Training Results Master Summary

Executive Snapshot

• Generated At: 2026-04-14T17:40:47

• Program State: active

• Current Completed Wave: Wave 1 structured-baseline familywise optimization pass

• Current Focus: the immediate implementation branch is now the offline

• Active Campaign Status: finished

• Active Campaign Name: track1_svm_open_cell_repair_campaign_2026_04_14_17_17_21

• Current Global Winner: te_hist_gbr_tabular | Family tree | Test MAE 0.002885

Main Takeaways

• Strongest current neural family: residual_harmonic_mlp

• Current plain MLP anchor: te_feedforward_stride1_high_compute_long_remote

• Active family-improvement branch count: 0

• Implemented and benchmarked family count: 5

Current Project Status

Implemented And Benchmarked Families

Family	Current Role	Best Run	Model Type	Test MAE [deg]	Params	Last Update
tree	Current Global Winner	te_hist_gbr_tabular	hist_gradient_boosting	0.002885	5	2026-04-04 13:03:55
residual_harmonic_mlp	Strongest Neural Family	te_residual_h12_deep_joint_wave1	residual_harmonic_mlp	0.003152	26,266	2026-04-04 12:18:37
feedforward	Current Plain MLP Anchor	te_feedforward_stride1_high_compute_long_remote	feedforward	0.003264	109,953	2026-04-04 13:02:09
periodic_mlp	Implemented Benchmark	te_periodic_mlp_h04_standard	periodic_mlp	0.003317	27,265	2026-03-20 15:10:02
harmonic_regression	Implemented Benchmark	te_harmonic_order12_linear_conditioned_recovery	harmonic_regression	0.020782	150	2026-03-20 16:41:05

Active Training Or Improvement Branches

• No campaign is currently in prepared or running state.

• The next active implementation branch should therefore be read from the live backlog focus and the next approved campaign plan.

Roadmap And Planned Work

Wave Or Track	Status
Wave 0. Shared Infrastructure	completed
Wave 1. Structured Static Baselines	planning report: completed; implementation: completed; smoke-tests: completed; validation checks: completed; campaign execution: completed; results report: completed
Wave 2. Temporal Models	planned after the harmonic-wise intermediate branch; temporal-model scope will start only after the harmonic-wise comparison
Intermediate Branch. Harmonic-Wise Comparison Pipeline	current primary implementation branch; focused scope: implement harmonic-wise prediction of A_k and phi_k; implement TE reconstruction from the predicted harmonic terms; add offline Robot and Cycloidal motion-profile playback; define comparable offline validation scenarios and TE-curve error metrics; close Target A
Wave 3. Hybrid Structured Models	pending; paper-reproduction scope:; compare hybrid structured predictors against the paper-style harmonic stack; prepare the repository-owned deployable predictor package
Wave 4. PINN Formulation And First PINN	pending; paper-reproduction scope:; implement the repository-side compensation-loop evaluation path in the; implement uncompensated vs compensated TE RMS / TE max measurements; prepare the final online benchmark harness
Wave 5. Cross-Wave Comparison And Best Solution	pending; paper-reproduction scope:; execute Table 9 style online compensation tests; evaluate Target B; finalize the real paper vs repository comparison with online results

Low-priority exploratory families currently listed in the backlog:

• Lightweight Transformer

• State-Space Sequence Model

• Neural ODE

• Hamiltonian-Inspired Model

• optional Kernel Ridge / Gaussian Process benchmark

Recent Campaign Changes

Campaign	Generated At	Completed	Failed	Winner	Impact
track1_svm_open_cell_repair_campaign_2026_04_14_17_17_21	2026-04-14 17:40:47	12	0	track1_svm_phase_repair_seed11	Canonical SVM row repaired across Tables 2-5; no family-best registry change
track1_full_matrix_family_reproduction_campaign_2026_04_14_13_50_51	2026-04-14 14:35:29	20	0	track1_rf_phase_full_matrix	Full paper-matrix row-reproduction surface prepared; no family-best registry change
track1_second_iteration_harmonic_wise_campaign_2026_04_09_18_56_03	2026-04-09 21:06:13	8	0	te_harmonic_wise_full_rcim_no_engineering_reference	No family-best change
targeted_remote_followup_campaign_2026_04_04_11_21_09	2026-04-04 13:03:55	5	0	te_hist_gbr_remote_refined	No family-best change
remote_training_validation_campaign_2026_04_03_17_54_21	2026-04-03 22:30:26	4	1	te_hist_gbr_remote_deep	No family-best change
wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00	2026-03-26 20:19:32	15	0	te_residual_h12_deep_joint_wave1	Updated residual_harmonic_mlp family best
wave1_structured_baseline_recovery_campaign_2026_03_20_15_40_42	2026-03-20 18:32:18	6	0	te_residual_h12_small_joint_recovery	No family-best change

