Skip to content

Fix selected TaskSpec grip evidence identity - #885

Draft
yuecideng wants to merge 1 commit into
codex/action-engine-v36-main-syncfrom
codex/gensim-e4-observer-order-upstream
Draft

yuecideng wants to merge 1 commit into
codex/action-engine-v36-main-syncfrom
codex/gensim-e4-observer-order-upstream

Conversation

@yuecideng

Copy link
Copy Markdown
Contributor

Description

Read the selected TaskSpec observer before the legacy audit-only observer. The legacy pass can query the native grip sensor; when it ran first, it advanced the selected receipt's fresh_grip.query_index by one. This change restores the selected evidence identity seen by the v15 baseline while keeping both evaluations in the same episode. It does not change motion, physics steps, commands, acceptance thresholds, or the legacy pass's audit-only role.

This is an independent fix for the PR #729 branch. The broader GenSim _task_program migration remains a separate change.

Type of change

  • Bug fix (non-breaking change which fixes an issue)

Validation

  • The existing dual-observer test now checks that selected evidence is collected before the legacy audit pass. The focused TaskSpec/acceptance tests passed (72 tests).
  • Paired v15 baseline (3e002eea) → fixed migration runs for E4 seeds 0, 1, 2 had zero formal E4 differences. Seeds 0 and 1 completed all 14 calls with strict TaskSpec success; seed 2 failed the same final stability check on both sides. Actions, 15 state records, planning records excluding wall time, videos, and rendered semantic graphs matched for each seed.
  • Paired fixed PR Integrate Task Engine runtime policies and robot profiles #729 → fixed migration E4 seed 0 completed 14/14 calls and 4221 control steps with strict TaskSpec success on both sides. The selected grip query index was 309 on both sides; actions, state, video, and rendered graph matched.
  • black ., python docs/scripts/check_api_docs.py (3399/3399), and project-context validation passed. No public API or project-context guidance changed.

Screenshots

Not applicable; the rendered semantic graph was checked in an offline browser on both sides.

Checklist

  • I have run the black . command to format the code base.
  • I reviewed affected documentation and agent context; the ownership guidance remains accurate.
  • Public API coverage was checked; no public API changed.
  • I verified focused behavioral coverage and the necessity of the added assertion.
  • Dependencies are unchanged.

@yuecideng yuecideng added bug Incorrect existing behavior or a fix restoring the intended behavior. gen-sim Scene Engine generation and editing, scene export and SimReady pipelines. task-program Task Program language, semantics, compiler, runtime and integration contracts. labels Oct 26, 2026

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug Incorrect existing behavior or a fix restoring the intended behavior. gen-sim Scene Engine generation and editing, scene export and SimReady pipelines. task-program Task Program language, semantics, compiler, runtime and integration contracts.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant