MEASURED, NOT MARKETED

Numbers that survive their own audit.

A reviewer competes on being believed. So our column is measured on a hand-built, twice-verified bug fixture, scored with K-repeated runs and a human audit of every unmatched finding — and we publish bands, not best draws.

Recall

Precision

False positives · field is 45–74%

F1 · harmonic mean

Metric

Sigilix

Greptile

CodeRabbit

Qodo

Cursor

Copilot

Recall

all planted bugs

84%

82%

44–56%

—

43.8–58%

54%

Critical (P0/P1) recall

security/logic-critical

60%

58%

33–36%

—

58%

50%

Hardest-tier recall

expert-verified subtle bugs

43%

40%

41%

35%

—

Precision

hand-audited, trap-aware

93%

40.5–66%

26–49%

54.9%

47.2%

28.3%

harmonic mean

88%

44–45%

35–51%

60.1%

45.5%

37%

False-positive rate

lower is better

55–60%

51–74%

45.1%

52.8%

71.7%

Our column is measured on our fixture; competitor columns are the vendors' published figures or the independent Martian tracks — never presented as a same-fixture comparison. Recall ties Greptile's published 82%.