Conversation
mahaloz
added
documentation
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
mahaloz
changed the title
mahaloz
changed the title
mahaloz
changed the title
mahaloz
changed the title
mahaloz
added a commit
that referenced
this pull request
Aug 27, 2026
[AUTOMATED] Generated by Codex for PR #48.
mahaloz
added a commit
that referenced
this pull request
Aug 27, 2026
…r docs PR #48 carried ~27k lines of calibration and prototype scaffolding around a ~4k-line metric core. The scaffolding is removed so the shipped change is reviewable; everything it documented about behavior that actually ships is migrated into docs/metrics.md first. Removed: decbench/experimental/ (6,782 lines), six scripts/ calibration and demo drivers (4,704), the four tests that only exercised them (2,201), and docs/experiments/ (13,616, of which a generated evidence.json is 12,640). Of decbench/experimental/, two files were one-line `import *` compat shims over production decbench/metrics/{variable_match,variable_features}.py -- which is why three test files imported "experimental" while really testing shipped code. Those are kept, with imports repointed and two renamed to match what they test. The other two experimental modules were substantial: local_variable_checkpoint.py (2,391 lines) and local_variable_semantic_audit.py (4,379). Nothing outside the deleted scripts and tests imported either. Migrated into docs/metrics.md, each value re-verified against the current code rather than copied from the old prose: the production policy constants and their validators; the tree-sitter parsing contract and why tree-sitter is a runtime dependency; the candidate-discovery rules; the sanitization applied to candidates recovered from rendered C, which is what keeps address-mode numbers honest for text-only backends; the feature inventory and the type-leakage exclusions; and the similarity and fusion math. Result tables, frozen replay anchors and "Reproducing" sections were not migrated -- they measure an unpublished local tree through now-deleted scripts. An adversarial review of the first draft of this migration caught three real defects, fixed here: it claimed overriding a matcher threshold needs a cache_version bump (it does not -- the policy is part of the per-function cache key, so an override re-keys itself); it stated the extractor abstains only on multiple mismatched definitions (any function_definition present with no matching name abstains, and no fragment wrapping is attempted); and it described the degree term as within-function (it is counted over the union of the source and decompiled variables being matched). It had also dropped the inferred_from_code sanitization rule and the candidate-discovery contract entirely, and inverted the threshold provenance -- the address pair was a pre-existing frozen baseline, only the usage and combined values were selected on the tuning partition. docs/metrics.md is also updated for cache_version 12, which the Thumb fix earlier on this branch introduced. Deleting the four scaffolding test files cost shipped code some coverage; test_variable_match_features.py regains the literal-recursion case (variable_features.py:309-312), which was their only reach into production code that no remaining test touched. Suite: 6 failed, 1116 passed, 29 skipped -- the same 6 failures that fail on main. ruff and black show no new debt (18 errors and 6 files before and after, identical lists). Residual-reference sweep for every deleted module, script and directory returns nothing. [AUTOMATED] change produced by Claude Code under maintainer direction. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016P13XTwaZ1HtpNU3F47nMg
mahaloz
force-pushed
the
experiment/local-variable-edit-distance
branch
from
September 10, 2026 19:12
c2871f8 to
4808a62
Compare
mahaloz
changed the title
[AUTOMATED] Limit native variable correspondence to the seven full-dataset decompilers and retain a labeled legacy fallback for unsupported producers.
mahaloz
force-pushed
the
experiment/local-variable-edit-distance
branch
from
September 11, 2026 16:01
4808a62 to
eb2d6d7
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
[AUTOMATED]
Adds TypeMatch variable correspondence using ABI argument position, calibrated stack offsets, and validated instruction-address overlap.
This PR remains a draft and must not be merged yet.