Methodology
perpfuse normalizes venue data. It does not invent prices, synthesize L1, or cap funding rates.
Market data may be delayed, incomplete or unavailable. Verify execution conditions on the trading venue.
Funding APR
Hourly equivalent = interval rate / interval hours. APR = hourly × 24 × 365. Extreme values are stored raw and flagged ANOMALOUS. They are excluded from public ranking and opportunities.
Funding normalization
Unknown intervals are flagged FUNDING_INTERVAL_UNKNOWN. Units, signs, timestamps and stuck rates have dedicated flags. Nothing is clipped to a display cap.
Open interest
OI is converted with venue contract specs (linear / inverse / quanto). Cross-checks (OI vs 24h volume, 1m/5m jumps) only add flags. Values are not rewritten.
Volume and price
Prices and volumes come from official venue APIs. USD turnover is shown only when the quote units can be identified. OKX turnover is an estimate: base volume multiplied by the latest trade price, not a trade-weighted daily total. CoinGecko/CMC are not trading sources.
Stale state
Rows stay visible with last-known values. A STALE badge shows field ages for price, funding, OI and L1.
Data quality
Quality scores combine freshness, L1, funding, OI method and integrity flags. Quarantined canonical mappings do not enter compare-across-venues or opportunity scans.
Coverage
Public coverage shows venue, markets, price, funding, OI, L1, history and quality. Transport, hosts and reconnect internals stay in admin.
Canonical assets
Symbol similarity is not enough to merge tickers (AVA ≠ KAVA, B ≠ B2). Numeric prefixes such as 1000PEPE are not auto-stripped. Mappings below confidence 70 require manual review.
Net Carry
Capital is average USD notional per leg, not total margin. Both legs hold equal base units. Net APR = (expected hourly funding income / capital × holding hours − round-trip costs / capital) × 8760 / holding hours. Income weights each leg by its observed mark. Entry walks received ask/bid levels; hypothetical exit walks the reverse sides of those same books. VWAP already includes spread, so spread is not charged twice. Books older than 30 seconds, insufficient received depth, incompatible contracts or missing applicable fees make net carry unavailable. No depth is extrapolated. Published current taker fees or an explicit user fee profile apply; extra round-trip bps can model builder, transfer and exchange-specific costs. Zero extra costs is an explicit assumption. Exit cost and future funding are estimates, never a guaranteed return. APR uses notional, not leveraged return on deposited margin.
Break-even
Hours of the current hourly spread needed to recover entry + estimated exit. If the hourly income is not positive, break-even is not reported as a tradable figure.
Confidence
Every derived figure is tagged OBSERVED, ESTIMATED, DEFAULT or UNAVAILABLE. Backtests do not invent basis PnL without a price path. Historical L2 percentiles are UNAVAILABLE until that archive exists. Quoted, predicted and settled funding remain separate. A forecast error is unavailable until a forecast is archived against the same settlement timestamp. Zero is a valid observed rate. Missing data is not zero.
Realized funding and historical replay
1h, 8h, 24h, 7d, 30d and 90d windows sum recorded settled rates, not current quotes multiplied by time. Completeness checks observed history boundaries, gaps and settlement freshness against recorded intervals. A changed or unknown historical schedule can make coverage conservative. Partial sums are labeled; an annualized realized rate requires complete coverage. Replay uses actual settlements with equal fixed base units and aligned hourly closing prices. Missing settlement marks use preceding historical observations. Current fee schedules and observed half-spreads are cost scenarios; historical market impact, fee tiers and transfer costs cannot be reconstructed where absent. The net breakdown and chart use identical costs. Incomplete periods return unavailable.
Funding Stability Score · observed-v1
The reference scenario is $100,000 average notional per leg held for 24 hours. The score is the arithmetic mean of six components, each bounded to 0–100. It is unavailable until all six components exist, with at least 30 observations over five minutes. This is an observed descriptive score, not a probability of survival or a trading recommendation.
- Persistence = 100 × observations with available positive net APR / all episode observations.
- Volatility = 100 / (1 + EMA of absolute changes in gross APR / 0.10).
- Liquidity = min(100, 100 × observed executable size at 10 bps / $100,000).
- OI stability = 100 / (1 + EMA of absolute relative changes in the smaller leg OI / 0.05).
- Spread stability = 100 / (1 + EMA of absolute changes in summed leg spreads, bps / 5).
- Data quality = the lower leg quality score. EMA uses α = 0.1. Missing OI or depth leaves that component unavailable.
Episodes, signal history and alerts
The durable observed-v2 journal separates repeated appearances of a pair, keeps closed episodes and negative net estimates, and records sample averages and peaks. An observation gap longer than 75 seconds ends the episode at its last observed time. The journal begins at rollout; legacy sampled data is not presented as a complete performance track record. Financial outcomes and break-even hit rates remain unavailable without complete settlement and execution evidence. Lifetime statistics use closed episodes; the one-hour cohort includes closures and episodes observed for at least one hour. Alert conditions use AND logic. Continuous-match timers reset on a failed sample or a gap longer than 75 seconds; opportunity age does not satisfy this timer.
New Market Radar
Discovery records listings, interval and contract specification changes with old/new values. A missing contract requires at least three observations across 30 minutes before delisting; an empty discovery response does not delist the venue. Fee changes track updates to the reviewed published fee catalogue, with source and checked date. They are not live account-specific tariffs.
Market-neutral prediction strategies
Separate from perpetual carry. Binary complete sets buy YES and NO on the same market from live CLOB/order books. Multi-outcome complete sets require mutually exclusive exhaustive outcomes. Cross-venue hedges are labeled HEDGED and require AUTO_EXACT or MANUAL_VERIFIED rule match — operator, strike, observation time, and resolution source must agree. Titles are not enough. LOCKED requires executable equal size, accounted fees, and no book gap. Net edge is the ranking figure. Annualized equivalent is secondary. Missing books stay empty.
Opportunity gate
The public scanner publishes opportunities only after the opportunity_ready gate passes: seven continuous live days plus coverage and quality checks. Missing realized 7d/30d, persistence and replay stay empty until settlements are stored. Archive backfill does not replace continuous live observation.