Sampling and median
How the Kimp Index is sampled in 60-second epochs, aggregated by median, finalized at quorum, and guarded against large single-epoch moves.
The Kimp Index is finalized in discrete 60-second epochs. Within each epoch, reporters submit independently and the KimpIndex contract aggregates their reports with a median. This page defines the epoch, the aggregation rule and the guards around it.
Epochs#
| Parameter | Value |
|---|---|
| Epoch length | 60 seconds |
| Epochs per day | 1,440 |
| Reports per reporter per epoch | One per asset |
| Quorum | 5 valid reports |
| Target active set at launch | At least 7 reporters |
| Aggregation | Median of valid reports |
| Circuit guard threshold | 300 bps from the previous finalized value |
| Staleness threshold | 5 minutes without a finalized value |
Epochs are aligned to the minute. Each epoch produces at most one finalized value per asset.
Valid reports#
A report is valid if all of the following hold:
- It is signed by a reporter in the active set of KimpReporterRegistry.
- It is for the current epoch and a listed asset.
- It is the first report from that reporter for that asset in that epoch.
- The reporter is not suspended.
Invalid reports are rejected by the contract and do not count toward quorum.
Median#
Once at least 5 valid reports are present, the finalized value is the median of all valid reports for the asset in the epoch.
- With an odd number of reports, the median is the middle value.
- With an even number of reports, the median is the mean of the two middle values, rounded to the nearest whole bps, halves away from zero.
Example with 7 reports for BTC, in bps: 245, 247, 248, 248, 249, 250, 262. Sorted, the middle value is the fourth, 248 bps. The 262 report deviates from the median by 14 bps, below the 25 bps flag threshold, and has no effect on the result.
The median tolerates up to half of the reports being wrong in the same direction without being moved outside the range of honest values.
Missed quorum#
If fewer than 5 valid reports arrive in an epoch, the epoch is not finalized. The previous finalized value is carried forward and the epoch is marked stale in the event data. Stale epochs do not count as settlement samples. See Settlement TWAP.
Staleness and reduce-only#
If an asset has no finalized update for 5 minutes, its market enters reduce-only mode. No new exposure can be opened or increased. Positions can still be closed. Normal trading resumes automatically at the next finalized update.
Circuit guard#
A single epoch can move the index by any amount if the market moves. Large moves receive extra scrutiny.
If an epoch's median moves more than 300 bps from the previous finalized value, it is finalized only if it is supported by reports from at least two thirds of the active set, rounded up. With an active set of 7, that is 5 reports. With 9, it is 6.
If that stronger quorum is not met, the value is held for one more epoch and the previous value is carried. In the next epoch the move is evaluated again against the previous finalized value. If the new median still moves more than 300 bps and meets the standard quorum of 5, it is finalized. A genuine market move is therefore delayed by at most one epoch, while a transient or erroneous spike from a partial reporter set is filtered.
Events#
Every finalization emits an event with the asset, the epoch, the finalized value in bps, the number of reports, and flags for stale epochs, circuit-guard holds and carried FX inputs. Interfaces and indexers read these events to display history and data quality.
Preconfirmation#
Finalization transactions are included in GIWA's 1-second blocks. Interfaces read them through Flashblocks preconfirmations, up to about 200 ms after submission, to display a new value before the block closes. See Flashblocks usage.