[Preparation 1/2] remove committed tests and retain configuration audits

.github/workflows/ci.yml

@@ -21,10 +21,10 @@ jobs:
         with:
           python-version: "3.11"
       - run: uv lock --check
-      - run: uvx ruff@${RUFF_VERSION} check egomimic tests scripts tools
-      - run: uvx ruff@${RUFF_VERSION} format --check egomimic tests scripts tools
+      - run: uvx ruff@${RUFF_VERSION} check egomimic scripts tools
+      - run: uvx ruff@${RUFF_VERSION} format --check egomimic scripts tools
 
-  cpu-and-wheel:
+  configuration-audit:
     runs-on: ubuntu-latest
     timeout-minutes: 40
     steps:
@@ -45,13 +45,6 @@ jobs:
           source .venv/bin/activate
           python scripts/audit_hydra_configs.py --output config-audit.json
           HF_HUB_OFFLINE=1 python -m scripts.audit_components --output component-audit.json
-      - name: CPU functional and isolated installed-wheel gates
-        env:
-          HF_HUB_OFFLINE: "1"
-          WANDB_MODE: disabled
-        run: |
-          source .venv/bin/activate
-          pytest tests -q
       - uses: actions/upload-artifact@v4
         if: always()
         with:

AGENTS.md

@@ -1,5 +1,8 @@
 # Repo Agent Rules
 
+See [the verification policy](docs/VERIFICATION_POLICY.md): keep tests and
+verification fixtures outside this repository; never commit them.
+
 ## Shell / Command Execution
 To run commands in the interactive shell, source `emimic/bin/activate` when it
 exists. In worktrees where `emimic` is absent and the checked project
@@ -72,12 +75,12 @@ not imply a separate runtime for that task or model.
 | Dataset filters, task prompts, camera/frame choices and experiment parameters | `egomimic/hydra_configs/`; calibration matrices in `hydra_configs/calibration/` |
 | Shared Eva/Yam collection, local graph rollout, Zarr replay and upload | [robot/AGENTS.md](egomimic/robot/AGENTS.md), [docs/YAM_RUNTIME.md](docs/YAM_RUNTIME.md) |
 | Offline normalization export for any model | [scripts/data/precompute_norm_stats.py](scripts/data/precompute_norm_stats.py), or `trainHydra.py norm_stats_only=true` |
-| Focused CPU regression checks | `tests/test_pipeline*.py`, `test_hpt*.py`, `test_arc*.py`, `test_e1*.py`, `test_pi05*.py`, `test_wrist6d_roundtrip.py`, `test_robot_runtime.py`, `test_robot_graph_policy.py` |
+| Offline configuration audits | `scripts/audit_hydra_configs.py`, `scripts/audit_components.py` |
 
 Useful searches from the repository root:
 
 ```bash
-rg --files egomimic/pipeline egomimic/rldb egomimic/eval egomimic/robot tests
+rg --files egomimic/pipeline egomimic/rldb egomimic/eval egomimic/robot
 rg -n '_target_:|action_mode:|coord_frame:|rotation_mode:' egomimic/hydra_configs
 rg -n 'class BimanualCartesianEval|class PI05Stage|class MultiDataset' egomimic
 ```
@@ -100,9 +103,11 @@ rg -n 'class BimanualCartesianEval|class PI05Stage|class MultiDataset' egomimic
 - `egomimic/campaigns/pi05/`, `e1_fold.py`, and `pi05_graph_eval.py` retain old
   import names. Their implementations now live at the canonical paths above.
   Use current YAML examples when migrating older serialized configuration APIs.
-- Read the relevant tests before changing normalization, pose conventions or
-  checkpoint loading. Verify behavior with small CPU fixtures before a cluster
-  run; hardware and pretrained-model validation are separate from those tests.
+- Do not commit tests, test fixtures, or local verification scripts. Keep any
+  regression checks outside the repository and report their results in the PR.
+  Verify normalization, pose conventions and checkpoint loading with local CPU
+  checks before a cluster run; hardware and pretrained-model validation remain
+  separate. Historical test sources remain available in Git history.
 
