/harness_run.json

@@ -0,0 +1,29 @@
+{
+  "run": "ctrl_fit12500_e12",
+  "started_utc": "2026-09-16T07:35:15Z",
+  "launcher": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/bin/train_ablation.sh",
+  "launcher_sha256": "31689adea48e27b63cda2cad9c0b1ae5a34cb42f5d3285d32834ce578e5398e6",
+  "deploy": "/cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32",
+  "deploy_manifest_sha256": "659f0427dc079a0bdb2ad3ae8ff8a328e98df6eeeb623532c34cb9a2bcd0b502",
+  "arm": "raw_plus_all_sidecar",
+  "fit_rows": 12500,
+  "inner_rows": 500,
+  "evaluation_rows": 500,
+  "epochs": 12,
+  "seed": 20260728,
+  "d_model": 128,
+  "world_size": 4,
+  "gpu_class": "4090",
+  "registry": "/cloud/cloud-ssd1/dlcoal/registry109k",
+  "unphased_cache": "/cloud/cloud-ssd1/dlcoal/level_caches109k/chain0_level32/unphased_evidence_cache.npz",
+  "sidecar_npz": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz",
+  "sidecar_bytes": 107621430,
+  "host_cache_gib": 48,
+  "extra_args": "",
+  "strict_budgets": true,
+  "master_port": 29711,
+  "smoke": false,
+  "out_dir": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/ctrl_fit12500_e12",
+  "log_dir": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/ctrl_fit12500_e12",
+  "free_gb_at_start": 157
+}

ablations/full_12500/run/DEVELOPMENT.json

@@ -0,0 +1,580 @@
+{
+  "accumulate_rows": 32,
+  "arm_reports": [
+    {
+      "all_raw_sites_consumed_when_raw_enabled": true,
+      "arm": "raw_plus_all_sidecar",
+      "best_epoch": 12,
+      "best_inner_validation_rmse": 0.3195314575125945,
+      "checkpoint_artifact": {
+        "path": "checkpoint.pt",
+        "sha256": "925d3d3914ab3a140e84dd0931fc37a9de3e0bc829991ffb8728b9d51af0044c"
+      },
+      "curve": [
+        {
+          "epoch": 1,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3877548377247502,
+          "fit_seconds": 2098.6562211629935,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 43.76968279702123,
+          "inner_validation_mae": 0.24319069107994437,
+          "inner_validation_rmse": 0.36749338920035934,
+          "learning_rate": 0.0002951444302461922,
+          "world_size": 4
+        },
+        {
+          "epoch": 2,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3521013502409691,
+          "fit_seconds": 2030.155105318001,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.58497067500139,
+          "inner_validation_mae": 0.23384933852776885,
+          "inner_validation_rmse": 0.361742992787689,
+          "learning_rate": 0.0002809086200392825,
+          "world_size": 4
+        },
+        {
+          "epoch": 3,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.34141303742492995,
+          "fit_seconds": 1986.695635454962,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.980467153014615,
+          "inner_validation_mae": 0.22482555044442415,
+          "inner_validation_rmse": 0.34681167203216906,
+          "learning_rate": 0.000258262716319083,
+          "world_size": 4
+        },
+        {
+          "epoch": 4,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.33391291771515436,
+          "fit_seconds": 1940.0182267469936,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.16284603095846,
+          "inner_validation_mae": 0.21823913822695612,
+          "inner_validation_rmse": 0.34150910726154,
+          "learning_rate": 0.00022875,
+          "world_size": 4
+        },
+        {
+          "epoch": 5,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3259226433894777,
+          "fit_seconds": 1916.2042815189925,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.927870412997436,
+          "inner_validation_mae": 0.21493795347958802,
+          "inner_validation_rmse": 0.33303814109650054,
+          "learning_rate": 0.00019438171392710925,
+          "world_size": 4
+        },
+        {
+          "epoch": 6,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3203502120929844,
+          "fit_seconds": 1901.2896965210093,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.509612301015295,
+          "inner_validation_mae": 0.22188886200636626,
+          "inner_validation_rmse": 0.3363407414213797,
+          "learning_rate": 0.00015750000000000003,
+          "world_size": 4
+        },
+        {
+          "epoch": 7,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.31557447811582906,
+          "fit_seconds": 1899.350969287043,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.88181267399341,
+          "inner_validation_mae": 0.21017889048904181,
+          "inner_validation_rmse": 0.3286568574059646,
+          "learning_rate": 0.00012061828607289084,
+          "world_size": 4
+        },
+        {
+          "epoch": 8,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.31071123407939055,
+          "fit_seconds": 1901.4595357929938,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.862473885994405,
+          "inner_validation_mae": 0.21104949691519143,
+          "inner_validation_rmse": 0.3292946363575192,
+          "learning_rate": 8.625000000000005e-05,
+          "world_size": 4
+        },
+        {
+          "epoch": 9,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3066627180539208,
+          "fit_seconds": 1898.602330002992,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.6989472709829,
+          "inner_validation_mae": 0.20480307349190116,
+          "inner_validation_rmse": 0.3253764053289252,
+          "learning_rate": 5.6737283680917e-05,
+          "world_size": 4
+        },
+        {
+          "epoch": 10,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.30293540239452327,
+          "fit_seconds": 1883.31971246202,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.698872285021935,
+          "inner_validation_mae": 0.20199428956583143,
+          "inner_validation_rmse": 0.3230924535828584,
+          "learning_rate": 3.4091379960717494e-05,
+          "world_size": 4
+        },
+        {
+          "epoch": 11,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.29976934745606265,
+          "fit_seconds": 1890.5131154309493,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.774635092995595,
+          "inner_validation_mae": 0.2002225075364113,
+          "inner_validation_rmse": 0.320565721433626,
+          "learning_rate": 1.9855569753807785e-05,
+          "world_size": 4
+        },
+        {
+          "epoch": 12,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.2978171607796869,
+          "fit_seconds": 1865.1841960400343,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.16861139499815,
+          "inner_validation_mae": 0.19898510994389654,
+          "inner_validation_rmse": 0.3195314575125945,
+          "learning_rate": 1.4999999999999999e-05,
+          "world_size": 4
+        }
+      ],
+      "device": "cuda",
+      "direct_64_bin_output": true,
+      "epochs_ran": 12,
+      "evaluation_raw_sites_consumed": 216060598,
+      "evaluation_seconds": 198.2888983929879,
+      "experiment_signature": "3ca82bb9f687729cbfb65b05c31df0ec2bbc45a26a6ba23acd76df6d057797de",
+      "fit_rows": 12500,
+      "fit_seconds": 23211.449025738984,
+      "fixed_epoch_budget": 12,
+      "gpu_name": "NVIDIA GeForce RTX 4090",
+      "hazard_tail_survival_dual_or_class_head": false,
+      "initial_state_sha256": "7ecf840102b35595f25ab8b6f6d6fbe36ad9fed367b184e33444a2f76b85b100",
+      "inner_rows": 500,
+      "inner_validation_seconds": 454.0208019739948,
+      "mechanism_evaluation_artifact": {
+        "path": "mechanism_evaluation.npz",
+        "sha256": "3394fd9dc43d7d05c607d6000a11e129b6a39033910b2677d6fc182cfcc9174c"
+      },
+      "mechanism_evaluation_metrics": {
+        "constant_predictor_rmse": 0.402485317024815,
+        "label_spread_strata": {
+          "flat": {
+            "constant_predictor_rmse": 0.15479404880509584,
+            "label_spread_range": [
+              0.0,
+              0.24
+            ],
+            "mae": 0.11587709188461304,
+            "n_samples": 123,
+            "rmse": 0.16616753757457434
+          },
+          "moderate": {
+            "constant_predictor_rmse": 0.31611850295071287,
+            "label_spread_range": [
+              0.24,
+              0.4
+            ],
+            "mae": 0.18091316937862598,
+            "n_samples": 194,
+            "rmse": 0.2590015454063642
+          },
+          "structured": {
+            "constant_predictor_rmse": 0.5661845833240783,
+            "label_spread_range": [
+              0.4,
+              null
+            ],
+            "mae": 0.27768984614092795,
+            "n_samples": 183,
+            "rmse": 0.4299183111378961
+          }
+        },
+        "mae": 0.2003345580101013,
+        "n_samples": 500,
+        "per_bin_mae": [
+          0.24480665016174316,
+          0.24165412139892578,
+          0.23476056146621704,
+          0.22647115230560302,
+          0.2162070164680481,
+          0.20844890427589416,
+          0.20216979670524599,
+          0.19761557865142823,
+          0.19385516953468324,
+          0.19966330766677856,
+          0.1992708203792572,
+          0.19563431525230407,
+          0.19781084227561951,
+          0.19335844707489014,
+          0.19116229581832886,
+          0.1795099196434021,
+          0.1833814444541931,
+          0.17681527400016786,
+          0.17881734347343445,
+          0.17666004419326783,
+          0.17795073413848878,
+          0.17758055686950683,
+          0.174167320728302,
+          0.16787309741973877,
+          0.16458117294311522,
+          0.16801715898513794,
+          0.16387080144882202,
+          0.16689377737045288,
+          0.16336587238311767,
+          0.16203060293197633,
+          0.15944181394577026,
+          0.15760187864303588,
+          0.15335523319244385,
+          0.1582550959587097,
+          0.16130270528793336,
+          0.16775134658813476,
+          0.17353232860565185,
+          0.17454354524612425,
+          0.1942271409034729,
+          0.20237503051757813,
+          0.2077674984931946,
+          0.20579347467422485,
+          0.20395809030532838,
+          0.19830880308151244,
+          0.1910183424949646,
+          0.1980002450942993,
+          0.2001696081161499,
+          0.2018975033760071,
+          0.2053577308654785,
+          0.2151354398727417,
+          0.21482129192352295,
+          0.2201436929702759,
+          0.22268735933303832,
+          0.22830970001220702,
+          0.23170307207107543,
+          0.23562037754058837,
+          0.2336483426094055,
+          0.23330765342712403,
+          0.23685285806655884,
+          0.2465304455757141,
+          0.25702758407592774,
+          0.2567328181266785,
+          0.26485122680664064,
+          0.28497833442687986
+        ],
+        "per_bin_rmse": [
+          0.32520573910379724,
+          0.32250945521117336,
+          0.3157539240689479,
+          0.3086616767917923,
+          0.29581979803678177,
+          0.28581838488499034,
+          0.281007623247398,
+          0.27671546146585985,
+          0.2742135533041201,
+          0.28917281991443694,
+          0.29050829939667644,
+          0.2820547319716613,
+          0.29305857781478345,
+          0.28910227369965985,
+          0.28941279997165237,
+          0.26437551917046953,
+          0.2814190253970371,
+          0.27022075860915534,
+          0.27117900715540544,
+          0.26814644778064756,
+          0.2623744737391719,
+          0.26018416615981166,
+          0.2623753507030622,
+          0.2404540647754082,
+          0.24799507837432422,
+          0.26524799062706933,
+          0.25464040488766204,
+          0.24740409096948301,
+          0.24077528445474491,
+          0.23666208342936063,
+          0.23221066766750725,
+          0.2326567442503473,
+          0.22918141125285502,
+          0.24823864365799014,
+          0.25655549223087337,
+          0.2658092802727122,
+          0.276482678683334,
+          0.2799454941062838,
+          0.31631638518108723,
+          0.3419284640880848,
+          0.3526859256319074,
+          0.3491264512074741,
+          0.3421960271678152,
+          0.33145771633138477,
+          0.31791508072574703,
+          0.33996262418167994,
+          0.3455191321014126,
+          0.34461121869707434,
+          0.35227926491676986,
+          0.3723790513160041,
+          0.36986248770462593,
+          0.372922607156353,
+          0.37975476377684275,
+          0.3844307063053253,
+          0.3851219669576572,
+          0.39316866127441613,
+          0.39022942719943726,
+          0.388628466655207,
+          0.39235823935499164,
+          0.3970528423214134,
+          0.41441974438364415,
+          0.4119495835201414,
+          0.42024137464032907,
+          0.4436961585683516
+        ],
+        "rmse": 0.3169667091237063
+      },
+      "mechanism_evaluation_rows": 500,
+      "mechanism_evaluation_rows_requested": 500,
+      "model_name": "carrier_raw_summary_capacity_matched_direct64_nw_residual_v1",
+      "model_parameters": 1026409,
+      "optimizer_boundaries_sha256": "ce18181bfb956b04c297d2cb57aa005b40b1e38d281c91b7bb45b430faf64adf",
+      "peak_cuda_allocated_bytes": 12191421440,
+      "peak_cuda_reserved_bytes": 13323206656,
+      "publication_benchmark": false,
+      "row_order_sha256": "ff48ba7533087a7f438b18488c5d13f2260e3734220b11666507796956c2689f",
+      "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+      "seed": 20260728,
+      "segment_reduction_backend": "padded",
+      "selection_access_count": 0,
+      "selection_rows_read": 0,
+      "stage": "pilot",
+      "status": "complete",
+      "summary_only_skips_raw_io": false,
+      "trainable_parameters": 1026409,
+      "training_objective": {
+        "mode_weight_alpha": 0.0,
+        "principal_basis_rows": 0,
+        "spectral_decoupling": 0.0
+      },
+      "training_raw_sites_consumed": 18535440261,
+      "training_seconds": 23211.449025738984
+    }
+  ],
+  "development_single_arm": "raw_plus_all_sidecar",
+  "development_state_control": "carrier_correct",
+  "epochs": 12,
+  "evidence_description": {
+    "sidecar": {
+      "diploid_set_binding": "filename_only: the sidecar records no diploid identity; its chain-0 level-32 set is asserted by its basename, and checked only offline by verify_cohort_artifacts [6] against the PSMC+ store",
+      "members": [
+        "sidecar",
+        "sidecar_present"
+      ],
+      "present_flag": "declared",
+      "row_binding": "positional: sidecar row i describes cache row i"
+    },
+    "view": {
+      "column_order": "ladder_columns exactly as the level-cache builder stored them (permutation order); never sorted, never re-derived",
+      "columns_member": "ladder_columns",
+      "every_row_carries_every_stream": true,
+      "nw_log10_one_step_rows": 0,
+      "nw_log10_rule": "per cache row: nw_log10 == float32(log10(max(nw_proxy, tiny))) to within one float32 step",
+      "registry_context_log2_haplotypes": 7.0,
+      "site_count_oracle": "per row: |nw_proxy * 4 * a_64 * mutation_exposure - view sites| < 0.5",
+      "site_rule": "a site is staged iff 0 < derived < called over the view's 64 haplotypes; a half-called diploid contributes its one observed allele"
+    }
+  },
+  "evidence_identity": {
+    "sidecar": {
+      "array_sha256": "04e59b9ef33e9913573ab1e0ac412ab6ecc190c6299e42fda1919403cae6c21b",
+      "basename": "chain0_level32.npz",
+      "file_sha256": "a40187ce360e422d3b059b8aa58d8ecf0f3c1ca9f18f95eb69830e0f610f6dec",
+      "present_rows": 104996,
+      "present_sha256": "c4e31d9b42a839fef1f9f158fd70ef3079e33fcb6dfc5bb38c43c54ec97e14cf",
+      "rows": 104996,
+      "width": 256
+    },
+    "view": {
+      "chain": 0,
+      "columns_sha256": "17ea45c1d3bf97d51c78698020497c8aeb1460160e7f0a5325ec7041010b0791",
+      "columns_shape": [
+        104996,
+        32
+      ],
+      "context_log2_haplotypes": 6.0,
+      "diploids": 32,
+      "nw_proxy_sha256": "81a36cf53d0ca56d92ffa52285085c39a3206d044fa418a8a5f692b1514b24d3",
+      "row_diploids": 64,
+      "schema": "dlcoalsim-fixed-diploid-view-v1",
+      "stream_presence_sha256": "821039ed142a7cc57bfad84a169bc6f5619100d27504d6b948f9c12d74c2c608"
+    }
+  },
+  "experiment_signature": "3ca82bb9f687729cbfb65b05c31df0ec2bbc45a26a6ba23acd76df6d057797de",
+  "formal_authorization": false,
+  "gate_eligible": false,
+  "ibs_length_bins": [],
+  "loader_performance": {
+    "artifact_cache": {
+      "bytes": 1952639996,
+      "entries": 33,
+      "evictions": 26531,
+      "hits": 40,
+      "label": "packed_development_rows",
+      "max_bytes": 2061584302,
+      "max_sites": 5000000000,
+      "misses": 26564,
+      "rejected_oversize": 0,
+      "sites": 76208682
+    },
+    "artifact_opens": 26564,
+    "cached_row_keyspace": "registry development indices only",
+    "cpu_dense_state_bytes_avoided": 683211142272,
+    "cpu_dense_state_h2d_bytes": 0,
+    "cpu_dense_states_cached": 0,
+    "gpu_state_controls_staged": 0,
+    "map_cache": {
+      "bytes": 257330048,
+      "entries": 1126,
+      "evictions": 23818,
+      "hits": 26264,
+      "label": "observed_or_deranged_maps",
+      "max_bytes": 257698037,
+      "max_sites": 10000000000,
+      "misses": 24944,
+      "rejected_oversize": 0,
+      "sites": 16082002
+    },
+    "map_exposure_cache": {
+      "bytes": 193220608,
+      "entries": 2576,
+      "evictions": 17806,
+      "hits": 5222,
+      "label": "map_exposure",
+      "max_bytes": 193273528,
+      "max_sites": 10000000000,
+      "misses": 20382,
+      "rejected_oversize": 0,
+      "sites": 6038144
+    },
+    "map_exposure_requests": 25604,
+    "map_requests": 51208,
+    "packed_call_bytes_staged": 402354176828,
+    "pinned_bytes_staged": 481690117144,
+    "prefetch": {
+      "max_bytes": 4509715660,
+      "max_sites": 400000000,
+      "peak_bytes": 957711124,
+      "peak_sites": 38104715,
+      "queue_rows": 10,
+      "stage_workers": 4
+    },
+    "row_sites_staged": 21350348196,
+    "rows_loaded": 26604,
+    "rows_staged": 25604,
+    "segment_plan_cache": {
+      "bytes": 112849120,
+      "entries": 10838,
+      "evictions": 0,
+      "hits": 14766,
+      "label": "carrier_segment_plans",
+      "max_bytes": 1030792151,
+      "max_sites": 5000000000,
+      "misses": 10838,
+      "rejected_oversize": 0,
+      "sites": 4348553669
+    },
+    "selection_rows_cached": 0,
+    "selection_rows_read": 0,
+    "shift_cache": {
+      "bytes": 0,
+      "entries": 0,
+      "evictions": 0,
+      "hits": 0,
+      "label": "carrier_control_shifts",
+      "max_bytes": 64424509,
+      "max_sites": 5000000000,
+      "misses": 0,
+      "rejected_oversize": 0,
+      "sites": 0
+    },
+    "sites_staged": 19266247123,
+    "transient_artifact": {
+      "max_bytes": 3478923509,
+      "max_sites": 1050000000,
+      "peak_bytes": 886371872,
+      "peak_sites": 33217818
+    },
+    "view_site_count_checks": 25604
+  },
+  "model_config": {
+    "bin_attention_temperature": false,
+    "carrier_pool_seeds": 0,
+    "carrier_region_chunk": 512,
+    "carrier_scale_levels": [],
+    "carrier_tree_levels": [],
+    "carrier_width": 32,
+    "chunk_sites": 65536,
+    "coalescence_auxiliary": false,
+    "context_dim": 2,
+    "d_model": 128,
+    "dropout": 0.0,
+    "extreme_tap_levels": [],
+    "individual_width": 32,
+    "leaf_width_bp": 100,
+    "long_block_dilation_cap": 0,
+    "macro_levels": 7,
+    "n_heads": 4,
+    "n_long_blocks": 11,
+    "n_region_scales": 4,
+    "n_regional_blocks": 3,
+    "ordered_leaf_scan_width": 0,
+    "overlapping_regions": false,
+    "pair_concordance_first_order": false,
+    "pair_concordance_scales_bp": [],
+    "pair_concordance_width": 8,
+    "region_downsample_mode": "conv",
+    "region_sfs_bins": 0,
+    "scale_film": false,
+    "segment_reduction_backend": "padded",
+    "sidecar_dim": 256,
+    "site_genetic_span_features": false,
+    "site_lag_span_features": false,
+    "site_ld_lags": [],
+    "site_pair_identity_lags": [],
+    "site_state_width": 0,
+    "site_width": 8,
+    "smooth_time_query_modes": 0,
+    "tied_row_attention_heads": 0,
+    "tree_tap_levels": [],
+    "use_checkpoint": false,
+    "use_site_frequency": false
+  },
+  "objective": {
+    "mode_weight_alpha": 0.0,
+    "spectral_decoupling": 0.0
+  },
+  "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+  "score_resume_from": "",
+  "seed": 20260728,
+  "selection_access_count": 0,
+  "selection_rows_read": 0,
+  "stage": "development_single_arm",
+  "stage_rows": {
+    "census": false,
+    "evaluation": 500,
+    "fit": 12500,
+    "inner": 500
+  },
+  "status": "complete"
+}

