Equity markets in most jurisdictions have a consolidated tape and rules requiring brokers to seek the best available price across venues. Crypto has neither. Each exchange runs an independent order book, prices are set by whoever is trading there, and nothing obliges anyone to reconcile them. The result is that at any moment the same asset has many prices, and a published figure is the output of a decision about which of them to use.
Small differences are ordinary and are closed by arbitrage within seconds. The arbitrageur buys where it is cheap, sells where it is dear, and is compensated for holding inventory on both venues and for the risk of the trade going wrong in between. What limits this is not intelligence but friction: capital must already be positioned on both sides, because moving it takes minutes on-chain and considerably longer through the banking system. The size of a persistent spread is therefore a measure of how difficult it is to move capital between two particular venues.
That reframes the well-known cases. Sustained premiums in markets with capital controls or restricted banking access are not mispricings that anyone can capture; they are the market pricing the cost and legal risk of moving money across a border. The Korean premium of 2017 and 2018 was the widely cited example, and it persisted precisely because the arbitrage was impractical rather than unnoticed. Discounts on a venue in difficulty are the same phenomenon inverted: the price reflects the risk that assets on that platform cannot be withdrawn, which is a real price for a real asset that is not quite the same asset.
Quote currency compounds the effect. A pair against a stablecoin prices the asset against that issuer's token, which is itself not always exactly a dollar. A pair against a local currency embeds that currency's own rate and its own convertibility. Comparing a stablecoin pair on one venue to a dollar pair on another and calling the difference an arbitrage is comparing two different trades.
For anyone constructing a reference price, the trade-offs are unavoidable and should be visible. A simple average treats a venue with negligible depth as equal to one with the deepest book in the market, which lets a stuck feed on a small exchange drag the answer. Volume weighting fixes that and imports whatever volume misreporting exists. A volume-weighted median with outliers excluded — and named — is more robust than either, which is the approach this site takes, and it is still a construction rather than a fact.
The reader's version of all this is short. A single quoted price is always from somewhere. Where a figure matters — a liquidation level, a tax basis, the price a piece of reporting is built on — the venue and the timestamp are part of the figure, and a number published without them is less precise than it looks.