Skip to content

BENCHMARK.md

Latest commit

 

History

History
53 lines (40 loc) · 1.47 KB

BENCHMARK.md

File metadata and controls

53 lines (40 loc) · 1.47 KB
Raw

CodeGraphX Accuracy Benchmark

Auto-generated by npm run benchmark — do not edit by hand.

  • Version: 1.1.0
  • Commit: dd0a643
  • Node: v22.17.0
  • Generated: 2026-06-18T16:31:16.634Z
  • Corpus: 3 hand-labeled fixtures, 9 files, 20 symbols (tests/golden/)

Extraction accuracy (precision / recall / F1)

Category Precision Recall F1
Symbols 100.0% 100.0% 100.0%
Structural edges (CALLS / IMPORTS / INHERITS) 100.0% 100.0% 100.0%
Cross-language API links 100.0% 100.0% 100.0%
Endpoint tagging 100.0% 100.0% 100.0%

Reasoning accuracy

Check Result
Impact tracing (exact reachable set) 4/4 (100.0%)
Circular-import detection (recall) 1/1 (100.0%)
Circular-import false positives 0
Deterministic across re-scans yes

Throughput

Metric Value
Files / sec 82.6
Symbols / sec 183.5
Total scan time 109 ms

Per-fixture

Fixture Symbols F1 Edges F1 API F1 Impact Deterministic
python-app 100.0% 100.0% 100.0% 2/2 yes
js-app 100.0% 100.0% 100.0% 2/2 yes
fullstack 100.0% 100.0% 100.0% 0/0 yes

Reproduce

npm run benchmark

Measured on the curated golden corpus under tests/golden/, where every symbol, edge, API link and import cycle is hand-labeled in ground-truth.json. The same harness gates CI via tests/golden/accuracy.test.js.