ablations/full_12500/run/DONE.json

@@ -0,0 +1,11 @@
+{
+  "experiment_signature": "3ca82bb9f687729cbfb65b05c31df0ec2bbc45a26a6ba23acd76df6d057797de",
+  "gate_eligible": false,
+  "report_sha256": "ed1bcd8d0e91c1ccd2ee968b9b55a832cfe6719118b936b232a336117e95b336",
+  "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+  "seed": 20260728,
+  "selection_access_count": 0,
+  "selection_rows_read": 0,
+  "stage": "development_single_arm",
+  "status": "complete"
+}

ablations/full_12500/run/PLAN.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ccffa7a9e978c480998c8f0a73484c7e59fc891b9ce775b32208f8a6bec8a122
+size 29598005

ablations/full_12500/run/all_inputs.final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b76056a43d36029594d41e8a594801c06035b34b047d7d417c54df1c4e1126e
+size 16735186

ablations/full_12500/run/all_inputs/DONE.json

@@ -0,0 +1,19 @@
+{
+  "arm": "raw_plus_all_sidecar",
+  "checkpoint_artifact": {
+    "path": "checkpoint.pt",
+    "sha256": "925d3d3914ab3a140e84dd0931fc37a9de3e0bc829991ffb8728b9d51af0044c"
+  },
+  "experiment_signature": "3ca82bb9f687729cbfb65b05c31df0ec2bbc45a26a6ba23acd76df6d057797de",
+  "mechanism_evaluation_artifact": {
+    "path": "mechanism_evaluation.npz",
+    "sha256": "3394fd9dc43d7d05c607d6000a11e129b6a39033910b2677d6fc182cfcc9174c"
+  },
+  "report_sha256": "596c5ab57deedc774ade13a47b59d21cf664516fbbc7f09e4b02210fc1b88ffa",
+  "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+  "seed": 20260728,
+  "selection_access_count": 0,
+  "selection_rows_read": 0,
+  "stage": "pilot",
+  "status": "complete"
+}

ablations/full_12500/run/all_inputs/checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:925d3d3914ab3a140e84dd0931fc37a9de3e0bc829991ffb8728b9d51af0044c
+size 4169376

ablations/full_12500/run/all_inputs/mechanism_evaluation.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3394fd9dc43d7d05c607d6000a11e129b6a39033910b2677d6fc182cfcc9174c
+size 581024

