BACKEND ARCHITECTURE
main.py FastAPI app, CORS middleware, mounts both routers
cli.py · the ldt command setup_logging(), then dispatch; unknown command exits 1
routes.py GET /api/champions POST /api/recommend plain def handlers (blocking psycopg) 500 if scoring raises
schemas.py Pydantic models: DraftStateRequest, ChampionInfo, RecommendationResponse
update_ladder.py update-challengers: replace the Challenger player list
ingest_and_process_matches.py run-match-sync: per player, fetch → aggregate → flush in one transaction
patch_maintenance.py daily-maintenance: wipe data on a new patch, refresh champions
services/draft_service.py recommend_champions(): candidates ≥ 20% role share, score = mean pair win rate (else role win rate)
domain/draft_state.py DraftState (position, banned, allies, enemies) Position enum
queries.py every read, upsert and delete
schema.py CREATE TABLE IF NOT EXISTS × 6
connection.py
db_connection(): pooled psycopg (1–4 conns), commit or roll back
external/riot_api.py Riot API: 20 req/s + 100 req/120 s limiter, 429 → Retry-After or backoff, up to 3 tries Data Dragon: plain requests, no key, no limiter
Neon Postgres metadata · champions · challenger_players matches · champion_role_stats champion_relationships Riot API LEAGUE-V4 + MATCH-V5
Data Dragon CDN versions + champion.json
core/ config.py: .env → RIOT_API_KEY, DATABASE_URL, FRONTEND_ORIGINS; na1 / americas, queue 420 · logging_config.py · imported by main, cli, connection, riot_api
request path · Cloud Run
batch path · GitHub Actions
include_router
dispatch subcommand
init-db
recommend_champions()
champion_dictionary_from_db()
candidates, win rates, relationships
queries.py (all three)
all three
HTTPS
HTTPS