Benchmark

Public snapshot of the current DataSitr detector benchmark: quality, performance, and claim boundary.

What the gates mean

Required quality gate: pass · Performance gate: pass · Overall benchmark gate: pass.

Performance snapshot

The public page shows the current 1K-character gating snapshot rather than the full internal performance report.

Current benchmark suites

• core_default_eval — pass, 207 cases, F1 99.7%, P 99.5%, R 100.0%.
• coding_session_eval — pass, 26 cases, F1 100.0%, P 100.0%, R 100.0%.
• technical_eval — pass, 88 cases, F1 100.0%, P 100.0%, R 100.0%.
• technical_real_sample — pass, 20 cases, F1 100.0%, P 100.0%, R 100.0%.
• technical_real_sample_wave3 — pass, 20 cases, F1 100.0%, P 100.0%, R 100.0%.
• technical_boundary_adversarial_wave4 — pass, 25 cases, F1 100.0%, P 100.0%, R 100.0%.
• technical_fp_regression — pass, 30 cases, F1 100.0%, P 100.0%, R 100.0%.
• support_dialogue_eval — pass, 14 cases, F1 100.0%, P 100.0%, R 100.0%.
• support_zero_noise_eval — pass, 35 cases, F1 100.0%, P 100.0%, R 100.0%.
• support_operational_noise_eval — pass, 14 cases, F1 100.0%, P 100.0%, R 100.0%.
• support_structural_obfuscated_eval — pass, 6 cases, F1 100.0%, P 100.0%, R 100.0%.
• quasi_risk_eval — pass, 49 cases, F1 100.0%, P 100.0%, R 100.0%.

Artifact and method

This page reads the published public JSON summary from /benchmark-artifact.json. It exposes the public benchmark snapshot without publishing internal provenance or runtime detail.

Last updated: 2026-04-29T14:40:52Z.

Claim boundary

This is not a blanket claim that every customer payload or every future detector change is perfect. It is a public benchmark snapshot tied to a fresh timestamp, a defined gate, and a frozen suite set.