Confirmed vs rumor vs market: how to grade an AI release signal

The three grades, defined

Every intel item is graded as one of three tags — Confirmed, Rumor, or Market — and the grade is a verdict on credibility, not a topic. Confirmed means a verifiable public artifact exists. Rumor means unverified chatter with no artifact yet. Market means the implied odds moved on Polymarket or Kalshi. The grade is what turns a raw leak into a weighted signal.

The point of grading is to stop treating every mention as equal. A vendor's own changelog and an anonymous screenshot can be about the same model on the same day, yet one is evidence and the other is noise. The three tags make that difference legible at a glance, so you can read a briefing without re-litigating each line. Each tag maps to a precise meaning:

• Confirmed — a verifiable public artifact exists: an official changelog entry, a published API SKU or model ID, release notes, or a documentation page you can open yourself.
• Rumor — unverified chatter: a claim circulating without a primary artifact behind it, such as an unsourced post or a screenshot that cannot be traced to its origin.
• Market — a move in the implied odds on Polymarket or Kalshi: the crowd repricing a release event, which is evidence of sentiment rather than a documentary fact.

These same three tags are defined in the intel-tag definitions alongside the rest of the site's vocabulary. The rest of this post is the rubric: how to apply each grade, and what each one does to a forecast once applied.

How to tell a confirmed signal from a rumor

A signal earns the Confirmed grade when you can open a primary public artifact that proves the claim — not when it merely sounds credible. The fastest test is whether there is something to click: a changelog line, a model ID in a public catalog, a documentation diff. If the only thing behind a claim is a paraphrase of a claim, it stays a Rumor until an artifact appears.

Use a short, repeatable checklist. It is deliberately evergreen — the same four questions work for any provider, any model, any cycle:

1. Is there an artifact you can open? A confirmed signal points to something durable and public — a changelog, release notes, a pricing page, a model card. A rumor points only to commentary about the thing.
2. Is the source primary? The provider's own surface outranks a re-share of a re-share. Trace a claim back toward its origin; the closer to the source, the higher the grade it can support.
3. Is it corroborated? Two independent surfaces converging on the same fact beat one loud voice. Corroboration upgrades confidence even before a formal artifact lands.
4. Does the timing fit the cadence? A claimed drop that lines up with a provider's known release rhythm is more plausible than one that contradicts it. Cadence is a sanity check, not proof.

Codenames are where this checklist gets tested most often, because a name like ⟨MYTHOS⟩ can sit in Rumor for weeks before an artifact confirms which public model it becomes. Walking a codename from first sighting to launch is its own discipline — verifying a codename signal covers how a name graduates from chatter to confirmed release.

Why a market move is its own grade

A move in the odds is its own grade because it is neither a documentary artifact nor idle chatter — it is informed participants putting conviction behind a view. When the implied probability of a release shifts on Polymarket or Kalshi, that repricing is evidence in itself: someone acted on information faster than the artifact arrived. It is sentiment with skin in it, which is why it sits apart from Confirmed and Rumor.

The crucial caveat is that a market move is evidence of belief, not of fact. Odds can climb on a credible whisper and fall again when it is denied; the crowd can be early, and the crowd can be wrong. That is why the Market grade is treated as a fast, directional signal rather than proof — it tells you the consensus is moving and roughly which way, without claiming the model has shipped. The mantra holds throughout: markets are signal, not stakes. We read the odds as a forecast input, never as a wager to place.

The two venues do not always agree, and the gap between them is itself informative. Learning to read that divergence — when Polymarket and Kalshi price the same release differently — is covered in reading a market move, which treats the cross-venue spread as a signal in its own right.

How grades change a forecast

Grades change a forecast through the intel-recency input, which carries 25% of the Drop Readiness score. The canonical formula is fixed: DR = .45 odds + .25 intel + .20 deadline + .10 volume. Within that 25% intel slice, the grade and freshness of supporting signals decide how much the input contributes — a fresh Confirmed item lifts it most, a Rumor lifts it less, and a stale signal of any grade decays toward zero.

Concretely, a newly Confirmed artifact can push a card from rumoured toward expected, because the intel component jumps while the other three inputs hold. A Market move tends to register first in the 45% odds input, then echo into intel as corroboration; a Rumor nudges intel only modestly until it is either confirmed or fades. Because every input is re-evaluated hourly, a card does not stay where a single leak left it — it climbs or stalls as new graded items land and old ones age out. (Any numbers here are illustrative of the mechanism, not live readings.)

The full computation — how each input is normalised and weighted, and how the stages imminent / expected / rumoured are assigned — lives in how grades feed Drop Readiness. The narrative behind the score, written for builders, is in the Drop Readiness explainer.

See graded intel live

The grading rubric is abstract until you watch it run, so the clearest way to understand it is to read the live briefing where every item already carries its Confirmed, Rumor, or Market tag. The tags update hourly, which means the verdict layer you see is at most an hour old — fresher than any multi-day editorial cadence.

Open the live graded briefing to see the three tags applied to real, source-linked items, then read how items are sourced for the provenance and trust posture behind each one. Grading answers "how much should I believe this?"; sourcing answers "where did it come from?" — two halves of the same credibility question.