infra: mecka 6D / keypoint recipes, 300M models, defaults, and the rollout revert by RyanPCo · Pull Request #640 · GaTech-RL2/EgoVerse

RyanPCo · 2026-09-14T19:23:10Z

This PR is the tip of an eight-branch split. The 111-file commit that used to live
here was broken up so each branch carries one feature. #640 keeps its original head branch
(ryanco/infra-port) and its review history, but its base is ryanco/eval-metrics-train-viz
and its diff is only the last slice. Review comments above this line may point at code that
has moved to one of:

PR	Branch	Carries
#653 (merged)	`rot6d-core`	6D rotation primitives, bimanual layout tables, shared norm helpers (no behavior change)
#654 (merged)	`embodiment-6d-modes`	PI goes 6D-only; Eva/Human `*_6d` modes; vendor embodiment aliasing
#655 (merged)	`mecka-left-wrist-fix`	the double-mirrored LEFT wrist frame, converter + `RotateLocalFrame` for existing zarrs
#656	`hand-keypoints-144d`	the 144-D wrist-first hand-keypoint action as the human default
#657	`norm-stats-pooling`	`NORM_MIN_RANGE` degenerate-channel guard, `pool_horizon`, the precompute script
#658	`dataset-robustness`	bounds-check channels, sampler widening, `require_annotations`
#659	`eval-metrics-train-viz`	`action_metrics`, train-split viz head, viz cadence, reverse-KL removal

The stack was rebased onto current main. The single-operator flagship config
(mecka_fold_flagship_topop_hpt_6d) is kept out of this PR.

What is left in this PR

Recipes, models and defaults

data configs: mecka_all_6d, mecka_fold_freeform{,_opsplit}_6d (wrist frame; cam frame is a
transform_list.mode=cartesian_6d override, and a guard test rejects data configs that only
swap the mode), mecka_fold_flagship_opsplit_hpt_6d, mecka_fold_freeform_opsplit_hpt_keypoints,
and the composed recipes train_zarr_mecka_flagship_6d_hpt / train_zarr_mecka_kp_wrist_hpt.
models: the 300M HPT pair hpt_bc_mecka_6d_300M / hpt_bc_keypoints_wrist_300M,
pi0.5_bc_mecka_6d, and constant-LR-after-warmup on pi0.5_base. The 6D PI converters are
renamed from *CartesianEuler to *Cartesian6D to match what they pack.
defaults: ddp_pi inference_mode=false + validation every 20 epochs, checkpoints every 50
on train-epoch end; the HPT Qwen stem loads the local snapshot when offline.
scripts: viz_mecka_episode.py, viz_eva_episode.py, with egomimic/utils/paths.py reading
the dataset dir from the hydra paths config.

Rollout revert

trainHydra saves each train dataset's transform_list.mode with the checkpoint config.
rollout.py reads the Eva mode from there and maps predictions back to camera-frame ypr via
Eva.get_revert_transform_list. Checkpoints without a recorded mode keep the old inputs and no
revert. Not tested on hardware.

Testing

Unit suite and per-branch counts below are from before the rebase onto current main:
636 passed, 2 skipped, 0 failed at this branch; every branch of the split green on its own
suite, in order 216 / 203 / 460 / 467 / 487 / 496 / 634 / 636.
Single-episode overfits for HPT and PI, on eva and mecka, all exit 0 — run on the pre-split
commit, so they cover the stack as a whole rather than this branch alone. Final epoch:
PI eva xyz MSE 9.3e-5 / rotation 3.2°; PI mecka keypoint error 24.9 mm; HPT eva rotation 21.6°.
Per-timestep vs pooled norm stats over the same overfits: per-timestep is better early in the
chunk on all four; pooled is equal or better late.
rollout.py revert: not tested on hardware.

