Files
PSO/benchmark_results/pso_v7_heavy_cross_split.json
jung-geun 813433000a feat: modernize PSO and add convergence research
Migrate the package and examples to the tensor-native PyTorch implementation, add benchmark evidence, and add the guarded post-training convergence protocol with TensorBoard progress monitoring and hash-verified recovery.

Constraint: Preserve one-shot official-test sealing and auditable research artifacts

Rejected: Commit local .omc runs and downloaded datasets | multi-gigabyte runtime state is machine-local

Confidence: high

Scope-risk: broad

Not-tested: Production CUDA run on pieroot-server
2026-09-07 22:03:25 +09:00

1440 lines
48 KiB
JSON

{
"protocol_version": "HEAVY-PSO-CROSS-SPLIT-PUBLISH 1.0.0",
"pso_version": "4.0.0",
"timestamp": "2026-09-04T07:38:14Z",
"mission_contract": {
"phase": "development",
"official_test_data_loaded": false,
"official_test_evaluations": 0,
"confirmation_executed": false,
"retained_policy": null
},
"official_test_data_loaded": false,
"official_test_evaluations": 0,
"confirmation_executed": false,
"retained_policy": null,
"verdict": {
"status": "NO_RETAINED_POLICY_NO_CONFIRMATION",
"retained_policy": null,
"confirmation_executed": false,
"official_test_data_loaded": false,
"official_test_evaluations": 0,
"best_observed_variant_id": "iteration-0005-development",
"best_observed_score": -185.61068572934795,
"description": "All 9 development candidates failed the frozen evaluator gates (specifically maximum accuracy regression and/or development wide CNN improvement). No candidate qualified for confirmation. Confirmation split 20260907 and official test data remained completely unexecuted and sealed."
},
"source_provenance": {
"source_dir": ".omc/autoresearch/heavy-pso-cross-split-robustness/runs/20260903T162504Z",
"n_variants": 9,
"candidate_files": [
"iteration-0001-development.json",
"iteration-0002-development.json",
"iteration-0003-replica1-development.json",
"iteration-0003-replica2-development.json",
"iteration-0004-development.json",
"iteration-0005-development.json",
"iteration-0006-development.json",
"iteration-0007-development.json",
"iteration-0008-development.json"
],
"evaluation_files": [
"iteration-0001-development.json",
"iteration-0002-development.json",
"iteration-0003-replica1-development.json",
"iteration-0003-replica2-development.json",
"iteration-0004-development.json",
"iteration-0005-development.json",
"iteration-0006-development.json",
"iteration-0007-development.json",
"iteration-0008-development.json"
]
},
"cumulative_resources": {
"n_variants": 9,
"total_runs": 432,
"total_queries": 414720,
"total_sample_evaluations": 4147200000,
"official_test_evaluations": 0,
"best_observed_variant_id": "iteration-0005-development",
"wall_time_sec": 3162.9717,
"best_observed_score": -185.61068572934795
},
"total_runs": 432,
"total_wall_time_sec": 3162.9717,
"total_queries": 414720,
"total_sample_evaluations": 4147200000,
"variants": [
{
"variant_id": "iteration-0001-development",
"candidate_file": "iteration-0001-development.json",
"evaluation_file": "iteration-0001-development.json",
"phase": "development",
"candidate_config": {
"ratio": 0.5,
"geometry_policy": "baseline_aligned",
"projection_scope": "global",
"projection_seed_mode": "explicit",
"projection_seed": {
"mnist_compact": 1800044939,
"mnist_wide": 592157828,
"fashion_compact": 1363313651,
"fashion_wide": 189641451
},
"geometry_multiplier": 1.0,
"particles": 12,
"epochs": 80,
"subset_size": 10000
},
"score": -289.2921035434933,
"pass": false,
"development_pass": false,
"eligible_for_confirmation": false,
"failed_hard_gate_count": 4,
"failed_gates": [
"maximum_accuracy_regression_percentage_points_each_split_workload",
"maximum_nll_regression_fraction_each_split_workload",
"development_mnist_wide_improvement",
"confirmation_executed"
],
"is_best_observed": false,
"summary_metrics": {
"development_cells": 8,
"confirmation_cells": 0,
"development_grand_mean_accuracy_gain_pp": 0.15333312499999963,
"development_grand_mean_nll_reduction_fraction": 0.005545633315067132,
"development_mnist_wide_accuracy_gain_pp": -0.3616670000000006,
"development_mnist_wide_nll_reduction_fraction": -0.024599621226530297
},
"state_ratios": {
"development": {
"mnist_compact": 0.4918032786885246,
"mnist_wide": 0.5,
"fashion_compact": 0.4918032786885246,
"fashion_wide": 0.5
},
"confirmation": null
},
"cell_metrics": [
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_compact",
"baseline_acc": 50.713333,
"candidate_acc": 50.723333,
"baseline_nll": 1.534298,
