The makeTeams(8) helper creates only a 70-pt Elo spread (1800→1730).
With the Champions Stage bracket math this gives team-0 a ~19.6% win
rate — right at the 0.2 threshold, causing the test to fail ~63% of
the time in CI despite 1000 iterations.
Use 100-pt steps (1800→1100) instead, giving team-0 a ~40% win rate
and raising the assertion threshold to 0.25 for a clear safety margin.
https://claude.ai/code/session_019w21Nkf5TvTZHH6oVHaQXR
Four separate postgres() clients were open simultaneously (app, timer,
snapshots, socket), each defaulting to 10 connections, exhausting the
database's max_connections limit. Replaced with a single shared lazy-
initialized client in server/db.ts using a Proxy to defer the
DATABASE_URL check until first use (preserving test compatibility).
Also bumps the CS2 Champions Stage stochastic test from 200 → 1000
iterations to eliminate flakiness.
https://claude.ai/code/session_019w21Nkf5TvTZHH6oVHaQXR
Implements a full CS2 Major tournament simulator with:
- 3-stage Swiss format (Opening Bo1, Elimination Bo1/Bo3, Decider all Bo3)
+ Champions Stage 8-team single-elimination (QF Bo3, SF Bo3, GF Bo5)
- Monte Carlo simulation (10,000 iterations) accumulating QP across 2 majors/season
- Sampled 24-team field per iteration: top 12 guaranteed, remaining weighted by 1/rank
- Stage 3 exits (placements 9-16) sub-ranked by W-L record (2-3 > 1-3 > 0-3)
- Stage assignments stored per-event so actual field composition drives simulation
- Admin CS Elo form for entering team Elo + HLTV world rankings
- Admin CS2 stage setup page for assigning teams to stages and tracking advancement
- Database migration: cs2_major_qualifying_points enum value + cs2_major_stage_results table
- 24 unit tests covering all exported pure functions
https://claude.ai/code/session_019w21Nkf5TvTZHH6oVHaQXR
