Practice vs AICopy Anchor Link

Lets a debater run a full timed round against an AI opponent. Setup is three choices: difficulty (Easy through Legends), topic (a suggested one or your own resolution, plus your side), and opponent (one of the 13 personas at that difficulty). Then the opponent preps: it finds evidence cards and caselist outlines for the topic and writes a short brief, a summary of the clash plus the main arguments for your side and the ones it plans to run. After that you debate through opening statements, cross-examination and closing statements, then get an AI-judged scorecard.

This page is also where the Practice Round Simulator lives now: its old address, /practice, redirects here, and its saved rounds, judge paradigms, replays and team matchups are in the collapsed Round simulator section below the round (see Practice Round Simulator).

• Route: /practice/versus-ai (/practice redirects here)
• Package: debate-practice-vs-ai, composed with debate-practice-rounds by debate-webview's PracticeVsAiSections
• API: POST /api/vsbot/create, /prep, /debate, /judge, /concede; GET /api/vsbot/history; prep also reads GET /api/search

What it showsCopy Anchor Link

A four-step setup (Difficulty, Topic, Opponent, Prep), the live round (two side-by-side panels with countdowns, typed or dictated turns), and a scorecard (each phase scored out of 10 for both sides, totals, a verdict, and follow-up drill recommendations). Per-phase clocks sit in a collapsed "Phase timings" box on the Prep step. During the round, the prep brief stays one click away in a collapsed panel above it.

The opponent's prepCopy Anchor Link

The Prep step searches the card corpus with the topic's key words (filler words such as "should" and "be" dropped) and, separately, the caselist outlines. Up to eight cards and four outlines go to POST /api/vsbot/prep, which asks the model for a brief in a fixed JSON shape: a summary, then two to four arguments per side, each citing the cards that back it. When no model key is set, or the model's reply can't be read, the brief is built from the cards alone: each card's tag becomes an argument on the side the corpus says read it. A card search that is rate-limited or fails counts as finding nothing, so prep still produces a brief.

A "View Debate History" button on the setup screen opens Debate History: every past round (topic, opponent, result, date), newest first, each expandable in place to its full transcript.

How it worksCopy Anchor Link

A port of the Go arguehub vs-bot server and its Vite/React client (the React client is kept as a reference source in the package; the Go server has been removed). The 13 personas carry over field for field from the Go GetBotPersonality switch. Text generation uses ANTHROPIC_API_KEY, falling back to GEMINI_API_KEY; with neither set, each persona answers with an in-character "my systems are offline" line.

Known gapsCopy Anchor Link

Gamification's score and badges are persisted (user_settings.practiceVsAiScore/ practiceVsAiBadges) and shown on the scorecard, but there is no real streak yet: getGamificationProfile always reports currentStreak: 0, since a real day-over-day streak needs a dated activity log this table doesn't have — so the Streak5 badge is currently unreachable. Fixed: user_settings now also carries practiceVsAiLastPlayedDayKey/practiceVsAiCurrentStreak, and debate-practice-vs-ai's advanceDailyStreak (apps/debate-ai.com/lib/practice-vs-ai/store.ts's applyGamificationAward) advances the streak by one on the calendar day after the last round, leaves it unchanged for a second round the same day, and restarts it at 1 after a missed day — so five days running now actually earns Streak5. getGamificationProfile (currentDisplayStreak) reports the streak as lapsed (0) once a day has been missed, without needing a round played to reset the stored value first. The Go original never actually incremented its own CurrentStreak field either (see debatevsbot_controller.go), so this isn't a regression from the port — the bug was already there. A reload during a round returns to the picker. Fixed: reloading mid-round now resumes the same round instead of dropping back to the picker. Only the vs-bot slice of the Go server is ported.

The prep brief is not yet fed into the opponent's turns: the bot argues from its persona and the topic, not from the cards it found. The brief is kept with the in-progress round in the browser only, so it is not part of Debate History.