"candidate_nll": 1.492805,
"acc_gain_pp": 0.00999999999999801,
"nll_reduction_fraction": 0.0270436381980554
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_wide",
"baseline_acc": 46.656667,
"candidate_acc": 46.04,
"baseline_nll": 1.677872,
"candidate_nll": 1.675644,
"acc_gain_pp": -0.6166669999999996,
"nll_reduction_fraction": 0.0013278724479579603
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_compact",
"baseline_acc": 46.366667,
"candidate_acc": 49.576667,
"baseline_nll": 1.497844,
"candidate_nll": 1.400574,
"acc_gain_pp": 3.210000000000001,
"nll_reduction_fraction": 0.06494000710354347
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_wide",
"baseline_acc": 48.516667,
"candidate_acc": 45.01,
"baseline_nll": 1.490304,
"candidate_nll": 1.584923,
"acc_gain_pp": -3.506667,
"nll_reduction_fraction": -0.06348973095422142
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_compact",
"baseline_acc": 46.81,
"candidate_acc": 51.663333,
"baseline_nll": 1.58986,
"candidate_nll": 1.436337,
"acc_gain_pp": 4.853332999999999,
"nll_reduction_fraction": 0.09656384838916639
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_wide",
"baseline_acc": 44.936667,
"candidate_acc": 44.83,
"baseline_nll": 1.665007,
"candidate_nll": 1.749135,
"acc_gain_pp": -0.10666700000000162,
"nll_reduction_fraction": -0.050527114901018556
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_compact",
"baseline_acc": 44.193333,
"candidate_acc": 48.753333,
"baseline_nll": 1.553097,
"candidate_nll": 1.375751,
"acc_gain_pp": 4.559999999999995,
"nll_reduction_fraction": 0.11418861796784104
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_wide",
"baseline_acc": 48.37,
"candidate_acc": 41.193333,
"baseline_nll": 1.490657,
"candidate_nll": 1.707819,
"acc_gain_pp": -7.176666999999995,
"nll_reduction_fraction": -0.14568207173078723
}
],
"resources": {
"runs": 48,
"queries": 46080,
"sample_evaluations": 460800000,
"wall_time_sec": 345.2731,
"official_test_evaluations": 0
}
},
{
"variant_id": "iteration-0002-development",
"candidate_file": "iteration-0002-development.json",
"evaluation_file": "iteration-0002-development.json",
"phase": "development",
"candidate_config": {
"ratio": 0.5,
"geometry_policy": "baseline_aligned",
"projection_scope": {
"mnist_compact": "global",
"mnist_wide": "balanced_global",
"fashion_compact": "global",
"fashion_wide": "balanced_global"
},
"projection_seed_mode": "explicit",
"projection_seed": {
"mnist_compact": 1800044939,
"mnist_wide": 592157828,
"fashion_compact": 1363313651,
"fashion_wide": 189641451
},
"geometry_multiplier": 1.0,
"particles": 12,
"epochs": 80,
"subset_size": 10000
},
"score": -186.3459040597986,
"pass": false,
"development_pass": false,
"eligible_for_confirmation": false,
"failed_hard_gate_count": 3,
"failed_gates": [
"maximum_accuracy_regression_percentage_points_each_split_workload",
"development_mnist_wide_improvement",
"confirmation_executed"
],
"is_best_observed": false,
"summary_metrics": {
"development_cells": 8,
"confirmation_cells": 0,
"development_grand_mean_accuracy_gain_pp": 1.0591663750000002,
"development_grand_mean_nll_reduction_fraction": 0.025949295652013885,
"development_mnist_wide_accuracy_gain_pp": -0.6550004999999963,
"development_mnist_wide_nll_reduction_fraction": -0.03737641704538702
},
"state_ratios": {
"development": {
"mnist_compact": 0.4918032786885246,
"mnist_wide": 0.5,
"fashion_compact": 0.4918032786885246,
"fashion_wide": 0.5
},
"confirmation": null
},
"cell_metrics": [
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_compact",
"baseline_acc": 50.713333,
"candidate_acc": 50.723333,
"baseline_nll": 1.534298,
"candidate_nll": 1.492805,
"acc_gain_pp": 0.00999999999999801,
"nll_reduction_fraction": 0.0270436381980554
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_wide",
"baseline_acc": 46.656667,
"candidate_acc": 45.193333,
"baseline_nll": 1.677872,
"candidate_nll": 1.733563,
"acc_gain_pp": -1.4633339999999961,
"nll_reduction_fraction": -0.03319144726176963
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_compact",
"baseline_acc": 46.366667,
"candidate_acc": 49.576667,
"baseline_nll": 1.497844,
"candidate_nll": 1.400574,
"acc_gain_pp": 3.210000000000001,
"nll_reduction_fraction": 0.06494000710354347
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_wide",
"baseline_acc": 48.516667,
"candidate_acc": 47.223333,
"baseline_nll": 1.490304,
"candidate_nll": 1.536216,
"acc_gain_pp": -1.2933340000000015,
"nll_reduction_fraction": -0.03080713733573818