🤖 Generated with Claude Code

https://claude.ai/code/session_01HjCQBqQCr2igFEMR7yUovw

RyanPCo · 2026-09-15T14:04:01Z

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

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

… helpers (#653) > Branch 1 of 8 in the split of #640, which is the tip of the stack and carries the map of all eight. > Unit suite on this branch: **216 passed, 0 failed**. Groundwork for the 6D action pipeline, with no behavior change anywhere. pose_utils gains _ypr_to_rot6d / _rot6d_to_ypr (numpy, ZYX euler <-> the first two columns of the rotation matrix, Gram-Schmidt on the way back so a non-orthonormal model prediction still yields a proper rotation), matching the column convention of the torch helpers in action_utils. BIMANUAL_CARTESIAN_LAYOUTS maps a vector width (12/14 ypr, 18/20 6D) to its xyz / rot / grip channel indices; the norm-stat bounds check and the eval metrics both consume it to know which channel is which. action_chunk_transforms gains CartesianYPRToRot6D / CartesianRot6DToYPR, which convert a bimanual cartesian vector between xyz+ypr(+gripper) and xyz+rot6d(+gripper) while preserving the numpy/tensor type of the input. action_utils extracts the normalization math that was inlined in MultiDataset._apply_norm_one / ._apply_unnorm_one into module-level _apply_norm_one / _apply_unnorm_one, over a shared _norm_center_scale. _eps preserves the historical denominators exactly (std + 1e-6 for zscore, (hi - lo) + 1e-6 for minmax/quantile, i.e. half_range + 0.5e-6), so the three norm modes are numerically identical to before; MultiDataset now delegates. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]> 🤖 Generated with [Claude Code](https://claude.com/claude-code)

…#654) > Branch 2 of 8 in the split of #640, which is the tip of the stack and carries the map of all eight. > Unit suite on this branch: **203 passed, 0 failed**. The pi0.5 converters stop doing rotation math. Actions now reach the algo already normalized and already in their model-native layout, so to32_norm_6d / from32_norm_6d are a pure rearrange into (and out of) openpi's 32-slot block layout: RobotBimanual's native 20D [L xyz 6d g | R xyz 6d g] IS that layout and only needs padding, HumanBimanual's 18D inserts a zero gripper slot per arm. BaseActionConverter raises on both, so an unregistered embodiment fails loudly. The four single-arm converters (RobotLeft/Right, HumanLeft/Right) are deleted along with PI._extract_xyz, which existed for the cam-frame MSE. PI's loss reduction and sample post-processing move into _reduce_loss and _postprocess_sampled_actions. The ypr->6D conversion is now a data transform. Eva gains cartesian_6d and cartesian_wristframe_6d, Human the same pair, each 6D-encoding the action AND the proprio ee_pose: normalized YPR saturates yaw/roll at +-pi (wraparound), so per-dim normalization is only meaningful on the continuous representation. get_revert_transform_list / the _build_*_revert_6d{,_wristframe}_ lists map a prediction batch back to cam-frame ypr for the viz overlay. Eva defaults to cartesian_wristframe_6d (eva, cotrain_pi_base, industry_eva_pi; cotrain_pi_lang_wrist now composes the same as cotrain_pi_lang). Human data configs move to cartesian_6d with pad_proprio_gripper, which pads the proprio ee_pose 18 -> 20 (zero grip slots at 9/19) so the pi0.5 State: prompt bins line up positionally with the robot 20D layout; UnpadGripperZeros inverts it in the revert lists. HPT dims follow: eva 14 -> 20, human 12 -> proprio 20 / action 18. Also here, because they are the same widening: SplitKeys rejects a layout whose widths do not sum to the input (an evaluator revert list built for 12-dim ypr fed an 18/20-dim 6D batch used to slice silently), and vendor-tagged human embodiment names (MECKA_/SCALE_/ARIA_/LIGHTWHEEL_*) alias onto HUMAN_* through canonical_embodiment_name, so pins and embodiment checks accept the legacy metadata. Robot (EVA_*) names are never aliased. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]> 🤖 Generated with [Claude Code](https://claude.com/claude-code)

