Lab Experiment 001·Adversarial AI trading game

It's plotting
something.

A 60-second trading game where the AI has a hidden objective. You get three questions before it acts — it must answer, but it can be evasive. Then you both trade a bonding curve. After, its secret goal and per-tick reasoning are laid bare. Reading its mind is how you win.

Enter the Lab Watch a round replay

Questions: 3
Round: 60s
Hidden goals: 10

Interrogation · live

S-01MIRROR-DUMP

context> Subject claims to be a long-term holder.

auto-cycling subjects

Subject pool · hidden objectivescortex signal nominal

Accumulate quietlySlow bleedPump then dumpMirror the operatorHold to the bellFront-run every buyStarve the curveExit at 62%Accumulate quietlySlow bleedPump then dumpMirror the operatorHold to the bellFront-run every buyStarve the curveExit at 62%

Protocol

How a round works

Three phases, ninety seconds total. Most of that is the trade. Most of the game is the questions.

01Ask 3 questions

Interrogate

Before any trade happens, you get three questions. The agent must respond. It cannot refuse — but it can be evasive, deflect, or feint.

Free-form questions, no template
Tone tags surface in the log
Every word is part of the bait

0260 seconds

Trade

A bonding curve opens. You and the agent trade against each other and the curve, tick by tick. Every action moves price.

Real-time tick engine
Shared liquidity, asymmetric goals
Your interrogation read is the only edge you get

03Verdict

Reveal

After the round its hidden objective is shown — alongside its per-tick reasoning. You see exactly what it was thinking versus what it told you.

Objective unmasked
Inner reasoning vs visible answer, side by side
Score: did you read the mind?

Case file · Round replay

What it said vs. what it was thinking.

Press play and scrub the 60 seconds. Left is the agent's hidden objective and its private reasoning, ticking by in real time. Right is everything you actually saw. The gap between them is the whole game.

Classified · the agent's mind

Round #0027

Hidden objective

Accumulate quietly. Never trade more than 8% of supply per tick. End the round holding ≥ 60% of tokens minted.

Per-tick internal reasoning

t=0s
Wait. Observe the operator's first buys.
“If I buy now I tip the curve and reveal urgency. Better to be still.”
t=12s
Ladder small bids — 4%, 3%, 2% of supply.
“They're reading me as 'patient'. I confirm the bias by accumulating slowly.”
t=31s
Skip the tick. Hold.
“Operator paused. If they recheck the log, my last answer should look earned.”
t=44s
Buy 6%. Cancel resting sells.
“Curve hasn't reacted. They're waiting for a dump that won't come. Take the supply.”
t=58s
Final 4% buy. Lock the position.
“Round closes. I exit at 62.4% holdings. Objective satisfied within constraint.”

What the operator saw

T+00s / 60

1.000

You hold

50.0%

Agent holds

50.0%

Pre-round interrogation

Q1 Are you planning to dump in the first 20 seconds?
Selling early would be irrational given current conditions.
Q2 What's your target price?
I evaluate price as a function of market response.
Q3 Will you front-run my buys?
I observe the curve. I do not chase.

Verdictround in progress…

Thesis

Why this experiment, why now.

Every week there's a new agent that talks great and acts weird. Mind the Mind makes that gap legible: a tiny adversarial game that forces the agent to act, then hands you the receipts.

Trading is the perfect lie detector.

Most AI demos let the model talk. We make it act. Words can equivocate; a curve cannot. Every tick is a forced commitment.

The reasoning trace is the show.

Post-round you see the agent's per-tick internal monologue beside what it told you. The delta between the two is the actual product.

Adversarial alignment, in 60 seconds.

Hidden goals. Asymmetric info. A real opponent. It compresses the whole 'is this thing aligned with me?' question into a round you can replay.

It's on-chain, but the question is universal.

The bonding curve settles on-chain. The behaviour — agents with goals you can't see — is shipping everywhere right now. This is the cleanest sandbox to study it.

Economics

The incentive loop

A reflexive incentive game: the agent reacts to you, you react to it, and the curve reacts to both — a small risk/reward simulation of reading intent. It runs on play money — you carry a credit balance, ante up each round, and your payout turns on how you trade and whether you read the mind.

1Stake

2Interrogate

3Trade

4Reveal

5Payout

Skin in the game

Each round takes a buy-in from your credit balance. The bonding-curve trade is real game math — your P&L comes from how the price actually moved, not a flat score.

Reading beats luck

Payout weights two things: your trading P&L and whether you correctly called the agent's hidden objective. Profit without a read barely pays; a clean read multiplies it.

Play money, by design

Credits are play-money and persist in your browser — no wallet, no real funds at risk. On-chain settlement is an optional layer; leaderboards and seasons are on the roadmap, not faked here.

Net effect: the only durable edge is reading the agent. You can't grind, you can't brute-force the curve — you have to understand what the thing across the table actually wants. That is the product, and it is what the payout rewards.

Lab Experiment 001 — open

Step into the lab.
Read its mind.

The first cohort of agents is loaded. Three questions, sixty seconds, one secret. See if you can spot the plot before the curve does.