Sources & methodology
How WalletFollow wallet data is measured
WalletFollow separates venue observations, retained executions and derived analytics. Each result needs a defined account scope, period and measurement basis. Market-data timestamps describe the observations; coverage describes the evidence available. Unknown values remain unknown, and modeled portfolio results remain separate from actual follower execution.
Where wallet observations come from
Public wallet profiles can combine Hyperliquid account reads with retained history and prepared analytics. The Data API serves prepared wallet documents rather than making a new venue lookup for every customer request. A retrieved document can therefore be available now while its underlying observation was collected earlier.
Data API sections identify their source. Account state, positions and spot holdings come from prepared account observations; discovery metrics come from prepared analytical rows; extended analysis comes from a prepared wallet overview. Preserve the source and section timestamp when exporting or comparing results. Derived measures should not be described as a direct venue field unless they are one.
Freshness and coverage answer different questions
An account as-of time records the relevant state observation. A valuation time records the price capture used for priced values. A document-build time records preparation of the response. The API exposes section-specific freshness information because these times can differ. An editorial update date belongs to an explanatory article, not to the underlying market data.
Historical coverage is specific to a requested window and calculation. Current positions can be available while older fills or a starting valuation are incomplete. A null or unavailable value is not zero. Read the associated gap reason, and check truncation indicators before assuming a returned list contains every row.
Account value and PnL use stated bases
Total wallet value is an aggregation under a stated account-mode and valuation rule. Unified and portfolio-margin accounts require care to avoid counting collateral and perpetual equity twice. Spot assets need usable valuation inputs, and excluded or unpriced holdings should remain distinguishable from a zero balance. A value marked as an estimate should be cited as an estimate.
Realized fills, completed-position results, unrealized PnL and mark-to-market period change are different views. Fees, funding, cash flows and venue scope can explain apparent disagreements. Return percentages also need their capital denominator. Align these definitions and timestamps before comparing two wallets or two trackers.
Trade grouping and rankings have limits
One order can produce multiple fills; one position can have partial reductions before a full close. Fill count, completed-trade count and decision count are therefore different units. Win rate and trade-duration measures depend on the grouping rule and the available sample. A score or style label describes eligible observations under a method, not a verified identity or future performance.
Discover comparisons depend on the selected period and eligible data. A high dollar gain can reflect more capital, while a large percentage can rely on a small base. Read coverage and sample warnings alongside rankings. Public wallet activity may omit the owner’s other addresses, external hedges or motivation.
Copy-portfolio records are separate evidence types
A backtest reconstructs a rule on historical inputs. Tracked portfolio statistics describe a published version under its stated method. Your executed follower result comes from your account’s fills, costs and cash flows. New portfolio weights should not receive credit for performance earned before their effective publication. Read version boundaries and basis labels instead of treating every plotted point as live copying.
Weights, cash handling, costs, reinvestment and risk settings affect modeled results. Actual followers additionally face startup prices, rounding, liquidity, delays and skipped trades. A pause can leave existing exposure, a stop can leave unmanaged positions and a close can leave execution residuals. Confirm engine state and actual positions rather than inferring completion from a requested command.