evidence_role	validation_boundary	model_family	run_label	perturbation_rep	model_label	split	model_rmse	model_cosine	zero_delta_rmse	train_mean_delta_rmse	beats_zero_delta_rmse	beats_train_mean_delta_rmse	gate_status	source_file	notes
formal_heldout_gene_validation	fewshot held-out genes	STATE	hepg2_gpu_smoke_500step_20260710_093124	onehot	STATE onehot 500-step smoke	val	0.1251339469105004	-0.0396329359733499	0.0930628716945648	0.0951926868408918	False	False	fail	results/run_comparisons/hepg2_gpu_smoke_500step_20260710_093124_validation_gate.tsv	Negative control for held-out identity generalization.
formal_heldout_gene_validation	fewshot held-out genes	STATE	hepg2_gpu_smoke_500step_20260710_093124	onehot	STATE onehot 500-step smoke	test	0.126819857884021	-0.0270984723742696	0.0937586594372987	0.0940848111751533	False	False	fail	results/run_comparisons/hepg2_gpu_smoke_500step_20260710_093124_validation_gate.tsv	Negative control for held-out identity generalization.
candidate_screen_auxiliary	candidate leave-in screen	STATE	hepg2_gpu_smoke_500step_20260710_093124	onehot	STATE onehot 500-step candidate screen	screen	0.1072707619695435	-0.012621616872604	0.073507354018234	0.0866221326092879	False	False	fail	results/run_comparisons/hepg2_gpu_smoke_500step_20260710_093124_candidate_gate.tsv	Candidate screen failed no-effect and train-mean baselines.
formal_heldout_gene_validation	fewshot held-out genes	STATE	hepg2_gpu_feature_probe_20step_20260710_1055	nonleaky_features_v1_63d	STATE feature v1 20-step	val	0.100385932996869	0.0054023461416363	0.0930628716945648	0.0951926868408918	False	False	fail	results/run_comparisons/hepg2_gpu_feature_probe_20step_20260710_1055_validation_gate.tsv	STATE objective/inference path did not preserve the non-leaky feature signal.
formal_heldout_gene_validation	fewshot held-out genes	STATE	hepg2_gpu_feature_probe_20step_20260710_1055	nonleaky_features_v1_63d	STATE feature v1 20-step	test	0.103635204867238	-5.180714651942385e-05	0.0937586594372987	0.0940848111751533	False	False	fail	results/run_comparisons/hepg2_gpu_feature_probe_20step_20260710_1055_validation_gate.tsv	STATE objective/inference path did not preserve the non-leaky feature signal.
candidate_screen_auxiliary	candidate leave-in screen	STATE	hepg2_gpu_feature_probe_20step_20260710_1055	nonleaky_features_v1_63d	STATE feature v1 20-step candidate screen	screen	0.0824693939870311	-0.0006768124155877	0.073507354018234	0.0866221326092879	False	True	partial_train_mean_only	results/run_comparisons/hepg2_gpu_feature_probe_20step_20260710_1055_candidate_gate.tsv	Auxiliary candidate screen only; not a validation gate.
formal_heldout_gene_validation	fewshot held-out genes	STATE	hepg2_gpu_feature_smoke_500step_20260710_1143	nonleaky_features_v1_63d	STATE feature v1 500-step	val	0.122860381938517	-0.0308879507745586	0.0930628716945648	0.0951926868408918	False	False	fail	results/run_comparisons/hepg2_gpu_feature_smoke_500step_20260710_1143_validation_gate.tsv	STATE objective/inference path did not preserve the non-leaky feature signal.
formal_heldout_gene_validation	fewshot held-out genes	STATE	hepg2_gpu_feature_smoke_500step_20260710_1143	nonleaky_features_v1_63d	STATE feature v1 500-step	test	0.1256858600037438	-0.0375888618374509	0.0937586594372987	0.0940848111751533	False	False	fail	results/run_comparisons/hepg2_gpu_feature_smoke_500step_20260710_1143_validation_gate.tsv	STATE objective/inference path did not preserve the non-leaky feature signal.
candidate_screen_auxiliary	candidate leave-in screen	STATE	hepg2_gpu_feature_smoke_500step_20260710_1143	nonleaky_features_v1_63d	STATE feature v1 500-step candidate screen	screen	0.1068246708739371	-0.0155812562443315	0.073507354018234	0.0866221326092879	False	False	fail	results/run_comparisons/hepg2_gpu_feature_smoke_500step_20260710_1143_candidate_gate.tsv	Auxiliary candidate screen only; not a validation gate.
