Almost Signed Track Record

What the scores can actually tell us

The 60–79 band completed 39 of 50 routes (78%), compared with 25 of 40 (63%) in the 40–59 band. These are the highest and lowest bands with at least 25 outcomes in this period. This describes this sample; it does not establish a future success rate.

104 of 118 resolved routes have a usable non-terminal score recorded strictly before a dated result. The other 14 stay in the outcome totals, with missing timing or scores clearly excluded from the comparison.

Why the final field is not our comparison score

For 27 comparable routes, the archived final-score field differs from the last usable history score before the recorded result. A field called “final” is not evidence that it was available before the outcome. We calculate the comparison from dated history and retain the old field separately in the CSV so the difference can be inspected.

Why a completion rate is not prediction accuracy

The 69% headline figure describes completed routes out of 118 resolved records. Completion rates answer one limited question: how many resolved routes are labelled completed? It does not measure how early we identified a move, whether every reported detail was correct, or the probability of a new transfer. There are also 154 open routes, excluded from that denominator. Treating unresolved routes as successes or failures would change the question.

Read the two score tables together

The first table uses the earliest qualifying history entry; the second uses the last one before the recorded result. They use the same routes but different points in time. Movement between their bands shows why a late-stage score cannot stand in for the first assessment. Scores of 100 and terminal events are excluded from both tables. These are recorded-event comparisons, not proof that the true transfer outcome was still unknown at the time.

Each rate needs at least 25 resolved routes in its own group. This is our display rule, not a statistical guarantee. Routes involving the same player or clubs can be related, the archive is not a random sample, and its outcome labels are not a new independent verification of each transfer. The records support an audit of this archive, not a league table of future certainty.

What is counted

The preserved snapshot contains tracked routes and their stored outcome labels. Provider-only Done Deals are not retroactively counted as successful predictions. The same route is counted once. The comparison uses only history entries dated strictly before an explicit result or completion timestamp, with a score of 0–99 and a recorded non-terminal event: interest, talks, bid, rejected bid, personal terms, agreement, medical or baseline.

Completion rate is completed predictions divided by completed, collapsed and expired predictions. This does not establish the accuracy of every claim in an article. Source groups use the first recorded source, not a claim that this outlet independently confirmed every report. Decision groups use the latest recorded scoring actor; missing attribution stays unknown.

Period filters use the outcome timestamp, falling back to the last recorded update. That fallback is not used to establish pre-result scoring. Missing scores remain missing. Every CSV export includes all records in the selected period, including open routes, the archived final field and the dated scores used in these comparisons.

How Almost Score works · Methodology changes · Report a correction