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_compact",
"baseline_acc": 46.81,
"candidate_acc": 51.663333,
"baseline_nll": 1.58986,
"candidate_nll": 1.436337,
"acc_gain_pp": 4.853332999999999,
"nll_reduction_fraction": 0.09656384838916639
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_wide",
"baseline_acc": 44.936667,
"candidate_acc": 45.09,
"baseline_nll": 1.665007,
"candidate_nll": 1.734207,
"acc_gain_pp": 0.1533330000000035,
"nll_reduction_fraction": -0.04156138682900441
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_compact",
"baseline_acc": 44.193333,
"candidate_acc": 48.753333,
"baseline_nll": 1.553097,
"candidate_nll": 1.375751,
"acc_gain_pp": 4.559999999999995,
"nll_reduction_fraction": 0.11418861796784104
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_wide",
"baseline_acc": 48.37,
"candidate_acc": 46.813333,
"baseline_nll": 1.490657,
"candidate_nll": 1.475127,
"acc_gain_pp": -1.5566669999999974,
"nll_reduction_fraction": 0.010418224984016995
}
],
"resources": {
"runs": 48,
"queries": 46080,
"sample_evaluations": 460800000,
"wall_time_sec": 356.2424,
"official_test_evaluations": 0
}
},
{
"variant_id": "iteration-0003-replica1-development",
"candidate_file": "iteration-0003-replica1-development.json",
"evaluation_file": "iteration-0003-replica1-development.json",
"phase": "development",
"candidate_config": {
"ratio": 0.5,
"geometry_policy": "baseline_aligned",
"projection_scope": {
"mnist_compact": "global",
"mnist_wide": "balanced_global",
"fashion_compact": "global",
"fashion_wide": "balanced_global"
},
"projection_seed_mode": "explicit",
"projection_seed": {
"mnist_compact": 1800044939,
"mnist_wide": 1170674349,
"fashion_compact": 1363313651,
"fashion_wide": 1196992285
},
"geometry_multiplier": 1.0,
"particles": 12,
"epochs": 80,
"subset_size": 10000
},
"score": -287.7250973286179,
"pass": false,
"development_pass": false,
"eligible_for_confirmation": false,
"failed_hard_gate_count": 4,
"failed_gates": [
"maximum_accuracy_regression_percentage_points_each_split_workload",
"maximum_nll_regression_fraction_each_split_workload",
"development_mnist_wide_improvement",
"confirmation_executed"
],
"is_best_observed": false,
"summary_metrics": {
"development_cells": 8,
"confirmation_cells": 0,
"development_grand_mean_accuracy_gain_pp": 0.5999997499999994,
"development_grand_mean_nll_reduction_fraction": 0.016749029213820897,
"development_mnist_wide_accuracy_gain_pp": -2.278334000000001,
"development_mnist_wide_nll_reduction_fraction": -0.029660445675769923
},
"state_ratios": {
"development": {
"mnist_compact": 0.4918032786885246,
"mnist_wide": 0.5,
"fashion_compact": 0.4918032786885246,
"fashion_wide": 0.5
},
"confirmation": null
},
"cell_metrics": [
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_compact",
"baseline_acc": 50.713333,
"candidate_acc": 50.723333,
"baseline_nll": 1.534298,
"candidate_nll": 1.492805,
"acc_gain_pp": 0.00999999999999801,
"nll_reduction_fraction": 0.0270436381980554
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_wide",
"baseline_acc": 46.656667,
"candidate_acc": 44.923333,
"baseline_nll": 1.677872,
"candidate_nll": 1.728906,
"acc_gain_pp": -1.7333339999999993,
"nll_reduction_fraction": -0.030415907768888223
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_compact",
"baseline_acc": 46.366667,
"candidate_acc": 49.576667,
"baseline_nll": 1.497844,
"candidate_nll": 1.400574,
"acc_gain_pp": 3.210000000000001,
"nll_reduction_fraction": 0.06494000710354347
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_wide",
"baseline_acc": 48.516667,
"candidate_acc": 47.176667,
"baseline_nll": 1.490304,
"candidate_nll": 1.56267,
"acc_gain_pp": -1.3399999999999963,
"nll_reduction_fraction": -0.048557878124194744
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_compact",
"baseline_acc": 46.81,
"candidate_acc": 51.663333,
"baseline_nll": 1.58986,
"candidate_nll": 1.436337,
"acc_gain_pp": 4.853332999999999,
"nll_reduction_fraction": 0.09656384838916639
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_wide",
"baseline_acc": 44.936667,
"candidate_acc": 42.113333,
"baseline_nll": 1.665007,
"candidate_nll": 1.713134,
"acc_gain_pp": -2.8233340000000027,
"nll_reduction_fraction": -0.028904983582651624
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_compact",
"baseline_acc": 44.193333,
"candidate_acc": 48.753333,
"baseline_nll": 1.553097,
"candidate_nll": 1.375751,
"acc_gain_pp": 4.559999999999995,
"nll_reduction_fraction": 0.11418861796784104
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_wide",
"baseline_acc": 48.37,