formal_heldout_gene_validation	fewshot held-out genes	STATE	hepg2_gpu_feature_smoke_v2_20step_20260710_1321	nonleaky_features_v2_280d	STATE feature v2 20-step	val	0.1010388681665062	0.0089529464021325	0.0930628716945648	0.0951926868408918	False	False	fail	results/run_comparisons/hepg2_gpu_feature_smoke_v2_20step_20260710_1321_validation_gate.tsv	STATE objective/inference path did not preserve the non-leaky feature signal.
formal_heldout_gene_validation	fewshot held-out genes	STATE	hepg2_gpu_feature_smoke_v2_20step_20260710_1321	nonleaky_features_v2_280d	STATE feature v2 20-step	test	0.1042032522105034	0.0064723079516884	0.0937586594372987	0.0940848111751533	False	False	fail	results/run_comparisons/hepg2_gpu_feature_smoke_v2_20step_20260710_1321_validation_gate.tsv	STATE objective/inference path did not preserve the non-leaky feature signal.
candidate_screen_auxiliary	candidate leave-in screen	STATE	hepg2_gpu_feature_smoke_v2_20step_20260710_1321	nonleaky_features_v2_280d	STATE feature v2 20-step candidate screen	screen	0.0843100272828624	0.0005798718027238	0.073507354018234	0.0866221326092879	False	True	partial_train_mean_only	results/run_comparisons/hepg2_gpu_feature_smoke_v2_20step_20260710_1321_candidate_gate.tsv	Auxiliary candidate screen only; not a validation gate.
seen_perturbation_diagnostic	seen perturbations, held-out cells	STATE	hepg2_seen_hepg2_gpu_seen_20step_20260710_1410	seen_identity	STATE seen perturbations 20-step	val	0.1697116585758825	0.0323945502132119	0.1567456731572747	0.1503798151388764	False	False	fail	results/state_nadig_trade_seen_split/infer/hepg2_seen_hepg2_gpu_seen_20step_20260710_1410_val_test_gate.tsv	Diagnostic split: perturbation identities are seen in training.
seen_perturbation_diagnostic	seen perturbations, held-out cells	STATE	hepg2_seen_hepg2_gpu_seen_20step_20260710_1410	seen_identity	STATE seen perturbations 20-step	test	0.1698357785741488	-0.0316663878459924	0.1526315800535182	0.1524882708520939	False	False	fail	results/state_nadig_trade_seen_split/infer/hepg2_seen_hepg2_gpu_seen_20step_20260710_1410_val_test_gate.tsv	Diagnostic split: perturbation identities are seen in training.
seen_perturbation_diagnostic	seen perturbations, held-out cells	STATE	hepg2_seen_hepg2_gpu_seen_20step_20260710_1410	seen_identity	STATE seen perturbations 20-step	all	0.1697737185750156	0.0003640811836097	0.1546886266053964	0.1514340429954851	False	False	fail	results/state_nadig_trade_seen_split/infer/hepg2_seen_hepg2_gpu_seen_20step_20260710_1410_val_test_gate.tsv	Diagnostic split: perturbation identities are seen in training.
seen_perturbation_diagnostic	seen perturbations, held-out cells	STATE	hepg2_seen_hepg2_gpu_seen_smoke_500step_20260710_1420	seen_identity	STATE seen perturbations energy 500-step	val	0.1793202604167163	0.0163224946356119	0.1567456731572747	0.1503798151388764	False	False	fail	results/state_nadig_trade_seen_split/infer/hepg2_seen_hepg2_gpu_seen_smoke_500step_20260710_1420_val_test_gate.tsv	Diagnostic split: perturbation identities are seen in training.
seen_perturbation_diagnostic	seen perturbations, held-out cells	STATE	hepg2_seen_hepg2_gpu_seen_smoke_500step_20260710_1420	seen_identity	STATE seen perturbations energy 500-step	test	0.1752737783050785	0.0217226349617097	0.1526315800535182	0.1524882708520939	False	False	fail	results/state_nadig_trade_seen_split/infer/hepg2_seen_hepg2_gpu_seen_smoke_500step_20260710_1420_val_test_gate.tsv	Diagnostic split: perturbation identities are seen in training.
