When a crypto index constituent becomes unavailable or invalid, the index provider may exclude it, change the surviving weights or switch calculation methods. The published index can keep updating while representing fewer sources. Check the exact product methodology, active constituents and fallback state; a recent index timestamp alone does not show that its original market coverage remains intact.
A live composite can have degraded coverage
An index constituent is one input to a reference price. It might be an exchange's spot pair, a converted pair or another permitted source. The composite is a calculation over those inputs, rather than a directly executable market.
Three observations answer different questions: whether the provider publishes a new index value, whether each constituent supplies acceptable data, and whether enough independent market coverage remains for your decision. The first can pass while the other two deteriorate.
This differs from an on-chain report becoming old. An oracle's heartbeat and publication timestamp concern its update path; an exchange index can be fresh at the output while an input has been removed. The oracle staleness and deviation guide covers that separate report-age question. Here the concern is a changing calculation set.
This article uses official methodology pages reviewed on October 23, 2026 and a hypothetical calculation. It does not reconstruct an actual outage or describe a KLINEO index-monitoring feature.
Exclusion, clipping and fallback are different actions
Exclusion removes a source's contribution under the applicable rule. Clipping keeps an input but limits the price used in the calculation. Smoothing can transition a contribution over time. A fallback can change the method itself. Treating all four as “bad exchange removed” hides meaningful differences.
OKX's current global index methodology says sources under maintenance or without updates for a specified interval are omitted. With at least three valid sources it ordinarily uses preset weights and adjusts prices that exceed a defined median-deviation threshold. With two sources it ordinarily weights them equally; with one, that source supplies the index. The page does not give one universal timeout or numerical deviation threshold for every index.
Bybit's index calculation documentation separately describes excluding a spot pair after more than 15 minutes without trading and re-including it when activity resumes. Its abnormal-price protection can gradually reduce an excluded component's weight and redistribute it through smoothing. Those are different conditions: a quiet market is not necessarily a broken connection, and a source that updates can still fail an abnormal-price rule.
Do not apply Bybit's no-trade interval as a generic network timeout. Nor should an instantaneous renormalization worksheet be presented as a reproduction of its smoothing algorithm.
Isolate the effect of losing one input
Consider a fictional asset/USDT index with four sources. All prices are USDT per asset; starting weights sum to 100%. Assume the example provider excludes unavailable inputs immediately and renormalizes the surviving starting weights proportionally. There is no clipping, smoothing or quote conversion in this teaching model.
| Source | Price | Starting weight | Starting contribution | Weight after B is excluded |
|---|---|---|---|---|
| A | 100 | 40% | 40.00 | 57.1429% |
| B | 102 | 30% | 30.60 | 0% |
| C | 98 | 20% | 19.60 | 28.5714% |
| D | 101 | 10% | 10.10 | 14.2857% |
| Total | — | 100% | 100.30 | 100% |
Now remove B while holding A, C and D's prices fixed. The surviving starting weight is 40% + 20% + 10% = 70%. Divide each surviving starting weight by 70%; A becomes 40 ÷ 70 = 57.1429% after rounding.
The surviving contributions before normalization total 40.00 + 19.60 + 10.10 = 69.70. The new index is 69.70 ÷ 0.70 = 99.571428…. Relative to 100.30, the composition-only change is approximately −0.7264%. No surviving source moved.
Three useful facts belong in the receipt:
- The source count fell from four to three.
- Removed starting weight was 30%, leaving 70% of the original allocation represented.
- The largest effective source weight rose from 40% to approximately 57.14%.
These describe different aspects of degradation. Losing one of four sources does not necessarily mean losing one quarter of the original weight. And “70% retained” is an accounting description of the original weights, not a probability that the index is correct.
The missing B price should become unavailable in the post-exclusion record. Retain its old 102 only as historical evidence for this counterfactual; do not label it current or insert it into the new weighted average.
A second loss can change the formula branch
Suppose D also becomes unavailable, leaving A at 100 and C at 98. Proportional renormalization of their original 40% and 20% weights produces 99.333333…. Equal weighting produces 99.00.
Both calculations are arithmetically valid, but only the method specified by the provider can reproduce its published index. The two-source rule in the reviewed OKX page illustrates why an observer must record the formula branch, rather than extend a three-source weighting assumption automatically.
If no valid inputs remain, the normalized weighted average is undefined: its weight denominator is zero. Do not turn that into an index value of zero. Record an unavailable state and consult the product's documented halt or fallback behavior. An application also needs to distinguish “no new index published” from “the old value was returned again.”
Preserve a composition receipt
A chart of index values alone cannot establish why the calculation changed. Capture the evidence below together, marking anything the venue does not disclose as unknown.
| Receipt field | Question it resolves |
|---|---|
| Venue, product, index identifier and quote currency | Which calculation does this evidence describe? |
| Methodology URL, revision and effective time | Which rules applied at the event? |
| Output timestamp and local receipt time | Was the index observation current when used? |
| Constituent prices, timestamps and inclusion states | Which inputs were actually eligible? |
| Starting and effective weights | Was influence redistributed, replaced or smoothed? |
| Exclusion reason and formula branch | Was this inactivity, invalid data, deviation or fallback? |
| Dependent mark and order-trigger sources | Which account or order conditions used this reference? |
OKX's component-weight FAQ directs readers to its index detail pages for current weights. A composition captured after recovery cannot prove which weights applied during an earlier interruption. If historical component state is unavailable, keep that reconstruction limit explicit.
Use the market-data freshness clocks to separate event time from local arrival time. A delayed composition response and a newer index response are not a synchronized snapshot merely because the requests completed together.
Check what the index change actually affects
An index movement need not be a matching move in an executable bid or ask. Whether it affects a particular derivative mark, funding calculation or conditional order depends on that product's rules. Follow the mark-price and liquidation guide before translating an index observation into an account-risk claim. Use order-book depth for the separate execution question.
A calculation change can also be phase-specific. Bybit's reviewed page uses an estimated opening price for pre-market perpetuals during their call auction. The call-auction guide explains why a phase transition deserves its own event record instead of being misdiagnosed as constituent failure.
Finally, source count does not establish independence. Two venues may depend on the same conversion reference, and a local premium can persist when transfers are impaired. The arbitrage transfer-latency guide explains why moving inventory is a separate constraint; a price gap alone is not an executable opportunity.
For an application consuming the index, write a response policy before an incident: permitted source states, acceptable concentration, conditions requiring review, and evidence needed for recovery. Those are application choices, not universal exchange defaults. A resumed connection or a fresh output should clear only the conditions it actually verifies.