"candidate_acc": 46.433333,
"baseline_nll": 1.490657,
"candidate_nll": 1.581386,
"acc_gain_pp": -1.936667,
"nll_reduction_fraction": -0.06086510847230454
}
],
"resources": {
"runs": 48,
"queries": 46080,
"sample_evaluations": 460800000,
"wall_time_sec": 333.3569,
"official_test_evaluations": 0
}
},
{
"variant_id": "iteration-0003-replica2-development",
"candidate_file": "iteration-0003-replica2-development.json",
"evaluation_file": "iteration-0003-replica2-development.json",
"phase": "development",
"candidate_config": {
"ratio": 0.5,
"geometry_policy": "baseline_aligned",
"projection_scope": {
"mnist_compact": "global",
"mnist_wide": "balanced_global",
"fashion_compact": "global",
"fashion_wide": "balanced_global"
},
"projection_seed_mode": "explicit",
"projection_seed": {
"mnist_compact": 1800044939,
"mnist_wide": 1993337128,
"fashion_compact": 1363313651,
"fashion_wide": 663492771
},
"geometry_multiplier": 1.0,
"particles": 12,
"epochs": 80,
"subset_size": 10000
},
"score": -286.97652369317115,
"pass": false,
"development_pass": false,
"eligible_for_confirmation": false,
"failed_hard_gate_count": 4,
"failed_gates": [
"maximum_accuracy_regression_percentage_points_each_split_workload",
"maximum_nll_regression_fraction_each_split_workload",
"development_mnist_wide_improvement",
"confirmation_executed"
],
"is_best_observed": false,
"summary_metrics": {
"development_cells": 8,
"confirmation_cells": 0,
"development_grand_mean_accuracy_gain_pp": 0.7808331249999991,
"development_grand_mean_nll_reduction_fraction": 0.02242643181828852,
"development_mnist_wide_accuracy_gain_pp": -2.0133335000000017,
"development_mnist_wide_nll_reduction_fraction": -0.056248381748622824
},
"state_ratios": {
"development": {
"mnist_compact": 0.4918032786885246,
"mnist_wide": 0.5,
"fashion_compact": 0.4918032786885246,
"fashion_wide": 0.5
},
"confirmation": null
},
"cell_metrics": [
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_compact",
"baseline_acc": 50.713333,
"candidate_acc": 50.723333,
"baseline_nll": 1.534298,
"candidate_nll": 1.492805,
"acc_gain_pp": 0.00999999999999801,
"nll_reduction_fraction": 0.0270436381980554
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_wide",
"baseline_acc": 46.656667,
"candidate_acc": 46.01,
"baseline_nll": 1.677872,
"candidate_nll": 1.72888,
"acc_gain_pp": -0.6466670000000008,
"nll_reduction_fraction": -0.0304004119503752
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_compact",
"baseline_acc": 46.366667,
"candidate_acc": 49.576667,
"baseline_nll": 1.497844,
"candidate_nll": 1.400574,
"acc_gain_pp": 3.210000000000001,
"nll_reduction_fraction": 0.06494000710354347
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_wide",
"baseline_acc": 48.516667,
"candidate_acc": 48.613333,
"baseline_nll": 1.490304,
"candidate_nll": 1.48049,
"acc_gain_pp": 0.09666599999999903,
"nll_reduction_fraction": 0.006585233616765431
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_compact",
"baseline_acc": 46.81,
"candidate_acc": 51.663333,
"baseline_nll": 1.58986,
"candidate_nll": 1.436337,
"acc_gain_pp": 4.853332999999999,
"nll_reduction_fraction": 0.09656384838916639
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_wide",
"baseline_acc": 44.936667,
"candidate_acc": 41.556667,
"baseline_nll": 1.665007,
"candidate_nll": 1.801698,
"acc_gain_pp": -3.3800000000000026,
"nll_reduction_fraction": -0.08209635154687045
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_compact",
"baseline_acc": 44.193333,
"candidate_acc": 48.753333,
"baseline_nll": 1.553097,
"candidate_nll": 1.375751,
"acc_gain_pp": 4.559999999999995,
"nll_reduction_fraction": 0.11418861796784104
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_wide",
"baseline_acc": 48.37,
"candidate_acc": 45.913333,
"baseline_nll": 1.490657,
"candidate_nll": 1.516614,
"acc_gain_pp": -2.456666999999996,
"nll_reduction_fraction": -0.017413127231817923
}
],
"resources": {
"runs": 48,
"queries": 46080,
"sample_evaluations": 460800000,
"wall_time_sec": 340.8067,
"official_test_evaluations": 0
}
},
{
"variant_id": "iteration-0004-development",
"candidate_file": "iteration-0004-development.json",
"evaluation_file": "iteration-0004-development.json",
"phase": "development",
"candidate_config": {
"ratio": 0.5,
"geometry_policy": "baseline_aligned",
"projection_scope": {
"mnist_compact": "global",
"mnist_wide": "two_hash_global",
"fashion_compact": "global",
"fashion_wide": "two_hash_global"
},
"projection_seed_mode": "explicit",
"projection_seed": {
"mnist_compact": 1800044939,
