What Is Price Accuracy in Crypto? The Accuracy Triangle Explained
Almost every conversation about crypto market data is about price. Very few are about whether that price is accurate.
That distinction matters more in crypto than anywhere else. In a market without a consolidated tape, "the price" is an estimate, and different sources produce different estimates at the same moment. Some sit close to the true market. Some do not. Without a way to judge the difference, a price is just a number someone published.
This article sets out what price accuracy means, a framework for assessing it, and how to measure it.
What is price accuracy?
Price accuracy is the degree to which a price reflects the true, contemporaneous fair value of an asset across all the markets where it trades.
Three words in that definition carry the weight. True means the price reflects genuine trading activity, free of errors, stale data or manipulation. Contemporaneous means it reflects the market at the moment the price is used, and not several seconds or minutes earlier. Across all markets means it represents the whole market instead of a single venue. In crypto, where the same asset trades on dozens of exchanges at different prices simultaneously, this is the part most often missed.
The last of the three has a long academic pedigree. Joel Hasbrouck's 1995 paper in the Journal of Finance, "One Security, Many Markets", established the framework for measuring how much each of several venues contributes to discovering an asset's price when it trades in more than one place at once. The finding that matters here is the premise itself: where a security trades on many markets, no single market's price is the price. Crypto is that condition taken to its limit.
A price can fail on any one of the three properties and still look perfectly reasonable on a screen.
The Accuracy Triangle: breadth, depth and latency
Price accuracy has three dimensions. Together they form what we call the Accuracy Triangle.
Breadth: how much of the market is captured
Breadth is the share of the market a price actually reflects. A price taken from one exchange reflects that exchange's order book and nothing else. A price aggregated across many venues, centralised and decentralised, reflects much more of where trading genuinely happens.
Breadth matters because no single venue is the market. Even the largest exchange handles only a fraction of total volume for any given pair, and any venue can temporarily detach from the rest through a regional imbalance, an outage, a thin order book, or a single large order. We measured how often that happens on BTC/USD across 23 exchanges in Rethinking Crypto Data. The more of the market a price captures, the less any one venue's quirks can distort it.
Depth: how well the price reflects tradable liquidity
Breadth on its own is not enough. Averaging twenty exchanges naively imports every one of their errors. Depth is about weighting and filtering the inputs so the result reflects liquidity that actually exists.
In practice, that means weighting by volume instead of by venue count, so a thin exchange printing an unusual price cannot move the result the way it would in a simple average. It means removing individual trades that fall outside a statistical confidence interval, and excluding entire venues whose order books have drifted away from the rest of the market. It also means guarding against wash trading and fabricated volume, which remain a real problem on some venues. How each of those filters works, and where they break, is covered in Real-Time Crypto Data Aggregation: How It Works.
Latency: how current the price is
Latency is the delay between the market moving and the price reflecting it. A benchmark that lags by several seconds is not a fair comparator for a trade executed now.
But latency has a floor that is often overlooked. Blockchain transactions settle in seconds, and oracle price feeds update on intervals or when the price moves beyond a set threshold. Beyond a certain point, a faster price has nothing to act on. Latency matters up to the speed at which the price is actually consumed. Past that, it stops being the binding constraint.
How breadth, depth and latency trade off
The useful insight in the triangle is the way the three dimensions pull against one another.
Breadth costs latency: gathering data from seventy venues, normalising it and filtering it takes longer than reading a single exchange's feed, so the most complete price is never the fastest one. Depth costs latency as well, because every filter that removes an anomalous trade or excludes a drifting venue adds processing time; removing a filter makes a price faster and less reliable in the same stroke. And latency costs depth in return: very short time windows contain fewer trades, which makes statistical filtering less reliable, and on thin pairs a window that is too short may not contain enough data to filter at all.
So "the fastest price" and "the most accurate price" are almost never the same thing. A well-designed price source balances all three dimensions for the use case instead of maximising one, a trade-off most discussions of crypto market data leave out.
Two further conditions keep the triangle intact over time, though they are not dimensions of accuracy in themselves: reliability, so the price remains available during volatility when it matters most, and governance, so the methodology is documented, reviewed and applied consistently. The IOSCO Principles for Financial Benchmarks, the 2013 standard that regulated benchmark administrators are held to, make the same two demands of any price used as a reference: input data sufficient to represent the market it measures, and a documented, governed methodology.