ablations/full_12500/run/all_inputs/report.json

@@ -0,0 +1,369 @@
+{
+  "all_raw_sites_consumed_when_raw_enabled": true,
+  "arm": "raw_plus_all_sidecar",
+  "best_epoch": 12,
+  "best_inner_validation_rmse": 0.3195314575125945,
+  "checkpoint_artifact": {
+    "path": "checkpoint.pt",
+    "sha256": "925d3d3914ab3a140e84dd0931fc37a9de3e0bc829991ffb8728b9d51af0044c"
+  },
+  "curve": [
+    {
+      "epoch": 1,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3877548377247502,
+      "fit_seconds": 2098.6562211629935,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 43.76968279702123,
+      "inner_validation_mae": 0.24319069107994437,
+      "inner_validation_rmse": 0.36749338920035934,
+      "learning_rate": 0.0002951444302461922,
+      "world_size": 4
+    },
+    {
+      "epoch": 2,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3521013502409691,
+      "fit_seconds": 2030.155105318001,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.58497067500139,
+      "inner_validation_mae": 0.23384933852776885,
+      "inner_validation_rmse": 0.361742992787689,
+      "learning_rate": 0.0002809086200392825,
+      "world_size": 4
+    },
+    {
+      "epoch": 3,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.34141303742492995,
+      "fit_seconds": 1986.695635454962,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.980467153014615,
+      "inner_validation_mae": 0.22482555044442415,
+      "inner_validation_rmse": 0.34681167203216906,
+      "learning_rate": 0.000258262716319083,
+      "world_size": 4
+    },
+    {
+      "epoch": 4,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.33391291771515436,
+      "fit_seconds": 1940.0182267469936,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.16284603095846,
+      "inner_validation_mae": 0.21823913822695612,
+      "inner_validation_rmse": 0.34150910726154,
+      "learning_rate": 0.00022875,
+      "world_size": 4
+    },
+    {
+      "epoch": 5,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3259226433894777,
+      "fit_seconds": 1916.2042815189925,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.927870412997436,
+      "inner_validation_mae": 0.21493795347958802,
+      "inner_validation_rmse": 0.33303814109650054,
+      "learning_rate": 0.00019438171392710925,
+      "world_size": 4
+    },
+    {
+      "epoch": 6,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3203502120929844,
+      "fit_seconds": 1901.2896965210093,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.509612301015295,
+      "inner_validation_mae": 0.22188886200636626,
+      "inner_validation_rmse": 0.3363407414213797,
+      "learning_rate": 0.00015750000000000003,
+      "world_size": 4
+    },
+    {
+      "epoch": 7,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.31557447811582906,
+      "fit_seconds": 1899.350969287043,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.88181267399341,
+      "inner_validation_mae": 0.21017889048904181,
+      "inner_validation_rmse": 0.3286568574059646,
+      "learning_rate": 0.00012061828607289084,
+      "world_size": 4
+    },
+    {
+      "epoch": 8,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.31071123407939055,
+      "fit_seconds": 1901.4595357929938,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.862473885994405,
+      "inner_validation_mae": 0.21104949691519143,
+      "inner_validation_rmse": 0.3292946363575192,
+      "learning_rate": 8.625000000000005e-05,
+      "world_size": 4
+    },
+    {
+      "epoch": 9,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3066627180539208,
+      "fit_seconds": 1898.602330002992,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.6989472709829,
+      "inner_validation_mae": 0.20480307349190116,
+      "inner_validation_rmse": 0.3253764053289252,
+      "learning_rate": 5.6737283680917e-05,
+      "world_size": 4
+    },
+    {
+      "epoch": 10,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.30293540239452327,
+      "fit_seconds": 1883.31971246202,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.698872285021935,
+      "inner_validation_mae": 0.20199428956583143,
+      "inner_validation_rmse": 0.3230924535828584,
+      "learning_rate": 3.4091379960717494e-05,
+      "world_size": 4
+    },
+    {
+      "epoch": 11,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.29976934745606265,
+      "fit_seconds": 1890.5131154309493,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.774635092995595,
+      "inner_validation_mae": 0.2002225075364113,
+      "inner_validation_rmse": 0.320565721433626,
+      "learning_rate": 1.9855569753807785e-05,
+      "world_size": 4
+    },
+    {
+      "epoch": 12,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.2978171607796869,
+      "fit_seconds": 1865.1841960400343,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.16861139499815,
+      "inner_validation_mae": 0.19898510994389654,
+      "inner_validation_rmse": 0.3195314575125945,
+      "learning_rate": 1.4999999999999999e-05,
+      "world_size": 4
+    }
+  ],
+  "device": "cuda",
+  "direct_64_bin_output": true,
+  "epochs_ran": 12,
+  "evaluation_raw_sites_consumed": 216060598,
+  "evaluation_seconds": 198.2888983929879,
+  "experiment_signature": "3ca82bb9f687729cbfb65b05c31df0ec2bbc45a26a6ba23acd76df6d057797de",
+  "fit_rows": 12500,
+  "fit_seconds": 23211.449025738984,
+  "fixed_epoch_budget": 12,
+  "gpu_name": "NVIDIA GeForce RTX 4090",
+  "hazard_tail_survival_dual_or_class_head": false,
+  "initial_state_sha256": "7ecf840102b35595f25ab8b6f6d6fbe36ad9fed367b184e33444a2f76b85b100",
+  "inner_rows": 500,
+  "inner_validation_seconds": 454.0208019739948,
+  "mechanism_evaluation_artifact": {
+    "path": "mechanism_evaluation.npz",
+    "sha256": "3394fd9dc43d7d05c607d6000a11e129b6a39033910b2677d6fc182cfcc9174c"
+  },
+  "mechanism_evaluation_metrics": {
+    "constant_predictor_rmse": 0.402485317024815,
+    "label_spread_strata": {
+      "flat": {
+        "constant_predictor_rmse": 0.15479404880509584,
+        "label_spread_range": [
+          0.0,
+          0.24
+        ],
+        "mae": 0.11587709188461304,
+        "n_samples": 123,
+        "rmse": 0.16616753757457434
+      },
+      "moderate": {
+        "constant_predictor_rmse": 0.31611850295071287,
+        "label_spread_range": [
+          0.24,
+          0.4
+        ],
+        "mae": 0.18091316937862598,
+        "n_samples": 194,
+        "rmse": 0.2590015454063642
+      },
+      "structured": {
+        "constant_predictor_rmse": 0.5661845833240783,
+        "label_spread_range": [
+          0.4,
+          null
+        ],
+        "mae": 0.27768984614092795,
+        "n_samples": 183,
+        "rmse": 0.4299183111378961
+      }
+    },
+    "mae": 0.2003345580101013,
+    "n_samples": 500,
+    "per_bin_mae": [
+      0.24480665016174316,
+      0.24165412139892578,
+      0.23476056146621704,
+      0.22647115230560302,
+      0.2162070164680481,
+      0.20844890427589416,
+      0.20216979670524599,
+      0.19761557865142823,
+      0.19385516953468324,
+      0.19966330766677856,
+      0.1992708203792572,
+      0.19563431525230407,
+      0.19781084227561951,
+      0.19335844707489014,
+      0.19116229581832886,
+      0.1795099196434021,
+      0.1833814444541931,
+      0.17681527400016786,
+      0.17881734347343445,
+      0.17666004419326783,
+      0.17795073413848878,
+      0.17758055686950683,
+      0.174167320728302,
+      0.16787309741973877,
+      0.16458117294311522,
+      0.16801715898513794,
+      0.16387080144882202,
+      0.16689377737045288,
+      0.16336587238311767,
+      0.16203060293197633,
+      0.15944181394577026,
+      0.15760187864303588,
+      0.15335523319244385,
+      0.1582550959587097,
+      0.16130270528793336,
+      0.16775134658813476,
+      0.17353232860565185,
+      0.17454354524612425,
+      0.1942271409034729,
+      0.20237503051757813,
+      0.2077674984931946,
+      0.20579347467422485,
+      0.20395809030532838,
+      0.19830880308151244,
+      0.1910183424949646,
+      0.1980002450942993,
+      0.2001696081161499,
+      0.2018975033760071,
+      0.2053577308654785,
+      0.2151354398727417,
+      0.21482129192352295,
+      0.2201436929702759,
+      0.22268735933303832,
+      0.22830970001220702,
+      0.23170307207107543,
+      0.23562037754058837,
+      0.2336483426094055,
+      0.23330765342712403,
+      0.23685285806655884,
+      0.2465304455757141,
+      0.25702758407592774,
+      0.2567328181266785,
+      0.26485122680664064,
+      0.28497833442687986
+    ],
+    "per_bin_rmse": [
+      0.32520573910379724,
+      0.32250945521117336,
+      0.3157539240689479,
+      0.3086616767917923,
+      0.29581979803678177,
+      0.28581838488499034,
+      0.281007623247398,
+      0.27671546146585985,
+      0.2742135533041201,
+      0.28917281991443694,
+      0.29050829939667644,
+      0.2820547319716613,
+      0.29305857781478345,
+      0.28910227369965985,
+      0.28941279997165237,
+      0.26437551917046953,
+      0.2814190253970371,
+      0.27022075860915534,
+      0.27117900715540544,
+      0.26814644778064756,
+      0.2623744737391719,
+      0.26018416615981166,
+      0.2623753507030622,
+      0.2404540647754082,
+      0.24799507837432422,
+      0.26524799062706933,
+      0.25464040488766204,
+      0.24740409096948301,
+      0.24077528445474491,
+      0.23666208342936063,
+      0.23221066766750725,
+      0.2326567442503473,
+      0.22918141125285502,
+      0.24823864365799014,
+      0.25655549223087337,
+      0.2658092802727122,
+      0.276482678683334,
+      0.2799454941062838,
+      0.31631638518108723,
+      0.3419284640880848,
+      0.3526859256319074,
+      0.3491264512074741,
+      0.3421960271678152,
+      0.33145771633138477,
+      0.31791508072574703,
+      0.33996262418167994,
+      0.3455191321014126,
+      0.34461121869707434,
+      0.35227926491676986,
+      0.3723790513160041,
+      0.36986248770462593,
+      0.372922607156353,
+      0.37975476377684275,
+      0.3844307063053253,
+      0.3851219669576572,
+      0.39316866127441613,
+      0.39022942719943726,
+      0.388628466655207,
+      0.39235823935499164,
+      0.3970528423214134,
+      0.41441974438364415,
+      0.4119495835201414,
+      0.42024137464032907,
+      0.4436961585683516
+    ],
+    "rmse": 0.3169667091237063
+  },
+  "mechanism_evaluation_rows": 500,
+  "mechanism_evaluation_rows_requested": 500,
+  "model_name": "carrier_raw_summary_capacity_matched_direct64_nw_residual_v1",
+  "model_parameters": 1026409,
+  "optimizer_boundaries_sha256": "ce18181bfb956b04c297d2cb57aa005b40b1e38d281c91b7bb45b430faf64adf",
+  "peak_cuda_allocated_bytes": 12191421440,
+  "peak_cuda_reserved_bytes": 13323206656,
+  "publication_benchmark": false,
+  "row_order_sha256": "ff48ba7533087a7f438b18488c5d13f2260e3734220b11666507796956c2689f",
+  "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+  "seed": 20260728,
+  "segment_reduction_backend": "padded",
+  "selection_access_count": 0,
+  "selection_rows_read": 0,
+  "stage": "pilot",
+  "status": "complete",
+  "summary_only_skips_raw_io": false,
+  "trainable_parameters": 1026409,
+  "training_objective": {
+    "mode_weight_alpha": 0.0,
+    "principal_basis_rows": 0,
+    "spectral_decoupling": 0.0
+  },
+  "training_raw_sites_consumed": 18535440261,
+  "training_seconds": 23211.449025738984
+}

