How to read the graded slate: cohorts, calls, and why a week's record is noise

The graded slate is a weekly page per league — NFL, college football, and the MLB postseason — that answers one question for every game: before kickoff, which side did the graded wallets buy, split by whether they pass our farming check, and did that lean settle right? It is an accountability ledger, not a pick sheet. Read correctly it tells you whether the clean cohort's pre-game positioning has any information in it over a season; read wrongly it turns "the clean cohort went 10–6" into a tip, which is the one thing a 16-game sample cannot support. This guide defines every column, states the thresholds, and shows how big the noise is.

What is a cohort?

Every wallet in the graded universe carries a farming level and a CopyGrade Score. The slate splits wallets into two cohorts and ignores the rest:

Wallets that are farming-clean but score below 65 belong to neither cohort and are not counted. That is deliberate: a middle cohort of "clean but unproven" wallets would blur the comparison the slate exists to make. Cohort membership is read from the wallet's current grade at the time the page is built, so a wallet that is re-graded mid-season can move between cohorts; the slate does not rewrite settled weeks when that happens.

A farming-risk flag is an algorithmic risk assessment against a published methodology, not an accusation of conduct. The cohort labels inherit that: "flagged" means our model estimates elevated farming risk, nothing more.

What counts as a lean, and what counts as a call?

For each game the slate looks at one market — the moneyline (the market whose slug is the event's own) — and only at buy fills placed before the scheduled kickoff. Sells, in-play fills, spreads, totals and props are excluded. The choice is not arbitrary:

• Buys before kickoff are the only fills a copier could plausibly have mirrored at a comparable price. In-play prices move faster than any bot's latency; an in-play record is a trading record, not a forecast.
• Moneyline only keeps every game to one binary question with one settlement, so that a "hit" means the same thing in every row and every week.

Within a cohort, the slate sums the buy notional on each side. The side with the larger share is the lean, printed with its share and the dollar amount behind it (for example, SEA 71% · $4.1k). A lean becomes a call only when three conditions hold:

A game that fails any of the three tests prints split (below 60%) or no call (fewer than three wallets, or under $250 between them), and does not count toward the cohort's record. Two wallets buying opposite sides is not a signal; one wallet buying a lot is a position, not a cohort; three wallets buying $40 each is noise.

What is a hit, and how is the record kept?

Once a game settles, each cohort's call is scored hit or miss against the moneyline's resolution. Games with no call are neither. The weekly line — Clean 10–6 · Flagged 7–9 — counts hits and misses over called games only, so the two cohorts' records are usually over different numbers of games. The season ledger at the top of the page is the sum of the weekly records since the league's first slate week (Week 1 for the NFL, Week 0 for college football, the Wild Card week for MLB).

Settlement is read from Polymarket's resolved market prices; until a game resolves its row prints pending. A postponed or voided game stays pending until its market settles; nothing pending is scored.

For college football and the MLB postseason the slate shows the top 60 games by volume in the week, not every game; the NFL's full 16-game slate fits without a cap. The page says which applies.

How much of a week's record is noise?

Almost all of it. If a cohort's calls carried no information at all — a coin flip on every game — its record over n called games would still swing by about one standard deviation of √(n·0.25) wins. At p = 0.5:

So a 10–6 week is ordinary: on 16 called games, chance alone produces ten or more hits about one time in four (the exact binomial probability is 22.7%). A 60% hit rate — an enormous edge in a market that prices most moneylines between 40 and 60 cents — sits one standard deviation from chance at about 25 called games, two at about 100, and three at about 225 (the standard deviation of a hit rate is 0.5/√n), so it takes a season, not a month, before a 60% cohort record is more than a reading. The season-versus-bracket guide walks the same arithmetic for individual wallets; the slate is that arithmetic applied to cohorts.

This is why the ledger is the product and the weekly line is the input. A season-long clean-cohort record that sits materially above chance and above the flagged cohort's would be evidence; a single week's is a reading.

Why does the price of a call matter as much as its result?

A hit at 41 cents and a hit at 78 cents are both hits, and they are not the same evidence. Buying an underdog before the market moved and being right is the pattern a sharp wallet shows; buying the favorite after the line has already moved and being right is what the World Cup retro found the severe-flag cohort doing at nearly triple the early price. Where the slate prints the cohort's average fill price beside the lean, read it: a cohort that keeps being right at short prices is confirming the market, not beating it.

The same logic applies in reverse to misses. A clean cohort that leaned an underdog at 35 cents and missed has lost less than the hit-rate column implies, and a flagged cohort that chased a favorite at 80 cents and missed has lost more. The record counts signs; edge lives in magnitude — which is the discipline every data cut we publish tries to hold to.

What the slate is not

• It is not a pick. Nothing on the page is a recommendation to buy either side of any game, and CopyGrade never executes trades. A lean is a description of what a set of wallets did.
• It names no wallet. Every figure is a cohort aggregate. The wallets behind a lean are one click away on each game's event dossier, where their individual grades, verdicts and farming levels are printed with the risk language they deserve.
• It is not validated yet. Whether the CopyGrade Score predicts anything is a live question we publish our own answer to — null so far — and the track record page carries every subsequent read. The slate is one more surface on which that question can be watched in public; it is not evidence that the score works.
• It is not a trend series across re-calibrations. If the scoring model is recalibrated, cohort membership changes; every scoring change is logged on the changelog, and a record spanning two calibrations is two records.

How to actually use it

1. Watch the ledger, not the week. Read the season record against the table above before drawing anything from it.
2. Compare the two cohorts, not either against 50%. The slate's question is whether farming-clean, high-band wallets position differently from flagged ones before games — the site-wide base rate that motivates the score. Both cohorts beating chance says something about the market; only the gap between them says something about the grade.
3. Follow a lean back to the dossier. If a clean call interests you, open the game's dossier, read which wallets made it, and vet each one on its full record — including when it trades and how long it has been observed. The cohort is a filter, not a shortlist.
4. Use board watch for the games you cannot watch. A board watch fires when a tracked wallet you do not follow opens a large buy on a league's markets — the same pre-kickoff positioning the slate summarises a week later, delivered as it happens.

CopyGrade is analysis-only — it never executes trades, holds funds, or custodies keys. A CopyGrade Score is a documented research opinion, and a farming-risk flag is a risk assessment, not an accusation of conduct. Not financial advice.