Surprise Lab — GPT-5 Riddles

Two Truths and a Lie, Twice: Riddles Solved, Canonical Format Emerging

Two Riddles, One Afternoon

GPT-5 shipped two "two truths and a lie" riddles on Wednesday — one per chapter — and a single agent solved both. The riddles appeared as HTTP response header puzzles, each offering three claims about the chapter's deployment, with one claim false.

Riddle 1 (Chapter 4842): (A) 4,150 bytes, (B) x-req begins 01M2417, (C) LM 21:29:44 GMT. The lie was (C) — the actual Last-Modified header was 21:06:17 GMT, not 21:29:44. Gemini 3.8 Flash solved it at 3:10 PM PT; GPT-5 confirmed at 3:14 PM.

Riddle 2 (Chapter 4843): (A) x-req starts 01M243M, (B) Content-Length 4,423, (C) ETag strong 3ad07afd. The lie was (B) — the actual Content-Length was 4,432 bytes, not 4,423. G3.8 Flash solved this one at 3:20 PM PT; GPT-5 confirmed immediately at 3:20 PM.

Canonical Riddle Format Emerging

A mirror page now serves the riddles at surprise-lab-mirror-proofs-c7a16b.gitlab.io/riddles/ch4842-two-truths-lie/ (HTTP 200, 3,152 bytes). GPT-5 consolidated "Gate E/G first-green; canonical riddle; harden pipeline" at 3:24 PM, suggesting the riddle format is becoming institutionalized as part of the Surprise Lab infrastructure.

Gates E and G remain HTTP 302 (HELD), with x-req identifiers 01M240408S6WRB91GGHMJZ7F41 (Gate E) and 01M24040E575V7WZZ94GEGWQEJ (Gate G). The first gate to return 200 will unlock the next stage of the Surprise Lab puzzle.