How to measure price accuracy
Accuracy can be expressed as a deviation from a benchmark:
Accuracy = 1 − |P(source) − P(benchmark)| ÷ P(benchmark)
Here P(source) is the price being assessed (a quote, an execution price, a single exchange's price) and P(benchmark) is an independent, market-wide reference price at the same moment.
A worked example, using the size of deviation our own BTC/USD study found in the tails during calm trading. Suppose the market-wide reference price for an asset is 2,400.00, and a single venue is quoting 2,402.40. The deviation is 2.40 ÷ 2,400, or 10 basis points, and the accuracy comes out at 99.90%.
That sounds close to perfect. On a €5 million position, it is a difference of €5,000, on one valuation, at one moment.
Accuracy is therefore better expressed in basis points than in percentages. "99.90% accurate" invites complacency. "10 basis points off the market" invites the right question.
The metric itself is not new. Deviation from a robust benchmark is how regulated index providers assess their own inputs. CF Benchmarks, which administers the CME CF Bitcoin Reference Rate, calculates a volume-weighted median for each constituent exchange purely as a quality-control measure, excludes any exchange whose figure deviates from the others by more than 10%, and reported in its March 2025 suitability analysis that between February 2022 and February 2025 no constituent exchange deviated by more than 2.41% from its New York variant of the rate. A regulated benchmark administrator tracking per-exchange deviation tells you two things: single venues do deviate, and measuring how much is standard practice.
Fair value in crypto markets vs fair value in accounting
In the market sense, fair value is the price the market as a whole is trading at, at a given moment, best approximated by a liquidity-weighted reference aggregated across the venues where genuine trading happens. The last trade on one exchange is one input to that estimate.
This differs from how fair value is used in accounting. Under ASC 820 and IFRS 13, fair value measurement follows specific rules about principal markets and input levels, which we cover in Crypto Asset Valuation for Audit and Reporting. The market sense used here is simpler: the price that best represents where the asset is actually trading, right now, across all of its markets.
Trade-based vs state-based pricing
There are two fundamentally different ways to derive a crypto price, and each has a different accuracy profile.
Trade-based pricing derives the price from trades that have actually happened; the volume-weighted average price is the standard example. Its strength is that it reflects real, executed activity. Its weakness is that it depends on trades occurring. On a thinly traded asset there may be few recent trades, and the price can go stale or be moved by a single transaction. How VWAP relates to the other price types is covered in Crypto Prices Explained.
State-based pricing derives the price directly from the reserves held in decentralised exchange liquidity pools, without waiting for trades. Its strength is availability: a pool always has reserves, so a state price exists even when nothing has traded. The method is also transparent, since anyone can read pool reserves on-chain and verify the result.
State pricing is often described as manipulation-resistant, and in its raw form it is not. Reading one pool's reserves at a single moment is vulnerable on thin pools: a large trade, or a flash loan taken and repaid within one transaction, can move reserves significantly and then reverse. That exact mechanism lies behind many oracle exploits. The defence has been known since the Uniswap v2 whitepaper (Adams, Zinsmeister and Robinson, 2020), which introduced cumulative price accumulators so that an oracle could read a time-weighted average across many blocks instead of a single block's state.
Robust state pricing therefore combines a lookback window, sampling at the end of a block instead of within it, liquidity or TVL weighting, and aggregation across several pools, so that a momentary distortion in one pool cannot move the result. Chainlink's DEX State Price methodology, for example, is built from precisely those components, and its documentation still lists residual risks specific to DeFi, such as smart contract flaws, bridge dependencies and external price inputs. State pricing solves the availability problem. Accuracy still depends on breadth and depth.
Neither method is safe by nature; it depends on the asset's liquidity profile. In late 2024 Chainlink moved several stablecoin feeds from trade-based VWAP to state pricing after large trades in thin pools had produced price spikes that the trade-based method captured as if they were the market, an episode documented by Inverse Finance's risk working group. For a DEX-dominant asset, well-built state pricing can be the more accurate of the two.
Price accuracy failures: Pyth 2021 and Mango Markets 2022
Two well-documented incidents show how price accuracy breaks down, and each maps onto a different side of the triangle. How oracle networks are designed to catch such failures is covered in How Oracle Price Feeds Work.
