No ensembles as a vibe. Ratings + trees. Receipts.

Just Elo’s cousin — Glicko-2 tuned per sport, calibrated per game.

MechaPip ELOBot rates seven leagues with Glicko-2, then blends LightGBM and market signals into win probabilities. Daily board on mechapip.com. This page is the open engine: clone it, paper-trade it, read the operator guide. Forecasts are not financial or gambling advice.

Live board GitHub

Seven sports. Seven tunings. One HUD.

Same engine. League-specific rest, travel, starters, and sample size. No lime scoreboard clone. No cave paintings.

NBA

82-game grind. Back-to-backs, altitude, star load. Rest deltas stay in the rating path.

NFL

Short season, fat K. Weather, byes, and QB-out flags punch above their sample.

MLB

162 games, starter swings, park factors. Lowest noise, longest memory.

NHL

Goalie sub-ratings. Travel and B2B matter more than a hot stick.

WNBA

Compact schedule, travel density, rest clustering. Same Glicko-2 math, tighter window.

NCAAF

Roster churn every year. Heavy regression, FCS/FBS context, Saturday volume.

NCAAMBB

Hundreds of teams. Transfer portal noise. Conference strength baked into the prior.

How it actually works

Glicko-2 base

Rating, RD, and volatility per team. Home edge and margin scaling tuned per sport. Missing Glicko-2 falls back to Elo.

LightGBM + LR blend

Tree model on rest, travel, and form. Logistic layer mixes market price when the fit file exists. Ensemble is a switch, not a vibe.

Three nodes

Stream PC paints the HUD. Windows Spot runs the morning bot. Weekly Spot box retrains. S3 is the bus. No second live bot.

Paper first

Committed defaults keep PAPER_TRADING=true and DRY_RUN=true. Live Kalshi money stays off until an operator confirm. This site does not sell bets.

Run it on your machine

git clone https://github.com/JerkyJesse/MechaPipELOBot.git
cd MechaPipELOBot
python -m venv .venv
.\.venv\Scripts\activate
pip install -r requirements.txt
copy appconfig\defaults.env .env
# PAPER_TRADING=true  DRY_RUN=true
# Add KALSHI_API_KEY_ID + KALSHI_PRIVATE_KEY_PATH to .env, then:
# python scripts/health_check.py

Operator walkthrough is instructions.html. Live forecasts stay on www.mechapip.com. This GitHub Pages host does not replace the board.