How we compute what we publish

Market-implied probability

For a fresh, valid, two-sided market, we use the midpoint between the best bid (the highest price a buyer will pay) and best ask (the lowest price a seller will accept) as the market-implied probability. "PH Composite" describes that probability aggregated across verified-equivalent venue contracts for the same canonical event — never a forecast, and never called a "true probability." With a single eligible source, we name that source directly rather than implying a broader consensus that doesn't exist yet.

Provenance

Every number on this page and on Predict Here Monitor carries its source, an as-of timestamp, and the methodology version that produced it, so a figure can be traced back to exactly where it came from and when. Our security and governance posture around that data is described on our trust & security page.

Event mapping

A canonical event is a real-world question; venue contracts are the tradable instruments that reference it. Every mapping between a canonical event and a venue contract carries the review status described below — and only VERIFIED mappings ever feed an aggregated figure or a derived probability distribution.

Semantic verification

Before any two venue contracts are compared or aggregated, the mapping between them is assigned a review state. Only VERIFIED mappings ever feed an aggregated figure.

VERIFIED

Reviewed and confirmed equivalent — safe to compare or aggregate.

PARTIAL

Related but not a clean equivalence — shown, but flagged, never silently aggregated.

NOT_EQUIVALENT

Reviewed and rejected as equivalent. Excluded from comparison.

UNREVIEWED

No review has happened yet. Treated as unverified, never assumed equivalent by default.

Derived distributions

A probability distribution is only derived for a verified family of mutually exclusive, exhaustive outcomes. We always show the raw summed probability alongside any normalized distribution, and suppress the normalized view — with a stated reason — when the raw sum drifts too far from 100% for the underlying quotes to be trusted.

Live, delayed, and stale data

Our data plane polls each source on a fixed interval rather than streaming continuously, so every figure is labeled delayed, with an as-of timestamp on every observation — never presented as "live" when it isn't. A source with no acceptable update within its freshness policy is marked stale and excluded from any aggregated figure or ranking, rather than silently shown as current.

Executable liquidity & capacity honesty

We compute price-at-size directly from visible order-book depth at fixed notional sizes ($10,000 / $100,000 / $1,000,000). When the book doesn't show enough depth to fill a size, the workstation states "insufficient visible depth" rather than extrapolating a number that the market never actually showed. A quoted top-of-book price is never presented as what a real position could execute.

Signals, not advice

Predict Here Monitor reports market-derived signals and the methodology behind them. It does not generate investment recommendations, and nothing on this product should be read as advice to buy, sell, or hold any contract.

Versioning & corrections

Our event-classification methodology is versioned as taxo_v1; our execution and liquidity methodology is versioned as intel-execution-r1@1. When a computation method changes, it gets a new version identifier and a documented backfill of affected records — we never silently rewrite historical figures. A mapping found to be wrong is marked withdrawn and any derived data recomputed, with the correction disclosed rather than hidden.

• taxo_v1
• intel-execution-r1@1