 ## Preserve work and verify the target
 

docs/VERIFICATION_POLICY.md

@@ -0,0 +1,27 @@
+# Verification without committed tests
+
+The rebuilt stack removes the tracked regression suites, their fixtures and
+test-only runners, including the vendored LeRobot suite. Do not commit new
+tests, fixtures, or local verification scripts. Run focused regression checks
+outside the application repository and include their results and limitations
+in each pull request.
+
+CI retains locked-environment checks, Ruff, recursive Hydra composition and
+offline component construction. It no longer runs pytest or the former
+installed-wheel regression gate. Passing CI therefore does not establish
+behavioral parity, wheel-install compatibility, hardware safety, or successful
+training. Those require separate evidence.
+
+The Qwen architecture metadata used by the constructor audit is retained at
+`scripts/resources/qwen3_embedding_06b_config.json` with the same checksum.
+This is audit input, not a test suite. Existing dependency pins, runtime guards,
+training/validation evaluators, and hardware-control examples are unchanged.
+The vendored ARX examples named `test_*` are operational robot programs, not
+automated regression tests.
+
+Historical test references and validation receipts in older documents describe
+the source revision that produced them. Their sources can be recovered from Git
+history (for example the pre-rebuild main commit
+`a525d73897bec64397048e7c8897fb886555567b`); they are no longer commands for the
+current checkout. Do not rewrite historical provenance or weaken checkpoint
+compatibility guards to accommodate a source cleanup.

docs/YAM_RUNTIME.md

@@ -438,6 +438,9 @@ directory. `--list` performs no upload and needs no AWS credentials.
 
 ## Validation scope
 
+The regression sources referenced below are historical and no longer shipped;
+see [the verification policy](VERIFICATION_POLICY.md). Keep new checks local.
+
 `tests/test_robot_runtime.py` checks collection, format compatibility, world-frame
 clutch control, stale tracking/cameras, previews, Eva/Yam interfaces, cleanup and
 Zarr replay. `tests/test_robot_graph_policy.py` checks strict checkpoint loading,

docs/integration/DATA_TOOLS.md

@@ -81,6 +81,9 @@ require SQL, cloud packages or a policy backend.
 
 ## Evidence and remaining gates
 
+The following regression evidence is historical; test sources are no longer
+shipped. See [the verification policy](../VERIFICATION_POLICY.md).
+
 `tests/test_data_tool_parity.py` converts an actual HDF5 fixture, checks unchanged
 source bytes and overwrite rejection, imports with cloud/model modules blocked,
 encodes a complete annotated episode, validates video frame count/FPS and saved

egomimic/eval/AGENTS.md

@@ -29,6 +29,5 @@ are YAML settings, including the `train_viz` group.
 
 Frame reversion must match the data configuration. Rotation encodings and
 gripper padding affect both actions and proprio; test both sides of the round
-trip. Focused regression tests are `test_shared_robot_components.py`,
-`test_eval_overlay_annotations.py`, `test_wrist6d_roundtrip.py`,
-`test_pi05_graph.py`, and the root ARC/tempo evaluation tests.
+trip. Keep regression checks outside the repository and report local results;
+do not commit tests or their fixtures.

egomimic/rldb/AGENTS.md

@@ -33,6 +33,5 @@ embodiment IDs when loading a cache. Live rollouts require the full
 `normalizer_state`, not just numeric bounds.
 
 Do not change stored episode metadata to rename an embodiment. Canonicalize the
-metadata at the loader boundary. Test geometric changes with
-`test_action_chunk_transforms.py`, `test_wrist6d_roundtrip.py`, `test_human_arc.py`,
-`test_abc_yam.py`, and the relevant ARC/PI numeric tests in the root `tests/`.
+metadata at the loader boundary. Verify geometric and ARC/PI changes with local
+CPU checks outside the repository; do not commit tests or their fixtures.

egomimic/robot/AGENTS.md

@@ -70,8 +70,7 @@ Use the same graph, data normalization and evaluation components for HPT and PI.
 Do not introduce an embodiment-specific inference server or upstream runtime
 launcher. Do not deprecate Eva when adding another robot implementation.
 
-Run `tests/test_gello_teleop.py`, `tests/test_robot_runtime.py`, and
-`tests/test_robot_graph_policy.py` for CPU validation. `--help`, GELLO
+Keep CPU verification outside the repository; do not commit tests. `--help`, GELLO
 `--check-config`/`--list-ports`, and uploader `--list` open no devices.
 Collection/rollout commands open hardware; operate a physical station only when
 the user requests that operation. Preserve source branches, user worktrees and