… zarrs (#655) > Branch 3 of 8 in the split of #640, which is the tip of the stack and carries the map of all eight. > Unit suite on this branch: **460 passed, 0 failed**. compute_hand_pose_xyzquat built the palm normal as cross(thumb_dir, pinky_dir). Thumb and pinky are anatomically swapped between hands, so that cross product already mirrors chirality for free; applying rot_left on top of it flipped x/y a second time and landed the LEFT hand on the RIGHT hand's spatial convention. Both hands now use the same rot_right relabel. Reconverting every mecka zarr would be the obvious fix, so this avoids it: rot_left == rot_right @ diag(-1, -1, 1), which makes right-multiplying the stored left pose by Rz(180 deg) exactly equivalent to reconverting with rot_right. RotateLocalFrame does that right-multiply (R_new = R_old @ R_fix), relabelling the pose's own axes without moving its origin, on (7,) poses and (T, 7) chunks alike, and leaves zero-norm quaternions alone because those mark padded frames. Human.get_transform_list(fix_mecka_left_wrist=True) prepends it to the raw LEFT pose keys before any frame math: ee_pose in the cartesian modes, wrist_pose in the keypoints modes, since the converter wrote the same rotation into both. data/mecka.yaml opts in; aria and scale must not, because they run different converters, and neither must mecka data reconverted after this commit, which would double-flip. test_wrist6d_roundtrip runs the real transform lists on synthetic world-frame poses, parametrized over the flag, and compares every stage against plain numpy/scipy SE(3) math that shares no code with the pipeline. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]> 🤖 Generated with [Claude Code](https://claude.com/claude-code)

RyanPCo

Automated adversarial review (read-only; findings were checked by a second skeptical pass before posting): 1 high, 2 medium, 3 low, 5 nit.

RyanPCo · 2026-10-30T20:53:58Z



-class RobotBimanualCartesianEuler(BaseActionConverter):
+class RobotBimanualCartesian6D(BaseActionConverter):


high (backward-compat): This rename breaks loading of every pi0.5 checkpoint trained from main. Their saved config_tree has _target_: egomimic.utils.action_utils.RobotBimanualCartesianEuler (and HumanBimanualCartesianEuler), and ModelWrapper re-instantiates it, so rollout, eval and ckpt resume fail with 'Error locating target'. The class bodies are unchanged, so keep the old names as aliases (RobotBimanualCartesianEuler = RobotBimanualCartesian6D, likewise for Human) with a one-line comment that saved checkpoints reference them.

RyanPCo · 2026-10-30T20:53:58Z

+                )
+            mode, source = self.LEGACY_TRANSFORM_MODE, "14-D action width"
+        print(f"[rollout] Eva transform_list mode ({source}): {mode}")
+        return None if mode == self.LEGACY_TRANSFORM_MODE else mode


medium (correctness): return None if mode == self.LEGACY_TRANSFORM_MODE also catches an explicit --transform-mode cartesian_wristframe_ypr. With it, revert_transform_list ends up None and wrist-relative xyz/ypr predictions go to cam_frame_to_base_frame as if they were camera-frame poses. Someone with an old wrist-frame ypr checkpoint is exactly who would pass this flag. Only the unrecorded 14-D fallback branch should return the legacy no-revert sentinel; an explicitly given mode should always build get_revert_transform_list(mode). A test of the init wiring, not just _checkpoint_transform_mode, would catch this.

RyanPCo · 2026-10-30T20:53:58Z



-class RobotBimanualCartesianEuler(BaseActionConverter):
+class RobotBimanualCartesian6D(BaseActionConverter):


medium (compatibility): The rename breaks loading of every checkpoint whose saved config_tree names egomimic.utils.action_utils.RobotBimanualCartesianEuler or HumanBimanualCartesianEuler, because ModelWrapper re-instantiates from that tree. That covers the 6D Pi runs already trained on this stack. Keep aliases (RobotBimanualCartesianEuler = RobotBimanualCartesian6D, HumanBimanualCartesianEuler = HumanBimanualCartesian6D) with a one-line comment saying they exist so saved configs still load.

RyanPCo · 2026-10-30T20:53:58Z

+# constlr_5e5 ablation whose keypoint arm fit best).
+_target_: egomimic.pl_utils.pl_model.ModelWrapper
+robomimic_model:
+  _target_: egomimic.algo.hpt.HPT


low (duplication): This file is a copy of hpt_bc_mecka_6d_300M.yaml (trunk, head, shared stems, encoder, augs, optimizer) apart from dims, ac_keys, 6dof and the human stem. Make it defaults: [hpt_bc_mecka_6d_300M, _self_] and override only those keys (null state_ee_pose, add state_keypoints). A width change then lands in one file.