seen_perturbation_diagnostic	seen perturbations, held-out cells	STATE	hepg2_seen_hepg2_gpu_seen_smoke_500step_20260710_1420	seen_identity	STATE seen perturbations energy 500-step	all	0.1772970193608974	0.0190225647986608	0.1546886266053964	0.1514340429954851	False	False	fail	results/state_nadig_trade_seen_split/infer/hepg2_seen_hepg2_gpu_seen_smoke_500step_20260710_1420_val_test_gate.tsv	Diagnostic split: perturbation identities are seen in training.
seen_perturbation_diagnostic	seen perturbations, held-out cells	STATE	hepg2_seen_hepg2_gpu_seen_mse_200step_20260710_1450	seen_identity	STATE seen perturbations MSE 200-step	val	0.1571176329938074	0.0353635653063975	0.1567456731572747	0.1503798151388764	False	False	fail	results/state_nadig_trade_seen_split/infer/hepg2_seen_hepg2_gpu_seen_mse_200step_20260710_1450_val_test_gate.tsv	Diagnostic split: perturbation identities are seen in training.
seen_perturbation_diagnostic	seen perturbations, held-out cells	STATE	hepg2_seen_hepg2_gpu_seen_mse_200step_20260710_1450	seen_identity	STATE seen perturbations MSE 200-step	test	0.1537543494875232	-0.0066207057498104	0.1526315800535182	0.1524882708520939	False	False	fail	results/state_nadig_trade_seen_split/infer/hepg2_seen_hepg2_gpu_seen_mse_200step_20260710_1450_val_test_gate.tsv	Diagnostic split: perturbation identities are seen in training.
seen_perturbation_diagnostic	seen perturbations, held-out cells	STATE	hepg2_seen_hepg2_gpu_seen_mse_200step_20260710_1450	seen_identity	STATE seen perturbations MSE 200-step	all	0.1554359912406653	0.0143714297782935	0.1546886266053964	0.1514340429954851	False	False	fail	results/state_nadig_trade_seen_split/infer/hepg2_seen_hepg2_gpu_seen_mse_200step_20260710_1450_val_test_gate.tsv	Diagnostic split: perturbation identities are seen in training.
positive_control_seen_split	seen perturbations, held-out cells	measured_delta_baseline	hepg2_seen_pseudobulk_delta	observed_train_delta	Seen pseudobulk measured delta	val	0.1367228428522745	0.3971671115335387	0.1567456731572747	0.1503798151388764	True	True	pass	results/state_nadig_trade_seen_split/pseudobulk_delta/hepg2_seen_pseudobulk_delta_gate.tsv	Positive control showing the seen split has recoverable signal.
positive_control_seen_split	seen perturbations, held-out cells	measured_delta_baseline	hepg2_seen_pseudobulk_delta	observed_train_delta	Seen pseudobulk measured delta	test	0.1380649305259188	0.3622678719111718	0.1526315800535182	0.1524882708520939	True	True	pass	results/state_nadig_trade_seen_split/pseudobulk_delta/hepg2_seen_pseudobulk_delta_gate.tsv	Positive control showing the seen split has recoverable signal.
positive_control_seen_split	seen perturbations, held-out cells	measured_delta_baseline	hepg2_seen_pseudobulk_delta	observed_train_delta	Seen pseudobulk measured delta	all	0.1373938866890967	0.3797174917223553	0.1546886266053964	0.1514340429954851	True	True	pass	results/state_nadig_trade_seen_split/pseudobulk_delta/hepg2_seen_pseudobulk_delta_gate.tsv	Positive control showing the seen split has recoverable signal.
positive_control_seen_split	seen perturbations, held-out cells	direct_delta_head	hepg2_seen_delta_target_embedding	learned_seen_delta_embedding	Seen delta-target embedding	val	0.1367228429298848	0.3971671150259984	0.1567456731572747	0.1503798151388764	True	True	pass	results/state_nadig_trade_seen_split/delta_target/hepg2_seen_delta_target_embedding_gate.tsv	Positive control; not held-out perturbation identity generalization.
positive_control_seen_split	seen perturbations, held-out cells	direct_delta_head	hepg2_seen_delta_target_embedding	learned_seen_delta_embedding	Seen delta-target embedding	test	0.1380649304483086	0.3622678721385455	0.1526315800535182	0.1524882708520939	True	True	pass	results/state_nadig_trade_seen_split/delta_target/hepg2_seen_delta_target_embedding_gate.tsv	Positive control; not held-out perturbation identity generalization.