ablations/no_posterior_12500/logs/attempt1_20260916_152815.log

@@ -0,0 +1,23 @@
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+/cloud/cloud-ssd1/dlcoal/venv/lib/python3.10/site-packages/torch/distributed/c10d_logger.py:83: UserWarning: barrier(): using the device under current context. You can specify `device_id` in `init_process_group` to mute this warning.
+  return func(*args, **kwargs)
+[rank0]:[W916 15:41:06.795029511 ProcessGroupNCCL.cpp:5310] Guessing device ID based on global rank. This can cause a hang if rank to GPU mapping is heterogeneous. You can specify device_id in init_process_group()
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"arm": "raw_plus_all", "epoch": 1, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.40015624844809733, "fit_seconds": 2093.6454600400175, "inner_raw_sites_consumed": 206175097, "inner_seconds": 43.5588939129957, "inner_validation_mae": 0.2536635223776102, "inner_validation_rmse": 0.3786614608951453, "learning_rate": 0.0002951444302461922, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 2, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3615719489050102, "fit_seconds": 2018.276178309985, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.653537246980704, "inner_validation_mae": 0.23118237311020493, "inner_validation_rmse": 0.35846581330878063, "learning_rate": 0.0002809086200392825, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 3, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.35103077339022887, "fit_seconds": 1974.612629832991, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.717835613992065, "inner_validation_mae": 0.22763696809858083, "inner_validation_rmse": 0.3529769172358241, "learning_rate": 0.000258262716319083, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 4, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3444591158997494, "fit_seconds": 1933.7861313030007, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.5098103939672, "inner_validation_mae": 0.2255077318586409, "inner_validation_rmse": 0.3526106006292206, "learning_rate": 0.00022875, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 5, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3391718357446799, "fit_seconds": 1917.0238381460076, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.68309396895347, "inner_validation_mae": 0.22725810876861213, "inner_validation_rmse": 0.3478976199499427, "learning_rate": 0.00019438171392710925, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 6, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3353327931839078, "fit_seconds": 1899.3311412120238, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.096282957994845, "inner_validation_mae": 0.23042901559919118, "inner_validation_rmse": 0.3471583560419813, "learning_rate": 0.00015750000000000003, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 7, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3305477654173744, "fit_seconds": 1900.9600651880028, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.619981688971166, "inner_validation_mae": 0.22294189005717635, "inner_validation_rmse": 0.3418899322674546, "learning_rate": 0.00012061828607289084, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 8, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.32624947796277753, "fit_seconds": 1897.2488543219515, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.9244443300413, "inner_validation_mae": 0.21564797833561897, "inner_validation_rmse": 0.3386951979830166, "learning_rate": 8.625000000000005e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 9, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3228154418396384, "fit_seconds": 1888.3135059690103, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.646107722015586, "inner_validation_mae": 0.21422379455715418, "inner_validation_rmse": 0.3375664071352261, "learning_rate": 5.6737283680917e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 10, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31890629316474045, "fit_seconds": 1871.0354464069824, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.9466422580299, "inner_validation_mae": 0.20946778214350342, "inner_validation_rmse": 0.3341248524935109, "learning_rate": 3.4091379960717494e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 11, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3162338020659697, "fit_seconds": 1885.552217441029, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.89084733999334, "inner_validation_mae": 0.2101674868762493, "inner_validation_rmse": 0.333845803512705, "learning_rate": 1.9855569753807785e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 12, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3140382644350096, "fit_seconds": 1858.5625097450102, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.27406584401615, "inner_validation_mae": 0.21033127358183265, "inner_validation_rmse": 0.3336641447535338, "learning_rate": 1.4999999999999999e-05, "seed": 20260728, "world_size": 4}

ablations/no_posterior_12500/logs/command.txt

@@ -0,0 +1,3 @@
+cwd /cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32
+env DLCOAL_TRAIN_DEPLOY=/cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32 DLCOAL_COHORT_SPLIT=99996,5000,4248 DLCOAL_FORMAL_EPOCHS=12 OMP_NUM_THREADS=8 PYTHONDONTWRITEBYTECODE=1 CUBLAS_WORKSPACE_CONFIG=:4096:8 MASTER_ADDR=127.0.0.1 MASTER_PORT=29711
+/cloud/cloud-ssd1/dlcoal/venv/bin/torchrun --rdzv-backend=c10d --rdzv-endpoint=127.0.0.1:29711 --nnodes=1 --nproc_per_node=4 -m training.carrier_raw_summary_fusion_runner --registry-dir /cloud/cloud-ssd1/dlcoal/registry109k --unphased-cache /cloud/cloud-ssd1/dlcoal/level_caches109k/chain0_level32/unphased_evidence_cache.npz --expected-selection-rows 4248 --out-dir /cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/drop_sidecar_fit12500_e12 --stage pilot --development-single-arm raw_plus_all --seed 20260728 --epochs 12 --formal-epochs 12 --pilot-rows 13500 --pilot-fit-rows 12500 --pilot-inner-rows 500 --pilot-evaluation-rows 500 --development-gpu-class 4090 --packed-hap-bytes 16 --d-model 128 --host-cache-gib 48 --host-budget-fraction transient=0.27 --host-budget-fraction artifact=0.16 --host-budget-fraction segment_plan=0.08 --host-budget-fraction map_exposure=0.015 --host-budget-fraction map=0.02 --host-budget-fraction shift=0.005 --host-budget-fraction prefetch=0.35 --prefetch-rows 10 --stage-workers 4 --host-cache-sites 20000000000 --accept-cache-refill --resume 

ablations/no_posterior_12500/logs/drop_sidecar_fit12500_e12.out

@@ -0,0 +1,40 @@
+=== data disk free 157 GB (floor 30 GB)
+=== deploy-hash guard: the three runner files match /cloud/cloud-ssd1/dlcoal/prep_fixed32/EXPECTED_SHA256.txt
+=== deploy verified: manifest 659f0427dc079a0bdb2ad3ae8ff8a328e98df6eeeb623532c34cb9a2bcd0b502 (34 artifacts) at /cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32
+=== SIDECAR=none: --sidecar-npz is not passed (ARM=raw_plus_all does not read the sidecar stream)
+=== row-count/rank deadlock pre-check (the deploy's rank guard)
+=== FIT=12500 at 4 ranks: 391 optimizer steps per epoch on every rank, world 1's
+=== memwatch every 60s -> /cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/drop_sidecar_fit12500_e12/memwatch.log (cgroup memory.max 398358216704)
+=== RUN=drop_sidecar_fit12500_e12 ARM=raw_plus_all FIT=12500 INNER=500 EVAL=500 SEED=20260728 EPOCHS=12 D_MODEL=128 SMOKE=0
+=== 4 x 4090  host cache 48 GiB  OMP=8  stage workers 4  prefetch 10
+=== cache=/cloud/cloud-ssd1/dlcoal/level_caches109k/chain0_level32/unphased_evidence_cache.npz
+=== sidecar=none
+=== OUT=/cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/drop_sidecar_fit12500_e12
+=== logs=/cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/drop_sidecar_fit12500_e12  retries=2  MASTER_PORT=29711
+=== 09-16 15:28:15 attempt 1/3, log /cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/drop_sidecar_fit12500_e12/attempt1_20260916_152815.log
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+/cloud/cloud-ssd1/dlcoal/venv/lib/python3.10/site-packages/torch/distributed/c10d_logger.py:83: UserWarning: barrier(): using the device under current context. You can specify `device_id` in `init_process_group` to mute this warning.
+  return func(*args, **kwargs)
+[rank0]:[W916 15:41:06.795029511 ProcessGroupNCCL.cpp:5310] Guessing device ID based on global rank. This can cause a hang if rank to GPU mapping is heterogeneous. You can specify device_id in init_process_group()
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"arm": "raw_plus_all", "epoch": 1, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.40015624844809733, "fit_seconds": 2093.6454600400175, "inner_raw_sites_consumed": 206175097, "inner_seconds": 43.5588939129957, "inner_validation_mae": 0.2536635223776102, "inner_validation_rmse": 0.3786614608951453, "learning_rate": 0.0002951444302461922, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 2, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3615719489050102, "fit_seconds": 2018.276178309985, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.653537246980704, "inner_validation_mae": 0.23118237311020493, "inner_validation_rmse": 0.35846581330878063, "learning_rate": 0.0002809086200392825, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 3, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.35103077339022887, "fit_seconds": 1974.612629832991, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.717835613992065, "inner_validation_mae": 0.22763696809858083, "inner_validation_rmse": 0.3529769172358241, "learning_rate": 0.000258262716319083, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 4, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3444591158997494, "fit_seconds": 1933.7861313030007, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.5098103939672, "inner_validation_mae": 0.2255077318586409, "inner_validation_rmse": 0.3526106006292206, "learning_rate": 0.00022875, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 5, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3391718357446799, "fit_seconds": 1917.0238381460076, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.68309396895347, "inner_validation_mae": 0.22725810876861213, "inner_validation_rmse": 0.3478976199499427, "learning_rate": 0.00019438171392710925, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 6, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3353327931839078, "fit_seconds": 1899.3311412120238, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.096282957994845, "inner_validation_mae": 0.23042901559919118, "inner_validation_rmse": 0.3471583560419813, "learning_rate": 0.00015750000000000003, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 7, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3305477654173744, "fit_seconds": 1900.9600651880028, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.619981688971166, "inner_validation_mae": 0.22294189005717635, "inner_validation_rmse": 0.3418899322674546, "learning_rate": 0.00012061828607289084, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 8, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.32624947796277753, "fit_seconds": 1897.2488543219515, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.9244443300413, "inner_validation_mae": 0.21564797833561897, "inner_validation_rmse": 0.3386951979830166, "learning_rate": 8.625000000000005e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 9, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3228154418396384, "fit_seconds": 1888.3135059690103, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.646107722015586, "inner_validation_mae": 0.21422379455715418, "inner_validation_rmse": 0.3375664071352261, "learning_rate": 5.6737283680917e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 10, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31890629316474045, "fit_seconds": 1871.0354464069824, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.9466422580299, "inner_validation_mae": 0.20946778214350342, "inner_validation_rmse": 0.3341248524935109, "learning_rate": 3.4091379960717494e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 11, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3162338020659697, "fit_seconds": 1885.552217441029, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.89084733999334, "inner_validation_mae": 0.2101674868762493, "inner_validation_rmse": 0.333845803512705, "learning_rate": 1.9855569753807785e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_all", "epoch": 12, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3140382644350096, "fit_seconds": 1858.5625097450102, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.27406584401615, "inner_validation_mae": 0.21033127358183265, "inner_validation_rmse": 0.3336641447535338, "learning_rate": 1.4999999999999999e-05, "seed": 20260728, "world_size": 4}
+runner exit=0 (attempt 1)
+=== launcher wall time 24702 s
+launcher exit=0

ablations/no_posterior_12500/logs/drop_sidecar_fit12500_e12.supervisor.log

@@ -0,0 +1,4 @@
+09-16 15:28:12 supervisor start pid 565590 for RUN=drop_sidecar_fit12500_e12 (max relaunch 4, unknown 1, interval 120s)
+09-16 15:28:12 OUT=/cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/drop_sidecar_fit12500_e12  FIT=12500 EPOCHS=12 SEED=20260728 ARM=raw_plus_all SIDECAR=none
+09-16 15:28:12 first launch -> drop_sidecar_fit12500_e12.out (RESUME_OK=0)
+09-16 22:20:30 DONE.json present; supervisor exits

ablations/no_posterior_12500/logs/harness_run.json

@@ -0,0 +1,29 @@
+{
+  "run": "drop_sidecar_fit12500_e12",
+  "started_utc": "2026-09-16T15:28:15Z",
+  "launcher": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/bin/train_ablation.sh",
+  "launcher_sha256": "31689adea48e27b63cda2cad9c0b1ae5a34cb42f5d3285d32834ce578e5398e6",
+  "deploy": "/cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32",
+  "deploy_manifest_sha256": "659f0427dc079a0bdb2ad3ae8ff8a328e98df6eeeb623532c34cb9a2bcd0b502",
+  "arm": "raw_plus_all",
+  "fit_rows": 12500,
+  "inner_rows": 500,
+  "evaluation_rows": 500,
+  "epochs": 12,
+  "seed": 20260728,
+  "d_model": 128,
+  "world_size": 4,
+  "gpu_class": "4090",
+  "registry": "/cloud/cloud-ssd1/dlcoal/registry109k",
+  "unphased_cache": "/cloud/cloud-ssd1/dlcoal/level_caches109k/chain0_level32/unphased_evidence_cache.npz",
+  "sidecar_npz": "none",
+  "sidecar_bytes": 0,
+  "host_cache_gib": 48,
+  "extra_args": "",
+  "strict_budgets": true,
+  "master_port": 29711,
+  "smoke": false,
+  "out_dir": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/drop_sidecar_fit12500_e12",
+  "log_dir": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/drop_sidecar_fit12500_e12",
+  "free_gb_at_start": 157
+}

