TOOLS & CONTEXT

Bullpen Outlook

Head-to-Head History

Open to scan up to 30 days of historical picks for the current matchups.

Team Predictability

Open to rank all teams by how often the model calls their games right.

Edge Calculator

                Model win prob

                %

Fair
-150
Min +4pp
-127
+8pp Goldi cap
-108
<4 noise
4-8 GOLDILOCKS
8-15 caution
>15 trap

                  Check a line

                  e.g. -120 or +150

Display-only -- the automated model gate stays frozen at 4-15pp; the 4-8pp Goldilocks tier is for your manual vetting. The 8-15pp band keeps logging into the OOS ledger for the July Upper-Band Edge Decay study. Edge vs Kalshi-fair; non-linear (a 4pp edge moves the line ~17 pts on a 54% fav, ~125 pts on an 84% fav). Never affects picks.

Pipeline health —

Loading health snapshot…

Gozorp's MLB edge

Daily MLB slate predictions with multi-rule grading

… loading visits

Daily MLB Slate · The Quant Terminal

Find the model's edge.

Multi-rule grading of every game on today's slate. Pre-game picks locked at first pitch, live win-prob trajectories during play, and a deep-analysis panel under every row.

0

Games today

0

A · A− picks

0

Live now

View today's slate ↓

Scroll

A-GRADES TODAY

–

of the full slate

BIGGEST EDGE

–

–

MODEL HEALTH

…

pipeline heartbeat

Browse the slate

Ask the Slate

Top Probable Outcomes

The Slate

Perfect Pair HEURISTIC

Loading…

PitcherList SP Report VOLATILE · UNVERIFIED

Third-party (pitcherlist.com) probable-SP extraction. Read-only reference — never feeds the model, picks, or [PROJ] badges. The source can flip values between requests; statsapi remains official truth.

Loading…

About this dashboard

Static GitHub Pages dashboard for the mlb_edge project. Slate data is baked into docs/data/ via a GitHub Action on every push, so the dashboard is self-contained — no cross-origin fetches, works whether the source repo is public or private.

Grades are produced by an eight-rule grader: SP edge with small-sample dampening, PQI (bullpen quality), team-quality modifier, F-signal cap, compound-small-sample cap, Odds-API guard, PLATINUM gating, and Stage 1/2 agreement. See the README for the full rubric.

Autonomous updates: a daily workflow runs predict.py at 07:00 UTC, generates the new slate, and commits it. When you're viewing today's slate, the dashboard polls every 3 minutes for changes (look for the LIVE indicator). When the model re-runs after a late SP scratch or line move, the table flashes briefly and updates in place — no page reload needed. Past slates are marked ARCHIVED and don't poll. Future slates show PENDING until predict.py can run for them.

Interactive query: ask the slate about specific games or categories using the query box above. Searches the loaded CSV + parlay data — no LLM call, no network.

Not financial or betting advice. Sports wagering is governed by local laws.

Dashboard help

Quick reference for new visitors. The model predicts every MLB game on the slate and grades its confidence in each pick. Click any slate row for the full reasoning behind a pick.

How to use it

Pick a date in the date picker. Today auto-refreshes; past dates are archived.
Scan the Grade column. A and A− are the model's most confident calls. C/D = skip.
Click a row to open the deep-analysis panel: pitching matchup, lineup edge, bullpen state, counter-signals, what would change the call.
Use Simple mode (default for new users) to hide expert columns. Toggle to Advanced when you want the full data.
Ask-the-Slate (the > prompt below the picker) lets you search the slate in plain English: "best pick", "show A grades", "parlay legs", or a team name.

What each column means

Matchup

Away team @ Home team. A "(G2 of 3)" tag means it's game 2 of a 3-game series.

Pick

Which team the model favors to win, plus the model's win probability for that team.

F5

First-5-innings win probability. Reflects the starting pitchers facing each other before bullpens take over.

Full

Full-game win probability. The model's headline number — combines F5 + bullpen quality + late-leverage signals.

Fair

Market-implied "fair" win probability, with the sportsbook's vig (juice) removed. Think of this as the market's honest opinion.

Edge

Gap between Model and Fair, in percentage points. Positive = model thinks market is underrating the pick. >5pp is meaningful.

Pred

Model's projected final score (away–home), in runs.

O/U

Over/Under recommendation: side (OVER/UNDER), the Vegas total line, and the edge in runs. Empty = no market signal.

Tier

Pre-grade conviction level: PLATINUM > GOLD > SILVER > BRONZE > SKIP. Assigned before the rule layer runs.

Grade

Final grade after the 8-rule grader applies caps and adjustments. A/A− = strong; B/B+/B− = moderate; C = skip (cap fired); D = avoid.

Claude

Claude's executive review of the model's call: CONFIRM (agrees), DOWNGRADE (less confident), OVERRIDE (disagrees with direction).

Result

Post-game outcome: ✓ WIN / ✗ LOSS / LIVE (with score) / PRE-GAME / POSTPONED.

Grade legend

A Strong conviction — multiple confirming layers.

A− Strong, just below A.

B+ Moderate signal.

B Marginal edge.

C Skip — a cap rule fired.

D Avoid — direction disagreement or stress warnings.

Anything else?

This is a research dashboard, not betting advice. The grades reflect the model's confidence in its own picks; results vary, and historical accuracy is no guarantee of future games. Treat A-grade picks as worth examining, not as locks.