What We Publish, and When

A publishing calendar is a claim about what a site can defend. Ours has two halves that behave differently, because the two halves of the fantasy year ask different questions. Draft season asks how to spend capital once, under a deadline, against a room that is guessing. The season asks a narrower question every week: what did the model get wrong, and did the reason survive contact with the box score.

The engine underneath both halves reruns on demand. A rerun that moves no median produces nothing worth publishing, so what follows is written as shapes and triggers. ADP moves daily and the boards read it daily; the model itself waits for a reason to run.

What Ships During Draft Season

Draft season carries two kinds of pages, and they age at different speeds. The boards are live surfaces. The PPR draft rankings and their half-PPR and standard siblings change underneath the prose whenever the model runs, which is why nothing written here quotes a current rank or a current gap. Strategy pieces are the slow layer: mechanism, thresholds, and how to read a board once it is in front of you.

The strategy library is built by argument, one page per operation. Replacement level and the arithmetic of positional scarcity live on the value-based drafting page. The spacing between projections, and how to find the point where a position falls off a cliff, is worked through on the tier breaks page. The round-by-round case gets its own piece; for a first read, the draft strategy guide is the entry point, and every other page narrows a slice of it.

Draft-season additions follow two triggers. A new argument earns a page when it changes how a reader should act at the board — a screen with a definable cutoff, a build shape with a stated cost, a format difference big enough to move a pick. A revision gets published when a backtest changes an answer we already gave, and the revision states what the old answer was and what moved it.

What The Ledger Covers, Week To Week

Every week in season, the Ledger prints the model's own scorecard: right and wrong calls in the same table, with the same columns and the same error figures. Coin-flips get named as coin-flips before the outcome is known, because a call the model was fifty-fifty on teaches nothing once it is filed as a win.

Each entry does three jobs. It reports what changed in the re-run since the last one, in medians rather than adjectives. It attributes every large error to a mechanism: usage the model read late, a band drawn too narrow, a replacement baseline that drifted while nobody was watching. It separates the errors the model can fix from the variance it has to live with.

The running version of that accounting is the public accuracy record, which holds the backtest and season-level figures the weekly post samples from. The Ledger is the narrated slice; the record is the arithmetic.

An Invented Week, Walked Through

Take an invented week of the ledger — numbers made up to show the shape, with no real players, no real week, and nothing to look up. Four graded calls: a lead back with a published median of 14.2 who scored 9.6; a second receiver at 11.0 who scored 11.4; a role tight end at 6.8 who scored 13.1; a committee back at 8.5 who scored 8.2.

Run the absolute errors and the week reads 4.6, 0.4, 6.3, and 0.3, for a mean absolute error of 2.9 across the four. The two calls inside half a point are the model doing the job it was built for, and they earn a row in the table and no prose at all. The 4.6 and the 6.3 are what the week's writing is about.

Those two misses fail in different ways. The lead back's 4.6 is a usage error: the median assumed a snap share that a committee split had already begun eating, and the entry says how many weeks of evidence the usage model needed before it moved. The tight end's 6.3 is a distribution question, since the median said 6.8 while the published P90 sat near 13, which puts the outcome at the top of a band the model had already drawn. A band that never gets caught out at its own P90 is a band too wide to be useful, so the ledger reports the error and the band in the same sentence and lets the reader weigh them.

Staying inside the same invented frame, one more number decides whether any of this becomes a mid-week post: the size of the move. A median that shifts 0.3 and leaves the start order untouched is a table update that ships silently. A median that shifts 2.4 and pushes a twelfth-ranked flex past the eighth has changed a decision, and changed decisions get paragraphs.

The Standing List Of Things We Skip

We skip breaking injury news on purpose: it is a speed business, and we would lose it. A report has nothing for the model to consume until usage actually moves, and when it moves, the boards absorb it on the next run and the ledger explains it after the fact.

Borrowed consensus is off the list for a different reason. A page that ranks by what the room already believes has no error of its own to report, and a site whose whole argument is public grading has no use for content that cannot be graded.

Hype churn is the third category: the weekly sleeper carousel, the annual roundup of players nobody can supposedly afford to miss, the recycled take under a new headline. Those pages exist to be indexed. One page per question, revised in place with its history left visible, is the shape we keep.

Two smaller omissions are worth stating plainly. There is no player or team photography anywhere on this site, which is a licensing fact and not an aesthetic one. Announcements of features that do not exist yet are the second, and leaving them out keeps a schedule honest in the only way a schedule can be.

Where This Puts You

For a draft that has not happened yet, this page is a map of what else is worth reading, and the strategy library is where the arguments themselves live. For a season already underway, the weekly Ledger is the thing to read against your own lineup calls, and the machinery behind every number in it is spelled out on the methodology page.

Read the record before you lean on any of it, including this piece. A schedule that advertises transparency is worth exactly what the published error figures support, and those figures sit on the record page for anyone who wants to check the claim against the arithmetic.