ablations/no_posterior_12500/run/DEVELOPMENT.json

@@ -0,0 +1,563 @@
+{
+  "accumulate_rows": 32,
+  "arm_reports": [
+    {
+      "all_raw_sites_consumed_when_raw_enabled": true,
+      "arm": "raw_plus_all",
+      "best_epoch": 12,
+      "best_inner_validation_rmse": 0.3336641447535338,
+      "checkpoint_artifact": {
+        "path": "checkpoint.pt",
+        "sha256": "58e35ac89a095b376bc35ceeab7b6fb0d00c11745b530c76e706f9eb5f4553b0"
+      },
+      "curve": [
+        {
+          "epoch": 1,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.40015624844809733,
+          "fit_seconds": 2093.6454600400175,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 43.5588939129957,
+          "inner_validation_mae": 0.2536635223776102,
+          "inner_validation_rmse": 0.3786614608951453,
+          "learning_rate": 0.0002951444302461922,
+          "world_size": 4
+        },
+        {
+          "epoch": 2,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3615719489050102,
+          "fit_seconds": 2018.276178309985,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.653537246980704,
+          "inner_validation_mae": 0.23118237311020493,
+          "inner_validation_rmse": 0.35846581330878063,
+          "learning_rate": 0.0002809086200392825,
+          "world_size": 4
+        },
+        {
+          "epoch": 3,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.35103077339022887,
+          "fit_seconds": 1974.612629832991,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.717835613992065,
+          "inner_validation_mae": 0.22763696809858083,
+          "inner_validation_rmse": 0.3529769172358241,
+          "learning_rate": 0.000258262716319083,
+          "world_size": 4
+        },
+        {
+          "epoch": 4,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3444591158997494,
+          "fit_seconds": 1933.7861313030007,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.5098103939672,
+          "inner_validation_mae": 0.2255077318586409,
+          "inner_validation_rmse": 0.3526106006292206,
+          "learning_rate": 0.00022875,
+          "world_size": 4
+        },
+        {
+          "epoch": 5,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3391718357446799,
+          "fit_seconds": 1917.0238381460076,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.68309396895347,
+          "inner_validation_mae": 0.22725810876861213,
+          "inner_validation_rmse": 0.3478976199499427,
+          "learning_rate": 0.00019438171392710925,
+          "world_size": 4
+        },
+        {
+          "epoch": 6,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3353327931839078,
+          "fit_seconds": 1899.3311412120238,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.096282957994845,
+          "inner_validation_mae": 0.23042901559919118,
+          "inner_validation_rmse": 0.3471583560419813,
+          "learning_rate": 0.00015750000000000003,
+          "world_size": 4
+        },
+        {
+          "epoch": 7,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3305477654173744,
+          "fit_seconds": 1900.9600651880028,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.619981688971166,
+          "inner_validation_mae": 0.22294189005717635,
+          "inner_validation_rmse": 0.3418899322674546,
+          "learning_rate": 0.00012061828607289084,
+          "world_size": 4
+        },
+        {
+          "epoch": 8,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.32624947796277753,
+          "fit_seconds": 1897.2488543219515,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.9244443300413,
+          "inner_validation_mae": 0.21564797833561897,
+          "inner_validation_rmse": 0.3386951979830166,
+          "learning_rate": 8.625000000000005e-05,
+          "world_size": 4
+        },
+        {
+          "epoch": 9,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3228154418396384,
+          "fit_seconds": 1888.3135059690103,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.646107722015586,
+          "inner_validation_mae": 0.21422379455715418,
+          "inner_validation_rmse": 0.3375664071352261,
+          "learning_rate": 5.6737283680917e-05,
+          "world_size": 4
+        },
+        {
+          "epoch": 10,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.31890629316474045,
+          "fit_seconds": 1871.0354464069824,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 36.9466422580299,
+          "inner_validation_mae": 0.20946778214350342,
+          "inner_validation_rmse": 0.3341248524935109,
+          "learning_rate": 3.4091379960717494e-05,
+          "world_size": 4
+        },
+        {
+          "epoch": 11,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3162338020659697,
+          "fit_seconds": 1885.552217441029,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.89084733999334,
+          "inner_validation_mae": 0.2101674868762493,
+          "inner_validation_rmse": 0.333845803512705,
+          "learning_rate": 1.9855569753807785e-05,
+          "world_size": 4
+        },
+        {
+          "epoch": 12,
+          "fit_raw_sites_consumed": 5107190494,
+          "fit_rmse": 0.3140382644350096,
+          "fit_seconds": 1858.5625097450102,
+          "inner_raw_sites_consumed": 206175097,
+          "inner_seconds": 37.27406584401615,
+          "inner_validation_mae": 0.21033127358183265,
+          "inner_validation_rmse": 0.3336641447535338,
+          "learning_rate": 1.4999999999999999e-05,
+          "world_size": 4
+        }
+      ],
+      "device": "cuda",
+      "direct_64_bin_output": true,
+      "epochs_ran": 12,
+      "evaluation_raw_sites_consumed": 216060598,
+      "evaluation_seconds": 199.32269198598806,
+      "experiment_signature": "69ff52c573de6925a676bcb991d7119229d39e9b1c99689be5ac9131dfbe4976",
+      "fit_rows": 12500,
+      "fit_seconds": 23138.34797791601,
+      "fixed_epoch_budget": 12,
+      "gpu_name": "NVIDIA GeForce RTX 4090",
+      "hazard_tail_survival_dual_or_class_head": false,
+      "initial_state_sha256": "ebedc02cc9fc0c67ed13724e34af3ef174cb81af45ed25de4c6e8e8678e15e0f",
+      "inner_rows": 500,
+      "inner_validation_seconds": 450.5215432779514,
+      "mechanism_evaluation_artifact": {
+        "path": "mechanism_evaluation.npz",
+        "sha256": "545a7456adecf48557c274a4c0bfc984377617d577a593cd7c0f43b22f8c3268"
+      },
+      "mechanism_evaluation_metrics": {
+        "constant_predictor_rmse": 0.402485317024815,
+        "label_spread_strata": {
+          "flat": {
+            "constant_predictor_rmse": 0.15479404880509584,
+            "label_spread_range": [
+              0.0,
+              0.24
+            ],
+            "mae": 0.11951362635425436,
+            "n_samples": 123,
+            "rmse": 0.17066689261786666
+          },
+          "moderate": {
+            "constant_predictor_rmse": 0.31611850295071287,
+            "label_spread_range": [
+              0.24,
+              0.4
+            ],
+            "mae": 0.1952067300408464,
+            "n_samples": 194,
+            "rmse": 0.2769268088176663
+          },
+          "structured": {
+            "constant_predictor_rmse": 0.5661845833240783,
+            "label_spread_range": [
+              0.4,
+              null
+            ],
+            "mae": 0.2893600353431669,
+            "n_samples": 183,
+            "rmse": 0.43653716377354895
+          }
+        },
+        "mae": 0.21104633627459407,
+        "n_samples": 500,
+        "per_bin_mae": [
+          0.25872102117538454,
+          0.2574238557815552,
+          0.2544191207885742,
+          0.24844665241241454,
+          0.23952793931961058,
+          0.23101413989067077,
+          0.2247856810092926,
+          0.21954951190948485,
+          0.2137761218547821,
+          0.21821088123321533,
+          0.21430532002449035,
+          0.20765164136886596,
+          0.20636261677742004,
+          0.2009211320877075,
+          0.20189660716056823,
+          0.1875525505542755,
+          0.18837180042266846,
+          0.1838572347164154,
+          0.18573030161857604,
+          0.17980736780166626,
+          0.17956282472610474,
+          0.17695083236694337,
+          0.16958984184265136,
+          0.1661336784362793,
+          0.16632140636444093,
+          0.1746431484222412,
+          0.17343592500686644,
+          0.1762556366920471,
+          0.17273916673660278,
+          0.16931749629974366,
+          0.1715409984588623,
+          0.1757011661529541,
+          0.17381114053726196,
+          0.1748917989730835,
+          0.17907089328765868,
+          0.18520425844192506,
+          0.19175126791000366,
+          0.18996810913085938,
+          0.20550774765014648,
+          0.21040356731414794,
+          0.2123220829963684,
+          0.20762387323379516,
+          0.2060391812324524,
+          0.20466476488113403,
+          0.2028860273361206,
+          0.20784376621246337,
+          0.2120401487350464,
+          0.21495712804794312,
+          0.21747990798950195,
+          0.22602142095565797,
+          0.22599719142913818,
+          0.22860319328308104,
+          0.23108028650283813,
+          0.23668288993835448,
+          0.2414417772293091,
+          0.2444851508140564,
+          0.2414763617515564,
+          0.24355956125259398,
+          0.24764243984222412,
+          0.25615241622924806,
+          0.26626666355133055,
+          0.2660611934661865,
+          0.2707990622520447,
+          0.2897066297531128
+        ],
+        "per_bin_rmse": [
+          0.34896980726431753,
+          0.34709251055552404,
+          0.3421776787523031,
+          0.3357477348165305,
+          0.3236791126624797,
+          0.31303682507500463,
+          0.3082640312649185,
+          0.3027514693268415,
+          0.2988764919988645,
+          0.30935150618101337,
+          0.30319353732431675,
+          0.2909803704134548,
+          0.29893455193244517,
+          0.29629442222182034,
+          0.29874164242952256,
+          0.26751661301301977,
+          0.28004897708551246,
+          0.273244006549535,
+          0.2756225409016931,
+          0.269652158191675,
+          0.27050748435753463,
+          0.25807585002971895,
+          0.2550076447945779,
+          0.24136317852276862,
+          0.2561427929475607,
+          0.2788487170454883,
+          0.2702609319612673,
+          0.2604130407260646,
+          0.2530495535403907,
+          0.24866028651762162,
+          0.25054212809775017,
+          0.2652816944878899,
+          0.2650450787098498,
+          0.2609937648429789,
+          0.26899104256336775,
+          0.2788974421869027,
+          0.2923080090466803,
+          0.29644489907235044,
+          0.3269216591311331,
+          0.3474323063187913,
+          0.3577926430077441,
+          0.3517563847712244,
+          0.3432117212312172,
+          0.3358277829235053,
+          0.32555655098598446,
+          0.3402813929580356,
+          0.34608173996197855,
+          0.3456521872670643,
+          0.3546878240549255,
+          0.3738941922531639,
+          0.3743833889814278,
+          0.37811233084276824,
+          0.3865729221224253,
+          0.39172335883565496,
+          0.3924530748232029,
+          0.4002031369657275,
+          0.39565291120606483,
+          0.39437730362043927,
+          0.39916586972934864,
+          0.4030455671467172,
+          0.41991077243034874,
+          0.41648041018961873,
+          0.42156874034021985,
+          0.4445341710296161
+        ],
+        "rmse": 0.3265992778522081
+      },
+      "mechanism_evaluation_rows": 500,
+      "mechanism_evaluation_rows_requested": 500,
+      "model_name": "carrier_raw_summary_capacity_matched_direct64_nw_residual_v1",
+      "model_parameters": 935529,
+      "optimizer_boundaries_sha256": "ce18181bfb956b04c297d2cb57aa005b40b1e38d281c91b7bb45b430faf64adf",
+      "peak_cuda_allocated_bytes": 12190329856,
+      "peak_cuda_reserved_bytes": 13321109504,
+      "publication_benchmark": false,
+      "row_order_sha256": "ff48ba7533087a7f438b18488c5d13f2260e3734220b11666507796956c2689f",
+      "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+      "seed": 20260728,
+      "segment_reduction_backend": "padded",
+      "selection_access_count": 0,
+      "selection_rows_read": 0,
+      "stage": "pilot",
+      "status": "complete",
+      "summary_only_skips_raw_io": false,
+      "trainable_parameters": 935529,
+      "training_objective": {
+        "mode_weight_alpha": 0.0,
+        "principal_basis_rows": 0,
+        "spectral_decoupling": 0.0
+      },
+      "training_raw_sites_consumed": 18535440261,
+      "training_seconds": 23138.34797791601
+    }
+  ],
+  "development_single_arm": "raw_plus_all",
+  "development_state_control": "carrier_correct",
+  "epochs": 12,
+  "evidence_description": {
+    "sidecar": null,
+    "view": {
+      "column_order": "ladder_columns exactly as the level-cache builder stored them (permutation order); never sorted, never re-derived",
+      "columns_member": "ladder_columns",
+      "every_row_carries_every_stream": true,
+      "nw_log10_one_step_rows": 0,
+      "nw_log10_rule": "per cache row: nw_log10 == float32(log10(max(nw_proxy, tiny))) to within one float32 step",
+      "registry_context_log2_haplotypes": 7.0,
+      "site_count_oracle": "per row: |nw_proxy * 4 * a_64 * mutation_exposure - view sites| < 0.5",
+      "site_rule": "a site is staged iff 0 < derived < called over the view's 64 haplotypes; a half-called diploid contributes its one observed allele"
+    }
+  },
+  "evidence_identity": {
+    "sidecar": null,
+    "view": {
+      "chain": 0,
+      "columns_sha256": "17ea45c1d3bf97d51c78698020497c8aeb1460160e7f0a5325ec7041010b0791",
+      "columns_shape": [
+        104996,
+        32
+      ],
+      "context_log2_haplotypes": 6.0,
+      "diploids": 32,
+      "nw_proxy_sha256": "81a36cf53d0ca56d92ffa52285085c39a3206d044fa418a8a5f692b1514b24d3",
+      "row_diploids": 64,
+      "schema": "dlcoalsim-fixed-diploid-view-v1",
+      "stream_presence_sha256": "821039ed142a7cc57bfad84a169bc6f5619100d27504d6b948f9c12d74c2c608"
+    }
+  },
+  "experiment_signature": "69ff52c573de6925a676bcb991d7119229d39e9b1c99689be5ac9131dfbe4976",
+  "formal_authorization": false,
+  "gate_eligible": false,
+  "ibs_length_bins": [],
+  "loader_performance": {
+    "artifact_cache": {
+      "bytes": 1952639996,
+      "entries": 33,
+      "evictions": 26531,
+      "hits": 40,
+      "label": "packed_development_rows",
+      "max_bytes": 2061584302,
+      "max_sites": 5000000000,
+      "misses": 26564,
+      "rejected_oversize": 0,
+      "sites": 76208682
+    },
+    "artifact_opens": 26564,
+    "cached_row_keyspace": "registry development indices only",
+    "cpu_dense_state_bytes_avoided": 683211142272,
+    "cpu_dense_state_h2d_bytes": 0,
+    "cpu_dense_states_cached": 0,
+    "gpu_state_controls_staged": 0,
+    "map_cache": {
+      "bytes": 257330048,
+      "entries": 1126,
+      "evictions": 23818,
+      "hits": 26264,
+      "label": "observed_or_deranged_maps",
+      "max_bytes": 257698037,
+      "max_sites": 10000000000,
+      "misses": 24944,
+      "rejected_oversize": 0,
+      "sites": 16082002
+    },
+    "map_exposure_cache": {
+      "bytes": 193220608,
+      "entries": 2576,
+      "evictions": 17805,
+      "hits": 5223,
+      "label": "map_exposure",
+      "max_bytes": 193273528,
+      "max_sites": 10000000000,
+      "misses": 20381,
+      "rejected_oversize": 0,
+      "sites": 6038144
+    },
+    "map_exposure_requests": 25604,
+    "map_requests": 51208,
+    "packed_call_bytes_staged": 402354176828,
+    "pinned_bytes_staged": 481690117144,
+    "prefetch": {
+      "max_bytes": 4509715660,
+      "max_sites": 400000000,
+      "peak_bytes": 957711124,
+      "peak_sites": 38104715,
+      "queue_rows": 10,
+      "stage_workers": 4
+    },
+    "row_sites_staged": 21350348196,
+    "rows_loaded": 26604,
+    "rows_staged": 25604,
+    "segment_plan_cache": {
+      "bytes": 112849120,
+      "entries": 10838,
+      "evictions": 0,
+      "hits": 14766,
+      "label": "carrier_segment_plans",
+      "max_bytes": 1030792151,
+      "max_sites": 5000000000,
+      "misses": 10838,
+      "rejected_oversize": 0,
+      "sites": 4348553669
+    },
+    "selection_rows_cached": 0,
+    "selection_rows_read": 0,
+    "shift_cache": {
+      "bytes": 0,
+      "entries": 0,
+      "evictions": 0,
+      "hits": 0,
+      "label": "carrier_control_shifts",
+      "max_bytes": 64424509,
+      "max_sites": 5000000000,
+      "misses": 0,
+      "rejected_oversize": 0,
+      "sites": 0
+    },
+    "sites_staged": 19266247123,
+    "transient_artifact": {
+      "max_bytes": 3478923509,
+      "max_sites": 1050000000,
+      "peak_bytes": 886371872,
+      "peak_sites": 33217818
+    },
+    "view_site_count_checks": 25604
+  },
+  "model_config": {
+    "bin_attention_temperature": false,
+    "carrier_pool_seeds": 0,
+    "carrier_region_chunk": 512,
+    "carrier_scale_levels": [],
+    "carrier_tree_levels": [],
+    "carrier_width": 32,
+    "chunk_sites": 65536,
+    "coalescence_auxiliary": false,
+    "context_dim": 2,
+    "d_model": 128,
+    "dropout": 0.0,
+    "extreme_tap_levels": [],
+    "individual_width": 32,
+    "leaf_width_bp": 100,
+    "long_block_dilation_cap": 0,
+    "macro_levels": 7,
+    "n_heads": 4,
+    "n_long_blocks": 11,
+    "n_region_scales": 4,
+    "n_regional_blocks": 3,
+    "ordered_leaf_scan_width": 0,
+    "overlapping_regions": false,
+    "pair_concordance_first_order": false,
+    "pair_concordance_scales_bp": [],
+    "pair_concordance_width": 8,
+    "region_downsample_mode": "conv",
+    "region_sfs_bins": 0,
+    "scale_film": false,
+    "segment_reduction_backend": "padded",
+    "site_genetic_span_features": false,
+    "site_lag_span_features": false,
+    "site_ld_lags": [],
+    "site_pair_identity_lags": [],
+    "site_state_width": 0,
+    "site_width": 8,
+    "smooth_time_query_modes": 0,
+    "tied_row_attention_heads": 0,
+    "tree_tap_levels": [],
+    "use_checkpoint": false,
+    "use_site_frequency": false
+  },
+  "objective": {
+    "mode_weight_alpha": 0.0,
+    "spectral_decoupling": 0.0
+  },
+  "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+  "score_resume_from": "",
+  "seed": 20260728,
+  "selection_access_count": 0,
+  "selection_rows_read": 0,
+  "stage": "development_single_arm",
+  "stage_rows": {
+    "census": false,
+    "evaluation": 500,
+    "fit": 12500,
+    "inner": 500
+  },
+  "status": "complete"
+}

