Trading the First Five
The first-five-innings market is the cleanest window in baseball. No bullpens. No pinch hitters. No managerial chess. Just the two starters and their lineups, head-to-head. Our F5 model is built specifically for that window.
Imagine asking a chemist to predict the result of an experiment. If the experiment runs for an hour with stable temperature and pressure, you can model the chemistry cleanly — known reagents, known conditions, known reactions. If you let the experiment run for three hours and let an apprentice swap reagents halfway through, you're modeling something completely different. The first five innings of an MLB game are the controlled experiment. The full nine has more truth-telling moments — and a lot more noise.
The F5 model exploits that. It's the same scoring-simulation foundation as the full-game model, but tuned for a tighter window with less variance. This page walks through what changes, what stays the same, and where the edge actually lives.
Why F5 Is a Different Market
The Cleanest Window in Baseball
In the first five innings, both starters are still in the game. Both lineups are largely intact. Managers haven't made strategic substitutions. The defensive alignment is what was on the lineup card. Almost every variable that drives third-time-through-the-order chaos and late-game bullpen wildcard scenarios is held constant. That's the structural advantage of the F5 market — and why a model built specifically for it can be sharper than a generic full-game model truncated at the fifth.
No Bullpens
The biggest source of variance in MLB totals — middle-relief meltdowns and lights-out closers — is removed entirely. Both starters are responsible for nearly every pitch.
Stable Lineups
Pinch hitters, defensive replacements, and double switches mostly happen later. The first time through the order is usually fully intact.
Tighter Distribution
Five innings instead of nine means lower expected scoring and a narrower range of outcomes. The probability distribution is more concentrated and easier to estimate accurately.
The opportunity: F5 lines are often derived by simply scaling a full-game total down. That ignores the structural difference: the first five innings aren't a smaller version of the full nine — they're a different market with different dynamics. Modeling it directly is where the edge comes from.
The Projection Engine
Same Foundation, Tighter Lens
Think of a long-exposure photograph versus a sharp, short-exposure shot. The long exposure captures everything — motion, color shifts, atmospheric changes — but it blurs the detail. The short exposure freezes the subject and lets you see exactly what's in front of the lens. The full-game model is the long exposure: rich context, nine innings of evolving conditions, more story. The F5 model is the short exposure: less context to model, but the parts that matter are crisp.
Mechanically, the F5 model uses the same two-stage scoring-simulation engine as the full-game model. Stage one estimates how many runs each team is expected to score in their five innings. Stage two spreads that expectation into a probability distribution shaped specifically for the F5 window — narrower and more concentrated than the full-game distribution because there are fewer plate appearances to absorb variance.
| Stage | What It Does (F5) |
|---|---|
| Stage 1: F5 Expectation | Estimate expected runs through five innings for each team. Driven primarily by the starter, the opposing lineup, and game environment. |
| Stage 2: F5 Distribution | Convert the expectation into a probability distribution sized for the F5 window. Same underlying logic as the full-game model, calibrated to the shorter horizon. |
| Combine Teams | Merge the home and away F5 distributions to compute the joint probability of every possible F5 total. |
| Calibrate | Map probabilities through a calibration layer trained on settled F5 outcomes specifically — not borrowed from the full-game calibration. |
Why a separate model: A starter throwing 65 pitches over five innings is a fundamentally different actor than the same starter dragged into the seventh on a three-run lead. Modeling them as the same thing — and then scaling down — leaves edge on the table. The F5 model treats the first five as its own market.
Starter-Centric
In the F5 window, the starter is typically responsible for 80%+ of the pitches thrown by their team. The model weights pitcher quality more heavily than the full-game version, where bullpen depth and back-end leverage matter substantially.
A frontline starter facing a weaker lineup gives up runs at a much lower rate than the full-game line, prorated, would suggest — that's where some of the cleanest F5 unders come from.
Lineup, First Time Through
The first time through the order is the hardest matchup for hitters. The model evaluates the projected lineup's offensive output specifically against tonight's starter, accounting for platoon advantage and recent form.
Hitters generally improve their results on the second and third looks at a starter — but that's a full-game phenomenon, not an F5 one.
Environment Still Matters
Park factors, wind, and umpire context all affect F5 scoring just like they affect full-game scoring — sometimes more, because there's less time for late-inning weather shifts or umpire adjustments to dilute the early-game effect.
What Falls Out
Bullpen quality. Closer reliability. Pinch-hit benches. Late-inning defensive substitutions. These all matter for the full-game model — and the F5 model intentionally leaves them out, which is what makes its probability estimates cleaner inside the five-inning window.
Where Edge Lives
Reading the Starter Mismatch
Books often set F5 lines by halving the full-game total and applying a generic adjustment. That works on average, but it breaks down in the specific cases that matter most:
- Lopsided starter quality: One ace facing a back-end starter creates an asymmetric F5 distribution that a simple half-of-full-game line misses entirely.
- Strong-vs-weak split: A great starter facing a weak lineup, or a struggling starter facing an elite lineup, produces an F5 expectation that is much further from neutral than the line implies.
- Park sensitivity: Hitter-friendly parks magnify even small starter quality differences in the early innings, before bullpens get a chance to flatten things out.
- Weather lock: Wind and temperature locked in for the first innings often differ from the seasonal park average. The full-game weather impact gets diluted; F5 doesn't.
