The slots

The money

The token is launched on Pons; its creator fees are moved by hand to the casino account. The casino is Stake, which takes ETH on Ethereum or BNB Chain, so fees are bridged to BNB Chain. Stake's terms forbid automated play and the account may be closed with its balance at any time. We know. Only one session's bankroll is ever there, the driver is a visible browser the fly looks at, and there is no hidden bot, proxy rotation or evasion.

The driver

A real Chromium under Playwright, signed in by hand. The fly sees its pixels; the loop reads the balance before and after a spin and treats the difference as the payout. Until the acceptance test in the task list is passed the driver never clicks (DRY_RUN=true), and during development the phone shows our own three-reel page instead:

A three-reel page served from a tiny in-process HTTP server (not Vite: it is one file and needs no build), driven by Playwright Chromium, headless by default, HEADED=1 to watch it. Seeded RNG (mulberry32) so a run replays. Strip of 32 stops per reel: 8 🍒, 8 🍋, 6 🔔, 5 ⭐, 3 7️⃣, 2 🪰.

Published RTP 94.57 %, computed exactly over all 32³ combinations (rtp()), and printed on the page. Hit rate on a 1,000-spin sample: 35.8 %.

The rules we chose

All of these are engineered, shown on the page's tile, and logged with every run:

• Bet. A fixed fraction of the balance, clamped between a minimum and a maximum, with a daily cap on the total wagered and a kill-switch file. The connectome never picks an amount.
• Spin. The mean rate of two DNa02 and two DNa03 descending neurons over one 50 ms observation, at or above 65 Hz, at least 2 s of neural time after the last press.
• Reward. After a paying spin, at least 20 mV and up to 60 mV (20 mV × multiplier) into the fifteen PAM11 cells for 500 ms. After a loss, nothing; a switch can send 20 mV into the two PPL101 aversive cells instead, off by default.
• Wake-up. White light before the phone until the Kenyon cells ignite, because from a cold start on a dark page the network sits in a state where the readout never fires.

What we are trying to do to it

Give it a gambling habit: a dopamine hit for every win, and a memory rule that turns those hits into a preference. The house edge is arithmetic and does not move, so the expected loss on the tile is wagered × (1 − return) whatever the synapses do; what the synapses can change is how often the fly presses, and that number is measured against controls.