positive_control_seen_split	seen perturbations, held-out cells	direct_delta_head	hepg2_seen_delta_target_embedding	learned_seen_delta_embedding	Seen delta-target embedding	all	0.1373938866890967	0.3797174935822719	0.1546886266053964	0.1514340429954851	True	True	pass	results/state_nadig_trade_seen_split/delta_target/hepg2_seen_delta_target_embedding_gate.tsv	Positive control; not held-out perturbation identity generalization.
formal_heldout_gene_validation	fewshot held-out genes	direct_delta_head	hepg2_fewshot_feature_delta_ridge_v1	nonleaky_features_v1	Feature-to-delta ridge v1	val	0.087503657117486	0.2978880886454135	0.0930628716945648	0.0951926868408918	True	True	pass	results/state_nadig_trade_fewshot/delta_feature/hepg2_fewshot_feature_delta_ridge_v1_gate.tsv	Direct delta supervision with train-only model selection.
formal_heldout_gene_validation	fewshot held-out genes	direct_delta_head	hepg2_fewshot_feature_delta_ridge_v1	nonleaky_features_v1	Feature-to-delta ridge v1	test	0.0877140478364058	0.2813188739597708	0.0937586594372987	0.0940848111751533	True	True	pass	results/state_nadig_trade_fewshot/delta_feature/hepg2_fewshot_feature_delta_ridge_v1_gate.tsv	Direct delta supervision with train-only model selection.
formal_heldout_gene_validation	fewshot held-out genes	direct_delta_head	hepg2_fewshot_feature_delta_ridge_v1	nonleaky_features_v1	Feature-to-delta ridge v1	all	0.0876114182174205	0.2894014177088648	0.0934192507823066	0.0946252383291721	True	True	pass	results/state_nadig_trade_fewshot/delta_feature/hepg2_fewshot_feature_delta_ridge_v1_gate.tsv	Direct delta supervision with train-only model selection.
formal_heldout_gene_validation	fewshot held-out genes	direct_delta_head	hepg2_fewshot_feature_delta_mlp_v1	nonleaky_features_v1	Feature-to-delta MLP v1	val	0.0864128831773996	0.2831788073992357	0.0930628716945648	0.0951926868408918	True	True	pass	results/state_nadig_trade_fewshot/delta_feature/hepg2_fewshot_feature_delta_mlp_v1_gate.tsv	STATE-adjacent neural delta head; direct delta supervision.
formal_heldout_gene_validation	fewshot held-out genes	direct_delta_head	hepg2_fewshot_feature_delta_mlp_v1	nonleaky_features_v1	Feature-to-delta MLP v1	test	0.0910465029023942	0.2854269224972952	0.0937586594372987	0.0940848111751533	True	True	pass	results/state_nadig_trade_fewshot/delta_feature/hepg2_fewshot_feature_delta_mlp_v1_gate.tsv	STATE-adjacent neural delta head; direct delta supervision.
formal_heldout_gene_validation	fewshot held-out genes	direct_delta_head	hepg2_fewshot_feature_delta_mlp_v1	nonleaky_features_v1	Feature-to-delta MLP v1	all	0.0887862005975188	0.2843302809860466	0.0934192507823066	0.0946252383291721	True	True	pass	results/state_nadig_trade_fewshot/delta_feature/hepg2_fewshot_feature_delta_mlp_v1_gate.tsv	STATE-adjacent neural delta head; direct delta supervision.
formal_heldout_gene_validation	fewshot held-out genes	direct_delta_head	hepg2_fewshot_feature_delta_ridge_v2	nonleaky_features_v2	Feature-to-delta ridge v2	val	0.0905455579981207	0.2973457721062004	0.0930628716945648	0.0951926868408918	True	True	pass	results/state_nadig_trade_fewshot/delta_feature/hepg2_fewshot_feature_delta_ridge_v2_gate.tsv	Direct delta supervision with train-only model selection.
formal_heldout_gene_validation	fewshot held-out genes	direct_delta_head	hepg2_fewshot_feature_delta_ridge_v2	nonleaky_features_v2	Feature-to-delta ridge v2	test	0.0869511488292898	0.2902384063317662	0.0937586594372987	0.0940848111751533	True	True	pass	results/state_nadig_trade_fewshot/delta_feature/hepg2_fewshot_feature_delta_ridge_v2_gate.tsv	Direct delta supervision with train-only model selection.
