Changelog

Brooklands sent 18 invites and 6 came back. The other 12 families see nothing that coach publishes, which makes the product look dead to them and pointless to him. Chasing them is the highest-value thing he can do and it was twelve separate clicks.

`row = data as typeof row` narrows to never after the null initialiser, so the next line failed to compile. Local `next build` had passed because it ran BEFORE this scenario was added, and tsx strips types without checking them — so the e2e went green and the Vercel build went red. Rebuild after the last edit, not before it.

The gate was enforced but invisible, which is how you end up with a coach who cannot take bookings and no idea why, and a submission sitting unreviewed for a week because the only way in was a URL you had to remember. Admin menu

Shipping 564d6e8 fixed the render race, but production proved the gate was still unsatisfiable. Measured on www.insidefooty.com AND the apex, headless and headed: render() returns a widget id, the wrapper and cf-turnstile-response input

A real coach could not create an account: the security check was invisible but the form still demanded he complete it. Root cause was an effect-ordering race. The Turnstile container is only mounted on step 1 (the form), but render() was called from the script-loading

Third rating mode. The per-player flow asks a coach to switch mental mode four times for every player — 56 judgements across 14 separate dialogs. Judging "who scanned well today" across the squad at once is a single mode, and much closer to how a coach actually remembers a session.

Second bug behind the retired-model one. With the model fixed, squad analysis still failed — "AI returned invalid JSON. Please try again." Current Gemini flash models spend internal "thinking" tokens out of the SAME maxOutputTokens budget before emitting any output. The real squad-analysis

- RF-DETR (COCO default) + supervision ByteTrack replaces YOLO26n + BoT-SORT - StableIDMapper: stable player IDs across occlusions - TeamClassifier: K-Means k=2 for outfield team assignment - PlayerTracker: p95 speed, distance, smoothed speed per player - Tactical alerts: team width, stretch, compactness, GK positioning

Generate a full squad analysis powered by AI. See team strengths, areas to develop, and tactical recommendations based on your actual rating data — not guesswork.

Add photos to player profiles to make your squad feel real. Simple upload from any device — no fuss.

Players earn badges based on their performance trends. Each badge comes with an AI-written insight explaining what the player is doing well and where they can improve — making feedback meaningful and personal.

Invite parents with one tap and they'll receive a branded email with a link to view their child's profile. Automated reminders keep everyone in the loop without extra effort from coaches.

Parents can now see their child's development over time. View Four Corner ratings, track trends across sessions, and stay connected to their progress — all from a simple, clean dashboard.

Create a full block of training sessions or matches in one go. Pick your day, set the time, and generate weeks of sessions instantly — no more creating them one by one.

Rate players across Technical, Physical, Psychological and Social corners — the same framework used in professional academies. Track individual metrics like scanning, first touch, decision making, and leadership.