Atres edge engineMethodology

Methodology

How the engine's signals are measured and how every Premium research view is built. Everything here is non-personalised research and education — not financial advice. Past performance does not guarantee future results.

1. Data sources

  • Signals and the backtest use Binance USDⓈ-M futures market data: prices, volume, open interest, funding and candles. Entry and exit prices are Binance futures prices. Open-interest changes (in signal reasons, scores and the OI columns) are measured in contracts, so they move only when positions open or close; the USD value of open interest, which also moves with price, is shown only as a level.
  • Hyperliquid is used for funding and open-interest comparisons across venues, for Hyperliquid prints on the large-trade tape, and for the whale-wallet tiers (public on-chain data we collect live).
  • Bybit is used only for cross-venue funding and open-interest comparisons.

2. Signal lifecycle

  • A signal fires only with at least one concrete directional reason and, by default, not against the market regime (see 3). The regime at fire time is stored with the signal.
  • The entry is the Binance futures price at the moment the signal is published — the price you could act on — not the close of the candle it was measured on. A signal found more than one candle of its timeframe plus 60 minutes after that candle closed (a slow scan, or a restart after downtime) is not published: it is kept for research only and never counts in signal results or the backtest. While it is open, its move, its best move in its favour (MFE) and its worst move against it (MAE) are updated every 5 minutes — so MFE / MAE are sampled, not tick-exact.
  • A signal closes by rule: an adverse stop (5% against it), a target when one is configured (none by default), or a time limit per timeframe — 5m: 2 h, 15m: 4 h, 1h: 12 h, 4h: 48 h, 1d: 5 days, 1w: 14 days. Scanner states close when their state ends. The exit price and time, the exit reason and the final move are stored.
  • Data gaps. If a market's price is older than 3 minutes (configurable), no new signal fires on it. Its open signals pause: MFE / MAE are not updated and they cannot close by rule until fresh prices return, and the gap is recorded on the signal. The backtest shows how many signals had gaps and for how long.
  • Moves are measured in the signal's own direction; a risk warning (e.g. "move extended") counts as a position against the prior move.

3. Market regime

RISK-ON, NEUTRAL or RISK-OFF from five votes: BTC's 24-hour and 4-hour change, BTC's RSI on 1H and 4H, and the share of the 100 largest markets that are up over 24 hours. Two votes in the same direction set the regime.

4. Signal backtest

Hypothetical results based on past signals. Computed only from stored closed signals that fired inside the chosen window (7, 30 or 90 days) and had already closed at the time of the calculation, with the exits recorded at the time. Nothing is re-fitted or optimised and there is no look-ahead. Trades are equal size and not compounded; the equity curve is shown before and after the assumed fee. Win rate: share of signals that closed with a positive move. Expectancy: mean move per signal; after fee: minus an assumed 0.09% round-trip fee (no funding, no slippage). Max drawdown: the deepest peak-to-trough fall of the equal-weight sequence, in percentage points.

5. Large-trade tape

The largest prints in the most traded futures markets on Binance and Hyperliquid. A print counts as large when it reaches 0.003% of the market's 24-hour volume, between $25,000 and $2,000,000 (configurable, with per-coin overrides). The venue is shown on every print.

6. Hyperliquid whale tiers

  • Wallets are discovered from large Hyperliquid trades as they happen (both sides of every trade of $50,000 or more) and grouped by account value: Whale $10M+, Shark $1M–$10M, Dolphin $250k–$1M. Up to 500 wallets are tracked.
  • A wallet is admitted only if it traded in the last 30 days and is not excluded. Excluded: vaults (including HLP) and protocol accounts, known market makers on our list, and wallets that look like market makers — more than 400 large prints in a day, open positions in more than 25 markets at once, or daily turnover above 40× account value.
  • Performance starts from zero at admission. PnL in a tier is marked to market from the positions held at admission, at that moment's prices; nothing earned before admission counts. A tier's total PnL and win rate count every wallet that has been in the tier, including wallets that later left, moved to another tier or were excluded — not only the wallets still in it, which are shown separately. A position that closes or flips side counts as one closed position for the win rate. Moving to another tier, or leaving and coming back, starts from zero again.
  • A wallet leaves its tier only after its account value stays below 80% of the tier floor for 24 hours.
  • Positions are read every 5 minutes, so PnL reflects those readings (trades in between, fees and funding are not counted). Addresses are shown shortened. This is public on-chain data presented as research — not a recommendation to copy anyone.

7. Weekly Premium report

Generated every Monday at 06:00 UTC for the week before, entirely from engine data: the regime through the week, whale-tier positioning changes and the largest prints. It is the same for every member, contains no personal content and makes no buy or sell calls. An optional introduction may be written by the operator; the data sections are never edited. Each report carries a disclosure: who produced it, when, this methodology, and a conflicts-of-interest statement.

8. Limits

Signals can be wrong, late or based on incomplete data. Statistics change as markets change, and small samples can mislead. Results shown do not include your own fees, funding, slippage or position sizing. Research and education only — not financial advice.