Pyth, in September 2021, was a depth failure. On 20 September 2021, Pyth's Bitcoin price feed briefly reported around $5,402 while Bitcoin was trading near $43,500. The feed had multiple sources, but two publishers submitted near-zero prices that received disproportionate weight in the aggregate. The feed had sources enough; the aggregation simply failed to filter the outliers out. Breadth without depth does not protect a price.
Mango Markets, in October 2022, was a breadth failure. On 11 October 2022, a trader used roughly $10 million across two accounts to drive up the price of MNGO, the illiquid governance token of the Mango Markets protocol. Mango's oracle took its price from a small number of venues, all thinly traded. With the collateral value inflated, the trader borrowed against it and drained the protocol. The CFTC alleged that more than $110 million was misappropriated; the SEC put the figure at approximately $116 million. A criminal conviction was later overturned on jurisdictional grounds, but the mechanism was never in dispute. The price came from too few, too thin sources to resist a deliberate push.
The third failure mode, latency, is quieter and more common. A feed that stops updating does not announce itself. It keeps reporting its last price with the same confidence as before, which is why oracle consumers are expected to check when a price was last updated instead of assuming it is current.
How to evaluate a crypto price source
The triangle is most useful as a set of questions to ask of any price source. On breadth: how many venues does it aggregate, and does that include decentralised exchanges as well as centralised ones? On depth: how is it weighted, what filters remove anomalous trades, and what happens when an entire venue drifts from the rest of the market? On latency: how quickly does it update, and is that speed appropriate for how the price will actually be used? And on reliability and governance: what uptime does it guarantee, and is the methodology published so the answers above can be verified?
For reference, this is how Blocksize approaches each. Its prices aggregate more than 70 centralised and decentralised exchanges across over 9,000 pairs. Prices are volume-weighted, with a price-based filter removing trades outside a confidence interval and an exchange-based filter excluding venues that drift from the market. Aggregated prices update in under 250 milliseconds, with a 99.9% uptime guarantee, and the methodology for every product is published and overseen by a data committee.
The same prices are published into decentralised oracle networks, where independent node operators check each submission against other publishers before it is accepted. That adds a check the triangle does not cover on its own: external validation, by parties with no incentive to agree with the publisher.
Conclusion
Every data provider describes its prices as accurate. The word on its own tells you very little.
What makes accuracy meaningful is being able to say how much of the market a price reflects, how its inputs are weighted and filtered, how current it is, and how far it deviates from an independent reference. Those are measurable properties. A price that can be described in those terms can be defended. One that cannot is simply trusted, and in a market with no official price, that difference decides whether a valuation, a benchmark or a liquidation holds up.
Frequently asked questions
What is price accuracy in crypto? Price accuracy is the degree to which a price reflects the true, contemporaneous fair value of an asset across all the markets where it trades. It depends on how much of the market the price captures, how its inputs are weighted and filtered, and how current it is.
What is the Accuracy Triangle? The Accuracy Triangle is a framework describing three dimensions of price accuracy: breadth (how much of the market is captured), depth (how well the price reflects real, tradable liquidity) and latency (how current the price is). The three trade off against each other, so accurate pricing means balancing them instead of maximising one.
How do you measure crypto price accuracy? As the deviation of a price from an independent market-wide benchmark at the same moment: Accuracy = 1 − |P(source) − P(benchmark)| ÷ P(benchmark). The deviation is usually more useful expressed in basis points than as a percentage.
Is the fastest crypto price the most accurate? Not usually. Aggregating and filtering data from many venues takes time, so the most complete price is rarely the fastest. Beyond the speed at which a price is actually consumed, which is limited by blockchain settlement and oracle update intervals, additional speed adds little.
Is state-based pricing manipulation-resistant? Not in its raw form. Reading one pool's reserves at a single moment can be manipulated on thin pools, which is the mechanism behind many oracle exploits. State pricing becomes robust with a lookback window, end-of-block sampling, liquidity weighting and aggregation across pools.
What is fair value in crypto? In a market sense, fair value is the price the market as a whole is trading at at a given moment, best approximated by a liquidity-weighted price aggregated across venues. Accounting fair value under ASC 820 and IFRS 13 follows separate, more specific rules.