Our reports are written by models. Here are the rules.

What the models actually do.

A Draftanomics report is a synthesis of two things: the written record of people who watch film for a living, and the statistical record of the player himself. Before a word is drafted, the pipeline builds a deterministic fact sheet — the prospect's consensus standing, production, athletic testing, age and program context, plus every scouting observation we've ingested, each tagged to a named scout at a named outlet (the full roster is on /sources). The model's job is to write the read that fact sheet supports — and nothing beyond it.

The hard rules.

Every draft passes through a validator before publication, and the validator is not a spell-checker. It enforces four rules with no exceptions:

• Film language requires a citation. Any claim about how a player looks on tape — bend, hand usage, route feel, play strength — must trace to a specific observation from a named scout. No citation, no sentence. That's why film claims on a dossier carry footnotes.
• Every number must be in the data. A forty time, a rank, a snap count, a dollar figure — each one is checked against the live fact sheet. A number the data can't produce is a rejected draft.
• No hedge-stacking. “Could potentially perhaps develop into” is how prose hides that it has nothing to say. The validator rejects it.
• No template constructions. Reports that read like a mail merge — the same skeleton with the name swapped — are rejected as a class.

Rejected drafts never publish. There is no fallback that quietly ships a weaker version; the report either clears validation or the dossier goes without one.

What we don't do.

We don't watch film, and we never pretend to. When you read a film claim on Draftanomics, you're reading a cited scout's eyes, not ours — we tell you whose. We don't invent observations to round out a profile. And we don't publish template fallbacks: early on we made a deliberate call that an empty dossier is more honest than mechanical prose, and we cut auto-generated filler from the site rather than let it dilute the reports that earn their place. When a player has no scouting coverage yet, his report says so and sticks to the numbers; when there isn't enough signal for a report at all, the dossier says that instead.

Why we work this way.

Draftanomics's whole pitch is receipts. The model's board is backtested out of sample and its methodology is public; the writing is held to the same standard. Machine composition is what makes that standard enforceable — every claim in every report is traceable to a source or a stat, checked by a process that doesn't get tired and doesn't get attached to a take. We'd rather tell you exactly how the sausage is made than ask you to trust a byline.

Who we cite and what data we build on: /sources. Questions or corrections: christopher.gardella@gmail.com.