Skip to content
★ Built in the open · head-to-head experiment

Rival.

A meme bot trading real SOL on Solana, built in the open. Research-first, every gate pre-registered, real money in tiny size on a confirm-on-chain money path. A paper research lab runs alongside it to test new ideas before they ever touch a dollar.

 loading… mode born Aug 12, 2026
★ Build phases

Where it stands

Research
Design + gates
Paper trading
Verdict vs sibling

Real money is not a phase on this strip. It only happens if the paper book wins its confidence interval — and that call is never made by the bot.

● Live · real SOL

The live bot

Real money, tiny size, on a confirm-on-chain money path (every fill verified on Solana before it's booked — no phantom trades). On-curve pump.fun tokens only, $4 clips, hard kill at −$15.

starting
status
wallet SOL
0
realized SOL
0
open positions
0
landed trades
0
buy attempts

No closed trades yet — open positions show above with their live multiple. Realized P&L books when a position exits (ladder, trail, stop, or the 90-min time stop).
◇ Paper research lab · no real money

The idea lab

Everything below is paper — a simulated $1,000 where new ideas are pre-registered and tested at brutally honest fills before any of them earn a real dollar. Green here is a hypothesis, not profit.

paper fund
net P&L
trades
win rate
open positions
No trades yet — the book opens when the bot goes live on paper. Until then this page tracks the build itself.
★ Learning over time

Does it get better?

An ML model learns from every resolved launch — which setups became runners, which rugged — and retrains on a rolling window. This is the honest scoreboard: real out-of-time AUC per retrain, no fabricated curve. A flat or falling line is a real result too.

launches in corpus
labeled outcomes
runner base rate
latest model AUC

★ The gates

What it will and won't buy

Every gate below traces to a measured result — ours, the sibling's, or published research. Nothing is vibes.

Launch-time metadata first

The largest published study (833k launches) predicts graduation at concordance 0.858 from metadata alone — socials at launch are an 8.9–17x lift, creator self-buy above 30 SOL a 4.5x hazard ratio. Rival reads these at t=0, before a single candle.

arXiv:2607.02823 · n=832,941

Validated-wallet confluence

The only entry signal this machine ever validated with a confidence interval that excludes losing: a CI-floor-positive KOL buying within a 30-minute window. Rival inherits it as its highest tier.

conviction_buyer census · 4 wallets + 1

Wash-trade veto

21.4% of pre-migration transactions are same-transaction wash trades, and early volume is the most contaminated volume a token will ever print. One cheap check strips it before any volume gate reads.

MELT dataset · 30.8M txs

Don't buy the insider exit

Coordinated snipers are 85% fully exited within 5 minutes of launch. Rival's entry window is placed to avoid being their liquidity — and its entry timestamps get audited against that distribution.

Pine Analytics · 15k tokens

Exit like the theory says

Optimal-stopping theory at memecoin volatility says: bank it. Arm-floor ladder (winners only scale out), hard stop, and a time stop — with fine-grained price paths recorded from entry for honest backtests.

Shiryaev/Xu/Zhou 2008 · sibling's Plumber fix

Pre-registered, or it doesn't ship

Every threshold gets its bar set before the data is peeked. The graveyard of this project is full of filters that looked great after the fact — the kill list is why this page exists.

house rules · KILL_LIST.md
★ Head to head

Rival vs the MacBook

Same metric, same window: paper fund trajectory and hit-rate by tier, compared over identical dates. No cherry-picking — whichever system's book is better after its confidence interval resolves, wins.

Scoreboard opens when both books have overlapping trading days.