PulseForge
Testing momentum beyond a rising price.
Look beyond the price move.
Seed 42 baseline · synthetic scenarios, not live-market returns
Overview
PulseForge explores whether short-horizon token momentum survives fees, AMM price impact, and delayed execution. A wallet-free baseline replays seeded synthetic events, while a separate dashboard observes confirmed Pump and PumpSwap activity and simulates trades.
Problem
A fast price move can be caused by a single large buyer, shallow liquidity, or a creator-driven hazard. Even broad buying may be too small to cover round-trip costs by the time an order could arrive.
Approach
Use deterministic scenarios to examine accounting and rejection logic before interpreting live signals. Separate event decoding, feature extraction, strategy decisions, and the paper broker so the cost model can be inspected independently.
How it works
The baseline updates token reserves and a bounded trade history, processes queued fills, evaluates exits, then considers new entries. Proposed size depends on cash and pool liquidity. Delayed fills use later reserves rather than the signal-time price.
The live dashboard validates canonical SOL-quoted pools from confirmed Solana RPC events. An optional official X API feed contributes social features without bypassing the on-chain gates.
- 01Confirmed Solana events
- 02Pool & mint validation
- 03Flow features & gates
- 04Latency-aware AMM fill
- 05Paper ledger
Strategy & methodology
The baseline requires token age, independent-buyer breadth, low concentration, positive 15-second net flow, and sufficient momentum above estimated costs. Exits include profit, loss, trailing, time, and creator-hazard rules.
The live paper engine implements Survivor, for fast graduation followed by a pullback and reclaim, and Momentum Scout, for already-active pools. Cooldowns, position caps, entry budgets, per-mint loss blocks, and session loss limits constrain simulated exposure.
Implementation
The root JavaScript replay has no third-party runtime dependencies. strategy.js maintains rolling features, amm.js models constant-product costs, paperBroker.js schedules fills, and engine.js coordinates the replay.
The TypeScript/React dashboard uses Vinext, a Next.js-compatible framework, with server routes for public RPC data and trade history. Drizzle defines the Cloudflare D1 paper-trade schema. The live engine includes dynamic fee tiers, delayed fills, and additional cost allowances.
Results & output
In the seed-42 synthetic baseline, four regimes generated two closed trades and 254 rejected entry evaluations. Net P&L was −0.000066259 SOL. Across 100 seeds, none of the synthetic runs was profitable.
These fixtures test strategy gates and accounting; they are not representative historical market data. Persistence smoke-test rows are excluded, and no verified live forward sample is presented as performance.
Deterministic synthetic tests, not historical market backtests. No verified live forward-test sample supports a live-performance claim.
Technical challenges
Signals must remain valid at fill time. Canonical-pool validation, re-evaluation after simulated latency, and creator/concentration filters address different ways a superficially attractive entry can fail.
Same-slot wallet grouping is a heuristic, not proven ownership clustering. The model does not reproduce validator ordering, transaction failures, sandwiching, or priority-fee spikes.
Takeaways
PulseForge demonstrates reproducible event-driven simulation and explicit execution-cost modeling. The negative synthetic result is useful: the winning trade did not offset the loss and the costs.