Practice Round SimulatorCopy Anchor Link

Lets a debater configure and track a simulated tournament round — format, side, AI judge paradigm, and AI opponent persona/difficulty.

Merged into Practice vs AI: /practice now redirects to /practice/versus-ai, where this panel and the team matchup simulator sit in the collapsed Round simulator section below the round setup.

• Route: /practice/versus-ai#round-simulator (/practice redirects there)
• Package: debate-round

What it showsCopy Anchor Link

A setup form, plus every persisted round as its own card: judge-paradigm/ opponent-persona/difficulty badges, submitted-speech progress, a "Generate AI opponent speech" action, a "Generate post-round feedback" form, and a "Get AI judge decision" action, next to which a Scoring rubric card checks the round's own judge paradigm's voting priorities (✅/⬜) against the rendered decision. A Compare your past attempts section shows a chronological win/loss trend across a debater's rounds.

When a round's setup carried an AI opponent persona, its post-round feedback gets an extra "Facing the <persona name> persona again" section — a short, numbered list of what to prep before facing that style again (e.g. pre-writing a framework defense before a Critique opponent, or drilling efficient tagline-first responses before a Fast Flow one). A custom persona gets a generic tip pointing back at its own described style notes, since its free-form instructions can't be turned into structured, style-specific advice the way a built-in persona's can.

The "AI opponent persona" picker now also reuses the AI Practice Opponent idea's custom-persona library: a "Custom opponent persona" option opens a name/style-notes form with "Save to my persona library"/"Share with my team" checkboxes, and "My persona library"/"Shared by your team" sections below it let a saved or team-shared entry prefill that form with one click — the same useCustomOpponentPersonaLibrary hook the AI Practice Opponent setup already uses.

Data flowCopy Anchor Link

opponent/opponent-personas.ts (debate-speech-writer)
  → buildOpponentPersonaFeedbackTips(persona)
  → buildCustomOpponentPersona({ name, notes })

round/practice-round-simulator.ts
  → resolvePracticeRoundOpponentPersonaChoice({ kind: "none" | "builtin" | "custom", ... })
  → buildPracticeRoundSetup({ styleKey, userSide, judgeParadigm, opponentPersona, opponentDifficulty })
  → buildPracticeRoundFeedback(flow, sideKey, judgeParadigm, { opponentPersona })

state/practiceRounds.ts (localStorage: practiceRounds)
  → buildAndSavePracticeRoundFeedback(flow, roundId, sideKey) passes the
    round's own saved setup.opponentPersona through automatically
  → buildPracticeRoundsPanelView()
  → panels/PracticeRoundSimulatorPanel.tsx
  → apps/debate-ai.com/app/practice-round/page.tsx

Matchup simulatorCopy Anchor Link

Pair any two ranked teams in a division and see who the ratings favour. The simulator reads each division's full-season rankings (<division>_full_rankings.csv), so every ranked team can be picked — filter by name or school, or hit Random opponent.

It shows the single-round win probability for each side, the odds of taking a 3- or 5-judge elim panel, and a Simulate 1,000 rounds button that plays the matchup out to show the spread a percentage hides.

The probability is Glicko-2's expected score, the same formula as upstream's simulate_round.py (skelo's Glicko2Model.compute_prob): ratings are converted back to Glicko-2's 1500 scale, and the result is 1 / (1 + exp(−g(φ_opp) · (μ − μ_opp))). Panel odds assume each ballot is independent at that probability.

It appears in two places:

• Practice round (/practice): both sides are free to pick.
• Every team profile (/teams/<team>): that team is locked in as side A, across each division it is ranked in, and any other ranked team can be its opponent.

The math lives in @debate/rankings-adapter's match-simulation.ts; the UI is debate-videos' MatchupSimulator, composed on the practice page by debate-practice-rounds' TeamMatchupSimulatorPanel.

Known gapsCopy Anchor Link

None open. One open Next item remains in TODO.md beyond this feature: a round replay/playback view.