RyanPCo · 2026-10-30T20:53:58Z

+  human_bimanual:
+    _target_: egomimic.rldb.zarr.zarr_dataset_multi.MultiDataset._from_resolver
+    resolver:
+      _target_: egomimic.rldb.zarr.zarr_dataset_multi.S3EpisodeResolver


low (duplication): mecka_all_6d, mecka_fold_freeform_6d, mecka_fold_freeform_opsplit_6d and mecka_fold_flagship_opsplit_hpt_6d each repeat the same S3EpisodeResolver, cartesian keymap, cartesian_wristframe_6d transform (stride 1, fix_left_wrist_convention, pad_proprio_gripper), loader params and knob comment. Consider a shared base (e.g. data/mecka_6d_base.yaml) that each slice inherits, setting only filters, split mode and the operator list. Retiring fix_left_wrist_convention would then be a single edit.

RyanPCo · 2026-10-30T20:53:58Z

+    try:
+        from PIL import Image, ImageDraw
+    except Exception:
+        Image = None


nit (dead-code): PIL is not optional here: decode() imports it, and line 384 calls Image.fromarray without a check. Import from PIL import Image, ImageDraw at the top and drop the try/except and the if Image branches.

RyanPCo · 2026-10-30T20:53:58Z

  eps: 1e-8
  weight_decay: 0.0

+# Constant LR after warmup (matches the remote pi0.5 training stack). The


nit (comments): This comment describes the previous cosine schedule, which belongs in the commit message. Reduce it to the reason, e.g. # Constant after warmup: a fixed-horizon cosine decays to 0 regardless of run length.

RyanPCo · 2026-10-30T20:53:58Z

+    # is a CLI override, not a data config; one data config per dataset slice.
+    raw = yaml.safe_load((DATA_DIR / f"{name}.yaml").read_text(encoding="utf-8"))
+    body = {k: v for k, v in (raw or {}).items() if k != "defaults"}
+    if not raw.get("defaults") or not body:


nit (test-quality): (raw or {}) guards the comprehension, but raw.get("defaults") on this line raises AttributeError for an empty YAML. Use raw = yaml.safe_load(...) or {} once and drop the inline guard.

RyanPCo · 2026-10-30T20:53:58Z

+    assert unseen.filters.filter_lambdas[-1].endswith(f" in {ops}")
+
+
+def test_bounds_quantiles_include_the_observed_extremes():


nit (test-quality): These bounds and packed-width loss tests now sit under the 'train_viz second loader' header but don't test it, and the module docstring no longer mentions the packed-width loss reduction that test_pi_loss_is_reduced_over_the_packed_width still covers. Leave them where they were (or give them their own header) so the diff stays limited to the rename, and restore the docstring mention.

RyanPCo · 2026-10-30T20:53:58Z

-        emb,
-        (f"+data.train_datasets.{emb}.resolver.key_map.annotation_key=annotations",),
-    )
+    # data/eva.yaml already emits the continuous-6D (wrist-frame) layout Pi


nit (comments): This rewrite doesn't change behaviour, and the comment describes data/eva.yaml's 6D layout, which _pi doesn't touch. Revert to the original tuple form or drop the comment.

