DarkTrade Intelligence
Every number on the Divergence Index comes from public Hyperliquid on-chain data through the fixed procedure below. Nothing is hand-edited; when the data is too thin to trust, the index says so instead of guessing.
For each coin we sum the USD value of long positions and short positions held by a cohort, then compute net bias = (long − short) ÷ (long + short), expressed as a percentage. +100% means the cohort is entirely long, −100% entirely short, 0% perfectly balanced. This normalisation makes a $40M cohort comparable with a $400M one.
The field is every large Hyperliquid account our collector risk-vets and tracks that day — the informed crowd.
The elite is the subset that passes our selection engine: account value ≥ $50k (L1), live trading activity that day (L2), a velocity cap that removes market-maker churn — daily volume must not exceed 10× account value (L3), a liquidation-distance check that drops accounts one wick from ruin (L4), and a composite score weighted toward sustained 30-day and 7-day return on equity plus volume and account size (L5). A wallet must rank into the top set for three consecutive days before it counts as elite, and short gaps are bridged by a grace period — the cohort cannot churn day to day.
divergence_score = elite_net_pct − field_net_pct in percentage points, from −200 to +200. The score is withheld (shown as “n/a — thin sample”) whenever fewer than 3 elite wallets or 10 field wallets hold the coin. Sample sizes are printed beside every row so you can judge the read yourself.
A call is recorded on any day a coin's |divergence score| reaches 40 points. It resolves at the first snapshot seven or more days later: a win if the price moved in the direction of the elite's relative lean, a loss otherwise. Every call is shown — a persistent divergence produces several calls, and losing streaks are displayed exactly like winning ones. A track record that can't show losses isn't a track record.
Data is pulled from Hyperliquid's public API twice daily (06:00 and 18:00 UTC); the later snapshot becomes the day's consolidated row. History is kept for roughly 400 days, and any “highest in N days” style comparison uses the number of days actually observed — never a nominal window the data doesn't cover.
It is the elite cohort’s net bias minus the broader field’s net bias, in percentage points. Net bias for a cohort in a coin is (long exposure − short exposure) ÷ gross exposure, so it runs from −100% (everyone short) to +100% (everyone long). A divergence score of +80 means the elite are far more long than the crowd; a negative score means they are more short. The score says nothing about direction being right — the scoreboard on the index page tracks that separately, wins and losses alike.
A wallet enters the elite set only by passing our daily selection engine on Hyperliquid’s public leaderboard: a minimum account value of $50,000, real trading activity that day, a velocity cap that removes market-maker-style churn (daily volume above 10× account value), a risk check that drops accounts within 2% of liquidation, and a composite ranking weighted toward sustained 30-day and 7-day return on equity rather than one lucky day. A wallet must qualify for three consecutive days before inclusion, so the set cannot whiplash daily.
All risk-vetted large accounts we track that hold a position in the coin that day — typically hundreds of the biggest Hyperliquid wallets — including the elite subset. The field is the informed crowd; the index asks where the proven performers lean differently from it.
Snapshots are collected twice daily (06:00 and 18:00 UTC) from Hyperliquid’s free public API — the same on-chain position data anyone can query. One consolidated row per coin per day is stored; the page always shows the latest day. No private data, no exchange partnerships, no manual edits.
Four honest ones. First, scores are withheld when fewer than 3 elite or 10 field wallets hold the coin — a thin sample is noise. Second, the index covers Hyperliquid perpetuals only; a whale’s spot or off-exchange hedges are invisible to it. Third, history depth is limited to what we have observed since the index started, and any "highest in N days" claim uses the actual observed window. Fourth, large accounts are often wrong — the scoreboard exists precisely to keep that measurable.
See it live: today's Divergence Index · daily analysis on the Whale Watch hub. Market observation, not financial advice. Large accounts are often wrong; position sizes here would be reckless at retail scale.