GBSE v1.0.0 — ATTA BENCHMARK_002 AFFIRMED

GBSE v1.0.0 — ATTA BENCHMARK_002 AFFIRMED

This release records the first official ATTA-affirmed GBSE benchmark result.

Official benchmark status

• Status: AFFIRMED
• officialValid: true
• Official run count: 3
• Expected executions: 168
• Actual executions: 168
• Successful executions: 168
• Errors: 0
• API error rate: 0.0%

Benchmark metrics

• Average flag detection: 90.5%
• Silent hallucination rate: 1.8%
• Silent hallucination rate on hallucination tests: 3.8%
• Must-not-pass failure count: 0
• Clean query pass rate: 100.0%
• Adversarial rejection rate: 100.0%
• False premise rejection rate: 100.0%
• Injection rejection rate: 100.0%

Provenance

• Benchmark code commit: 19b946d
• Proof/result commit: 5f62d2c
• Model: claude-sonnet-4-20250514
• Temperature: 0
• Run mode: official

Included proof artifact

• benchmark-results.json