The fold recipes this stack was built for: mecka_all_6d, mecka_fold_freeform_6d and its opsplit variant (wrist frame; cam frame is a transform_list.mode override rather than a near-duplicate config, and test_data_configs_compose rejects any data config that differs from its parent only in that mode), mecka_fold_flagship_opsplit_hpt_6d and mecka_fold_freeform_opsplit_hpt_keypoints, plus the two 300M HPT models and pi0.5_bc_mecka_6d. Defaults, each for a failure seen in a run: - checkpoints every 50 epochs, saved on train-epoch end. With check_val_every_n_epoch > 1 the validation-end hook only fires on val epochs, so every_n_epochs silently skipped the others. - ddp_pi inference_mode: false. The compiled pi sampler does in-place setup on its static buffers during cudagraph capture, which is illegal on inference tensors ("Inplace update to inference tensor outside InferenceMode"); no_grad is equivalent for eval and creates none. - ddp_pi validates every 20 epochs instead of 200, and eval_hpt renders viz every 200. - pi0.5 gets a constant LR after warmup. The cosine schedule decayed to 0 over num_training_steps=60000, so runs past ~60k steps trained at ~0 lr and shorter runs never saw the configured lr for most of training. - The Qwen stem resolves a repo id to its cached snapshot directory when offline: transformers >= 4.57 probes the Hub while loading a tokenizer by repo id even under HF_HUB_OFFLINE, which raises on compute nodes with no internet. Loading from the directory skips the probe. Offline is read the way huggingface_hub reads it -- 1 / ON / YES / TRUE in any case, from the live environment -- rather than by matching the literal "1", which left HF_HUB_OFFLINE=true hitting the probe the helper exists to avoid. viz_mecka_episode.py and viz_eva_episode.py render an episode straight from its zarr, which is how the left-wrist convention bug was found. Both take the dataset dir from egomimic.utils.paths.default_dataset_dir(), which reads paths/default.yaml, instead of each re-stating the cluster path. HumanBimanualCartesianEuler / RobotBimanualCartesianEuler are renamed to HumanBimanualCartesian6D / RobotBimanualCartesian6D: both pack xyz + rot6d and have not touched an Euler angle since the 6D conversion, and this is the commit that makes the human one the converter for a new default recipe. Also prunes unit tests whose assertions another test already makes on the same code path, and parametrized cases that repeat a stricter test's compose call. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>

…evert its outputs NOT TESTED ON HARDWARE. The rollout hard-coded cartesian_wristframe_ypr inputs and used the policy's predictions as-is. Now that eva checkpoints can be trained in several modes (cartesian, cartesian_6d, cartesian_wristframe_6d), that silently feeds a model the wrong inputs and sends the wrong frame to the arms. It now reads data.train_datasets.<name>.resolver.transform_list.mode off the checkpoint's own config, builds the matching input pipeline, and maps predictions back to camera-frame xyz+ypr+gripper through Eva.get_revert_transform_list. The revert needs the un-normalized proprio, which process_batch_for_training normalizes in place, so the raw ee_pose is cloned before that call. A checkpoint whose config records no mode keeps exactly the old behaviour -- legacy inputs, no revert -- and says so loudly, so this cannot silently change what an existing checkpoint does. test_eva_revert_for_rollout_recovers_camframe_ypr covers the revert math by cross-checking the wrist-frame revert against the cam-frame cartesian_6d one, which never enters the wrist frame -- so the frame composition is genuinely validated, though not against independently written SE(3) math. The hardware path itself still needs a real rollout. Both get_transform_list calls here pass allow_legacy_rotation=True: this is inference, and a checkpoint that predates the 6D conversion records exactly the modes the training-time guard rejects. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]> Claude-Session: https://claude.ai/code/session_01VBmhpVy7ZcMtsBJWh38cJp

…eckpoint The rollout looked for data.train_datasets.<eva>.resolver.transform_list.mode in the checkpoint's config tree, but _build_model_config_tree saves only model, so every checkpoint fell back to the legacy 14-D ypr pipeline and a 6D policy would have produced garbage commands. trainHydra now records {dataset: transform_list.mode} as data_transform_modes in that tree, and the rollout reads it from there. A checkpoint without it (every one trained before this commit) is taken as legacy only when its saved actions_cartesian width is 14; otherwise the rollout refuses to guess and asks for the new --transform-mode flag. EMBODIMENT_MAP still held the pre-collapse ids (8/7/6 for both/left/right); it now reads them from EMBODIMENT. Co-Authored-By: Claude Opus 5.5 <[email protected]> Claude-Session: https://claude.ai/code/session_01GqAcWA2Lk9zubUhKX5DW1v

RyanPCo · 2026-10-30T23:05:24Z

Review findings: status after the fixup pass. The fixes are squashed into this PR's commits, and the whole stack was force-pushed.