Ranking Policy

• Primary metric: test_mae

• First tie-breaker: test_rmse

• Second tie-breaker: val_mae

• Third tie-breaker: trainable_parameter_count

• Direction: minimize

Best Result Per Family

Family	Best Run	Model Type	Val MAE [deg]	Test MAE [deg]	Test RMSE [deg]	Params	Artifact Size	Training Cost	Current Role
tree	te_hist_gbr_tabular	hist_gradient_boosting	0.002719	0.002885	0.003607	5	0.62 MB	Unknown	Current Global Winner
residual_harmonic_mlp	te_residual_h12_deep_joint_wave1	residual_harmonic_mlp	0.003024	0.003152	0.003640	26,266	0.32 MB	Medium	Strongest Neural Family
feedforward	te_feedforward_stride1_high_compute_long_remote	feedforward	0.003044	0.003264	0.003679	109,953	1.28 MB	Medium	Current Plain MLP Anchor
periodic_mlp	te_periodic_mlp_h04_standard	periodic_mlp	0.003097	0.003317	0.003793	27,265	0.33 MB	Medium	Implemented Benchmark
harmonic_regression	te_harmonic_order12_linear_conditioned_recovery	harmonic_regression	0.017004	0.020782	0.022405	150	0.01 MB	Low	Implemented Benchmark

Cross-Family Interpretation

• Current global reference winner: te_hist_gbr_tabular from family tree.

• Strongest current neural family: residual_harmonic_mlp.

• Current plain-MLP comparison anchor: te_feedforward_stride1_high_compute_long_remote.

• Predictive quality and deployment suitability must stay separate: the best leaderboard entry is not automatically the best TwinCAT/PLC candidate.

• Large tree artifacts should be treated cautiously even when tree-based accuracy remains strong, because model weight and memory footprint can dominate deployment feasibility.

Paper Reference Benchmark

The repository benchmark paper is reference/RCIM_ML-compensation.pdf. At the current repository state, the comparison is explicitly offline-only. A real paper-equivalent comparison still requires repository-owned online compensation tests.

Extracted Paper Targets

• Paper dataset size: 1026 operating-condition samples.

• Paper input axes: input speed, applied torque, oil temperature.

• Offline prediction target: TE-curve mean percentage error at or below 4.7% on unseen validation scenarios.

• Online robot compensation target: at least 83.6% TE RMS reduction.

• Online cycloidal compensation target: at least 94.0% TE RMS reduction and 91.7% TE max reduction.

• Paper compensation harmonics baseline: 0, 1, 39 with additional checks on 40, 78.

Paper Vs Repository

Comparison Item	Paper Reference	Repository Status	Current Verdict
Offline model-selection direction	Boosting/tree-heavy deployed harmonic predictors	Current winner te_hist_gbr_tabular from family tree with model type hist_gradient_boosting	aligned
Strongest neural branch role	Neural models are evaluated, but not the primary deployed winners	Strongest repository neural family is residual_harmonic_mlp and still trails the tree winner	aligned
Track 1 canonical closure rule	Paper Tables 2-6 replicated per target, per harmonic, and now per family row	Full-matrix benchmark now includes the repaired SVM row with Table 2: 53/37/10, Table 3: 52/35/13, Table 4: 52/37/1, Table 5: 43/41/6; Track 1 still open	not_yet_met
Supporting harmonic-wise TE metric	Mean percentage error over full TE curves	Latest harmonic-wise validation reports 8.707% mean percentage error on held-out curves using harmonics 0, 1, 3, 39, 40, 78, 81, 156, 162, 240	supporting_only_not_yet_met
Online robot-profile compensation	TE RMS reduction 83.6%	No repository-owned online compensation result yet	not_yet_comparable
Online cycloidal-profile compensation	TE RMS reduction 94.0%, TE max reduction 91.7%	No repository-owned online compensation result yet	not_yet_comparable
Table 9-style end-to-end benchmark	PLC-integrated motion-profile compensation benchmark	Missing in the repository at the current state	not_yet_comparable

