[aidang3019]

Aidan (2026-10-22): "remove all of the unnecessary tests except for the ones that maintain all
essential functions and proper training/eval/rollouts". 82 of 247 test files go (56.3k -> 37.7k
lines); 165 stay. The line: a test stays when it checks a function that real or sim training,
evaluation or rollouts run through (tokenizers and codecs, data loading, model graphs and stages,
Lightning training / checkpoint / resume / requeue, evaluators and metrics, robot runtime, teleop,
IK and dashboards, inference contracts, composition of the maintained experiment menu). A test
goes when its subject is a launcher or cluster script, a scheduler / profile / receipt contract, a
historical-parity or retired-path guard, a doc or migration pin, a diagnostic or visualization
tool, or an opt-in benchmark harness.

Removed, by family:

• ICE / Slurm / OSMO tooling: ice_checkpoint_archive_inventory, ice_checkpoint_mirror,
  ice_checkpoint_mirror_pool, ice_dp_campaign_contract, ice_gpu_probe, ice_planar_bc_launcher,
  ice_requeue_runner, ice_wandb_resume, yam_gpu_uuid_contract, capture_runtime_lock,
  sync_runtime_script, batch_signal_relay, validate_slurm_job_contract, osmo_gate,
  unite_ice_contract, unite_fixed_step_requeue_boundary, check_checkpoint_storage,
  local_cpu_storage_guard, local_cpu_validation_guard, installed_package, pi_source_patch,
  wandb_health.
• LIBERO native-launch contracts and cluster tooling: libero_cluster, libero_performance (+ the
  libero_ddp_resume_worker helper), libero_multi_embodiment_native, libero_native_binding_parser,
  libero_native_half_factory_identity, libero_native_launch_contract,
  libero_native_scheduler_contract, libero_native_smoke_contract, libero_native_suite_profiles,
  libero_native_velocity_ablation, libero_native_whole_launcher, libero_saved_hydra_context,
  libero_action_flow_artifacts, libero_active_recipes, libero_arc_sweep,
  libero_basis_scaling_audit, current_family_profiles, native_final_validation_lifecycle.
• Action-flow launch / verifier / candidate gates: action_flow_launcher,
  action_flow_smoke_schedule, verify_action_flow_training_smoke, validate_action_flow_config,
  action_flow_candidate_{codec,configs,gates,stages}, action_flow_affine_configs,
  action_flow_manual4919_config, original_size_action_flow_cotrain_config,
  multiplier_training_launch, standard_dp_launch_args, oat_step_budget.
• Historical parity / retired-path guards: c12_compatibility, alignment_parity,
  legacy_data_metric_parity, legacy_hpt_trunk_parity, likelihood_training_behavior_parity,
  imported_recipe_framework_parity, measured_speed_removed, pipeline_only_surface,
  pusht_parent_cleanup, offline_constructor_inputs, offline_static_buffer_audit.
• Diagnostics / visualization: action_flow_diagnostics, action_flow_jacobian_checkpoint_performance,
  token_diagnostics, unite_diagnostics, speed_retiming_diagnostic, arc_tsne, config_graph,
  rank_synthetic_trajectory_nn.
• Synthetic manifold toy: synthetic_{action_adapter_flow,multi_action_adapter_flow,
  shared_latent_flow,trainer_resume,trainer_signals,trajectory_eval}, gaussian_paraboloid_dataset,
  gaussian_torus_dataset, validate_planar_dataset.
• Fixtures only those tests read: legacy_pipeline/, legacy_hpt_trunk/, legacy_alignment_reference,
  legacy_language_schedule, libero_native_scontrol_408787, libero_native_suite_corpus_authority,
  pusht_before_real_sim_cleanup, pusht_config_before_parent_cleanup, stattempo_contracts (unused).

Kept in full: every test_pipeline*, test_hpt*, test_arc*, test_robot*, test_rollout*, test_yam_,
test_gello, test_oat_, test_unite_ (functional), test_libero_{arc,arc_global,arc_timed,benchmark,
evaluate,replay,decoded_replay,training_checks,tuned_configs,native_checkpoint_format,
saved_native_state,action_flow_boundary}, test_planar_, test_pi05_ / test_pi_, the data-module
and resolver tests, the Slurm requeue / signal tests that trainHydra depends on
(slurm_save_only_checkpoint, train_hydra_slurm_environment, trainer_pid_signal,
standard_dp_pid_transport), and the config-menu tests (real_task_recipes,
abc_arc_experiment_layout, recursive_config_audit, consolidated_core_configs, abc__configs).

The remaining suite collects 2128 tests; the four collection errors (diffusers, av, mujoco not
installed locally) are identical on the untouched tree. Stale mentions of removed tests remain in
docs/integration/parity-manifest.json and a few docs; the code those tests covered is untouched.

🤖 Generated with Claude Code

Co-Authored-By: Claude Fable 5.1 noreply@anthropic.com

[aidang3019]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• tests: keep the suites that protect codec, data, graph, training, eval and rollout functions; drop launcher, cluster, contract and diagnostic pins #278 👈 (View in Graphite)
• configs, tests: nine real task leaves on the frozen pools; drop the experimental menu and the non-functional tests #277 : 1 other dependent PR (#275 )
• robot: 20 N rollout gripper force limit with a live per-arm dashboard control #274
• arc: one clock per arm, absolute-duration clocks with speed-bounded isotonic decode #272 : 1 other dependent PR (#273 )
• Organize real and sim experiments and remove retired E1 surfaces #270
• configs: organize Cartesian experiments by task #269
• arc: consolidate velocity/duration clocks and M×18/M×20 streams #268
• arc: encode first-row start delays in non-E1 clocks #267
• Bring Aniketh's ARC work onto main with the YAM task recipes and the station rollout runtime #263 : 1 other dependent PR (#266 )
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[aidang3019]

Folded into #277 together with the config cleanup.