Methodology

Engine weights

The five 30-day engines below blend into one composite score at /api/thesis using this table. The dashboard itself no longer displays that blended score directly — each engine is shown independently instead — so these weights currently only matter if you query the API directly.

Data sources

• Hyperliquid Info API (api.hyperliquid.xyz/info) — funding, open interest, volume, mark/oracle price, spot supply, validator stakes, per-wallet delegation and positions, HIP-3 deployments. Public, no key required.
• Hyperliquid stats leaderboard (stats-data.hyperliquid.xyz) — full account leaderboard by value/PnL/volume, used to find whale candidates. Public, no key required.
• burn.a1ntel.com — Assistance Fund buyback activity and protocol fee history. Powers the Value Capture engine directly, as specified.
• hypurrscan.io — an independent, keyless Hyperliquid indexer used for the network-wide unstaking queue and top on-chain HYPE holders (neither is available from Hyperliquid’s own API at that completeness).

Next-hour momentum signal

The headline long/short call is deliberately separate from the 30-day thesis above — it does not enter the weight table and the two can disagree. It combines four real, live inputs from the Hyperliquid Info API: 15m/60m price momentum (1-minute candles), order book imbalance within 1% of mid price (l2Book), volume-weighted trade pressure from the last 15 minutes of candles, and current funding rate. The first three are trend-following; funding is scored as contrarian (crowded positioning tends to mean-revert). When they disagree, confidence drops rather than the model picking a side. Component weights are a published heuristic, not backtested — see the signal’s own assumptions for the exact figures. It is a read of public order flow and momentum only, not a guarantee.

We separately tried building a real 4-hour-ahead price-direction model (a logistic regression trained on historical price momentum, volume-weighted pressure, and funding) and backtested it before considering whether to show it here. It did not pass: out-of-sample accuracy on independent, non-overlapping 4-hour windows was 58.5% (55 correct of 94), but the 95% confidence lower bound on that figure is 48.4% — not statistically distinguishable from a coin flip at this sample size. Nothing based on it is shown anywhere on this site.

Glossary

Terms used across the engine pages that assume Hyperliquid-specific or derivatives-trading background:

APR

Annual Percentage Rate — the annualized reward rate a validator is currently projected to pay its delegators, before compounding.

Funding rate

A periodic payment between long and short perpetual-futures holders that keeps the perp price anchored to spot. Positive means longs pay shorts (bullish crowding); negative means shorts pay longs.

Open interest (OI)

The total notional value of all open perpetual-futures positions on HYPE right now — a proxy for how much leveraged exposure is active in the market.

Notional

The full dollar value a position controls, independent of how much margin (collateral) backs it — e.g. a $50k position at 10× leverage uses $5k of margin.

Leverage

How large a position is relative to the margin backing it. Higher leverage means a smaller adverse price move triggers liquidation.

Liquidation price

The mark price at which a leveraged position runs out of margin and is force-closed by the exchange.

Net exposure

Long notional minus short notional across tracked whale positions. Positive = tracked whales lean long in aggregate; negative = they lean short.

Concentration

The share of a total (stake, notional, listed assets) held by just the top 3 participants — a rough proxy for how centralized that layer is.

Unlock

The moment previously staked HYPE finishes Hyperliquid's 7-day unstaking queue and becomes freely liquid — i.e. transferable and sellable. A scheduled unlock is not a confirmed sale.

Indexed performance

Two assets' prices rebased to the same starting value (=100) at the start of the window, so their percentage moves can be compared on one chart regardless of each asset's actual price.

Assistance Fund (AF)

Hyperliquid's protocol-owned fund that uses collected fees to buy HYPE on the open market and burn it — the mechanism the Value Capture engine measures.

Basis / mark price

The mark price is the reference price derivatives settle against (blended from spot and the perp order book), distinct from the last-traded price.

Upcoming unlock dates

The Staking engine’s “upcoming unlocks” chart is built from hypurrscan.io’sunstakingQueue endpoint — the complete, network-wide pending-unstake queue with a real computed unlock timestamp per entry, not a sample. (We separately verified live, on a real wallet, that Hyperliquid’s own delegator-withdrawal “finalized” phase follows “initiated” at exactly +7 days — 604,800,229ms measured — confirming the underlying 7-day queue mechanics hypurrscan is reporting against.) A scheduled unlock is never presented as confirmed intent to sell.

Known gaps

This platform shows “insufficient evidence” rather than fabricating a number when the data doesn’t support one. As of this version:

• No public liquidations feed exists. Liquidation pressure is only inferred indirectly from open interest drops and funding extremes, and is never presented as a liquidation count.
• No confirmed public API for HyperEVM contract-level activity. The Ecosystem engine covers HIP-3 market launches and deployer concentration; HyperEVM activity, new collateral types, and deployer/treasury fee revenue are excluded rather than estimated.
• Staking, whale-position, and HIP-3 launch trends have no historical API. Hyperliquid only exposes current state for these — this platform’s own polling history is the only source of trend data, so these signals report “insufficient evidence” and mature in confidence as more polling cycles accumulate.
• Whale classification is a published heuristic, not a confirmed fact. Accounts are excluded from the whale set if registered as a vault or if their day volume exceeds 3× their account value (likely market-making/automated). This heuristic can misclassify.
• Fee history beyond 7 days is extrapolated, not measured. burn.a1ntel.com exposes granular fee data for a maximum of 7 days; the 30-day fee figure is the measured 7-day daily rate multiplied by the window length, always labeled as an estimate. Assistance Fund buyback totals, by contrast, are measured from full history, not extrapolated.