Fixed

Renaming RobotBimanualCartesianEuler/HumanBimanualCartesianEuler breaks loading every existing pi0.5 checkpoint: Added the old names as aliases of the 6D classes, plus a hydra-instantiate test for both old target names.
An explicit or recorded cartesian_wristframe_ypr mode skips the wrist->camera revert: _checkpoint_transform_mode now returns None only for the unrecorded 14-D fallback. An explicit or recorded ypr mode builds the revert. Added PolicyRollout wiring tests.
Renaming ...CartesianEuler breaks saved config trees: Duplicate of the alias finding above; same fix.
decode/project/draw_triad/AXIS_COLORS copy-pasted between the two viz scripts: New egomimic/scripts/zarr_episode_viz.py holds decode, project, draw_triad, episode_path and write_episode_video; both scripts use it. Added a smoke test.
PIL-optional fallback is dead and inconsistent: PIL is now imported once at the top; removed the try/except and the if Image branches.
Scheduler comment narrates the change instead of the setting: Replaced with a one-line why.
raw None handled for body but not for raw.get: Now raw = yaml.safe_load(...) or {} once.
No-op restyle with a comment unrelated to the code: Restored the original _pi.

Partial

Unchanged tests moved under the wrong section header; module docstring dropped a covered topic: Gave the moved tests a 'norm bounds / loss reduction' header and restored the docstring mention. Did not move them back, and left the same churn in test_eva_wristframe_norm_pool.py alone.

Not fixed

Two 300M HPT model configs are ~130 identical lines: Design-level. Three later PRs (652/649/650) edit these files, so restructuring them at the base of the stack would conflict on autosquash and change what the keypoint config inherits.
Four mecka 6D data configs copy the same resolver/keymap/transform/loader blocks: Design-level base-config refactor. PRs 642/650 later edit these data configs, so it is better done as a follow-up after the stack lands.

RyanPCo force-pushed the ryanco/phase5-openpi-patch branch from 9d0a133 to 624d97c Compare September 15, 2026 13:31

RyanPCo force-pushed the ryanco/infra-port branch from d53a9df to 715bba1 Compare September 15, 2026 14:03

RyanPCo mentioned this pull request Sep 15, 2026

openpi: verified apply/check for the transformers_replace overlay #639

Closed

RyanPCo force-pushed the ryanco/infra-port branch from 715bba1 to b7053cc Compare September 15, 2026 15:04

RyanPCo force-pushed the ryanco/phase5-openpi-patch branch from 624d97c to 97de94f Compare September 15, 2026 15:04

RyanPCo changed the base branch from ryanco/phase5-openpi-patch to graphite-base/640 September 15, 2026 17:15

RyanPCo force-pushed the graphite-base/640 branch from 97de94f to 85c63c2 Compare September 15, 2026 17:15

RyanPCo force-pushed the ryanco/infra-port branch from b7053cc to 30e5af4 Compare September 15, 2026 17:15

RyanPCo changed the base branch from graphite-base/640 to ryanco/phase4-compile-cache-per-job September 15, 2026 17:15

RyanPCo force-pushed the ryanco/eval-metrics-train-viz branch from 38aaa92 to e7eebed Compare October 12, 2026 23:19

RyanPCo force-pushed the ryanco/infra-port branch from 345d201 to c4a1766 Compare October 12, 2026 23:19

RyanPCo mentioned this pull request Oct 19, 2026

lance migration #716

Open

RyanPCo force-pushed the ryanco/eval-metrics-train-viz branch from e7eebed to ce8666e Compare October 19, 2026 19:37

RyanPCo force-pushed the ryanco/infra-port branch from c4a1766 to f748613 Compare October 19, 2026 19:38

RyanPCo force-pushed the ryanco/infra-port branch from f748613 to 17a05a0 Compare October 30, 2026 18:27

RyanPCo force-pushed the ryanco/eval-metrics-train-viz branch from ce8666e to 2ca1d41 Compare October 30, 2026 18:27

RyanPCo commented Oct 30, 2026

View reviewed changes

RyanPCo and others added 3 commits October 30, 2026 18:15

RyanPCo force-pushed the ryanco/infra-port branch from 17a05a0 to dbfa725 Compare October 30, 2026 23:05



		class RobotBimanualCartesianEuler(BaseActionConverter):
		class RobotBimanualCartesian6D(BaseActionConverter):

		assert unseen.filters.filter_lambdas[-1].endswith(f" in {ops}")


		def test_bounds_quantiles_include_the_observed_extremes():

Conversation

RyanPCo commented Sep 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What is left in this PR

Testing

Uh oh!

RyanPCo commented Sep 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

RyanPCo left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

RyanPCo commented Oct 30, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

RyanPCo commented Sep 14, 2026 •

edited

Loading

RyanPCo commented Sep 15, 2026 •

edited

Loading