claude/llws-simulator-logic-0aqbjb #142

Merged

chrisp merged 2 commits from claude/llws-simulator-logic-0aqbjb into main

2026-08-23 01:57:38 +00:00

chrisp commented

2026-08-23 00:11:58 +00:00

Owner

No description provided.

chrisp added 2 commits 2026-08-23 00:11:59 +00:00

Make LLWS simulator bracket-aware and calibrate its futures model d4df0b65fb

The LLWS simulator was overestimating favorites and ignoring games that had
already been played. Two separate causes:

1. Championship futures were used directly as single-game strength
   (p1 / (p1 + p2)). A future already compounds the ~6 wins needed to take
   the title, so this made every individual game as lopsided as the whole
   tournament and re-compounded that edge round after round. Against a
   representative 20-team board the favorite priced at 21.8% simulated at
   44.9%, and the longest shot fell to ~0%.

   Futures are now decompressed to single-game Elo via convertFuturesToElo,
   the same pipeline the other bracket simulators use, and games are played
   with eloWinProbabilityWithParity. The parity factor was calibrated by
   sweeping it until a randomized-draw simulation reproduces the board it
   was fed: at 1000 the favorite simulates at 21.8% and field-wide RMSE
   drops from 0.062 to 0.003. It is overridable per season via config.

2. The simulator never read playoff_matches, so it re-ran the tournament
   from an empty bracket every time and shuffled the draw at random each
   iteration. A recorded loss changed nothing.

   It now loads the seeded llws_20 bracket, places teams in their real
   slots, and replays completed games from their recorded result instead of
   re-simulating them, so an eliminated team correctly drops to zero. When
   no bracket exists (or it has no participants seeded) it falls back to the
   previous randomized-draw behavior, and a seeded bracket is authoritative
   about which side a team is on, so externalId is only required on the
   pre-bracket path.

Guards: a recorded result is only honored when its two participants are the
ones the simulation routed into that game, so a corrupt or out-of-order row
cannot desynchronize the rest of the bracket; brackets seeding an unknown or
duplicated participant now fail loudly.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X5vQZMPeokzfMqHQjq1RDZ

Address review findings in LLWS simulator

🚀 Deploy / 🧪 Test (pull_request) Successful in 3m48s

Details

🚀 Deploy / ʦ🔍 Typecheck & Lint (pull_request) Successful in 1m20s

Details

🚀 Deploy / 🐳 Build (pull_request) Has been skipped

Details

🚀 Deploy / 🚀 Deploy (pull_request) Has been skipped

Details

3fae78c521

Four fixes from code review of the previous commit.

1. Preserve the futures board's dispersion (llws-simulator.ts).
   convertFuturesToElo finishes by rescaling any field onto a fixed
   1250-1750 Elo span, discarding how spread out the board actually is: a
   board with a 22%-priced favorite and one with a 6%-priced favorite both
   came out 500 Elo wide. On a tight board that inflated the favorite from
   6% to 13% -- worse than the raw-futures model it replaced (RMSE 0.025 vs
   0.005), so the previous commit was a regression in that regime.

   buildLLWSElos now maps decompressed strengths by their log-ratio to the
   field's geometric mean, so Elo span tracks real dispersion. Re-swept the
   parity factor across three board shapes rather than one: 550 minimizes
   total error. Simulated vs priced favorite, with RMSE:
     wide       21.8% -> 20.7%  (0.0055), elo span 1355-1682
     top-heavy  28.2% -> 25.6%  (0.0085), elo span 1354-1733
     tight       6.0% ->  5.8%  (0.0026), elo span 1484-1518

2. Rate an unpriced team at the field's median, not 1500.
   DEFAULT_ELO is the midpoint of the Elo output range, not of the field;
   on a typical board it ranked an unpriced team ~6th of 20, so blanking a
   team's odds promoted it. buildLLWSElos now returns the priced field's
   median alongside the ratings (ranks 11th of 20 on the same board).

3. Pick the bracket event deterministically.
   scoringEvents.findFirst with no ordering returned an arbitrary row when a
   season had more than one llws_20 playoff event; landing on a stale one
   silently reverted to a randomized draw that ignored all recorded results.
   Now takes the most recent, matching world-cup-simulator.ts.

4. Fail on a partially seeded bracket instead of discarding it.
   readBracketSlots returned null on any single missing participant, throwing
   away the draw and every completed result with no warning. Since
   playoff_matches participant columns are ON DELETE SET NULL, removing and
   re-adding one participant mid-tournament was enough to put eliminated
   teams back in contention. It now distinguishes "generated but not seeded"
   (0 slots filled -> randomized draw) from "partially seeded" (throws).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X5vQZMPeokzfMqHQjq1RDZ