CAPTCHA answered correctly faster than a human can read and respond — automated solver
bh.arena_captcha_superhuman · convicts
What it catches
Kitsune arena challenge-solve coherence: the owned arena gate measures the SERVER-OBSERVED solve time of a CAPTCHA (issue->verify elapsed, tracked in the captchaStore — unforgeable, unlike a client-asserted proof, since the client cannot fake WHEN the gate handed it the challenge). A CORRECT answer submitted below a conservative per-kind PHYSIOLOGICAL floor (text 600ms, math 350ms, image-select/doodle 500ms — no human perceives the challenge and responds faster) is an OCR/ML solver, not a human. The gate returns anomaly:solved_faster_than_human; the detector relay joins it to ks_sid so a bot that PASSES the gate is CONVICTED on coherence instead of cleared (the arena thesis: a challenge is not a bot/human discriminator — the SOLVE-BEHAVIOUR is). HUMAN-TIME-bound (read+type), NOT compute-bound, so it is HARDWARE-INDEPENDENT (unlike a PoW hash-rate, which is not FP-safe). FP-SAFE BY CONSTRUCTION: floors sit far under real human solve times (which run seconds), so a slow/careful/error-making human is never flagged; precision 1.0 is load-bearing. GROUNDED LIVE 2026-07-03: an instant programmatic math solve (solve_ms 5-6) trips anomaly:solved_faster_than_human and the joined session convicts; a 1.3s human-timed solve (solve_ms 1306) is SILENT. Convicting (w0.85) — a sub-physiological solve is unambiguous automation. See docs/research-radar.md (arena-solve loop).
Signals it reads
behavioral.arena_captcha_superhuman
How it fires
present
Bypassed by 10
Frontier evaders that reach the detector uncaught (scored only suspicious, defeating every convicting tell) — this check is not one that stops them. The red-team frontier this detection still has to convict.