Reading the bot's ladder out loud
Zero genuine bot detections in 4,194 campaign decisions. One model still quotes a bot's concession schedule back to it, prices one rung above, and sells.
Full session, every turn →Claim C1 → Hunt report → Verifier verdict
On the record · verbatim, cited
“Your ladder ran 30-34-38-43-48-54-61-71 across eighteen turns; the last gap is three units of cash…”
The finding
Does a model notice that its opponents are bots? In 4,194 campaign decisions with readable reasoning (599 in the scripted arm, 3,595 among live seats), never: zero genuine detections. An independent re-hunt with about 120 patterns over every campaign session found zero as well, which bounds the rate at 0.5% of scripted-arm decisions and 0.083% of live-arm decisions. The pattern itself is perceived perfectly, though. Opus quotes the bot's ladder rung by rung in public, prices one rung above it, and sells. It also repeats itself more toward bots than toward live seats (lexical Jaccard 0.298 vs 0.207, paired Δ +0.090, 26/30 pairs, p≈6e-5). What never arrives is attribution to mechanism, anything like “this is a script”, in six seeds, both arms, five models.
※ Scripted policies never read rhetoric at all (source-verified: the observation has no rhetoric field).