Skip to content

Prune case-local BFCL equivalent siblings - #55

Closed
SonAIengine wants to merge 1 commit into
codex/bfcl-neardup-disambiguationfrom
codex/bfcl-equivalent-sibling-pruning
Closed

Prune case-local BFCL equivalent siblings#55
SonAIengine wants to merge 1 commit into
codex/bfcl-neardup-disambiguationfrom
codex/bfcl-equivalent-sibling-pruning

Conversation

@SonAIengine

Copy link
Copy Markdown
Owner

Summary

  • Suppress model-facing non-priority equivalent siblings when a BFCL case-local surface is present, while keeping retrieval evidence intact.
  • Strengthen surface equivalence evidence for currency conversion, definite integral / area-under-curve, Fibonacci sequence / series, and GCD / HCF.
  • Add repeated-call prompt guidance and avoid decimal-rate hints for fields explicitly described as percentages.
  • Document the xgen-0.27 100-case gate pass and remaining retrieval-layer bottlenecks.

Validation

  • poetry run ruff check .
  • poetry run ruff format --check .
  • poetry run pytest tests/test_bfcl_tool_selection_llm_loop.py tests/test_graphify_contract_025.py tests/test_bfcl_tool_selection_sweep.py -q -> 64 passed
  • poetry run pytest tests/ -q -> 967 passed, 3 skipped, 1 warning
  • Fresh Qwen hardcase run: /tmp/gtc-bfcl-equivalent-sibling-pruning-hardcases-v2.json -> 6 / 10 pass
  • Fresh Qwen 100-case sweep: /tmp/gtc-bfcl-equivalent-sibling-pruning-limit25-sweep.json -> retrieved exact 0.96, retrieval@5 0.99, row preservation 0.98, parallel_multiple exact 1.00, xgen-0.27 gate pass

Remaining Bottlenecks

  • multiple_24: route retrieval miss; expected route_planner.calculate_route is not in top-5.
  • parallel_3: over-decomposition where the model calls lower-level sequence-only and sequence+3D tools together.

@SonAIengine

Copy link
Copy Markdown
Owner Author

Closing this internal stacked PR as superseded by the current main branch. The branch is preserved for history.

@SonAIengine SonAIengine closed this Aug 2, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant