Check completeness first
For Binance and OKX public snapshots on BTC, ETH and SOL, calculate expected observations, successful observations, gaps and longest outage. Do not compare averages when coverage is inadequate.
Preserve venue, contract, rate unit, observed time, fetched time, next settlement and source URL. Failed attempts remain separate and never overwrite the last successful snapshot.
Core metrics
Report median, mean, 10th and 90th percentile, maximum positive, minimum negative, positive/negative interval share and longest same-sign run for each market.
Median describes a typical state while quantiles and extremes reveal tails. A monthly mean can hide a few spikes among many near-zero intervals. Cross-venue differences require observations from aligned windows.
Report field specification
A monthly review that others can check needs its fields fixed in advance. The table lists what every market and venue should produce, and what question each field answers. A missing field is worse than an unflattering number, because the reader cannot judge how far the conclusion extends.
The first three rows are data-quality fields and belong above the statistics. When coverage is short, every average and percentile below them describes an incomplete sample and cannot carry a cross-venue comparison.
| Field | Question it answers |
|---|---|
| Expected vs successful observations | Is the sample complete |
| Longest continuous outage | Are gaps scattered or concentrated |
| Failed attempts and error types | Is the fault upstream or in the sync |
| Median | The typical state |
| Mean | The overall level once outliers count |
| 10th / 90th percentile | The bounds of the normal range |
| Largest positive / smallest negative | Tail strength |
| Share of positive intervals | Directional skew |
| Longest same-sign run | Whether the skew persists |
The step-by-step procedure
Write the review as a repeatable procedure rather than a one-off analysis. Run these seven steps in the same order each month, and when a step fails, stop there and record why instead of skipping ahead to a conclusion.
The constraint in step seven is the one most often dropped. The review describes observed data, not next month's expectation. The moment it starts saying 'will' or 'is expected to', it has become a forecast, and forecasts need an entirely different standard of validation.
- One: fix the data cut-off, and record generation time and formula version.
- Two: compute coverage per market and venue, marking anything below threshold as not comparable.
- Three: compute median, mean, 10th/90th percentiles and extremes separately.
- Four: count the share of positive and negative intervals and the longest same-sign run.
- Five: convert the median, the 90th percentile and the actual interval series into funding amounts on a fixed 10,000 USDT position.
- Six: list every official source URL with its check date.
- Seven: describe only observed data, publish no directional forecast, and link back to the funding calculator so readers can recompute on their own position.
Convert observations to holding scenarios
For a fixed 10,000 USDT position, convert the monthly median, 90th percentile and actual interval series into funding cost. The series is closest to a historical ledger; fixed quantiles support stress testing.
State direction clearly: positive funding means longs pay and shorts receive, with the reverse for negative. When frequency changes, count actual intervals instead of assuming three per day.
Publication rules
Attach data cutoff, generation time, gap notes, formula version and official sources. If either venue lacks sufficient coverage, retain the data but withhold a comparative conclusion.
Describe observed data only; do not claim next month will rise or a token is best to long. Link back to the calculator so readers can apply their own position.
Official sources and calculation boundary
OKX's funding FAQ defines what the report is counting. This article offers a method rather than a conclusion. The thing most worth checking is therefore not any single figure, but whether your own sample coverage can carry a comparison at all.
Next checks in this series
Funding rates
Long funding cost over three intervals, seven days and thirty days
Reproducible scenario guide
Funding rates
Who pays when perpetual funding is negative: longs or shorts?
Reproducible scenario guide
Liquidation and margin risk
BTC at 5x, 10x, 20x and 50x: how liquidation risk changes
Reproducible scenario guide
Frequently asked questions
Why not rank annualized funding?
Annualization magnifies a short snapshot and assumes persistence. The monthly report prioritizes the actual interval distribution.
Can the monthly average predict next month?
No. It only describes the observed window; market structure and settlement rules can change.
Why emphasise the median over the mean?
Because funding distributions are usually spiky. A handful of extreme intervals can lift a monthly mean while most of the period sits close to zero. The median describes the typical state and the mean describes the level once tails count; both are needed. Reporting only a mean invites readers to misjudge ordinary holding cost.
How much coverage is enough for a cross-venue comparison?
There is no universal threshold, but it should be defined and published in advance. The point is to fix it before seeing the results, otherwise it becomes sample selection. Below the threshold, keep the data, mark it not comparable, and draw no conclusion — which is far more honest than adjusting the threshold afterwards.