"mnist_wide": 592157828,
"fashion_compact": 1363313651,
"fashion_wide": 189641451
},
"geometry_multiplier": 1.0,
"particles": 12,
"epochs": 80,
"subset_size": 10000
},
"score": -390.14004501856266,
"pass": false,
"development_pass": false,
"eligible_for_confirmation": false,
"failed_hard_gate_count": 5,
"failed_gates": [
"maximum_accuracy_regression_percentage_points_each_split_workload",
"maximum_nll_regression_fraction_each_split_workload",
"development_grand_mean_accuracy_gain_minimum_pp",
"development_mnist_wide_improvement",
"confirmation_executed"
],
"is_best_observed": false,
"summary_metrics": {
"development_cells": 8,
"confirmation_cells": 0,
"development_grand_mean_accuracy_gain_pp": -0.4275001249999999,
"development_grand_mean_nll_reduction_fraction": 0.002874551064373243,
"development_mnist_wide_accuracy_gain_pp": -3.941666999999999,
"development_mnist_wide_nll_reduction_fraction": -0.05273851995206307
},
"state_ratios": {
"development": {
"mnist_compact": 0.4918032786885246,
"mnist_wide": 0.5,
"fashion_compact": 0.4918032786885246,
"fashion_wide": 0.5
},
"confirmation": null
},
"cell_metrics": [
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_compact",
"baseline_acc": 50.713333,
"candidate_acc": 50.723333,
"baseline_nll": 1.534298,
"candidate_nll": 1.492805,
"acc_gain_pp": 0.00999999999999801,
"nll_reduction_fraction": 0.0270436381980554
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_wide",
"baseline_acc": 46.656667,
"candidate_acc": 44.763333,
"baseline_nll": 1.677872,
"candidate_nll": 1.698999,
"acc_gain_pp": -1.8933339999999959,
"nll_reduction_fraction": -0.012591544527830428
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_compact",
"baseline_acc": 46.366667,
"candidate_acc": 49.576667,
"baseline_nll": 1.497844,
"candidate_nll": 1.400574,
"acc_gain_pp": 3.210000000000001,
"nll_reduction_fraction": 0.06494000710354347
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_wide",
"baseline_acc": 48.516667,
"candidate_acc": 44.816667,
"baseline_nll": 1.490304,
"candidate_nll": 1.582588,
"acc_gain_pp": -3.6999999999999957,
"nll_reduction_fraction": -0.06192293652838617
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_compact",
"baseline_acc": 46.81,
"candidate_acc": 51.663333,
"baseline_nll": 1.58986,
"candidate_nll": 1.436337,
"acc_gain_pp": 4.853332999999999,
"nll_reduction_fraction": 0.09656384838916639
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_wide",
"baseline_acc": 44.936667,
"candidate_acc": 38.946667,
"baseline_nll": 1.665007,
"candidate_nll": 1.819662,
"acc_gain_pp": -5.990000000000002,
"nll_reduction_fraction": -0.09288549537629572
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_compact",
"baseline_acc": 44.193333,
"candidate_acc": 48.753333,
"baseline_nll": 1.553097,
"candidate_nll": 1.375751,
"acc_gain_pp": 4.559999999999995,
"nll_reduction_fraction": 0.11418861796784104
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_wide",
"baseline_acc": 48.37,
"candidate_acc": 43.9,
"baseline_nll": 1.490657,
"candidate_nll": 1.658117,
"acc_gain_pp": -4.469999999999999,
"nll_reduction_fraction": -0.11233972671110803
}
],
"resources": {
"runs": 48,
"queries": 46080,
"sample_evaluations": 460800000,
"wall_time_sec": 341.0849,
"official_test_evaluations": 0
}
},
{
"variant_id": "iteration-0005-development",
"candidate_file": "iteration-0005-development.json",
"evaluation_file": "iteration-0005-development.json",
"phase": "development",
"candidate_config": {
"ratio": 0.5,
"geometry_policy": "baseline_aligned",
"projection_scope": {
"mnist_compact": "global",
"mnist_wide": "largest_tensor_hash",
"fashion_compact": "global",
"fashion_wide": "largest_tensor_hash"
},
"projection_seed_mode": "explicit",
"projection_seed": {
"mnist_compact": 1800044939,
"mnist_wide": 592157828,
"fashion_compact": 1363313651,
"fashion_wide": 189641451
},
"geometry_multiplier": 1.0,
"particles": 12,
"epochs": 80,
"subset_size": 10000
},
"score": -185.61068572934795,
"pass": false,
"development_pass": false,
"eligible_for_confirmation": false,
"failed_hard_gate_count": 3,
"failed_gates": [
"maximum_accuracy_regression_percentage_points_each_split_workload",
"development_mnist_wide_improvement",
"confirmation_executed"
],
"is_best_observed": true,
"summary_metrics": {
"development_cells": 8,
"confirmation_cells": 0,
"development_grand_mean_accuracy_gain_pp": 0.43999974999999925,
"development_grand_mean_nll_reduction_fraction": 0.03949314520652041,
"development_mnist_wide_accuracy_gain_pp": -2.6633340000000025,
