Docs/Core concepts/Accuracy & drift backtest

Accuracy & drift backtest

Our numbers measured against Birdeye, GeckoTerminal & DexScreener — published, not asserted.

We don't ask you to take "reconciled" on faith: we compared our numbers against Birdeye, GeckoTerminal and DexScreener and publish the drift as plain percentages — including the one number where we read lower, and why.

Methodology

Snapshot generated 2026-06-03 against one live mint — $COLLAT (C7heQqfNzdMbUFQwcHkL9FvdwsFsDRBnfwZDDyWYCLTZ), dominant Raydium pool resolved via DexScreener (DeAWLtRGAqCW7imV1jSC9NZkRKKGStFC1U7eDFBRxryR):

  • OHLCV close drift — 1h candles aligned by unix hour; drift = |ourClose − refClose| / refClose over the overlapping hours. Candles loaded: ours 423 · Birdeye 500 · GeckoTerminal 300.
  • Point-in-time — spot price, holders, liquidity and market cap / FDV compared at the same moment.
  • Volume — our 24h USD volume vs the vendors' USD figures only (Birdeye's OHLCV v is base-token volume and is excluded by design).

OHLCV close drift (1h candles)

ReferenceHours overlappedMedianp90Max
vs Birdeye2870.44%1.09%3.81%
vs GeckoTerminal1750.35%1.11%3.81%

Point-in-time comparisons

MetricOursReferenceDrift
Spot price vs Birdeye$0.001022$0.001011+1.07%
Spot price vs DexScreener$0.001022$0.001012+0.96%
Holders vs Birdeye12,43012,432−0.02%
Liquidity vs Birdeye$235,526$235,403+0.05%
Liquidity vs DexScreener$235,526$235,560−0.01%
Market cap / FDV vs DexScreener$1,021,605$1,012,738+0.88%

For scale: the reference vendors disagree with each other — Birdeye vs DexScreener liquidity differs by −0.07% on the same pool at the same moment — so a few hundredths of a percent is the noise floor, not signal.

Where we read lower: 24h volume

Source24h volume (USD)Drift vs ours
Ours (/metrics, window=24)$15,998
Birdeye (v24hUSD)$17,814−10.19%
DexScreener (volume.h24)$17,820−10.22%

Expected and honest, not a defect: we attribute volume from the pools we index while aggregators sum every venue; swaps we can't price to USD contribute $0 by design instead of a best-effort guess; and the candle tail can briefly lag the head between catch-ups.

Caveats

  • One mint. The backtest covers a single live token ($COLLAT) and its dominant pool — not a fleet-wide average.
  • Bounded windows. Birdeye overlap: 2026-05-14T02Z → 2026-06-03T19Z. GeckoTerminal overlap: 2026-05-21T00Z → 2026-06-03T02Z.
  • A snapshot, not a live feed. Generated 2026-06-03; it is not (yet) regenerated continuously.

Don't trust the snapshot either — verify any mint yourself: the per-candle confirmed flag (Intervals & candles) plus the per-mint coverage receipt tell you exactly which bars are settled, live.