Manifesto

We got tired of watching tokens fail for reasons that had nothing to do with the token.

A project ships. The repository is active, the team is real, the idea holds up. Then a KOL who never read the whitepaper decides the timing is wrong and posts FUD. A sniper bundle from launch week starts unwinding. The top-10 holders are too concentrated and one of them panics — which makes two more panic — and the arc collapses before organic demand had a chance to find it. The idea wasn't the problem. The structure was.

Supply distribution. LP lock status. Dev wallet size relative to float. Holder concentration at the top. These are not unknowable variables. They are visible on-chain, and they determine how a token responds to pressure. A project with a tight top-10 and an unlocked LP does not need bad news to die — it just needs one nervous wallet.

The thing is, most of this is fixable before launch. Or at least, it is diagnosable. You can look at a contract and say: this supply structure will produce a panic cascade the moment sentiment dips. This LP size cannot absorb the sell pressure that will come when the snipers exit. This holder concentration means three wallets can determine the outcome for everyone else.

We built chumsim because we wanted to see the simulation before we committed. Paste a contract address. 300 synthetic agents — snipers, KOLs, paperhands, diamond hands, rug-sniffers, normies — seeded from real on-chain data, play out the next 24 hours. You see where the arc breaks. You see which archetype tips the cascade. You see what a tighter supply or a locked LP actually changes about the outcome.

It is not a prediction machine. A simulation is not the market. But it is a rehearsal room — a place to understand the dynamics you are actually dealing with before you decide how to deal with them, or whether to deal with them at all.

We built it for the same reason anyone builds a tool: because we needed it and it did not exist. If you are a builder, use it to stress-test your supply structure before you launch. If you are a buyer, use it to understand what you are actually buying into. If it helps you avoid one bad position or fix one structural flaw, it did its job.

Manifesto Simulation Technology Engine GitHub X
CA: 856WSpqa2AMcfwo375mzNx8QwFSiD6FXxni5XV86pump
kol
sniper
diamond
rugsniffer
paperhand
normie
ready
No simulation running.
Paste a contract address above and run, or click Demo for a synthetic simulation.
Technology

Rehearse the trench before you enter it.

Paste a contract address. 300 AI agents — snipers, KOLs, paperhands, diamond hands, rug-sniffers, normies — seeded with real on-chain data, simulate the next 24 hours and return a verdict.

From contract address to prediction

Every simulation runs the same pipeline. On-chain data drives the agent population; the swarm does the rest.

01
Seed from on-chain reality
Helius RPC + DAS pulls holder distribution, dev wallet %, LP size and lock status, mint and freeze authority, token age, and recent swap velocity.
02
Map signals to a crowd
High top-10 concentration raises rugsniffers. Fresh + high velocity raises snipers and KOLs. Unlocked LP lowers the paperhand threshold. Signals set population counts directly.
03
Run the MiroFish swarm
300 LLM-driven agents interact in a synthetic environment for 24 simulated hours. Each archetype has distinct behaviour, language, and reaction thresholds.
04
Score the arc
Emergent sentiment is scored at each step. Send and rug probabilities are derived from the sentiment arc weighted against on-chain risk penalties.
05
Record and resolve on-chain
Every prediction is hashed and stored on Solana. After the resolution window, the outcome is marked and the accuracy leaderboard updates — hits and misses both.

Six roles. One trench.

Each agent is seeded with a named persona, a follower count, and a disposition shaped by on-chain signals.

KOL
Influential callers. Posts move markets. They shill what they hold, never mention risk, followers ape in blindly.
Cupsey · Rambo · Ansem · Hsaka · Murad
Sniper
On-chain bot operators. In within seconds of launch, out on the first stall. Pure price action.
0xViper · BundleKing · MEVGhost · JitoSniper
Diamond Hand
Bought early, not selling. Dismisses all FUD. Dips are just an opportunity to average down.
NeverSelling_Chad · WAGMI_Dave · ZoomOutZach
Paperhand
Anxious at the first red candle. Panic sells, FUDs loudly to justify it, goes quiet when it recovers.
PanicPete · RektRon · CopiumCarl · ShookSusan
Rug Sniffer
On-chain security analysts. Cite block explorers, check LP locks, flag dev wallets, post cold receipts.
OnChainSleuth · BubbleMaps_Rex · RugRadar
Normie
Found crypto on TikTok. Follows KOLs blindly, buys after the 10x, asks "wen moon?" in every group.
WenMoon_Gary · JustBoughtTop_Jenny · PumpFun_Phil

What shapes the population

Direct mappings from Helius on-chain data to agent weights. The simulation responds to what is actually on-chain.

SignalEffect on swarm
High top-10 concentrationRugsniffers +20, baseline trust lowered, normies more nervous
Dev holds large %Rugsniffers primed, paperhands +10, rug prior elevated
Mint authority liveRisk flag added, paperhand jitter +10
LP small or unlockedPaperhand trigger threshold drops, volatility multiplier up
Old token, steady velocityDiamond hands +15, normies +20, calmer arc
Fresh (<6h) + high velocitySnipers +25, KOLs +8, paperhands +10, spiky arc
Freeze authority activeRisk flag, rugsniffers cite it in the first steps
Engine

Chum Engine

Architecture, API reference, build phases, and deployment guide. All code is open source.

jconstantine627752-maker / Chum
Components
A
Swarm Engine
Runs the MiroFish simulation. Accepts a SeedBundle, spawns agent personas, drives 24 simulated hours, returns Report JSON.
Python · FastAPI · camel-oasis · Redis
B
Seeding Pipeline
Pulls on-chain data via Helius RPC + DAS. Maps signals to agent population weights. Generates the situation brief.
Python · Helius · httpx
C
On-chain Program
Access gating, credit accounting, fee routing (burn / pool / treasury), immutable prediction log, resolution oracle.
Rust · Anchor · Token-2022
D
Transfer Hook
Token-2022 hook that routes 0.5% of every transfer to the shared credit pool vault. Fail-open by design.
Rust · Anchor · spl-transfer-hook-interface
E
Blink / Actions
Solana Actions endpoint. Paste a CA in a tweet, sign a transaction, get a report back — no frontend required.
TypeScript · @solana/actions · Express
F
X Bot
Auto-sims trending tokens, posts verdict threads. Resolution cron posts outcome receipts — hits and misses both.
TypeScript · twitter-api-v2 · node-cron
API
POST
/simulate
Body: { mint } or { narrative, name }. Returns Report JSON. Cached by seed hash.
GET
/report/{seed_hash}
Fetch cached report by blake2b seed hash. Returns 404 after TTL (default 10 min).
GET
/health
Engine status and pinned MiroFish commit SHA.
Build Phases
0
Engine proof — MiroFish fork, SeedBundle → Report JSON
1
Seeding pipeline — Helius pulls, signal→population mapping
2
On-chain core — credits, fee routing, prediction log
3
Blink MVP — paste CA in tweet → sign → report
4
X bot + resolver — auto-posting, accuracy leaderboard
5
Transfer hook — Token-2022, 0.5% per trade to credit pool
6
Web dashboard — leaderboard, report explorer
Token Economics
Hold ≥ threshold → daily free credits
Pay-per-query → burn / credit pool / treasury split
Transfer hook → every trade tops up the pool
Public accuracy leaderboard — hits and misses
Oracle transparency: the backend key writing predictions is a centralization point, documented openly. Decentralization path: multi-signer commit-reveal post-launch.