Skip to content
PROBORA

Market data

Building an independent price index

A prediction is only as trustworthy as the price that settles it. How the Probora Price Index is constructed, and why a median of independent sources is the right primitive.

Probora Research3 min read

Why settlement data deserves its own design

Every Probora market asks a simple question: will the asset finish the window above or below the price it started at? The answer depends entirely on two numbers: the Price To Beat and the Settlement Price. If either can be distorted by one venue's glitch, a thin order book or a single erroneous print, the entire track record inherits the error.

Probora therefore never settles markets with prediction-market odds or with any single exchange's last trade. It builds its own reference: the Probora Price Index.

Why a median

For most crypto assets the index samples independent spot venues (Binance, OKX and Coinbase) at the same instant and takes the median. With three sources, the median is simply the middle value.

The median has a property an average lacks: it ignores one extreme. If a single venue prints a spike, the average moves toward it; the median does not. To move a three-source median, at least two independent venues must move together, which is what a genuine market move looks like.

Freshness, outliers and quorum

A median is only robust if its inputs are valid. Three checks run before any index value exists:

  • Freshness. Quotes older than 5 seconds relative to the target instant are stale and excluded. A venue that stopped updating cannot anchor a settlement.
  • Outliers. After a provisional median, any source deviating by more than 50 basis points is excluded and the median is recomputed.
  • Quorum. At least 2 valid sources are required. Without quorum there is no index value, and a market whose start or settlement lacks quorum resolves as VOID instead of guessing.

Getting the currency right

Venues quote in different currencies. Some quote against USD; others against stablecoins such as USDT or USDC, which can trade slightly away from one dollar. Mixing them naively introduces a small but systematic bias, so stablecoin-quoted prices are normalized to USD with a live conversion rate before the median is taken.

Ticks, not floating point

Index values are rounded to each asset's tick size, and outcomes are decided by comparing whole numbers of ticks. It sounds like a detail until you consider that a floating-point rounding difference could otherwise flip a window from UP to DOWN. Exact equality resolves as FLAT. It is recorded as a push and never forced into a direction.

ptb_ticks = round(Price To Beat / tick) settle_ticks = round(Settlement Price / tick) UP if settle_ticks > ptb_ticks · DOWN if < · FLAT if =

When sources fail

Venues go down, rate-limit or return malformed data. The index is designed to degrade visibly rather than silently. If one of three sources is stale or erroneous, the median of the remaining valid sources is used and the excluded quote is recorded together with the reason it was dropped.

If quorum cannot be met at either end of a window, the market resolves as VOID. Every prediction for that window stays on the record, marked VOID and excluded from scoring, instead of being resolved against a price nobody should trust. Each exclusion is also logged as a data incident, so a source that keeps failing is visible in monitoring long before it could affect a settlement.

Extending to commodities

Gold, Silver and Oil use the same architecture with different inputs: a spot reference plus perpetual markets that track the same underlying, each normalized to USD. Because the provider layer is abstracted, sources can be added or a faulty one removed without changing how markets resolve. Commodity windows follow the exchange session, so no market opens while the underlying is closed.

Auditable by design

Every index value is stored with its individual source quotes, their timestamps, their deviation from the median and whether each was used or excluded. Any settlement can be inspected source by source, and the same inputs always reproduce the same output.

That is what independence means in practice: not trusting any single venue, and not asking anyone to trust Probora blindly either.

Research notes are educational and describe Probora's methodology at a high level. They are not financial advice. Examples are illustrative unless stated otherwise.