Skip to content

PLAN.md

Latest commit

 

History

History
62 lines (52 loc) · 2.82 KB

PLAN.md

File metadata and controls

62 lines (52 loc) · 2.82 KB
Raw

Kawkab AI — Next Build Plan (v0.12.0)

Context

Project: Kawkab AI — AI Football Coach for Amateur Teams Current version: v0.12.0 Goal: Continue building toward Phase 2/3 (Analyst → Detective)

Note: Previous versions of this plan referenced v0.5.5/v0.6.0. As of v0.12.0 the project has shipped all 25 audit gaps closed, production hardening, 900+ tests, 57 services.

What We Build Next

1. Match Type Detection + Kalman Smoother Integration

Why: Kalman smoother exists (v0.5.3) but is NOT wired into the main pipeline. STATUS.md says it "degrades on fragmented highlight reels" but would help full-match analysis. We need to detect match type and only enable Kalman for full matches.

Implementation:

  • Add match_type inference to CVService.process_video() based on:
    • Duration >= 80 min → "full_match"
    • Duration < 20 min → "highlight"
    • In between → "unknown"
    • Also consider fragmentation rate and avg track span
  • Add match_type field to MatchTrackData
  • Modify AnalysisService._compute_player_stats():
    • When match_type == "full_match", use PlayerPositionSmoother per track
    • When match_type == "highlight", use existing raw delta cap approach
    • Log which method is used
  • Update Bridge.analyze_match() to pass match_type
  • Add use_kalman toggle to AnalysisService.__init__

2. Knowledge Base Expansion

Why: Current KB has ~30 rules + ~19 drills. PLAN.md targets 100 rules + 100 drills by Phase 3.

Implementation:

  • Add 10 new tactical rules covering gaps:
    • Defensive: poor_pressing_shape, weak_aerial_defense, zonal_marking_gaps, defensive_third_errors
    • Transitions: slow_defensive_transition, poor_counter_pressing
    • Individual: midfielder_positioning, striker_pressing, winger_defensive_work_rate
    • Meta: goalkeeper_communication
  • Add 5 new drills:
    • pressing_shape_8v8, defensive_transition_6v6, aerial_defense_circles, zonal_marking_game, counter_pressing_4v4

3. Update graphify after changes

  • Run graphify update . or document the structural changes
  • Verify new connections appear in graph

Success Criteria

  • Kalman produces smoother trajectories for full-match test videos
  • Distance estimates improve for full matches (no broadcast-cut fragmentation)
  • Highlight reels continue using existing cap-based approach (no regression)
  • Knowledge base loads all new rules without errors
  • All tests pass
  • graphify updated to reflect new code structure

Files to Modify

  • src/kawkab/services/cv_service.py — match_type detection
  • src/kawkab/services/analysis_service.py — Kalman integration
  • src/kawkab/ui/bridge.py — pass match_type
  • src/kawkab/knowledge/tactics/*/ — new rules
  • src/kawkab/knowledge/drills/ — new drills
  • tests/ — add tests for new features
  • STATUS.md — update with v0.6.0 changes