formal_heldout_gene_validation	fewshot held-out genes	direct_delta_head	hepg2_fewshot_feature_delta_ridge_v2	nonleaky_features_v2	Feature-to-delta ridge v2	all	0.0887045191555488	0.2937054140266122	0.0934192507823066	0.0946252383291721	True	True	pass	results/state_nadig_trade_fewshot/delta_feature/hepg2_fewshot_feature_delta_ridge_v2_gate.tsv	Direct delta supervision with train-only model selection.
formal_heldout_gene_validation	fewshot held-out genes	direct_delta_head	hepg2_fewshot_feature_delta_mlp_v2	nonleaky_features_v2	Feature-to-delta MLP v2	val	0.0894924679771065	0.282441780064255	0.0930628716945648	0.0951926868408918	True	True	pass	results/state_nadig_trade_fewshot/delta_feature/hepg2_fewshot_feature_delta_mlp_v2_gate.tsv	STATE-adjacent neural delta head; direct delta supervision.
formal_heldout_gene_validation	fewshot held-out genes	direct_delta_head	hepg2_fewshot_feature_delta_mlp_v2	nonleaky_features_v2	Feature-to-delta MLP v2	test	0.0911172983192262	0.2870045838256677	0.0937586594372987	0.0940848111751533	True	True	pass	results/state_nadig_trade_fewshot/delta_feature/hepg2_fewshot_feature_delta_mlp_v2_gate.tsv	STATE-adjacent neural delta head; direct delta supervision.
formal_heldout_gene_validation	fewshot held-out genes	direct_delta_head	hepg2_fewshot_feature_delta_mlp_v2	nonleaky_features_v2	Feature-to-delta MLP v2	all	0.0903246981523385	0.2847788258932712	0.0934192507823066	0.0946252383291721	True	True	pass	results/state_nadig_trade_fewshot/delta_feature/hepg2_fewshot_feature_delta_mlp_v2_gate.tsv	STATE-adjacent neural delta head; direct delta supervision.
candidate_screen_auxiliary	candidate leave-candidate-out	direct_delta_head	hepg2_candidate_screen_feature_delta_ridge_v1_lco	nonleaky_features_v1	Candidate LCO feature-to-delta ridge v1	candidate_screen_leave_candidate_out	0.0769212198044572	0.1217221799306571	0.073507354018234	0.0867055945453189	False	True	partial_train_mean_only	results/state_nadig_trade_actual/candidate_screen/hepg2_candidate_screen_feature_delta_ridge_v1_lco_gate.tsv	Auxiliary hypothesis triage only; gate failed no-effect baseline.
candidate_screen_auxiliary	candidate leave-candidate-out	direct_delta_head	hepg2_candidate_screen_feature_delta_ridge_v2_lco	nonleaky_features_v2	Candidate LCO feature-to-delta ridge v2	candidate_screen_leave_candidate_out	0.0757097490131855	0.1248591411193566	0.073507354018234	0.0867055945453189	False	True	partial_train_mean_only	results/state_nadig_trade_actual/candidate_screen/hepg2_candidate_screen_feature_delta_ridge_v2_lco_gate.tsv	Auxiliary hypothesis triage only; gate failed no-effect baseline.
candidate_screen_auxiliary	candidate leave-candidate-out	direct_delta_head	hepg2_candidate_screen_feature_delta_mlp_v1_lco	nonleaky_features_v1	Candidate LCO feature-to-delta mlp v1	candidate_screen_leave_candidate_out	0.0801358852712881	0.1021740809082984	0.073507354018234	0.0867055945453189	False	True	partial_train_mean_only	results/state_nadig_trade_actual/candidate_screen/hepg2_candidate_screen_feature_delta_mlp_v1_lco_gate.tsv	Auxiliary hypothesis triage only; gate failed no-effect baseline.
candidate_screen_auxiliary	candidate leave-candidate-out	direct_delta_head	hepg2_candidate_screen_feature_delta_mlp_v2_lco	nonleaky_features_v2	Candidate LCO feature-to-delta mlp v2	candidate_screen_leave_candidate_out	0.0789002850651741	0.1050844832456537	0.073507354018234	0.0867055945453189	False	True	partial_train_mean_only	results/state_nadig_trade_actual/candidate_screen/hepg2_candidate_screen_feature_delta_mlp_v2_lco_gate.tsv	Auxiliary hypothesis triage only; gate failed no-effect baseline.