"development_mnist_wide_nll_reduction_fraction": -0.02187403189962274
},
"state_ratios": {
"development": {
"mnist_compact": 0.4918032786885246,
"mnist_wide": 0.5,
"fashion_compact": 0.4918032786885246,
"fashion_wide": 0.5
},
"confirmation": null
},
"cell_metrics": [
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_compact",
"baseline_acc": 50.713333,
"candidate_acc": 50.723333,
"baseline_nll": 1.534298,
"candidate_nll": 1.492805,
"acc_gain_pp": 0.00999999999999801,
"nll_reduction_fraction": 0.0270436381980554
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_wide",
"baseline_acc": 46.656667,
"candidate_acc": 43.723333,
"baseline_nll": 1.677872,
"candidate_nll": 1.675155,
"acc_gain_pp": -2.933334000000002,
"nll_reduction_fraction": 0.0016193130346057866
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_compact",
"baseline_acc": 46.366667,
"candidate_acc": 49.576667,
"baseline_nll": 1.497844,
"candidate_nll": 1.400574,
"acc_gain_pp": 3.210000000000001,
"nll_reduction_fraction": 0.06494000710354347
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_wide",
"baseline_acc": 48.516667,
"candidate_acc": 45.45,
"baseline_nll": 1.490304,
"candidate_nll": 1.441667,
"acc_gain_pp": -3.0666669999999954,
"nll_reduction_fraction": 0.03263562333591002
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_compact",
"baseline_acc": 46.81,
"candidate_acc": 51.663333,
"baseline_nll": 1.58986,
"candidate_nll": 1.436337,
"acc_gain_pp": 4.853332999999999,
"nll_reduction_fraction": 0.09656384838916639
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_wide",
"baseline_acc": 44.936667,
"candidate_acc": 42.543333,
"baseline_nll": 1.665007,
"candidate_nll": 1.740544,
"acc_gain_pp": -2.393334000000003,
"nll_reduction_fraction": -0.04536737683385127
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_compact",
"baseline_acc": 44.193333,
"candidate_acc": 48.753333,
"baseline_nll": 1.553097,
"candidate_nll": 1.375751,
"acc_gain_pp": 4.559999999999995,
"nll_reduction_fraction": 0.11418861796784104
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_wide",
"baseline_acc": 48.37,
"candidate_acc": 47.65,
"baseline_nll": 1.490657,
"candidate_nll": 1.454402,
"acc_gain_pp": -0.7199999999999989,
"nll_reduction_fraction": 0.02432149045689245
}
],
"resources": {
"runs": 48,
"queries": 46080,
"sample_evaluations": 460800000,
"wall_time_sec": 342.0566,
"official_test_evaluations": 0
}
},
{
"variant_id": "iteration-0006-development",
"candidate_file": "iteration-0006-development.json",
"evaluation_file": "iteration-0006-development.json",
"phase": "development",
"candidate_config": {
"ratio": 0.5,
"geometry_policy": "baseline_aligned",
"projection_scope": {
"mnist_compact": "global",
"mnist_wide": "largest_tensor_row_hash",
"fashion_compact": "global",
"fashion_wide": "largest_tensor_row_hash"
},
"projection_seed_mode": "explicit",
"projection_seed": {
"mnist_compact": 1800044939,
"mnist_wide": 592157828,
"fashion_compact": 1363313651,
"fashion_wide": 189641451
},
"geometry_multiplier": 1.0,
"particles": 12,
"epochs": 80,
"subset_size": 10000
},
"score": -186.8633001166316,
"pass": false,
"development_pass": false,
"eligible_for_confirmation": false,
"failed_hard_gate_count": 3,
"failed_gates": [
"maximum_accuracy_regression_percentage_points_each_split_workload",
"development_mnist_wide_improvement",
"confirmation_executed"
],
"is_best_observed": false,
"summary_metrics": {
"development_cells": 8,
"confirmation_cells": 0,
"development_grand_mean_accuracy_gain_pp": 0.37833324999999984,
"development_grand_mean_nll_reduction_fraction": 0.027583666333684136,
"development_mnist_wide_accuracy_gain_pp": -4.3733334999999975,
"development_mnist_wide_nll_reduction_fraction": -0.04283278305566589
},
"state_ratios": {
"development": {
"mnist_compact": 0.4918032786885246,
"mnist_wide": 0.5,
"fashion_compact": 0.4918032786885246,
"fashion_wide": 0.5
},
"confirmation": null
},
"cell_metrics": [
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_compact",
"baseline_acc": 50.713333,
"candidate_acc": 50.723333,
"baseline_nll": 1.534298,
"candidate_nll": 1.492805,
"acc_gain_pp": 0.00999999999999801,
"nll_reduction_fraction": 0.0270436381980554
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_wide",
"baseline_acc": 46.656667,
"candidate_acc": 42.006667,
"baseline_nll": 1.677872,
"candidate_nll": 1.749895,
"acc_gain_pp": -4.649999999999999,
"nll_reduction_fraction": -0.04292520525999596
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_compact",
"baseline_acc": 46.366667,
"candidate_acc": 49.576667,
"baseline_nll": 1.497844,
