Methodology | Brierpoint

What is AI-generated

Three things are written by AI models: explainers, Gotcha ratings and Why It Moved explanations. Everything else, including prices, order-book depth, tradability, movement detection and postmortem prices, is captured or calculated directly from data, with no AI.

AI output is never fed back into a price, a score or any other number. It is written in advance and stored, so every visitor sees the same version. On market pages each piece is labelled as AI-generated with the time it was written, and explainers and Gotcha ratings also name the model. Models can misread rules or miss context, so check a market's own rules on Polymarket before relying on anything here.

Where the data comes from

Polymarket's public APIs supply each market's question, rules and close date (checked every 15 minutes, with a full pass each day), its prices and order book (every 5 minutes), and whether it has closed and how it resolved (every 15 minutes).
News headlines for Why It Moved come from a news search service, Webz.io. Brierpoint stores each headline's title, link and publication time, not the article.
Crypto prices for Why It Moved come from Binance's public market data.

Brierpoint follows Polymarket markets above a minimum level of liquidity, so not every market is covered. Price history is kept at 5-minute detail for 7 days, then hourly for 90 days, then as one point per day.

Prices

A market's price on Brierpoint is the midpoint between the best bid and the best ask when it was captured, never the last trade. On a thin market the last trade can be one small order far from where anyone could trade now; the midpoint reflects the book as it stands.

Prices are shown in cents: a price of 42¢ means the market implies a 42% chance. The 24-hour change compares the latest price with the latest one captured at least 24 hours earlier. A price captured more than an hour ago is labelled stale.

Explainers

An AI model reads the market's question, resolution rules and close date and writes a plain-language summary: how the market resolves, which source decides it, and the dates that matter. Event pages, which group related markets, carry an event-level explainer written the same way.

A new explainer is written when the question, the rules or the close date changes; earlier versions are kept for the record. Not every market has one yet.

Gotcha risk

Gotcha risk rates the rules, not the outcome. An AI model reads the question and rules and answers one question: how likely is someone who read only the question to be surprised by how the market actually resolves? The answer is Low, Medium or High. It is not a probability, and a High rating does not mean the market is mispriced; it means the rules deserve a careful read.

Four factors are rated alongside it, each Low, Medium or High. The overall rating is a separate judgement, not an average of them.

Wording ambiguity: Can the question or rules be read two ways? Vague terms, undefined thresholds, conditions that could describe more than one outcome.
Source discretion: How much judgement does whoever decides the result have? A named official source publishing one figure leaves little; “a consensus of credible reporting” leaves a lot.
Timing risk: Could the result be delayed, or come out differently depending on when it is checked? Deadlines, “as of” dates, runoffs, results that take weeks to certify.
Exceptional clauses: Rules that can catch a reader out: early resolution, timeouts that resolve to a definite outcome, tie-breaks, things that do not count.

A market is marked can't assess when it has no rules, when the rules are too thin to judge, or when they name no resolution source. The rating is redone whenever the question or rules change.

Tradability

Tradability is a fixed formula with no AI. Once a day at 03:00 UTC it reads each market's most recent order book from the previous 24 hours and measures two things: the spread (the gap between the best bid and the best ask) and the depth (the value of orders on both sides within 5¢ of the midpoint).

High: A spread of 1¢ or less and at least $50K of orders within 5¢ of the midpoint.
Medium: Anything between High and Low.
Low (thin book): A spread of 5¢ or more, or no quote at all, or less than $2K within 5¢ of the midpoint. Large orders will move the price.

Order books change through the day, so the rating describes the book at the daily check, not right now.

Why It Moved

Every 15 minutes Brierpoint checks prices for moves. A move of 2 points or more within 24 hours is recorded. A move of more than 10 points gets a written explanation, but only on markets with Medium or High tradability: on a thin book one order can move the price, so no explanation is written.

Two stages

Soon after a move is detected, a first explanation is written from data alone: the shape of the move, how close the market is to its deadline, whether related markets in the same event moved together, the coin's price for crypto markets, and the order book.

For politics, geopolitics and economy markets, Brierpoint then searches news published from 3 hours before to 4 hours after the move. About five hours after the move began, a news-backed explanation replaces the first one and links the headlines it relies on. It also says how many independent sources were reviewed.

What the label means

Each explanation carries one label:

News: A cited headline, published around the move, reports a development that bears on the market’s outcome.
Coin price: For a crypto market, the coin’s own price move accounts for the market’s move.
Close to deadline: The market is near its deadline and the move fits it settling toward its likely outcome.
Related outcomes: This move is part of the event’s outcomes repricing together.
Thin market: A thin order book explains the move better than new information does.
Several factors: A cited headline and at least one price fact both plausibly contributed.
No clear catalyst: The news search was completed, and neither a headline nor a price fact supports a driver.
Not enough evidence: Any other case, including when no headline was found and no fact supports a driver.

“No clear catalyst” is a real result: sometimes nothing in the evidence explains a move. Explanations describe a likely driver, never a proven cause.

Postmortems

When a market closes, its page becomes a record of what happened: the outcome; the price 7 days, 24 hours and 1 hour before close; the explainer, Gotcha rating and Why It Moved explanations as they stood before close; the rules at close; and the price path. “Close” is the earlier of the actual close and the listed close date.

Only analysis written before the market closed is shown as “before close”, so the record cannot be improved with hindsight. Nothing on a postmortem is written by AI after close.

Limits

If Polymarket's data is unavailable, prices stop updating. Pages say when data is stale rather than hide it.
Coverage is partial: some markets have no explainer, rating or explanation yet.
AI makes mistakes. Ratings and explanations are a starting point for reading the rules, not a substitute for them.

Spotted something wrong? Contact Brierpoint with the page link.