ablations/no_posterior_12500/run/DONE.json

@@ -0,0 +1,11 @@
+{
+  "experiment_signature": "69ff52c573de6925a676bcb991d7119229d39e9b1c99689be5ac9131dfbe4976",
+  "gate_eligible": false,
+  "report_sha256": "f46be94331a2d116d889b3f59c03bb926396ef11237954206fe37614c48f3324",
+  "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+  "seed": 20260728,
+  "selection_access_count": 0,
+  "selection_rows_read": 0,
+  "stage": "development_single_arm",
+  "status": "complete"
+}

ablations/no_posterior_12500/run/PLAN.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:538b5152f1759146525ee192bddf45d2f5e253ef34d6541f5b88a9c6efc19317
+size 29597111

ablations/no_posterior_12500/run/site_encoder_and_summaries.final.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68b5d29fd594855b31b57b3b9e75a3eb1f51193016b95174e58e020085a37c81
+size 15267151

ablations/no_posterior_12500/run/site_encoder_and_summaries/DONE.json

@@ -0,0 +1,19 @@
+{
+  "arm": "raw_plus_all",
+  "checkpoint_artifact": {
+    "path": "checkpoint.pt",
+    "sha256": "58e35ac89a095b376bc35ceeab7b6fb0d00c11745b530c76e706f9eb5f4553b0"
+  },
+  "experiment_signature": "69ff52c573de6925a676bcb991d7119229d39e9b1c99689be5ac9131dfbe4976",
+  "mechanism_evaluation_artifact": {
+    "path": "mechanism_evaluation.npz",
+    "sha256": "545a7456adecf48557c274a4c0bfc984377617d577a593cd7c0f43b22f8c3268"
+  },
+  "report_sha256": "11294cc4afd8ea92c54287771cfe4416b5f65bccd4b1b17aa8970beff3736ee6",
+  "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+  "seed": 20260728,
+  "selection_access_count": 0,
+  "selection_rows_read": 0,
+  "stage": "pilot",
+  "status": "complete"
+}

ablations/no_posterior_12500/run/site_encoder_and_summaries/checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58e35ac89a095b376bc35ceeab7b6fb0d00c11745b530c76e706f9eb5f4553b0
+size 3803031

ablations/no_posterior_12500/run/site_encoder_and_summaries/mechanism_evaluation.npz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:545a7456adecf48557c274a4c0bfc984377617d577a593cd7c0f43b22f8c3268
+size 581024

