There is no “the” price
Ask what Bitcoin costs and the honest answer is a question back: where? Every exchange runs its own order book, matching its own buyers against its own sellers. Those books are linked by arbitrage but never identical, so at any instant the same asset trades at slightly different prices in different places.
Most of the time the spread between venues is small enough to ignore. The moments it is not are exactly the moments that matter: a sudden move, a venue with thin liquidity, an exchange with a technical problem printing trades nobody else sees. A settlement that reads one venue inherits every one of that venue’s bad seconds.
So the real question for anything that settles on price is not “is the price correct” — there is no single correct — but “which construction of the price does this thing read, and who can influence it”.
What an oracle actually does
An oracle is the machinery that turns many disagreeing quotes into one published number. The naive version averages them, and the naive version is a mistake: an average is dragged by outliers, so a single venue printing a nonsense quote moves the result for everyone.
The standard construction instead uses a median, usually weighted by volume. A median ignores how extreme an outlier is and only counts how many sources agree — so one broken venue moves nothing, and an attacker has to corrupt a majority of the sources rather than one of them. That single choice is most of what makes an oracle robust.
- Multiple independent sources, so no one venue is load-bearing.
- Median rather than mean, so an outlier is outvoted rather than averaged in.
- Volume weighting, so a thin market cannot speak as loudly as a deep one.
- Independent reporters, so the aggregation itself is not one machine anyone can lean on.
- A published record, so the number can be checked afterwards rather than taken on trust.
That last property is the one players should care about most, and it is the one least often discussed. A feed you can look up after the fact is a feed you can audit. A number a house quotes only to itself is not a price, it is a claim.
Why this decides whether a bet is fair
A price-settled game has exactly one place where the house could cheat without anyone noticing: the price it settles on. Not the payout table, which is published; not the rules, which are visible; the number.
Consider what a dishonest desk could do with a self-quoted price. It does not need to lie by much. A tenth of a percent, applied only at the moment of settlement, only in the direction that suits it — and at high leverage a tenth of a percent is the entire position. The manipulation would be invisible in any single case and devastating in aggregate.
This is why the meaningful commitment is not “our prices are accurate” but a structural one: the number the chart draws and the number the settlement reads are the same published number, and it is not ours to pick. Display and settlement fed from one source removes the gap where the trick would live.
What to ask any price-settled venue
You do not need to audit an oracle to be a careful player. You need to ask four questions and notice whether the answers exist at all.
| ASK | A BAD SIGN |
|---|---|
| Which feed settles my position? | “Our internal price” — the one construction nobody else can check |
| Is the chart the same source as the settlement? | Two different numbers, or an unwillingness to say |
| What happens when the feed stalls? | No published policy, decided case by case |
| Can I look the price up independently? | A feed that exists nowhere but on their screen |
One published feed for the chart, the fill and the bust priceBTC ×1000 →
Why do different exchanges show different Bitcoin prices?
Each runs its own order book with its own buyers and sellers. Arbitrage keeps them close but never identical, and the gaps widen exactly when the market moves fastest.
Why do oracles use a median instead of an average?
An average is dragged by outliers, so one broken or manipulated venue moves the result. A median only counts how many sources agree, so an attacker must corrupt a majority rather than a single feed.
Can a casino manipulate the price a leveraged bet settles on?
It can if it quotes the price itself, and at high leverage it would only need a fraction of a percent. That is why the meaningful question is whether the settlement reads an independently published feed — one you could look up yourself.
What stops a stale price from being used to settle?
A freshness rule. On our desk a tick past its freshness window is not treated as a price at all: it blocks fills and liquidations equally until a live one arrives.
- Betkyo engine source: CryptoProEngine strictPrice() and the shared tick buffer behind the chart, the fill and the liquidation
