Prediction contract formats: mentions, rankings and thresholds

Start from the measurement, not the headline

A prediction contract format is the shape of the test a resolver applies to evidence: a recorded mention, a rank position, a numeric threshold, or an event occurrence. The same headline can hide different formats, so the first useful question is which of these 4 tests the rules actually describe. Read the individual market's rules rather than the parent event page, because an event groups 1 or more markets, while events, conditions and outcome token identifiers are separate things. Tags also overlap rather than forming an exclusive hierarchy, and each market carries its own flags for active, closed and acceptingOrders states.

4 formats need 4 different pieces of evidence. A recorded mention needs a speaker, a wording list, a medium and a time window. A ranking needs a population, a metric and a tie rule. A threshold needs an exact value, an operator and a named source. An event needs an actor, an action and a window. When a question combines a threshold and a ranking source, extract both conditions from that contract and evaluate them together.

Which rule answers each question

The table compares 4 contract formats and their settlement criteria. It reflects the contract structures documented in the source snapshots retrieved on 10 October 2026. Some dates in those examples are past, so treat each one as a documented historical rule snapshot rather than as a current tradable opportunity or a statement about present prices or outcomes.

Use the row that matches the question you are trying to answer. If the market settles on whether a phrase was spoken aloud and recorded, that is a mention format. If it settles on which item finishes ahead of others, that is a ranking format with a tie rule. If it settles on whether a number crosses a value, that is a threshold format. If it settles on whether a specified actor did a specified thing inside a window, that is an event format.

Recorded phrase: speaker, wording, medium and time

The mention format has 4 moving parts, and each one can decide the result independently. The speaker must match the person named in the rules. The wording must match the listed term under whatever matching rule the contract states. The medium must qualify: in the October 2026 snapshot, a mention counted only if it was a verbal mention recorded in audio or video and publicly accessible. Written usages such as social media posts or written statements explicitly did not count. A video posted on social media did not qualify either if it was filmed outside the market's time frame.

The time window attaches to the recording, not only to the posting. The snapshot set the window between 1 October 2026 at 12:00 AM ET and 31 October 2026 at 11:59 PM ET, and the rules also required that qualifying video be filmed inside that same window. This matters for the authored example below: a phrase recorded today about an event yesterday fits the contract when the rules specify the statement time, because the measurement is when the phrase was spoken and captured, not when the underlying event occurred.

Do not assume every mention market uses an exact-match test. Word lists, capitalization, nicknames and spoken variants are handled by each market's own rules, and the same headline pattern can be drafted differently elsewhere. The route to a full treatment of phrase and wording rules is the guide on mentions markets, which covers how quoted terms are matched against recorded speech.

The later worked case uses invented inputs to show a decision procedure. It provides no empirical accuracy estimate.

Ranking: population, metric and the tie rule

A ranking contract does not ask whether an item was mentioned or whether a number crossed a line. It asks which item ranks first across a defined population under a defined metric. The 99th Academy Awards snapshot defined the population as films, the metric as the number of Oscar nominations, and it explicitly counted nominations of a film's crew, cast or production as nominations of that film. A nomination for work related to the film counted toward the film's total.

That market also states what happens when 2 films finish level: a tie for the greatest number of nominations resolves in favor of the film whose title comes first in alphabetical order. Because the tie rule is written into the contract, the question has a determinate answer even when the nomination counts are equal. Other ranking markets may use different tie rules or none at all.

2 further clauses limit the ranking format. If no nominations were announced by 30 June 2027 at 11:59 PM ET, that market would resolve to Other rather than to any film. The named resolution source was official information from the Academy of Motion Picture Arts and Sciences and its website, with a consensus of credible reporting permitted as a fallback. A ranking market therefore needs its population, its metric, its tie rule, its deadline and its source read together.

The rankings guide examines the population, ranking and tie rules in detail.

Threshold and event: exact value, operator and source

A threshold format fixes 3 things: the exact value, the comparison operator and the data source. The benchmark snapshot resolved Yes if any model on the Chatbot Arena LLM Leaderboard reached at least the specified Arena Score by 31 December 2026 at 11:59 PM ET. The score came from the Score section of the Text Arena Leaderboard tab with style control unchecked, so the example value 1510 attaches to a particular display mode of a particular leaderboard page.

Threshold markets also define what happens when the source itself is unavailable. That snapshot kept the market open while the leaderboard was temporarily inaccessible, and directed a No resolution if the source became permanently unavailable. This is a rule about evidence availability rather than about the measured number, and it can decide a market that never sees a qualifying score published.

Event formats combine an actor, an action and a deadline. Identify the observable action and its required source before treating an announcement as a completed event. A numeric threshold instead asks whether the documented value satisfies its comparison operator at the stated time.

Binary share payout is a separate rule from the format itself. Ordinary binary markets pay 1 to winning shares and 0 to losing shares, and a rare unknown or 50-50 resolution can pay 0.50 to each side rather than canceling or refunding universally. The resolution guide explains how sources, deadlines and these edge cases interact once the format question is settled.

Worked case: a recorded phrase about yesterday

The case below is explicitly illustrative. The inputs are invented to demonstrate how the stated rules determine each illustrative outcome.

Read the decisive line: if the rules say the phrase must be mentioned inside the window, the measurement is the time the phrase was spoken and recorded, so a statement made today about yesterday is a mention today. If the rules instead fix on the date of the underlying event, the same audio would fail. No generic rule resolves this for every mentions market, and none of the inspected snapshots supports a universal claim that all mentions use exact matching.

The 2 threshold rows test the comparison operator: 1510 is at least 1510, while 1509 is not, and a strict above-1510 draft would flip the first row. The thresholds guide takes the price-window and operator mechanics further.

Limits of the contract examples

4 limits apply to these examples. First, the mention, ranking and benchmark rules are snapshots retrieved on 10 October 2026, and some of their dates are past; they document historical contract language and do not describe current listings, prices or outcomes. Second, the benchmark example uses 1 leaderboard tab and 1 display mode, so another view can show a different number. Third, a source, time window or operator from 1 contract cannot be transferred to another. Fourth, the worked case is hypothetical and provides no estimate of how often a resolver would reach a given decision.

The mention and ranking snapshots also carry no information about other markets with similar titles, no fee values and no authorization claims for any jurisdiction. Where a rule is silent on a variant, such as a nickname or a co-production credit, the contract text rather than the headline decides. Those are questions for the specific market, not for a general format description.

Your next decision

Write the settlement test as 1 sentence before you do anything else. Name the measured item, the source and the window. Then say what could make the test fail: wrong speaker, wrong medium, wrong date of recording, a tie with no stated rule, a number below the operator, or a source that stays offline. If any of those failure points is unstated in the rules you are reading, you have found the part that needs a closer look rather than an answer from a general guide.

Recorded speech goes to mentions markets; position against a population goes to rankings; a numeric comparison goes to thresholds; and payout behavior after the test is answered goes to resolution.