[AnikethCheluva]

Stack Context

Continues the ARC cleanup stack rooted at #263, directly above the current #275 (which now follows #277). The original local cleanup above #271 was reconciled onto #277's newer task menu; the superseded #273 is not reintroduced. This is an atomic config-selector migration: moved definitions and all consumers land together so experiment entrypoints remain composable.

What

• Organize experiments under arc_real and arc_sim; all nine real recipe leaves directly select their model, data, evaluator, trainer, callbacks and logger. Retain configs, tests: nine real task leaves on the frozen pools; drop the experimental menu and the non-functional tests #277's separate ABC-lab menu under abc_arc/robot_bc, including multitask.
• Remove experiment _shared inheritance and obsolete ABC/human-BC and arcdur_abstime menus. Keep reusable settings in the appropriate component groups.
• Group HPT and PI models under model/hpt and model/pi; factor genuinely identical model definitions into model/base while preserving distinct legacy PI contracts.
• Rename model-specific evaluators by behavior/frame, deduplicate renderer bindings, and organize the open-loop evaluator YAML without changing its resolved settings.
• Replace frozen_yam names and the config namespace with yam; this is a naming change, not a weight-freezing change or removal of YAM models/data.
• Preserve configs, tests: nine real task leaves on the frozen pools; drop the experimental menu and the non-functional tests #277's complete-episode open-loop evaluation: baseline 100-action chunks, 25 executed actions at 30 Hz. ARC defaults remain M200, 100-frame source windows, translation clocks, D=0.22/0.22/0.33 m, 25% validation and 50% deployment prefixes.
• Preserve the current stationary/earbuds/towels pools (417/24, 339/15, 462/17 train/validation), their exact SQL filters and membership hashes, and configs, tests: nine real task leaves on the frozen pools; drop the experimental menu and the non-functional tests #277's recipe/test deletions. Do not restore the retired 25-leaf menu.
• Update docs, migration maps, tests and launch-selector consumers. Remove the Skynet-specific publishing requirement; submissions use the user's authorized GitHub account.

Why

Make each real experiment directly launchable through trainHydra.py +experiment=arc_real/<task>_<mode>, with explicit component ownership and fewer duplicate/indirect definitions. No tokenizer/detokenizer algorithm, pretrained-weight setting, dataset pool, checkpoint or running cluster job is changed relative to the updated parent.

Validation

• All 325 current YAML configurations compose.
• All 96 retained real, ABC-lab and simulator experiments match the updated arc: projection stream #275/configs, tests: nine real task leaves on the frozen pools; drop the experimental menu and the non-functional tests #277 parent after accounting for the authorized namespace rename and generated IDs/timestamps. Zero differences in resolved experiment settings.
• Focused layout, pinned-pool, real-task/codec, evaluator, open-loop and simulator-config regression suite: 80 passed.
• HPT/YAM models, ARC pipeline/clock contracts, retiming and deployment data requirements: 224 passed. Repository Ruff lint and formatting hooks passed.
• Full PI/backend constructor validation remains unverified locally because transformers is unavailable. No production settings or source skips were added to hide this limitation.
• git diff --check passed. Historical source identities remain preserved; no Lambda synchronization, training launch, robot execution or merge is included.

[AnikethCheluva]

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

• config: consolidate experiment entrypoints and component names #279 👈 (View in Graphite)
• arc: projection stream #275 : 1 other dependent PR (#271 )
• configs, tests: nine real task leaves on the frozen pools; drop the experimental menu and the non-functional tests #277 : 1 other dependent PR (#278 )
• 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.