Backtesting Studio

Replay a strategy on real Pons V2 history from Robinhood Chain: curve-exact fills with fees and creator tax, walk-forward and purged cross-validation, and paper trading through the same code. Every result is labelled BACKTEST or PAPER.

BACKTESTSimulated on recorded Pons V2 history from Robinhood Chain (4663). Nothing here is live performance, and these results are never combined with live or PAPER results.

1 · STRATEGY

Live agent preset · balanced

Replays the live agent's own code: verdict() (every rug-guard rule and the score), sizeFor() and exitDecision() from @intelo/agent, at the live cadence. Checks that cannot be replayed from history are listed under assumptions below.

MAX PER TRADE (ETH)sizeFor() caps every buy at thisMAX POSITIONSopen positions at onceMAX EXPOSURE (ETH)total cost basis across open positionsMIN SCOREguard score needed to buyMIN AGE (S)seconds since launchMIN BUYERSdistinct curve buyersMIN PROGRESS %curve progress window, low endMAX PROGRESS %curve progress window, high endMAX TOP HOLDER %largest wallet's share of circulating supplySTOP LOSS %exitDecision stop-lossTAKE PROFIT 1 %first take-profit triggerTRAILING %trailing stop off the peakMAX HOLD (MIN)stale exit when gain < 10%SLIPPAGE %orders that would move more than this are rejected

2 · HISTORY (ROBINHOOD CHAIN)

reading chain…

FROM BLOCKTO BLOCK

RPC (OPTIONAL: A PAID ENDPOINT HAS HIGHER GETLOGS LIMITS)

3 · RUN BACKTEST

BACKTEST

STARTING ETHLATENCY (MS)decision to execution; history in between moves the priceGAS PER TX (ETH)DECISION STEPSNIPE WINDOW

fetch history first

ASSUMPTIONS AND LIMITS

read before trusting a number

· Fills use the curve's constant-product maths with its fee and creator tax, read from the events; reserves are recovered from each curve's own trades (verified by replaying them) and simulated forward. Our own trades' impact stays on the curve for later fills; recorded trades by others keep their amounts.
· Orders execute 2s (configurable) after the decision and are rejected beyond the slippage limit.
· No entries in a token's first 15s (snipe tax), measured conservatively.
· Positions still open when a curve graduates continue on the token's Uniswap v4 pool when the history holds every swap of that pool (priced on its virtual reserves at the last swap, LP fee included, hook fees and range crossings not modelled); otherwise they are written down to 0 with the reason.
· Positions open at the end of the range are marked at their curve liquidation value and labelled open-at-end.
· Only tokens launched inside the range are simulated: older tokens' buyer and holder history is not in the data.

live check	live	backtest
Sell simulation (honeypot)	eth_call of a real sell from the agent's address with storage overrides	Not replayable (needs historical state). Pons V2 curves sell through the curve contract's fixed code, so the sell is modelled with the curve maths; it passes when the curve holds enough real ETH.
Buy simulation and hidden buy tax	eth_call of the buy, compared with the curve formula at the posted fee + tax	Curve maths at the reconstructed reserves. Hidden tax is measured on the latest recorded buy (tokens received vs. posted-fee output) inside the snipe window, and taken as 0 after it.
Top holder	largest wallet's net curve position from CurveFlow (Pons tokens)	Same computation over the recorded CurveBuy/CurveSell events; circulating = totalSupply − token reserve.
Creator holdings	balanceOf(deployer) read at check time	Approximated by the creator's net curve purchases (bought − sold); transfers outside the curve are invisible without archive state.
Grok veto / boost	optional xAI read on cleared tokens	Skipped: not reproducible. Scores are the deterministic guard score only.
DEX (Uniswap v2/v3) tokens	DexScreener discovery plus pool checks	Out of scope: the dataset is Pons V2 curve history only.
Daily loss pause	autopilot pauses until re-armed (local midnight)	sizeFor blocks entries until the next UTC day.
Gas	measured per transaction	Fixed per transaction (engine setting), plus one approval after each buy.

PAPER TRADING (FORWARD TEST)

PAPER stopped

Feeds new Pons launches and curve trades, polled every 5s, through the same strategy and engine as the backtest, after replaying the last 30 minutes without trading (warm-up). No transaction is ever sent. The same events always give the same decisions as a backtest over them.

PAPER trade	strategy	entry (UTC)	in / out ETH	P&L	outcome
no PAPER trades recorded in this browser yet