Methodology: How Drop Readiness Works

What Drop Readiness is

Drop Readiness is a 0–100 score estimating how close an AI model is to shipping, combining 45% blended Polymarket and Kalshi odds, 25% intel recency, 20% deadline proximity, and 10% market volume. It collapses four independent signals into one number you can sort a forecast board by.

The score is a forecast, not a guarantee. A high reading means the available evidence points to a near-term release; it does not promise a date. Drop Readiness refreshes hourly, so the number moves as markets reprice, intel lands, and expected windows close in. Watch a release climb or stall over a few cycles and you learn more than any single snapshot tells you. The live forecast board ranks every tracked release by this score.

The formula

The canonical Drop Readiness formula is a fixed-weight blend of four normalised components, each scaled to the same 0–100 range before weighting:

DR = .45 odds + .25 intel + .20 deadline + .10 volume

The output runs 0–100, where higher means closer and likelier to ship. The weights never drift: odds carry the most influence because public markets are the fastest consensus signal, and volume carries the least because it modifies confidence rather than direction. Every component is probabilistic, so the final score is a calibrated estimate, never a certainty. These exact weights are the single source of truth across the site — they read identically on this page, in the glossary definitions of odds spread, T−Nd, and forecast window, and in the FAQ.

Component 1 — Odds (45%)

Odds is the blended implied probability, drawn from Polymarket and Kalshi, that a model ships inside its forecast window. It is the heaviest input because prediction markets are the fastest public consensus on timing — they reprice within minutes of a credible leak or denial.

We read each venue's market price as an implied probability and blend the two into one figure. When Polymarket and Kalshi disagree, that gap is itself informative, so we surface it as the cross-venue spread rather than hiding it inside the average — see the cross-venue spread section below and the glossary entry for odds spread. Markets are used here strictly as a forecasting signal: we read the odds, we do not place or take positions.

Component 2 — Intel (25%)

Intel measures how recent and how corroborated the supporting evidence is for a given release. Fresher signals from more independent sources score higher; a single stale mention scores low.

What counts as intel: official changelogs and release notes, public posts on X from people close to a launch, deltas in public model-catalog listings, and codename mentions such as ⟨MYTHOS⟩ or ⟨EMBER-ALPHA⟩. A claim corroborated across two or three of those carries more weight than the same claim from one. Intel decays with age — a confirmation from yesterday outweighs the same wording from three weeks ago. Each item is source-linked so you can judge it yourself; see our sources and freshness policy.

Component 3 — Deadline (20%)

Deadline scores proximity to the expected release window: the closer the expected date, the higher this component climbs. It encodes the simple intuition that a model expected this month is readier than one expected next quarter.

We express remaining time as T−Nd — days to the expected release, so a model fourteen days out reads as T−14d. As T−Nd shrinks toward zero, the deadline component rises; once a window passes without a ship, it decays and the release is re-evaluated against fresh intel and odds. T−Nd is defined alongside the other coined terms in the glossary.

Component 4 — Volume (10%)

Volume measures how much market activity stands behind the odds, and it earns the smallest weight as a confidence modifier rather than a direction signal. A confident price set by deep trading is worth more than the same price on a thin, illiquid market.

Heavier volume tightens our confidence in the 45% odds component without overriding it — it tells us the consensus is well-supported, not which way the consensus leans. Low volume does not zero out a release; it simply caps how much the odds alone can move the score.

An illustrative worked example

Here is how the four components combine into one score. The numbers below are illustrative, chosen to show the arithmetic — they are not live data and do not describe any real release.

Suppose a tracked model carries a blended odds component of 80, an intel component of 70, a deadline component of 90 (it is close to its expected window), and a volume component of 60. Applying the fixed weights:

DR = .45(80) + .25(70) + .20(90) + .10(60)
   = 36 + 17.5 + 18 + 6
   = 77.5  ->  78 / 100  (example only)

An illustrative score near 78 would place the release high on the board — strong markets, a near deadline, and decent corroboration, tempered slightly by moderate volume. For the live, real numbers, always read the live forecast board; static documentation never carries current odds, dates, or scores.

Stages: imminent, expected, rumoured

Every tracked release sits in one of three stages that summarise its Drop Readiness band and signal quality. They map directly to the groupings on the homepage forecast board.

• Imminent — high readiness with a near-term window. Markets are confident, the deadline is close, and intel corroborates a ship soon.
• Expected — a credible window backed by solid signal, but further out or with more uncertainty than an imminent drop.
• Rumoured — early or single-source signal and low readiness. The release is on the radar but lacks the corroboration or market depth to rank higher.

Intel tags: Confirmed, Rumor, Market

Each individual intel item carries one of three tags describing where it came from, so you can weigh a signal at a glance.

• Confirmed — from an official or directly verifiable source, such as a changelog or a vendor's own announcement.
• Rumor — credible but unverified, such as a well-placed post that has not yet been confirmed.
• Market — derived from a prediction-market move on Polymarket or Kalshi, where a price shift is itself the signal.

Every intel item is source-linked regardless of tag, so the underlying evidence is one click away.

Cross-venue spread: Polymarket vs Kalshi

The cross-venue spread is the gap between Polymarket's and Kalshi's implied odds for the same release, surfaced as its own signal rather than averaged away. No competing tracker exposes it, because most surface zero market data at all.

The two venues often resolve questions differently: Kalshi markets frequently resolve to a specific named model, while a Polymarket question may resolve at the company level. That structural difference can make their odds diverge even when both are right about their own question. We blend both into the 45% odds component and publish the spread alongside it, so a widening or narrowing gap reads as information about disagreement and resolution scope. Polymarket and Kalshi are sources we build on, not rivals — the spread is a synthesis of both. The deep how-to-read-the-spread tutorial lives in our explainer on how to read the Polymarket vs Kalshi odds spread, and the broader question of whether prediction markets forecast releases accurately is covered in how prediction markets forecast AI releases.

Forecast windows and refresh cadence

A forecast window is the date range a model is most likely to ship, derived by reconciling expected dates from intel with the resolution dates of the underlying Polymarket and Kalshi markets. T−Nd is simply the distance from today to that window.

When intel and market resolution dates align, the window tightens; when they conflict, it widens to reflect the uncertainty rather than forcing a false-precision date. The whole model — odds, intel, deadline, volume, and the resulting Drop Readiness scores — refreshes hourly. That freshness is the core edge: the picture you read is at most an hour old, not a multi-day-old snapshot.

What we do not do

We do not fabricate odds or dates, we do not present examples as live data, and we do not take bets or offer financial advice. Any sample numbers in documentation — including the worked example above — are illustrative only; the live figures live solely on the app-driven home page.

Drop Readiness is a planning tool for builders, not a wager and not investment guidance. Prediction-market odds enter the score purely as a forecasting signal — markets are signal, not stakes. Our public sources for odds are Polymarket and Kalshi, and we describe intel only in general terms: official changelogs, public posts, public model-catalog listings, and market moves. For the full position on what the score is and is not, see forecasts are signal, not stakes, and the frequently asked questions answer common follow-ups.