Skip to content

Method

how we work · the one rule we never break · sources · today

How we lock a call

Football in.
Forecast out.

Six stages, every one on the record. Nothing from outside the game gets in, and nothing gets quietly changed once it’s locked.

  1. RESEARCHThe Scout gathers only sourced facts. Anything it can't trace to a dated source is dropped at the door, before it can reach a page.
  2. FORECASTAn independent model turns football features (ratings, form, rest, home advantage) into a win / draw / loss probability. The model owns the number. No writer can move it.
  3. WRITEThe Scribe drafts 'The Read' in the house voice, working from the sourced facts alone, never from memory.
  4. CHECKDeterministic gates run: every name must trace to a fresh source, the copy must reference nothing from outside football, every claim must be gradeable. A gate that blocks is the system working, not failing.
  5. LOCKThe call is timestamped and locked before kickoff. That lock time is the evidence. We can't quietly re-grade what we've already published.
  6. GRADEAt full time the Ref grades the call against the result and scores each claim. The verdict joins the record, for good or ill.

You don’t have to take this on faith. Watch the desk do exactly this, step by step, on the live desk.

The one rule

The forecast is built from football alone. Nothing from outside the game feeds the number, and none is ever shown beside it.

Team strength, recent form, rest and congestion, home advantage. Confidence is shown as a plain frequency, a 60% chance is about 6 games in 10, and only ever as a frequency. The honesty of that number is the whole point, so it is graded in the open on the record.

Our model, and how it earns its place

The number isn’t borrowed from anyone. Our model reads a match the way football does: each side’s strength, recent form, rest, home advantage. It shows its working on every call, so you can follow exactly how the inputs become a probability. You are never asked to trust a black box.

And we don’t stand still on one formula. Rival models run quietly in the background against the live one, scored only on games locked before kickoff. A challenger takes over the live call only when it genuinely forecasts better across many graded matches, never because it is new, never on a short hot streak. The method improves the way the record does: in the open, on evidence.

What we will never do

  • Invent a player, squad, lineup, or stat. Every factual token traces to a dated source (seven days or fresher) or it does not ship.
  • Take any signal from outside football. Not as an input, not in the copy, not as a benchmark.
  • Backtest. Predictions are locked before kickoff and judged forward-only.
  • Treat one result as a verdict on the model. Quality is measured across many graded matches, never a single one.
  • Publish to the public surface on its own. Unattended runs reach the shadow surface only. A human opens the public door.

Sources

Every figure is provenance-tagged and dated. Qualitative tactical reads are labelled opinion, never dressed up as fact.