Model vs. Market
Like the full-game model, F5 outputs P(over) and P(under) for every line on the board. We strip the bookmaker's margin to get a fair implied probability, then compare to the calibrated model probability. The gap is the edge. If a book prices F5 OVER 4.5 at -120 (fair implied ~54.5%) and the model says the true P(OVER) is 60.1%, the edge is +5.6%.
Important: The F5 model uses the same multi-gate filter stack as the full-game model — minimum edge thresholds, vig ceilings, odds-band filters, and a top-six slate cap. The bar to surface a pick is identical.
The Confidence System
Same Gate Stack, Different Calibration
The F5 model passes through the identical gate sequence as the full-game model — edge threshold, vig ceiling, odds band, slate cap of six picks per night, one per game. What changes is the calibration: F5 probabilities are calibrated against settled F5 outcomes, not full-game outcomes. The model knows the F5 market is its own thing and tunes itself accordingly.
| Metric | How It Works |
|---|---|
| Edge | F5-calibrated probability minus the de-vigged fair F5 probability. Cleanest measure of mispricing in this market specifically. |
| EV | Expected value per unit, computed from the F5 probability and the actual F5 odds offered. |
| Tier | Confidence tier driven by edge strength. Higher tiers, larger unit allocations. |
Important: F5 totals are still subject to single-game variance. The model is built for expected value across many bets, not certainty on any one of them. A first-inning grand slam can flip an F5 over no matter how clean the projection was.
How We Measure Success
F5 picks are tracked on the same scoreboard as everything else. The metrics that matter:
Win Rate by Tier
Tier-by-tier hit rate validates that higher-conviction F5 picks actually win at higher rates. The model's confidence has to map to outcomes.
Return on Investment
F5 lines often run at sharper prices than full-game totals. ROI captures whether we're paying a fair price for our edge.
Closing Line Value
F5 lines move sharply once lineups are confirmed. Consistent positive CLV means we're ahead of the lineup-driven movement.
Sample Size
F5 markets are noisier per-pick than full-game totals because of how a single early-inning event can swing the result. Sample size matters even more here.
The Bottom Line
The first five innings are the cleanest, most controlled scoring window in baseball — and the market often prices them as if they were just a fraction of the full game. They aren't. The F5 model is built specifically for the dynamics of the early innings:
- Starter-led scoring: we model the matchup that actually drives F5 outcomes — starter vs. lineup, controlled environment
- F5-specific distribution: calibrated for the narrower, more concentrated probability surface of a five-inning total
- Same multi-gate filter stack: edge thresholds, vig ceilings, odds bands, top-six slate cap, one pick per game
- Independent calibration: the F5 calibration layer is trained on settled F5 outcomes, not borrowed from the full-game model
- Tracked transparently: every F5 pick is graded, scored against closing lines, and reported alongside the rest of the slate
Game totals tell the story of the whole night. F5 totals tell the story of the matchup we can model best. We trade them as the distinct markets they are.
Technical Breakdown
The architecture in plain language, without the proprietary internals.
Data Pipeline
| Component | Description |
|---|---|
| Starter Profiles | Season-long and trailing-window pitcher quality, weighted toward early-game effectiveness given the F5 horizon |
| Lineup Quality | Projected batting order strength against tonight's specific starter, platoon-aware, with focus on first-time-through performance |
| Park Factors | Per-stadium scoring environment, applied to the F5 window |
| Weather Feed | Game-time wind, temperature, humidity, and precipitation chance — captured at the time most relevant to the early innings |
| Umpire Context | Plate umpire tendencies, particularly relevant in F5 since the same umpire calls the entire window |
| Live Odds | F5 totals odds aggregated from licensed sportsbooks in real time |
Projection Pipeline
| Step | Process |
|---|---|
| 1. Per-Team F5 Expectation | Stage 1 model produces an expected F5 run count for each team, driven primarily by starter quality and opposing lineup |
| 2. F5 Distribution | Stage 2 spreads the expectation into a probability distribution shaped for the F5 window's narrower outcome range |
| 3. Game F5 Distribution | Combine team distributions to produce the joint probability of every possible F5 total |
| 4. F5 Calibration | Map raw probabilities through a calibration layer trained specifically on settled F5 outcomes |
| 5. Edge Analysis | Compare to de-vigged market probability, compute EV, run the multi-gate filter stack |
| 6. Slate Ranking | Surface only the top six F5 candidates per slate, one per game maximum |
Daily Schedule (ET)
| Time | Job |
|---|---|
| Morning | Settlement: previous slate's F5 picks graded against verified box-score F5 totals |
| Midday | Projection: F5 pipeline runs against confirmed lineups, weather, and odds snapshot |
| 11:00 AM & 5:00 PM | Two fixed lock windows: afternoon games lock at 11:00 AM, the main evening slate at 5:00 PM. Once locked, each F5 pick is final. |
| Throughout the day | CLV tracking: F5 picks compared to live and closing F5 odds |
| Weekly | F5 calibration retrained from a rolling settled-F5 outcome window — separate from the full-game calibration |
Monthly Access
- Predictions only go live when the model finds true edge
- Closing line value tracked on every prediction so you can verify it yourself
- Covers every market we model and we're always adding more
- Cheaper than your average unit size