"candidate_nll": 1.400574,
"acc_gain_pp": 3.210000000000001,
"nll_reduction_fraction": 0.06494000710354347
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_wide",
"baseline_acc": 48.516667,
"candidate_acc": 49.08,
"baseline_nll": 1.490304,
"candidate_nll": 1.466855,
"acc_gain_pp": 0.5633330000000001,
"nll_reduction_fraction": 0.015734373657991962
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_compact",
"baseline_acc": 46.81,
"candidate_acc": 51.663333,
"baseline_nll": 1.58986,
"candidate_nll": 1.436337,
"acc_gain_pp": 4.853332999999999,
"nll_reduction_fraction": 0.09656384838916639
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_wide",
"baseline_acc": 44.936667,
"candidate_acc": 40.84,
"baseline_nll": 1.665007,
"candidate_nll": 1.73617,
"acc_gain_pp": -4.0966669999999965,
"nll_reduction_fraction": -0.04274036085133581
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_compact",
"baseline_acc": 44.193333,
"candidate_acc": 48.753333,
"baseline_nll": 1.553097,
"candidate_nll": 1.375751,
"acc_gain_pp": 4.559999999999995,
"nll_reduction_fraction": 0.11418861796784104
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_wide",
"baseline_acc": 48.37,
"candidate_acc": 46.946667,
"baseline_nll": 1.490657,
"candidate_nll": 1.508747,
"acc_gain_pp": -1.4233329999999995,
"nll_reduction_fraction": -0.012135588535793386
}
],
"resources": {
"runs": 48,
"queries": 46080,
"sample_evaluations": 460800000,
"wall_time_sec": 371.7913,
"official_test_evaluations": 0
}
},
{
"variant_id": "iteration-0007-development",
"candidate_file": "iteration-0007-development.json",
"evaluation_file": "iteration-0007-development.json",
"phase": "development",
"candidate_config": {
"ratio": 0.5,
"geometry_policy": "baseline_aligned",
"projection_scope": {
"mnist_compact": "global",
"mnist_wide": "adjacent_pair",
"fashion_compact": "global",
"fashion_wide": "adjacent_pair"
},
"projection_seed_mode": "explicit",
"projection_seed": {
"mnist_compact": 1800044939,
"mnist_wide": 592157828,
"fashion_compact": 1363313651,
"fashion_wide": 189641451
},
"geometry_multiplier": 1.0,
"particles": 12,
"epochs": 80,
"subset_size": 10000
},
"score": -391.33742460463645,
"pass": false,
"development_pass": false,
"eligible_for_confirmation": false,
"failed_hard_gate_count": 5,
"failed_gates": [
"maximum_accuracy_regression_percentage_points_each_split_workload",
"maximum_nll_regression_fraction_each_split_workload",
"development_grand_mean_accuracy_gain_minimum_pp",
"development_mnist_wide_improvement",
"confirmation_executed"
],
"is_best_observed": false,
"summary_metrics": {
"development_cells": 8,
"confirmation_cells": 0,
"development_grand_mean_accuracy_gain_pp": -1.462500125000001,
"development_grand_mean_nll_reduction_fraction": 0.0012507552036354948,
"development_mnist_wide_accuracy_gain_pp": -4.938333500000002,
"development_mnist_wide_nll_reduction_fraction": -0.06456567243825684
},
"state_ratios": {
"development": {
"mnist_compact": 0.4918032786885246,
"mnist_wide": 0.5,
"fashion_compact": 0.4918032786885246,
"fashion_wide": 0.5
},
"confirmation": null
},
"cell_metrics": [
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_compact",
"baseline_acc": 50.713333,
"candidate_acc": 50.723333,
"baseline_nll": 1.534298,
"candidate_nll": 1.492805,
"acc_gain_pp": 0.00999999999999801,
"nll_reduction_fraction": 0.0270436381980554
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_wide",
"baseline_acc": 46.656667,
"candidate_acc": 41.05,
"baseline_nll": 1.677872,
"candidate_nll": 1.794261,
"acc_gain_pp": -5.606667000000002,
"nll_reduction_fraction": -0.06936703157332626
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_compact",
"baseline_acc": 46.366667,
"candidate_acc": 49.576667,
"baseline_nll": 1.497844,
"candidate_nll": 1.400574,
"acc_gain_pp": 3.210000000000001,
"nll_reduction_fraction": 0.06494000710354347
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_wide",
"baseline_acc": 48.516667,
"candidate_acc": 42.513333,
"baseline_nll": 1.490304,
"candidate_nll": 1.615206,
"acc_gain_pp": -6.003333999999995,
"nll_reduction_fraction": -0.08380974619943303
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_compact",
"baseline_acc": 46.81,
"candidate_acc": 51.663333,
"baseline_nll": 1.58986,
"candidate_nll": 1.436337,
"acc_gain_pp": 4.853332999999999,
"nll_reduction_fraction": 0.09656384838916639
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_wide",
"baseline_acc": 44.936667,
"candidate_acc": 40.666667,
"baseline_nll": 1.665007,
"candidate_nll": 1.764515,