Track 1 Canonical Status

• Latest exact-paper repair campaign report: doc/reports/campaign_results/2026-04-14-17-40-47_track1_svm_open_cell_repair_campaign_results_report.md

• Full paper-matrix row package status: 20/20 family-row runs completed, plus 12/12 targeted SVM repair runs completed

• Table 2 amplitude MAE full matrix: 53 green, 37 yellow, 10 red cells

• Table 3 amplitude RMSE full matrix: 52 green, 35 yellow, 13 red cells

• Table 4 phase MAE full matrix: 52 green, 37 yellow, 1 red cell

• Table 5 phase RMSE full matrix: 43 green, 41 yellow, 6 red cells

• Strongest current rows: track1_rf_phase_full_matrix, track1_ert_amplitude_full_matrix, track1_hgbm_amplitude_full_matrix, plus the repaired merged SVM row

• Track 1 verdict: full-matrix replication surface exists and SVM is no longer a blocker row, but the paper rows are not yet fully reproduced

Latest Harmonic-Wise Validation Support

• Latest harmonic-wise validation summary: output/validation_checks/paper_reimplementation_rcim_harmonic_wise/2026-04-13-15-11-49__track1_hgbm_h01_wide_depth_2_campaign_run/validation_summary.yaml

• Harmonic-wise test mean percentage error: 8.707%

• Target A status from the latest harmonic-wise run: not_yet_met

Online Compensation Tracking Placeholder

• Repository online compensation status: not yet available.

• When online compensation tests are implemented, update this master summary with TE RMS, TE max, and reduction percentages for both robot and cycloidal motion profiles.

• Until those tests exist, present the paper comparison as offline-only rather than end-to-end equivalent.

Gap Summary

• Track 1 remains open primarily because the canonical Tables 3-6 are not yet fully matched.

• Offline benchmark scope remains partially comparable rather than like-for-like.

• Partially aligned: the current repository winner is tree-based (hist_gradient_boosting / family tree), which is consistent with the paper’s boosting/tree-heavy deployed predictors.

• Neural models remain secondary in the repository (residual_harmonic_mlp), which is also consistent with the paper not promoting a plain neural winner for deployment.

• End-to-end paper comparison remains not yet comparable until repository-owned online compensation tests exist.

