Every call dated. Every call graded.

Four things, on a fixed cadence

One claim per post. Numbers open it, one opinion closes it, and the reasoning is published with the call rather than reconstructed after the result. The board behind every one of them is a format-matched consensus of public sources — public rankings, ADP and box-score data, matched to the scoring format. We do not run a forecast of our own, and nothing here is sold as one. This is the publishing plan, written down before the first post rather than after it. None of it has run in public yet.

Waiver call

Weekly, in season. One waiver target per video: the usage, the market price, and the bid.

Our board vs market

Daily through draft season. The player where our board and the market's ADP disagree most that day.

The receipt

Weekly. The graded record for the week that just closed — hits, misses, and anything that could not be resolved.

Tier wall

As the data moves. Where our board's tier break falls against public ADP at one position.

A call is written so a machine can decide it

Nothing is graded by opinion, and nothing is graded generously. Each call carries the test that will decide it before the week it is measured over begins. The machinery that does this is built and tested; what it has not done yet is grade a published call, because none has been published.

The claim

One named player, one call type, one sentence, published on a date.

The criterion

A metric, a comparator, a threshold and the weeks it is measured over — for example mean position rank over weeks 1–17 is at most 12.

The push band

A margin around the threshold inside which neither side is credited, so a call decided by a rounding artefact is published as a push and not as a win.

Unresolved

When the result data cannot establish an answer — an identity that never mapped to a player id, a week that never arrived — the call is published as unresolved and counted neither way. It stays in the record.

The ledger

Append-only and hash-chained. A grade that turns out to be wrong is corrected by a superseding entry; no entry is edited and none is deleted.

The record

Empty. The 2026 ledger holds no calls, so there is nothing to grade and nothing to show — not a season that went badly, and not a record being withheld until it looks good.

This page is a read of that ledger and cannot be anything else: it can only print calls the ledger holds. That constraint is the point of publishing a record at all, and it is worth more before the first call than after it — an empty table here today is what makes a full one mean something later.

The record page says the same thing at length: what a call has to carry, what decides it, and what appears here the moment one posts.

Plug your league into your own model

The machinery that will produce the calls on this page is a harness, and the harness is the product: you point it at your own model account and your own league, and it runs the same week for you that it will run here.

Bring your own LLM

Your model account, your key, your quota. Nothing here resells inference, and the harness has no model of its own to sell you.

Your league, read directly

The same reads that will produce this account's calls: your scoring rules, your rosters, your waiver wire — not a generic ranking list.

Dev-native shape

A CLI first, an MCP server next. It is not an app, there is no dashboard planned, and it is not a picks service.

Status

Not released. No date, no price, no beta programme. The account it is meant to run has not launched either, and neither one is for sale.

There is nowhere to put your address yet