New in FencingLab 3.0
1. AI Daily Training Plans
- Every day, the AI builds a personalized training target based on your recent video analyses.
- Two parts:
- Fencing drills — e.g. footwork ladders, bladework exercises, each with a description and duration/reps.
- Physical conditioning — e.g. jump rope, push-ups, agility work.
- The next day, the AI reviews your latest video to mark the previous target as met / partial / missed, and adapts the new plan from your actual progress.
2. Resumable Video Uploads
- Videos now get a stable, content-derived ID, so if an upload fails or is interrupted, retrying continues the same upload instead of starting over or creating a duplicate.
- Larger, more reliable uploads via chunked transfer.
3. One-Tap Video Import
- Share a video straight into FencingLab from Photos, Files, or any app using the iOS share sheet.
- Pick the title, session type (training/competition/drill), and visibility before uploading.
4. Improved AI Analysis
- Technical breakdown with coaching moments (screenshot thumbnails of key actions) and an interactive timeline so you can jump straight to the moment that needs work.

Fixes for things that were actively breaking video import/playback:
- Session expiry no longer breaks video import. The app now refreshes your login token every time it comes to the foreground, not just on a
fresh launch — previously a session left backgrounded past the 7-day token lifetime would silently fail every action (including "Import
Video") until you manually logged out and back in.
- Uploads survive network blips instead of restarting. A failed chunk (timeout, dropped connection, server hiccup) now resumes from the last
successfully-uploaded chunk when you tap Retry, instead of re-sending the whole video from scratch. Bad requests (not just transient errors)
still fail cleanly rather than retrying forever.
- Fixed the "kicked back to Training Videos" bug. Watching a video (or viewing its details) no longer gets silently interrupted and dumped
back to the video list whenever an analysis finishes in the background — traced to a SwiftUI view-identity bug in the tab badge, now fixed.
- Video captions no longer cut off mid-sentence. The on-screen caption during playback showed at most 2 lines and truncated with "…"; now it
grows to fit the full description.
- Coaching moments now show a screenshot of the video frame at that moment (backend already generated these; the UI just wasn't displaying
them).
- New: scoreboard & touch-indicator detection. The model now explicitly reports whether it can see a physical scoreboard or the
touch-indicator light apparatus in the shot (these are real objects in frame, not video overlays, and can drift in/out as the camera moves) —
and states plainly when it doesn't see one, instead of silently leaving score fields empty.
- New: automatic retry on a bad sample. Same video, same prompt can occasionally produce a degenerate, repetitive response purely from model
sampling variance — this is now detected and retried once automatically before being accepted.