Draft Lab — Direborn
Dota 2 Captain's Mode draft trainer. Next.js (App Router) on Vercel, data from the OpenDota API.
Modes
- Draft Simulator — full 24-step Captain's Mode vs an AI (Easy / Hard / Team style).
- Live feedback: S–F grade, pros, risks and better options after every pick and ban.
- Blind: no hints; full graded review at
/review after the draft.
- Ranked helper (
/ranked) — fully manual: enter bans, your team's picks (by position) and enemy picks
as they happen in a real ranked All Pick game. Suggests heroes for your position (or any open slot):
pro position data, Immortal win rate, counters vs enemy picks, synergy. Grades your pick and lanes.
- Multiplayer (
/play) — captain vs captain in a shared room with a 6-character code. Spectators can watch.
The server runs the turns, timers, reserve time and timeouts; each captain gets their own live or blind grades.
- Draft Quiz — a recent pro match from OpenDota; guess each of the 10 picks (100 exact / 60 strong alternative / 0).
- Teams — shared team database (Redis): players, positions, hero pools, and imported pro matches.
Import by match IDs or by OpenDota team ID (recent matches). The team's pick/ban habits per draft slot
become an AI profile, so the simulator opponent can role-play that team. Cards show a scouting summary.
Data
| Route | Source | Cache |
|---|
/api/heroes | OpenDota /heroStats (names, roles, portraits, pro + Immortal stats) | 12 h |
/api/matchups/[id] | OpenDota /heroes/{id}/matchups | 24 h |
/api/quiz | OpenDota /proMatches + /matches/{id} | 1 h / 7 d |
Hero portraits load from Valve's CDN using the image path OpenDota returns.
Optional env var: OPENDOTA_API_KEY for higher rate limits.
Multiplayer setup
Rooms are stored in Redis. In Vercel: Project → Storage → add Upstash for Redis (free tier is fine).
It sets KV_REST_API_URL / KV_REST_API_TOKEN automatically; redeploy afterwards.
Without it the app falls back to in-memory rooms, which reset between server instances (local dev only).
Clients poll /api/rooms/[code] once a second.
Storage
Redis keys: dl:teams (all teams, JSON), dl:history (latest 400 drafts/quizzes), dl:room:* (multiplayer, 24 h).
API: /api/teams, /api/teams/[id]/import, /api/teams/[id]/match, /api/history, /api/health.
Data saved in a browser by older versions is moved to the database on first visit.
Grading (lib/grade.js)
Score = 30 + 0.6 × meta strength + counter value + composition needs, clamped to 0–100.
- Meta: pro contest rate (picks + bans), pro win rate, Immortal-bracket win rate.
- Counters: public matchup win rates vs enemy picks (picks) or vs your picks (bans), min 40 games.
- Synergy: pro-match win rate of hero pairs on the same team (last 120 days, OpenDota explorer, min 8 games).
Picks gain for pairing with your heroes; bans gain for breaking the enemy's pairs.
- Composition: support/core balance, carries, initiation, lockdown.
Tiers: S ≥ 85, A ≥ 75, B ≥ 65, C ≥ 55, D ≥ 45, F below.
Draft order & timer
lib/draft.js — current order (patch 7.40, unchanged through 7.41f): bans F F S S F S S · picks F S · bans F F S · picks S F F S S F · bans F S F S · picks F S. 15 s per ban in ban phase 1, 30 s otherwise, 130 s reserve. A timed-out ban is lost; a timed-out pick is random.
Re-check against the current patch notes when Valve changes the draft.
Positions & lanes (lib/positions.js)
Captains set Pos 1–5 or Flex for each picked hero (on the board or in the review). Unset heroes are auto-assigned to the best open position.
- Position fit: estimated from OpenDota hero roles (Carry, Support, Initiator, Durable, ...).
- Flex: hero is placed at its best open position; a real flex (fits 2+ positions) earns a small bonus.
- Lanes: safe (1+5 vs 3+4), mid (2 vs 2), off (3+4 vs 1+5) graded from head-to-head win rates. Opponent positions are estimated.
- Final grade = 70% picks & bans + 15% positions + 15% lanes.
Next steps
- Scheduled data refresh + hero synergy data.
- Team-style AI trained on a team's real pro drafts.
