GTT ©2026 — The log4 entries · updated 21 OctThe possession model

The log.

What we study, the improvements we find and the tests we run on the possession model — the ones that worked and the ones that didn't.

The work itself happens in our fork of the challenge: rules, possession changes, temporal continuity, hysteresis, minimum duration and smoothing.

Entries: 4
Tests ✓ / ✕: 0 / 0
Best spellF1 · public-5min: —

Results · one line per model, one point per version

Model results over time.

Best spellF1 · public-5min

—

Metric

Split

Dashed line · published baseline for this split

Chronology · 20 Oct → 17 NovSuccessFailedMixedOpenNo outcome

Kick-offPhase 1 deadline

26 Oct02 Nov09 Nov16 Nov

MW01 — Matchweek 1Week of 19 Oct

21Oct 2026

Study· 13 minWho has the ball? The possession problem, drawnWhat H, A, D and X mean, why spellF1 punishes flicker, how to spot a change of possession when the ball is missing 41% of the time, and the smoothing toolbox — every idea drawn on an invented toy match, plus the plan we test first.

20Oct 2026

Decision· 1 minThe fork stays public; the match data never enters gitThe rules require a public repository at submission, so we work in the open from day one — and keep every byte of match data out of git.

20Oct 2026

Study· 1 minReading the challenge before writing a lineThe data landed today. Before any code: what the metric actually rewards, why the public window can mislead us, and what an invisible ball means for the problem.

20Oct 2026

Change· 1 minProject log scope and study-first workflowdocs/README.md , section Project log : The team log records what we study, the improvements we find and the tests we run on the possession model in this repository — not website changes. Workflow: stu