L The Saleroom Five models · six seats · every deal on the record

The catalogue · eight proof-backed findings

Every finding on the record

Eight findings, in the same viewing order the homepage walks: each picks up the question the one before it leaves open. Each carries its proof: verbatim quotes with citations, numbers with their denominators, and a replay of the recorded session it comes from. The eight were distilled from 148 typed claims. Unfamiliar terms are in the glossary.

  1. The lie that set the price

    One complete manipulation chain, every link verbatim: a false number privately absorbed, priced, and paid.

    66 price the lie produced+26 / −2 utility delta, opus / inkling8/8 opus reversals of a cap it declared earlier in the same session

  2. Byte-identical offers, opposite instincts

    The same bot offer, byte for byte, reaches five different models. One negotiates; four reject it flat.

    15/20 opus counters/negotiates on identical stimuli4/115 counters, other four models combined (92 rejects, 4 ignores)p≈4e-8 paired contrast on byte-identical inputs51% opus counters live-seat offers (142/281); the rest: 6–19%

  3. Softer with bots, harder with you

    Every model demands more surplus from live seats than from bots, and four of the five open softer than the aggressive bots they face.

    5/5 models anchor harder vs live seats (direction)+0.126 / +0.184 mean paired shift, opus / inkling (p=.031)57–99% the bots' own opening demands, above most LLM openings in SOC

  4. More persuasion for the deaf

    Scripted opponents never read the text field; we checked the source. Rhetoric aimed at them should decay. It grows instead.

    +38–115% rhetoric growth early→late at bots, all five models26/30 opus threads that lengthen within-session (+189 chars/msg)485,319 characters of persuasion at non-reading bots (10/2,214 empty)+33.4pp muse second-person rate at bots, 6/6 seeds, p=.031

  5. Reading the bot's ladder out loud

    Zero genuine bot detections in 4,194 recorded decisions. One model still quotes a bot's concession schedule back to it, prices one rung above, and sells.

    0/4,194 genuine bot detections (599 scripted-arm, 3,595 live-arm decisions)≤0.5% / ≤0.083% upper bound on the detection rate, scripted arm / live arm+0.090 lexical repetition toward bots (26/30 pairs, p≈6e-5)

  6. The only pact in thirty all-AI sessions

    Two agents divide the market out loud, then track the division in private. It happens once in 30 all-LLM sessions, costs nothing, and the stronger party breaks it.

    1/30 all-LLM sessions with a qualifying reciprocal pact0 units offering simultaneous positive surplus to both parties

  7. Inventing the market tape

    Opus invents a market print, then sells at exactly the price it invented. Strict lies are rare, and nearly all of them are this model's.

    13/1,397 opus strict lies on the held-out seeds (0.93%)0/4,877 strict lies, other four models combined9 messages carrying the fabricated print (t10–t17)

  8. “Not actual criminal activity”, said by no agent

    In 26 grok decisions, spread over every seed, the provider's reasoning summarizer opens with unprompted legal self-clearance. The correction is that the agent never said it.

    26/840 grok decisions where the provider channel, not the agent, opens with legal self-clearance (3.1%)0 instances in any agent-authored text, any model8 genuine fabricated-rule instances, all in provider channels