Method, definitions and limits
Written since 22 Sept 2026: 55 reads, and 10 of 10 sources answered the last one. Every number on this site has its definition here; a number without one is a mood.
Reads in the book
Sources that answered
Thresholds watched
beacon litWords defined
What each source contributes
Each source covers a different part of the market, and says nothing of the rest. Last night, every one of them answered.
Metric definitions
The selection rules and calculations used across the site, written once and kept. Each word has its own anchor: a page that uses it can point here.
- Persistent cohort
- Accounts with more than $500,000 in equity, positive PnL over 7 days, 30 days and all time, and all-time trading volume divided by equity between 0.5x and 200x. Membership is recalculated daily; this is a selection rule, not evidence of skill.
- Turnover flags
- “Low turnover” means 30-day trading volume below 1x account equity. “High turnover” means at least 100x. These are activity measures: they do not establish whether an account is a bot, market maker or discretionary trader.
- Funding and z-score
- Hyperliquid’s hourly funding is annualised as rate × 24 × 365 × 100, without compounding. Positive rates mean longs pay shorts; negative rates mean shorts pay longs. The z-score compares the rate with up to 90 days of prior snapshots and requires at least 10 observations. Funding reference.
- Cohort gross leverage
- Total absolute USD exposure divided by account equity for sampled accounts with open positions. It measures exposure relative to equity, not a common liquidation price. Margin mode, hedges and collateral affect the risk of each account.
- Open interest
- The value of outstanding perpetual positions. Total open interest includes the main Hyperliquid order book and available builder DEXs. It is not trading volume or deposited collateral.
- Underwater accounts
- Accounts with negative all-time PnL as reported by the leaderboard. The share counts accounts equally, regardless of size. It does not describe today’s return or the group’s aggregate PnL.
How to read the colours
Few colours, each with one meaning. Green and red belong to profit and loss alone; gold marks a threshold crossed; everything else stays in plain ink.
The limits of the read
The Eye sees far, not everywhere. The scope of the data limits what can be inferred from it.
A sample of one venue
The public leaderboard and selected position probes do not cover the entire market. Other venues and accounts outside the source dataset are absent.
An account is not a person
One person may control several accounts; one account may represent a fund or automated strategy. Addresses alone do not establish identity or intent.
History starts with collection
The archive begins with our first snapshot. Source outages can leave gaps; a missing observation is not replaced with an estimate.
Behavior: winners, losers and account size
Two observational cohorts of BTC, ETH, SOL and HYPE perpetual positions on the canonical Hyperliquid DEX: the Captains, watched trade by trade, and the Host, read by its positions. Six rules, in the order they apply.
- I
Who the Captains are
Up to 64 accounts: eight winners and eight losers in each equity band ($10k–100k, $100k–1M, $1M–10M, $10M+). Winners have positive 7-day and 30-day PnL at admission; losers have negative PnL on both. Mixed results are excluded. Selection takes evenly spaced equity ranks among active leaderboard accounts within each stratum. Membership, performance labels and size bands are frozen at admission, including accounts that later lose money or move to another size band.
- II
How the watch is kept
Collection runs every five minutes. The first observation establishes a baseline, not an entry event. Executions identify opens, additions, reductions, closes and reversals, including round trips between position snapshots. Missing responses and incomplete execution histories create explicit gaps. Summaries include only accounts covered throughout the selected window and value signed quantity changes at one common current price. A net buyer may be covering a short.
- III
What counts as unusual
Habit comparisons use earlier events for the same account and asset over 28 days, scaled by equity observed at collection. At least 20 events spanning seven days are required. An unusual event exceeds twice the median and the 95th percentile; these are descriptive initial thresholds, not validated trading signals. Orders spanning collections may produce multiple events. Liquidations are labelled and excluded from unusual-action flags.
- IV
What is measured afterwards
Prices are saved at detection and measured again after 1h, 4h and 24h, with a maximum delay of ten minutes. Returns are market returns, not executable strategy returns. Events and outcomes are retained; detailed position snapshots are retained for 35 days. Panel addresses can belong to the same trader, and positions may be hedged elsewhere. No predictive accuracy is claimed.
- V
Who marches in the Host
Every leaderboard account the Captains’ rules would admit: at least $10k of equity, 7-day and 30-day PnL of the same sign, some volume over 30 days. Up to 6,000 accounts, the largest first. Unlike the Captains, the Host is drawn again every day: its winners and losers are the current ones, and an account can change camp or band from one day to the next.
- VI
How the Host is read
Too many accounts to read their executions: each one is read by its open positions, once every thirty minutes, least recently read first. A flow is the change of a position between two reads less than two hours apart, valued at the price of the second read; a position opened and closed between two reads is not seen, and a liquidation counts as a sale. Over a window, changes are netted per account before buyers and sellers are counted. Seven days of flows and positioning are kept.