"acc_gain_pp": -4.270000000000003,
"nll_reduction_fraction": -0.059764313303187405
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_compact",
"baseline_acc": 44.193333,
"candidate_acc": 48.753333,
"baseline_nll": 1.553097,
"candidate_nll": 1.375751,
"acc_gain_pp": 4.559999999999995,
"nll_reduction_fraction": 0.11418861796784104
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_wide",
"baseline_acc": 48.37,
"candidate_acc": 39.916667,
"baseline_nll": 1.490657,
"candidate_nll": 1.609595,
"acc_gain_pp": -8.453333,
"nll_reduction_fraction": -0.07978897895357565
}
],
"resources": {
"runs": 48,
"queries": 46080,
"sample_evaluations": 460800000,
"wall_time_sec": 356.7982,
"official_test_evaluations": 0
}
},
{
"variant_id": "iteration-0008-development",
"candidate_file": "iteration-0008-development.json",
"evaluation_file": "iteration-0008-development.json",
"phase": "development",
"candidate_config": {
"ratio": 0.5,
"geometry_policy": "baseline_aligned",
"projection_scope": {
"mnist_compact": "global",
"mnist_wide": "adjacent_difference",
"fashion_compact": "global",
"fashion_wide": "adjacent_difference"
},
"projection_seed_mode": "explicit",
"projection_seed": {
"mnist_compact": 1800044939,
"mnist_wide": 592157828,
"fashion_compact": 1363313651,
"fashion_wide": 189641451
},
"geometry_multiplier": 1.0,
"particles": 12,
"epochs": 80,
"subset_size": 10000
},
"score": -390.14811518493707,
"pass": false,
"development_pass": false,
"eligible_for_confirmation": false,
"failed_hard_gate_count": 5,
"failed_gates": [
"maximum_accuracy_regression_percentage_points_each_split_workload",
"maximum_nll_regression_fraction_each_split_workload",
"development_grand_mean_accuracy_gain_minimum_pp",
"development_mnist_wide_improvement",
"confirmation_executed"
],
"is_best_observed": false,
"summary_metrics": {
"development_cells": 8,
"confirmation_cells": 0,
"development_grand_mean_accuracy_gain_pp": -0.7670833750000003,
"development_grand_mean_nll_reduction_fraction": 0.006189681900629345,
"development_mnist_wide_accuracy_gain_pp": -5.428333500000001,
"development_mnist_wide_nll_reduction_fraction": -0.050613618488390404
},
"state_ratios": {
"development": {
"mnist_compact": 0.4918032786885246,
"mnist_wide": 0.5,
"fashion_compact": 0.4918032786885246,
"fashion_wide": 0.5
},
"confirmation": null
},
"cell_metrics": [
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_compact",
"baseline_acc": 50.713333,
"candidate_acc": 50.723333,
"baseline_nll": 1.534298,
"candidate_nll": 1.492805,
"acc_gain_pp": 0.00999999999999801,
"nll_reduction_fraction": 0.0270436381980554
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "mnist_wide",
"baseline_acc": 46.656667,
"candidate_acc": 41.14,
"baseline_nll": 1.677872,
"candidate_nll": 1.748292,
"acc_gain_pp": -5.516666999999998,
"nll_reduction_fraction": -0.041969828449369154
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_compact",
"baseline_acc": 46.366667,
"candidate_acc": 49.576667,
"baseline_nll": 1.497844,
"candidate_nll": 1.400574,
"acc_gain_pp": 3.210000000000001,
"nll_reduction_fraction": 0.06494000710354347
},
{
"phase": "development",
"split_seed": 20260905,
"workload_id": "fashion_wide",
"baseline_acc": 48.516667,
"candidate_acc": 45.726667,
"baseline_nll": 1.490304,
"candidate_nll": 1.558163,
"acc_gain_pp": -2.789999999999999,
"nll_reduction_fraction": -0.04553366293051612
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_compact",
"baseline_acc": 46.81,
"candidate_acc": 51.663333,
"baseline_nll": 1.58986,
"candidate_nll": 1.436337,
"acc_gain_pp": 4.853332999999999,
"nll_reduction_fraction": 0.09656384838916639
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "mnist_wide",
"baseline_acc": 44.936667,
"candidate_acc": 39.596667,
"baseline_nll": 1.665007,
"candidate_nll": 1.763671,
"acc_gain_pp": -5.340000000000003,
"nll_reduction_fraction": -0.059257408527411654
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_compact",
"baseline_acc": 44.193333,
"candidate_acc": 48.753333,
"baseline_nll": 1.553097,
"candidate_nll": 1.375751,
"acc_gain_pp": 4.559999999999995,
"nll_reduction_fraction": 0.11418861796784104
},
{
"phase": "development",
"split_seed": 20260906,
"workload_id": "fashion_wide",
"baseline_acc": 48.37,
"candidate_acc": 43.246667,
"baseline_nll": 1.490657,
"candidate_nll": 1.649349,
"acc_gain_pp": -5.123332999999995,
"nll_reduction_fraction": -0.10645775654627461
}
],
"resources": {
"runs": 48,
"queries": 46080,
"sample_evaluations": 460800000,
"wall_time_sec": 375.5616,
"official_test_evaluations": 0
}
}
]
}