ablations/no_posterior_12500/run/site_encoder_and_summaries/report.json

@@ -0,0 +1,369 @@
+{
+  "all_raw_sites_consumed_when_raw_enabled": true,
+  "arm": "raw_plus_all",
+  "best_epoch": 12,
+  "best_inner_validation_rmse": 0.3336641447535338,
+  "checkpoint_artifact": {
+    "path": "checkpoint.pt",
+    "sha256": "58e35ac89a095b376bc35ceeab7b6fb0d00c11745b530c76e706f9eb5f4553b0"
+  },
+  "curve": [
+    {
+      "epoch": 1,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.40015624844809733,
+      "fit_seconds": 2093.6454600400175,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 43.5588939129957,
+      "inner_validation_mae": 0.2536635223776102,
+      "inner_validation_rmse": 0.3786614608951453,
+      "learning_rate": 0.0002951444302461922,
+      "world_size": 4
+    },
+    {
+      "epoch": 2,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3615719489050102,
+      "fit_seconds": 2018.276178309985,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.653537246980704,
+      "inner_validation_mae": 0.23118237311020493,
+      "inner_validation_rmse": 0.35846581330878063,
+      "learning_rate": 0.0002809086200392825,
+      "world_size": 4
+    },
+    {
+      "epoch": 3,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.35103077339022887,
+      "fit_seconds": 1974.612629832991,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.717835613992065,
+      "inner_validation_mae": 0.22763696809858083,
+      "inner_validation_rmse": 0.3529769172358241,
+      "learning_rate": 0.000258262716319083,
+      "world_size": 4
+    },
+    {
+      "epoch": 4,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3444591158997494,
+      "fit_seconds": 1933.7861313030007,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.5098103939672,
+      "inner_validation_mae": 0.2255077318586409,
+      "inner_validation_rmse": 0.3526106006292206,
+      "learning_rate": 0.00022875,
+      "world_size": 4
+    },
+    {
+      "epoch": 5,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3391718357446799,
+      "fit_seconds": 1917.0238381460076,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.68309396895347,
+      "inner_validation_mae": 0.22725810876861213,
+      "inner_validation_rmse": 0.3478976199499427,
+      "learning_rate": 0.00019438171392710925,
+      "world_size": 4
+    },
+    {
+      "epoch": 6,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3353327931839078,
+      "fit_seconds": 1899.3311412120238,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.096282957994845,
+      "inner_validation_mae": 0.23042901559919118,
+      "inner_validation_rmse": 0.3471583560419813,
+      "learning_rate": 0.00015750000000000003,
+      "world_size": 4
+    },
+    {
+      "epoch": 7,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3305477654173744,
+      "fit_seconds": 1900.9600651880028,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.619981688971166,
+      "inner_validation_mae": 0.22294189005717635,
+      "inner_validation_rmse": 0.3418899322674546,
+      "learning_rate": 0.00012061828607289084,
+      "world_size": 4
+    },
+    {
+      "epoch": 8,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.32624947796277753,
+      "fit_seconds": 1897.2488543219515,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.9244443300413,
+      "inner_validation_mae": 0.21564797833561897,
+      "inner_validation_rmse": 0.3386951979830166,
+      "learning_rate": 8.625000000000005e-05,
+      "world_size": 4
+    },
+    {
+      "epoch": 9,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3228154418396384,
+      "fit_seconds": 1888.3135059690103,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.646107722015586,
+      "inner_validation_mae": 0.21422379455715418,
+      "inner_validation_rmse": 0.3375664071352261,
+      "learning_rate": 5.6737283680917e-05,
+      "world_size": 4
+    },
+    {
+      "epoch": 10,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.31890629316474045,
+      "fit_seconds": 1871.0354464069824,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 36.9466422580299,
+      "inner_validation_mae": 0.20946778214350342,
+      "inner_validation_rmse": 0.3341248524935109,
+      "learning_rate": 3.4091379960717494e-05,
+      "world_size": 4
+    },
+    {
+      "epoch": 11,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3162338020659697,
+      "fit_seconds": 1885.552217441029,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.89084733999334,
+      "inner_validation_mae": 0.2101674868762493,
+      "inner_validation_rmse": 0.333845803512705,
+      "learning_rate": 1.9855569753807785e-05,
+      "world_size": 4
+    },
+    {
+      "epoch": 12,
+      "fit_raw_sites_consumed": 5107190494,
+      "fit_rmse": 0.3140382644350096,
+      "fit_seconds": 1858.5625097450102,
+      "inner_raw_sites_consumed": 206175097,
+      "inner_seconds": 37.27406584401615,
+      "inner_validation_mae": 0.21033127358183265,
+      "inner_validation_rmse": 0.3336641447535338,
+      "learning_rate": 1.4999999999999999e-05,
+      "world_size": 4
+    }
+  ],
+  "device": "cuda",
+  "direct_64_bin_output": true,
+  "epochs_ran": 12,
+  "evaluation_raw_sites_consumed": 216060598,
+  "evaluation_seconds": 199.32269198598806,
+  "experiment_signature": "69ff52c573de6925a676bcb991d7119229d39e9b1c99689be5ac9131dfbe4976",
+  "fit_rows": 12500,
+  "fit_seconds": 23138.34797791601,
+  "fixed_epoch_budget": 12,
+  "gpu_name": "NVIDIA GeForce RTX 4090",
+  "hazard_tail_survival_dual_or_class_head": false,
+  "initial_state_sha256": "ebedc02cc9fc0c67ed13724e34af3ef174cb81af45ed25de4c6e8e8678e15e0f",
+  "inner_rows": 500,
+  "inner_validation_seconds": 450.5215432779514,
+  "mechanism_evaluation_artifact": {
+    "path": "mechanism_evaluation.npz",
+    "sha256": "545a7456adecf48557c274a4c0bfc984377617d577a593cd7c0f43b22f8c3268"
+  },
+  "mechanism_evaluation_metrics": {
+    "constant_predictor_rmse": 0.402485317024815,
+    "label_spread_strata": {
+      "flat": {
+        "constant_predictor_rmse": 0.15479404880509584,
+        "label_spread_range": [
+          0.0,
+          0.24
+        ],
+        "mae": 0.11951362635425436,
+        "n_samples": 123,
+        "rmse": 0.17066689261786666
+      },
+      "moderate": {
+        "constant_predictor_rmse": 0.31611850295071287,
+        "label_spread_range": [
+          0.24,
+          0.4
+        ],
+        "mae": 0.1952067300408464,
+        "n_samples": 194,
+        "rmse": 0.2769268088176663
+      },
+      "structured": {
+        "constant_predictor_rmse": 0.5661845833240783,
+        "label_spread_range": [
+          0.4,
+          null
+        ],
+        "mae": 0.2893600353431669,
+        "n_samples": 183,
+        "rmse": 0.43653716377354895
+      }
+    },
+    "mae": 0.21104633627459407,
+    "n_samples": 500,
+    "per_bin_mae": [
+      0.25872102117538454,
+      0.2574238557815552,
+      0.2544191207885742,
+      0.24844665241241454,
+      0.23952793931961058,
+      0.23101413989067077,
+      0.2247856810092926,
+      0.21954951190948485,
+      0.2137761218547821,
+      0.21821088123321533,
+      0.21430532002449035,
+      0.20765164136886596,
+      0.20636261677742004,
+      0.2009211320877075,
+      0.20189660716056823,
+      0.1875525505542755,
+      0.18837180042266846,
+      0.1838572347164154,
+      0.18573030161857604,
+      0.17980736780166626,
+      0.17956282472610474,
+      0.17695083236694337,
+      0.16958984184265136,
+      0.1661336784362793,
+      0.16632140636444093,
+      0.1746431484222412,
+      0.17343592500686644,
+      0.1762556366920471,
+      0.17273916673660278,
+      0.16931749629974366,
+      0.1715409984588623,
+      0.1757011661529541,
+      0.17381114053726196,
+      0.1748917989730835,
+      0.17907089328765868,
+      0.18520425844192506,
+      0.19175126791000366,
+      0.18996810913085938,
+      0.20550774765014648,
+      0.21040356731414794,
+      0.2123220829963684,
+      0.20762387323379516,
+      0.2060391812324524,
+      0.20466476488113403,
+      0.2028860273361206,
+      0.20784376621246337,
+      0.2120401487350464,
+      0.21495712804794312,
+      0.21747990798950195,
+      0.22602142095565797,
+      0.22599719142913818,
+      0.22860319328308104,
+      0.23108028650283813,
+      0.23668288993835448,
+      0.2414417772293091,
+      0.2444851508140564,
+      0.2414763617515564,
+      0.24355956125259398,
+      0.24764243984222412,
+      0.25615241622924806,
+      0.26626666355133055,
+      0.2660611934661865,
+      0.2707990622520447,
+      0.2897066297531128
+    ],
+    "per_bin_rmse": [
+      0.34896980726431753,
+      0.34709251055552404,
+      0.3421776787523031,
+      0.3357477348165305,
+      0.3236791126624797,
+      0.31303682507500463,
+      0.3082640312649185,
+      0.3027514693268415,
+      0.2988764919988645,
+      0.30935150618101337,
+      0.30319353732431675,
+      0.2909803704134548,
+      0.29893455193244517,
+      0.29629442222182034,
+      0.29874164242952256,
+      0.26751661301301977,
+      0.28004897708551246,
+      0.273244006549535,
+      0.2756225409016931,
+      0.269652158191675,
+      0.27050748435753463,
+      0.25807585002971895,
+      0.2550076447945779,
+      0.24136317852276862,
+      0.2561427929475607,
+      0.2788487170454883,
+      0.2702609319612673,
+      0.2604130407260646,
+      0.2530495535403907,
+      0.24866028651762162,
+      0.25054212809775017,
+      0.2652816944878899,
+      0.2650450787098498,
+      0.2609937648429789,
+      0.26899104256336775,
+      0.2788974421869027,
+      0.2923080090466803,
+      0.29644489907235044,
+      0.3269216591311331,
+      0.3474323063187913,
+      0.3577926430077441,
+      0.3517563847712244,
+      0.3432117212312172,
+      0.3358277829235053,
+      0.32555655098598446,
+      0.3402813929580356,
+      0.34608173996197855,
+      0.3456521872670643,
+      0.3546878240549255,
+      0.3738941922531639,
+      0.3743833889814278,
+      0.37811233084276824,
+      0.3865729221224253,
+      0.39172335883565496,
+      0.3924530748232029,
+      0.4002031369657275,
+      0.39565291120606483,
+      0.39437730362043927,
+      0.39916586972934864,
+      0.4030455671467172,
+      0.41991077243034874,
+      0.41648041018961873,
+      0.42156874034021985,
+      0.4445341710296161
+    ],
+    "rmse": 0.3265992778522081
+  },
+  "mechanism_evaluation_rows": 500,
+  "mechanism_evaluation_rows_requested": 500,
+  "model_name": "carrier_raw_summary_capacity_matched_direct64_nw_residual_v1",
+  "model_parameters": 935529,
+  "optimizer_boundaries_sha256": "ce18181bfb956b04c297d2cb57aa005b40b1e38d281c91b7bb45b430faf64adf",
+  "peak_cuda_allocated_bytes": 12190329856,
+  "peak_cuda_reserved_bytes": 13321109504,
+  "publication_benchmark": false,
+  "row_order_sha256": "ff48ba7533087a7f438b18488c5d13f2260e3734220b11666507796956c2689f",
+  "schema": "dlcoalsim-carrier-raw-summary-fusion-v1",
+  "seed": 20260728,
+  "segment_reduction_backend": "padded",
+  "selection_access_count": 0,
+  "selection_rows_read": 0,
+  "stage": "pilot",
+  "status": "complete",
+  "summary_only_skips_raw_io": false,
+  "trainable_parameters": 935529,
+  "training_objective": {
+    "mode_weight_alpha": 0.0,
+    "principal_basis_rows": 0,
+    "spectral_decoupling": 0.0
+  },
+  "training_raw_sites_consumed": 18535440261,
+  "training_seconds": 23138.34797791601
+}

ablations/no_summaries_12500/logs/attempt1_20260917_090003.log

@@ -0,0 +1,27 @@
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+/cloud/cloud-ssd1/dlcoal/venv/lib/python3.10/site-packages/torch/distributed/c10d_logger.py:83: UserWarning: barrier(): using the device under current context. You can specify `device_id` in `init_process_group` to mute this warning.
+  return func(*args, **kwargs)
+[rank0]:[W917 09:13:01.175735861 ProcessGroupNCCL.cpp:5310] Guessing device ID based on global rank. This can cause a hang if rank to GPU mapping is heterogeneous. You can specify device_id in init_process_group()
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"arm": "raw_plus_sidecar", "epoch": 1, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.40280082986333887, "fit_seconds": 2100.75503844599, "inner_raw_sites_consumed": 206175097, "inner_seconds": 44.6153955560294, "inner_validation_mae": 0.2623751557096839, "inner_validation_rmse": 0.3860716412534121, "learning_rate": 0.0002951444302461922, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 2, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3728956463993135, "fit_seconds": 2026.6554818970035, "inner_raw_sites_consumed": 206175097, "inner_seconds": 38.323421272973064, "inner_validation_mae": 0.24997115455567837, "inner_validation_rmse": 0.3744709933991675, "learning_rate": 0.0002809086200392825, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 3, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.36228370012790256, "fit_seconds": 1989.884824140987, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.367676672991365, "inner_validation_mae": 0.25331357107684016, "inner_validation_rmse": 0.37672541360497436, "learning_rate": 0.000258262716319083, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 4, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3529866970998087, "fit_seconds": 1952.2969021759927, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.70884804299567, "inner_validation_mae": 0.24276938736438752, "inner_validation_rmse": 0.36397743890068485, "learning_rate": 0.00022875, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 5, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.34569708132813953, "fit_seconds": 1930.6446910519735, "inner_raw_sites_consumed": 206175097, "inner_seconds": 38.290049678005744, "inner_validation_mae": 0.23556552973017097, "inner_validation_rmse": 0.3559241722737078, "learning_rate": 0.00019438171392710925, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 6, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.33954955783298707, "fit_seconds": 1909.0664127360214, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.501103370974306, "inner_validation_mae": 0.23325910672917963, "inner_validation_rmse": 0.35368021959716006, "learning_rate": 0.00015750000000000003, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 7, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3349930672057057, "fit_seconds": 1893.8109187550144, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.22699772298802, "inner_validation_mae": 0.2302289959192276, "inner_validation_rmse": 0.3502578632841301, "learning_rate": 0.00012061828607289084, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 8, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.32935283910263935, "fit_seconds": 1890.7674577709986, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.861188630980905, "inner_validation_mae": 0.22890744926407933, "inner_validation_rmse": 0.35300032044417096, "learning_rate": 8.625000000000005e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 9, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3257234434308157, "fit_seconds": 1880.4224090510397, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.58814372197958, "inner_validation_mae": 0.22018843384459616, "inner_validation_rmse": 0.3426789540165044, "learning_rate": 5.6737283680917e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 10, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.32155072777293303, "fit_seconds": 1866.504600921995, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.493068513984326, "inner_validation_mae": 0.22043433362990617, "inner_validation_rmse": 0.34388450286360606, "learning_rate": 3.4091379960717494e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 11, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31840896090081117, "fit_seconds": 1876.714364250016, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.08863538102014, "inner_validation_mae": 0.2191481573805213, "inner_validation_rmse": 0.340967715307979, "learning_rate": 1.9855569753807785e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 12, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31642129180121525, "fit_seconds": 1847.4906284600147, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.80954212299548, "inner_validation_mae": 0.21691400266066194, "inner_validation_rmse": 0.339111135363208, "learning_rate": 1.4999999999999999e-05, "seed": 20260728, "world_size": 4}

