Hit Rate Versus Projection: What Each Number Can Tell You

Open any prop research tool and two numbers compete for your attention. One looks backward: this player has gone over in eight of his last ten games. One looks forward: a model expects 6.8 strikeouts against tonight's lineup. They are usually displayed side by side as if they were two opinions about the same question. They are not. They answer different questions, they fail in different ways, and confusing them is one of the most common mistakes in prop research.

The distinction is worth being precise about, because a trailing hit rate is the single most persuasive number in this space. It is simple, it is concrete, and it arrives with the emotional weight of a streak. A projection, by contrast, is an abstraction. It asks you to trust a process instead of a scoreboard. Knowing exactly what each number measures is how you keep the persuasive one from overruling the informative one.

What a trailing hit rate actually measures

A hit rate is a count of past settlements. Eight of ten means that on ten specific past dates, against ten specific opponents, at whatever lines were posted on those dates, the over cashed eight times. Every word in that sentence is a condition. The lines were different. The opponents were different. The player's role, the weather, the ballpark, and the platform's settlement rules may all have been different. The hit rate compresses all of that into a single fraction and throws the conditions away.

That compression is the problem. A hit rate is honest about what happened and silent about why. It cannot tell you whether the player beat a line of 5.5 that has since moved to 7.5, whether six of those ten games came against the two weakest lineups in the league, or whether the streak began exactly when a teammate's injury handed the player a role he no longer holds. The number survives even when every condition that produced it has changed.

There is also a quieter issue: you rarely encounter hit rates at random. Tools surface streaks because streaks attract attention, which means the eight of ten you are looking at was often selected for you precisely because it is extreme. Scan enough players and enough stats and pure chance will always hand you a wall of impressive fractions. Selection is doing the work that skill appears to be doing.

What a projection measures instead

A projection is a forward estimate for one specific offer: this player, this opponent, this line, tonight. A good one starts from expected opportunity, applies rates that blend recent play with a longer baseline, and adjusts for the context of this particular game. The mechanics are covered in how projections are built; the point here is what the output claims. A projection does not say what happened. It says what a disciplined process expects, given everything knowable before the game starts.

That makes a projection answerable in a way a hit rate is not. It is pinned to today's line, so it can be graded against today's result. It is produced by a process, so the process can be audited over hundreds of estimates. And because it speaks in probability rather than streaks, it can be checked for calibration: when the process says 58 percent, does the event happen about 58 percent of the time? A trailing hit rate offers no equivalent handle. There is no process behind it to audit, only an arithmetic fact about a window someone chose.

Why eight of the last ten is weaker than it feels

Stacked against each other, the specific failure modes of a trailing hit rate form a list worth keeping visible whenever the fraction starts to feel like evidence.

• Moving lines. The platform saw the same ten games you did. If the line has climbed from 5.5 to 7.5, the streak was earned against numbers nobody is selling anymore.
• Opponent mix. Ten games is not a schedule, it is a sample of one. A soft stretch of opponents can manufacture a streak that the next two weeks quietly dismantle.
• Role changes. Streaks often begin with a change in opportunity, a lineup promotion, an injured teammate, a new starting job. When the role reverts, the hit rate keeps advertising a player who no longer exists.
• Small samples. Ten trials of a roughly even proposition produce eight or more successes far more often than intuition expects. The law of large numbers works over hundreds of trials, not ten.
• Selection effects. You are shown the streaks that survived, never the hundreds of players whose streaks just ended. The display itself is a filter.

When a hit rate does carry signal

None of this makes trailing results worthless. A hit rate earns attention in the situations where its hidden conditions are actually stable, or where it captures something a projection is likely to miss.

Settlement quirks are the clearest case. If a platform settles a stat in a slightly unusual way, or a scorer in one building credits a stat generously, past settlements encode that quirk while a model built on official box scores may not. A player who keeps clearing a line by amounts a projection cannot explain is sometimes a hint that the thing being settled is not quite the thing being modeled.

Persistent roles are the second case. When a player's opportunity has been genuinely stable, same lineup spot, same minutes, same usage, for the full window, the hit rate's conditions stop drifting and the fraction starts to approximate a real frequency. It is still a small sample, but it is at least a sample of one consistent thing. The further a player's context has moved during the window, the less the window means.

And at genuine volume, trailing results stop being a streak and become a record. Two hundred graded settlements of a market is evidence about that market. Ten is an anecdote. The line between the two is not sharp, but the direction is: every additional condition held constant and every additional trial makes a hit rate more like data and less like a story.

How to combine them without fooling yourself

The productive relationship between the two numbers is interrogation, not averaging. Do not blend a 70 percent hit rate with a 54 percent projection and split the difference. Instead, use each number to ask questions of the other.

1. Start from the projection, because it is the only number aimed at today's line. Treat it as the working estimate.
2. Use the hit rate as a discrepancy alarm. If past results and the forward estimate point in sharply different directions, something specific explains it. Find it.
3. Check the boring explanations first: a line that moved, a role that changed, a soft schedule, a window chosen to flatter the streak.
4. If the discrepancy survives, look for settlement or data quirks, the one place trailing results routinely know something models do not.
5. Let volume arbitrate. A market's graded record over hundreds of settlements outranks both a ten game streak and any single projection.

This ordering keeps each number in its lane. The projection proposes. The hit rate objects. The graded record, once it is large enough, decides. What you should never do is let the streak stand in for the probability, because a streak is not a probability. Reading probabilities honestly is its own discipline, covered in the probability article.

How Slateline publishes both

Slateline shows graded hit rates alongside projections rather than instead of them, and the pairing is deliberate. The projection is the product: a forward estimate for the posted line, produced by a simulation process that commits before the game. The hit rate is the receipt: what that same process actually settled at the lines it actually faced, counted misses included. Neither number is allowed to stand alone, and the full graded record, calibration included, is public in the Model Room.

One honesty rule matters enough to state plainly: a useful published hit rate must count results at the lines that were really offered, under the settlement rules that really applied, with the losing days included. Any tool can display a flattering fraction over a chosen window. The question to ask of every tool, ours included, is whether the window, the lines, and the misses were chosen before the results came in.

So keep both numbers, and keep them in role. The hit rate tells you what happened under conditions that no longer fully exist. The projection tells you what a process expects under the conditions that do. Research well with the second, audit with the first, and let neither one talk you into certainty. If you want to see the pairing in practice, the current slate's projections sit on the Signal Board with their graded history one click away.