Back to roadmap

Roadmap

Build checklist for players and for us. Live means it is on production. Nested rows show how close a family is to finished. Use the arrows or area tiles to move between sections.

Coaching AI

Match-linked VOD analysis: facts first, narrative second. Valorant sets the quality bar.

50%
9 live · 3 done · 12 planned

Pipeline

  • Async VOD analysis (video + match data)LiveValorant deepest; CS2/Deadlock VOD paths exist; LoL is stats-led for now
    VALCS2DL
  • Web YouTube submit for ValorantLive
    VAL
  • Knowledge bank & guide generationLive
    All
  • Analysis feedback (thumbs up/down)Live
    All
  • League match-stats coaching pipeline (web)LiveProcessLoLAnalysis + /lol/analyze; not full VOD parity yet
    LoL

Coaching quality

  • Round-linked top issue, verdict, coaching tagsLive
    VAL
  • Rank tone, feedback loop, eval harnessLive
    VAL
  • Claim integrity: golden fixtures + contract testsDoneclaim_contract suite live in CI; expand fixtures as leaks appear
    VAL
  • Behaviour catalogue foundationsLivebehaviours/catalogue + valorant.yaml
    VAL
  • Event normaliser + detector enginesLiveintelligence/events + rule/stat/vision engines; framework default live
    VAL
  • Multi-cohort comparison layer (rank + own history)Donecomparison/providers; more cohorts still planned
    VAL
  • Deterministic measured facts; LLM explainsDoneIntelligence Framework live mode feeds Claude measured facts
    VAL
  • Improvement modelling + climbers cohortPlanned
    VAL
  • One analysis job flow; server-owned scoringPlanned
    All

Next few years

  • CS2 coaching depth toward Valorant barPlanned
    CS2
  • League VOD + stats coaching (desktop parity)PlannedStats pipeline shipped; VOD-linked coaching still open
    LoL
  • Deadlock coaching depth beyond record + logsPlanned
    DL
  • Structured “missed this” / wrong-call feedback on insightsPlanned
    VAL
  • Patch-aware coaching when meta shiftsPlanned
    VAL
  • Auto moment reel from top issues (share + coach brief)Planned
    VAL
  • Domain model: text/distillation first, video laterPlannedTrigger: funding + consented corpus at scale
    VAL
  • Frozen holdout eval set before any model swapPlanned
    VAL
  • Between-round focus prompts (not full live callouts)Planned
    VAL
  • Shared coaching language across FPS titlesPlanned
    VALCS2DL