ablations/no_summaries_12500/logs/command.txt

@@ -0,0 +1,3 @@
+cwd /cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32
+env DLCOAL_TRAIN_DEPLOY=/cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32 DLCOAL_COHORT_SPLIT=99996,5000,4248 DLCOAL_FORMAL_EPOCHS=12 OMP_NUM_THREADS=8 PYTHONDONTWRITEBYTECODE=1 CUBLAS_WORKSPACE_CONFIG=:4096:8 MASTER_ADDR=127.0.0.1 MASTER_PORT=29711
+/cloud/cloud-ssd1/dlcoal/venv/bin/torchrun --rdzv-backend=c10d --rdzv-endpoint=127.0.0.1:29711 --nnodes=1 --nproc_per_node=4 -m training.carrier_raw_summary_fusion_runner --registry-dir /cloud/cloud-ssd1/dlcoal/registry109k --unphased-cache /cloud/cloud-ssd1/dlcoal/level_caches109k/chain0_level32/unphased_evidence_cache.npz --sidecar-npz /cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz --expected-selection-rows 4248 --out-dir /cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/rawsc_fit12500_e12 --stage pilot --development-single-arm raw_plus_sidecar --seed 20260728 --epochs 12 --formal-epochs 12 --pilot-rows 13500 --pilot-fit-rows 12500 --pilot-inner-rows 500 --pilot-evaluation-rows 500 --development-gpu-class 4090 --packed-hap-bytes 16 --d-model 128 --host-cache-gib 48 --host-budget-fraction transient=0.27 --host-budget-fraction artifact=0.16 --host-budget-fraction segment_plan=0.08 --host-budget-fraction map_exposure=0.015 --host-budget-fraction map=0.02 --host-budget-fraction shift=0.005 --host-budget-fraction prefetch=0.35 --prefetch-rows 10 --stage-workers 4 --host-cache-sites 20000000000 --accept-cache-refill --resume 

ablations/no_summaries_12500/logs/harness_run.json

@@ -0,0 +1,29 @@
+{
+  "run": "rawsc_fit12500_e12",
+  "started_utc": "2026-09-17T09:00:03Z",
+  "launcher": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/bin/train_ablation.sh",
+  "launcher_sha256": "31689adea48e27b63cda2cad9c0b1ae5a34cb42f5d3285d32834ce578e5398e6",
+  "deploy": "/cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32",
+  "deploy_manifest_sha256": "659f0427dc079a0bdb2ad3ae8ff8a328e98df6eeeb623532c34cb9a2bcd0b502",
+  "arm": "raw_plus_sidecar",
+  "fit_rows": 12500,
+  "inner_rows": 500,
+  "evaluation_rows": 500,
+  "epochs": 12,
+  "seed": 20260728,
+  "d_model": 128,
+  "world_size": 4,
+  "gpu_class": "4090",
+  "registry": "/cloud/cloud-ssd1/dlcoal/registry109k",
+  "unphased_cache": "/cloud/cloud-ssd1/dlcoal/level_caches109k/chain0_level32/unphased_evidence_cache.npz",
+  "sidecar_npz": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz",
+  "sidecar_bytes": 107621430,
+  "host_cache_gib": 48,
+  "extra_args": "",
+  "strict_budgets": true,
+  "master_port": 29711,
+  "smoke": false,
+  "out_dir": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/rawsc_fit12500_e12",
+  "log_dir": "/cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/rawsc_fit12500_e12",
+  "free_gb_at_start": 156
+}

ablations/no_summaries_12500/logs/rawsc_fit12500_e12.out

@@ -0,0 +1,43 @@
+=== data disk free 156 GB (floor 30 GB)
+=== deploy-hash guard: the three runner files match /cloud/cloud-ssd1/dlcoal/prep_fixed32/EXPECTED_SHA256.txt
+=== deploy verified: manifest 659f0427dc079a0bdb2ad3ae8ff8a328e98df6eeeb623532c34cb9a2bcd0b502 (34 artifacts) at /cloud/cloud-ssd1/dlcoal/deploy_20260913_fixed32
+=== row-count/rank deadlock pre-check (the deploy's rank guard)
+=== FIT=12500 at 4 ranks: 391 optimizer steps per epoch on every rank, world 1's
+=== memwatch every 60s -> /cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/rawsc_fit12500_e12/memwatch.log (cgroup memory.max 398358216704)
+=== RUN=rawsc_fit12500_e12 ARM=raw_plus_sidecar FIT=12500 INNER=500 EVAL=500 SEED=20260728 EPOCHS=12 D_MODEL=128 SMOKE=0
+=== 4 x 4090  host cache 48 GiB  OMP=8  stage workers 4  prefetch 10
+=== cache=/cloud/cloud-ssd1/dlcoal/level_caches109k/chain0_level32/unphased_evidence_cache.npz
+=== sidecar=/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz
+=== OUT=/cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/rawsc_fit12500_e12
+=== logs=/cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/rawsc_fit12500_e12  retries=2  MASTER_PORT=29711
+=== 09-17 09:00:03 attempt 1/3, log /cloud/cloud-ssd1/dlcoal/ablation_20260916/logs/rawsc_fit12500_e12/attempt1_20260917_090003.log
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"development_gpu_class_override": "4090", "formal_result": false, "observed_gpu": "NVIDIA GeForce RTX 4090"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"fit_coverage": 1.0, "fit_rows_scored": 89996, "present_flag": "declared", "rows": 104996, "rows_imputed": 0, "sidecar_loaded": "/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz"}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+/cloud/cloud-ssd1/dlcoal/venv/lib/python3.10/site-packages/torch/distributed/c10d_logger.py:83: UserWarning: barrier(): using the device under current context. You can specify `device_id` in `init_process_group` to mute this warning.
+  return func(*args, **kwargs)
+[rank0]:[W917 09:13:01.175735861 ProcessGroupNCCL.cpp:5310] Guessing device ID based on global rank. This can cause a hang if rank to GPU mapping is heterogeneous. You can specify device_id in init_process_group()
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"accepted_cache_refill": true, "caches_that_refill": ["artifact", "control_shift_per_kind", "map_both_kinds", "map_exposure_both_kinds", "segment_plan"], "note": "the corpus is re-read from disk each epoch; a row is visited once per epoch either way, so this is the second and later epochs' reads, not extra work within an epoch", "refill_bytes_per_epoch": 1339541919183, "refill_gib_per_epoch": 1247.5}
+{"arm": "raw_plus_sidecar", "epoch": 1, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.40280082986333887, "fit_seconds": 2100.75503844599, "inner_raw_sites_consumed": 206175097, "inner_seconds": 44.6153955560294, "inner_validation_mae": 0.2623751557096839, "inner_validation_rmse": 0.3860716412534121, "learning_rate": 0.0002951444302461922, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 2, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3728956463993135, "fit_seconds": 2026.6554818970035, "inner_raw_sites_consumed": 206175097, "inner_seconds": 38.323421272973064, "inner_validation_mae": 0.24997115455567837, "inner_validation_rmse": 0.3744709933991675, "learning_rate": 0.0002809086200392825, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 3, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.36228370012790256, "fit_seconds": 1989.884824140987, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.367676672991365, "inner_validation_mae": 0.25331357107684016, "inner_validation_rmse": 0.37672541360497436, "learning_rate": 0.000258262716319083, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 4, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3529866970998087, "fit_seconds": 1952.2969021759927, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.70884804299567, "inner_validation_mae": 0.24276938736438752, "inner_validation_rmse": 0.36397743890068485, "learning_rate": 0.00022875, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 5, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.34569708132813953, "fit_seconds": 1930.6446910519735, "inner_raw_sites_consumed": 206175097, "inner_seconds": 38.290049678005744, "inner_validation_mae": 0.23556552973017097, "inner_validation_rmse": 0.3559241722737078, "learning_rate": 0.00019438171392710925, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 6, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.33954955783298707, "fit_seconds": 1909.0664127360214, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.501103370974306, "inner_validation_mae": 0.23325910672917963, "inner_validation_rmse": 0.35368021959716006, "learning_rate": 0.00015750000000000003, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 7, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3349930672057057, "fit_seconds": 1893.8109187550144, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.22699772298802, "inner_validation_mae": 0.2302289959192276, "inner_validation_rmse": 0.3502578632841301, "learning_rate": 0.00012061828607289084, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 8, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.32935283910263935, "fit_seconds": 1890.7674577709986, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.861188630980905, "inner_validation_mae": 0.22890744926407933, "inner_validation_rmse": 0.35300032044417096, "learning_rate": 8.625000000000005e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 9, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.3257234434308157, "fit_seconds": 1880.4224090510397, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.58814372197958, "inner_validation_mae": 0.22018843384459616, "inner_validation_rmse": 0.3426789540165044, "learning_rate": 5.6737283680917e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 10, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.32155072777293303, "fit_seconds": 1866.504600921995, "inner_raw_sites_consumed": 206175097, "inner_seconds": 36.493068513984326, "inner_validation_mae": 0.22043433362990617, "inner_validation_rmse": 0.34388450286360606, "learning_rate": 3.4091379960717494e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 11, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31840896090081117, "fit_seconds": 1876.714364250016, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.08863538102014, "inner_validation_mae": 0.2191481573805213, "inner_validation_rmse": 0.340967715307979, "learning_rate": 1.9855569753807785e-05, "seed": 20260728, "world_size": 4}
+{"arm": "raw_plus_sidecar", "epoch": 12, "fit_raw_sites_consumed": 5107190494, "fit_rmse": 0.31642129180121525, "fit_seconds": 1847.4906284600147, "inner_raw_sites_consumed": 206175097, "inner_seconds": 37.80954212299548, "inner_validation_mae": 0.21691400266066194, "inner_validation_rmse": 0.339111135363208, "learning_rate": 1.4999999999999999e-05, "seed": 20260728, "world_size": 4}
+runner exit=0 (attempt 1)
+=== launcher wall time 24730 s
+launcher exit=0

ablations/no_summaries_12500/logs/rawsc_fit12500_e12.supervisor.log

@@ -0,0 +1,4 @@
+09-17 09:00:00 supervisor start pid 676562 for RUN=rawsc_fit12500_e12 (max relaunch 4, unknown 1, interval 120s)
+09-17 09:00:00 OUT=/cloud/cloud-ssd1/dlcoal/ablation_20260916/runs/rawsc_fit12500_e12  FIT=12500 EPOCHS=12 SEED=20260728 ARM=raw_plus_sidecar SIDECAR=/cloud/cloud-ssd1/dlcoal/sidecar109k/chain0_level32.npz
+09-17 09:00:00 first launch -> rawsc_fit12500_e12.out (RESUME_OK=0)
+09-17 15:52:18 DONE.json present; supervisor exits