Family-By-Family Result Breakdowns

feedforward

• Best run: te_feedforward_stride1_high_compute_long_remote

• Best test MAE: 0.003264

• Completed tracked runs: 20

• Known failed campaign attempts: 0

Rank	Run	Model Type	Test MAE [deg]	Test RMSE [deg]	Val MAE [deg]	Params	Duration	Artifact Size	Model Complexity	Training Heaviness	Campaign
1	te_feedforward_stride1_high_compute_long_remote	feedforward	0.003264	0.003679	0.003044	109,953	30m 09s	1.28 MB	High	Medium	targeted_remote_followup_campaign_2026_04_04_11_21_09
2	te_feedforward_high_compute_remote	feedforward	0.003274	0.003873	0.003059	109,953	10m 24s	1.28 MB	High	Low	remote_training_validation_campaign_2026_04_03_17_54_21
3	te_feedforward_stride1_big_remote	feedforward	0.003278	0.003671	0.003019	109,953	29m 55s	1.28 MB	High	Medium	remote_training_validation_campaign_2026_04_03_17_54_21
4	te_feedforward_stride5_long_large_batch	feedforward	0.003301	0.003791	0.003109	26,241	N/A	0.32 MB	Medium	Unknown	standalone_or_unknown
5	te_feedforward_stride1_long_large_batch_big_model	feedforward	0.003308	0.003779	0.003090	109,953	N/A	1.28 MB	High	Unknown	standalone_or_unknown
6	te_feedforward_high_compute	feedforward	0.003319	0.003915	0.003198	109,953	N/A	1.28 MB	High	Unknown	standalone_or_unknown
7	te_feedforward_high_epoch	feedforward	0.003335	0.003767	0.003007	26,241	N/A	0.32 MB	Medium	Unknown	standalone_or_unknown
8	te_feedforward_stride1_long_large_batch	feedforward	0.003358	0.003769	0.003104	26,241	N/A	0.32 MB	Medium	Unknown	standalone_or_unknown
9	te_feedforward_best_training	feedforward	0.003409	0.003948	0.003039	26,241	N/A	0.32 MB	Medium	Unknown	standalone_or_unknown
10	te_feedforward_stride10_long_large_batch_big_model	feedforward	0.003413	0.004063	0.003040	109,953	N/A	1.28 MB	High	Unknown	standalone_or_unknown
11	te_feedforward_stride10_long_large_batch	feedforward	0.003433	0.004123	0.003066	26,241	N/A	0.32 MB	Medium	Unknown	standalone_or_unknown
12	te_feedforward_stride5_long_large_batch_big_model	feedforward	0.003472	0.004004	0.003104	109,953	N/A	1.28 MB	High	Unknown	standalone_or_unknown
13	te_feedforward_stride10_long	feedforward	0.003483	0.004050	0.003053	26,241	N/A	0.32 MB	Medium	Unknown	standalone_or_unknown
14	te_feedforward_baseline	feedforward	0.003504	0.003969	0.003148	26,241	N/A	0.32 MB	Medium	Unknown	standalone_or_unknown
15	te_feedforward_high_density	feedforward	0.003519	0.004046	0.003077	26,241	N/A	0.32 MB	Medium	Unknown	standalone_or_unknown
16	te_feedforward_trial	feedforward	0.003535	0.004211	0.003618	26,241	N/A	0.32 MB	Medium	Unknown	standalone_or_unknown
17	te_feedforward_high_compute_long_remote	feedforward	0.003542	0.004228	0.003058	109,953	13m 24s	1.28 MB	High	Low	targeted_remote_followup_campaign_2026_04_04_11_21_09
18	te_feedforward_stride5_long	feedforward	0.003580	0.004008	0.003178	26,241	N/A	0.32 MB	Medium	Unknown	standalone_or_unknown
19	te_feedforward_stride1_long	feedforward	0.003646	0.003990	0.003126	26,241	N/A	0.32 MB	Medium	Unknown	standalone_or_unknown
20	te_feedforward_trial	feedforward	0.003671	0.004418	0.003706	26,241	N/A	0.32 MB	Medium	Unknown	standalone_or_unknown

harmonic_regression

• Best run: te_harmonic_order12_linear_conditioned_recovery

• Best test MAE: 0.020782

• Completed tracked runs: 3

• Known failed campaign attempts: 3

Rank	Run	Model Type	Test MAE [deg]	Test RMSE [deg]	Val MAE [deg]	Params	Duration	Artifact Size	Model Complexity	Training Heaviness	Campaign
1	te_harmonic_order12_linear_conditioned_recovery	harmonic_regression	0.020782	0.022405	0.017004	150	9m 45s	0.01 MB	Very Low	Low	wave1_structured_baseline_recovery_campaign_2026_03_20_15_40_42
2	te_harmonic_order12_static_recovery	harmonic_regression	0.039404	0.042797	0.040524	25	10m 53s	0.01 MB	Very Low	Low	wave1_structured_baseline_recovery_campaign_2026_03_20_15_40_42
3	te_harmonic_order06_static_recovery	harmonic_regression	0.039406	0.042796	0.040529	13	9m 05s	0.01 MB	Very Low	Low	wave1_structured_baseline_recovery_campaign_2026_03_20_15_40_42

Known failed campaign attempts for this family:

• te_harmonic_order06_static | campaign wave1_structured_baseline_campaign_2026_03_17_21_01_47 | model type harmonic_regression | error 'hidden_size'

• te_harmonic_order12_static | campaign wave1_structured_baseline_campaign_2026_03_17_21_01_47 | model type harmonic_regression | error 'hidden_size'

