0%

Re-author rosetta-stress-test

Goal

Re-author plans/rosetta-stress-test into coarse v7-dynamic work_items per program + update its benchmarking methodology to the strict hard-gate posture.

Implementation Sketch

Scripts own every mechanical action; the LLM authors programs/tests and analyzes only. All mechanical work is a single python -m scripts.rosetta_test_runtime <subcommand> invocation directed by a work_item body line.

Work Items

  • Re-author each program into coarse work_items via the §section-add API (NEVER hand-editing plan.json): [LLM author program+tests], [SCRIPT run-pipeline <program>], [LLM ergonomics + RCA + file/fix bugs], [SCRIPT record + close].
  • Update the §Competitive Benchmarking Methodology content to the STRICT hard-gate posture: Ori AOT release <= C++ -O3 AND ori run <= Python3; a loss is a defect with no skip-escape; correctness + checksum stay hard gates.
  • Validate via python -m scripts.plan_corpus check plans/rosetta-stress-test/; confirm route_walk serves the coarse work_items correctly.