• te_harmonic_order12_linear_conditioned | campaign wave1_structured_baseline_campaign_2026_03_17_21_01_47 | model type harmonic_regression | error 'hidden_size'

periodic_mlp

• Best run: te_periodic_mlp_h04_standard

• Best test MAE: 0.003317

• Completed tracked runs: 3

• Known failed campaign attempts: 0

Rank	Run	Model Type	Test MAE [deg]	Test RMSE [deg]	Val MAE [deg]	Params	Duration	Artifact Size	Model Complexity	Training Heaviness	Campaign
1	te_periodic_mlp_h04_standard	periodic_mlp	0.003317	0.003793	0.003097	27,265	16m 22s	0.33 MB	Medium	Medium	wave1_structured_baseline_campaign_2026_03_17_21_01_47
2	te_periodic_mlp_h08_standard	periodic_mlp	0.003395	0.003951	0.003086	28,289	16m 46s	0.35 MB	Medium	Medium	wave1_structured_baseline_campaign_2026_03_17_21_01_47
3	te_periodic_mlp_h08_wide	periodic_mlp	0.003590	0.004143	0.003089	47,745	17m 22s	0.57 MB	Medium	Medium	wave1_structured_baseline_campaign_2026_03_17_21_01_47

residual_harmonic_mlp

• Best run: te_residual_h12_deep_joint_wave1

• Best test MAE: 0.003152

• Completed tracked runs: 19

• Known failed campaign attempts: 2

Rank	Run	Model Type	Test MAE [deg]	Test RMSE [deg]	Val MAE [deg]	Params	Duration	Artifact Size	Model Complexity	Training Heaviness	Campaign
1	te_residual_h12_deep_joint_wave1	residual_harmonic_mlp	0.003152	0.003640	0.003024	26,266	28m 48s	0.32 MB	Medium	Medium	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
2	te_residual_h12_small_joint_high_dropout_wave1	residual_harmonic_mlp	0.003230	0.003704	0.003001	4,890	21m 29s	0.07 MB	Low	Medium	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
3	te_residual_h16_small_joint_wave1	residual_harmonic_mlp	0.003274	0.003747	0.003020	4,898	20m 09s	0.07 MB	Low	Medium	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
4	te_residual_h12_wide_joint_low_lr_long_wave1	residual_harmonic_mlp	0.003278	0.003814	0.002924	17,946	22m 45s	0.22 MB	Medium	Medium	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
5	te_residual_h12_small_joint_medium_dense_large_batch_wave1	residual_harmonic_mlp	0.003302	0.003909	0.002935	4,890	18m 07s	0.07 MB	Low	Medium	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
6	te_residual_h12_small_joint_low_dropout_wave1	residual_harmonic_mlp	0.003359	0.003852	0.003027	4,890	21m 04s	0.07 MB	Low	Medium	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
7	te_residual_h12_small_joint_no_layer_norm_wave1	residual_harmonic_mlp	0.003360	0.003835	0.003089	4,634	12m 49s	0.07 MB	Low	Low	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
8	te_residual_h12_deep_dense_remote	residual_harmonic_mlp	0.003365	0.003868	0.003018	26,266	13m 28s	0.32 MB	Medium	Low	targeted_remote_followup_campaign_2026_04_04_11_21_09
9	te_residual_h12_small_frozen_wave1	residual_harmonic_mlp	0.003368	0.003898	0.003036	4,865	23m 21s	0.07 MB	Low	Medium	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
10	te_residual_h12_wide_joint_wave1	residual_harmonic_mlp	0.003376	0.003906	0.002884	17,946	31m 23s	0.22 MB	Medium	Medium	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
11	te_residual_h12_deep_long_remote	residual_harmonic_mlp	0.003384	0.003908	0.002973	26,266	15m 58s	0.32 MB	Medium	Medium	targeted_remote_followup_campaign_2026_04_04_11_21_09
12	te_residual_h08_small_frozen_wave1	residual_harmonic_mlp	0.003384	0.003912	0.003007	4,865	18m 38s	0.07 MB	Low	Medium	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
13	te_residual_h08_small_joint_wave1	residual_harmonic_mlp	0.003385	0.003862	0.003030	4,882	11m 22s	0.07 MB	Low	Low	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
14	te_residual_h12_medium_joint_wave1	residual_harmonic_mlp	0.003406	0.003863	0.002968	9,498	22m 14s	0.13 MB	Low	Medium	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
15	te_residual_h12_small_joint_dense_wave1	residual_harmonic_mlp	0.003410	0.003790	0.002962	4,890	26m 57s	0.07 MB	Low	Medium	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
16	te_residual_h12_small_joint_low_lr_long_wave1	residual_harmonic_mlp	0.003465	0.003944	0.002987	4,890	27m 44s	0.07 MB	Low	Medium	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00
17	te_residual_h12_small_joint_recovery	residual_harmonic_mlp	0.003466	0.003967	0.003016	4,890	16m 51s	0.07 MB	Low	Medium	wave1_structured_baseline_recovery_campaign_2026_03_20_15_40_42
18	te_residual_h12_small_frozen_recovery	residual_harmonic_mlp	0.003554	0.004061	0.003030	4,865	17m 29s	0.07 MB	Low	Medium	wave1_structured_baseline_recovery_campaign_2026_03_20_15_40_42
19	te_residual_h12_small_joint_anchor_wave1	residual_harmonic_mlp	0.003557	0.004064	0.003090	4,890	11m 20s	0.07 MB	Low	Low	wave1_residual_harmonic_family_campaign_2026_03_26_13_52_00

Known failed campaign attempts for this family:

• te_residual_h12_small_frozen | campaign wave1_structured_baseline_campaign_2026_03_17_21_01_47 | model type residual_harmonic_mlp | error 'hidden_size'

• te_residual_h12_small_joint | campaign wave1_structured_baseline_campaign_2026_03_17_21_01_47 | model type residual_harmonic_mlp | error 'hidden_size'

tree

• Best run: te_hist_gbr_tabular

• Best test MAE: 0.002885

• Completed tracked runs: 5

• Known failed campaign attempts: 2

Rank	Run	Model Type	Test MAE [deg]	Test RMSE [deg]	Val MAE [deg]	Params	Duration	Artifact Size	Model Complexity	Training Heaviness	Campaign
1	te_hist_gbr_tabular	hist_gradient_boosting	0.002885	0.003607	0.002719	5	N/A	0.62 MB	Light Artifact	Unknown	standalone_or_unknown
2	te_hist_gbr_remote_deep	hist_gradient_boosting	0.002920	0.003644	0.002749	5	1m 55s	0.91 MB	Light Artifact	Very Low	remote_training_validation_campaign_2026_04_03_17_54_21
3	te_hist_gbr_remote_refined	hist_gradient_boosting	0.003101	0.003781	0.002809	5	1m 46s	0.84 MB	Light Artifact	Very Low	targeted_remote_followup_campaign_2026_04_04_11_21_09
4	te_random_forest_tabular_recovery	random_forest	0.003833	0.004809	0.003792	5	1h 16m 53s	7.09 GB	Extreme Artifact	High	wave1_structured_baseline_recovery_campaign_2026_03_20_15_40_42
5	te_random_forest_remote_medium	random_forest	0.003865	0.004861	0.003808	5	N/A	85.40 GB	Extreme Artifact	Unknown	standalone_or_unknown

Known failed campaign attempts for this family:

• te_random_forest_remote_aggressive | campaign remote_training_validation_campaign_2026_04_03_17_54_21 | model type random_forest | error could not allocate 536870912 bytes

• te_random_forest_tabular | campaign wave1_structured_baseline_campaign_2026_03_17_21_01_47 | model type random_forest | error could not allocate 134217728 bytes

Source Of Truth

• Live backlog: doc/running/te_model_live_backlog.md

• Active campaign state: doc/running/active_training_campaign.yaml

• Program registry: output/registries/program/current_best_solution.yaml

• Family registries root: output/registries/families

• Training campaign root: output/training_campaigns

• Training run root: output/training_runs

• Paper reference report: doc/reports/analysis/RCIM Paper Reference Benchmark.md

This document is repository-generated. Regenerate it after new campaign results so the cross-family snapshot stays aligned with the canonical registries and campaign artifacts.