Two engines: predict, then referee

There are two entirely different questions a handloader asks, and it is tempting to imagine that one clever program answers both. It does not. What will this load do? is a question about physics — powder, pressure, air, gravity. What did my ammunition actually do? is a question about evidence — a handful of chronograph numbers and what they honestly prove. Vegan’s Loadbook answers each with its own engine, and it keeps them apart on purpose. This piece is about why that separation is the whole design, not a technicality.

Prediction and evidence are not the same kind of truth

A prediction is a model’s best guess about the future. Evidence is a record of the past. They fail in different ways and they earn trust in different ways, so blurring them corrupts both.

• A prediction can be precise and wrong. A model can hand you a muzzle velocity to the nearest foot per second and still be off, because it doesn’t know your barrel, your brass, or your lot of powder. The honest form of a prediction is therefore never a bare number — it is a number with an interval and a grade that says how much to trust it.
• Evidence can be true and uninformative. Five chronograph readings are real measurements, but five of anything is a thin basis for a confident claim. The honest form of evidence is a number with a confidence interval that admits how little five shots pin down.

If you let a prediction quietly stand in for a measurement, you get false confidence — a modeled number wearing the authority of something you actually observed. If you let measurements feed back into a prediction without discipline, you get a model that flatters your data instead of testing it. Keeping the two engines physically separate makes that contamination structurally impossible, not just discouraged.

The creed underneath everything: the statistics engine never predicts, and the ballistics engine never touches your measurements. One referees; one predicts. Neither pretends to be the other.

Why interior and exterior ballistics are separate engines

Even inside “prediction,” there are two distinct physics problems, and they share almost nothing.

Interior ballistics — the problem inside the barrel

From the instant the primer fires to the instant the bullet leaves the muzzle, you are solving a problem in thermodynamics and chemistry. Solid propellant turns to hot gas, pressure climbs to tens of thousands of psi in about a millisecond, and that pressure drives the bullet down the bore while the expanding volume behind it fights back. The unknowns are burn rate, gas energy, pressure, and the bullet’s acceleration inside a few centimetres of steel. The output that matters is the muzzle velocity — and, reported honestly as a percentage of the SAAMI maximum, the peak pressure.

Exterior ballistics — the problem in the open air

The moment the bullet clears the muzzle, the physics changes entirely. There is no more powder, no more pressure, no more barrel. Now it is a problem in aerodynamics and mechanics: a spinning mass coasting through air, pulled down by gravity and slowed by drag, over a flight lasting a second or more across hundreds or thousands of metres. The unknowns are drag versus speed, air density, wind, and a family of subtle spin- and rotation-of-the-Earth effects. The output is the trajectory — where the bullet is, and how much to dial.

These are different equations, different time scales (a millisecond versus a second), different length scales (centimetres versus kilometres), and different sources of error. Trying to model them in one lump would make each worse. Two engines, each doing one job well, is the honest architecture.

Why the engines still combine

Separate does not mean disconnected. They hand off in a single, clean place:

$\text{interior ballistics} \;\longrightarrow\; v_0 \;\longrightarrow\; \text{exterior ballistics}$

Interior ballistics produces the muzzle velocity $v_0$ — the speed the bullet is going as it leaves the barrel. That one number is the seed the exterior model needs to start the trajectory. Everything downrange — drop, drift, time of flight — grows from it. A small error in $v_0$ blooms into a large error at distance, which is exactly why the muzzle velocity is carried with its interval into the trajectory, not passed along as if it were exact.

And your measurements calibrate both:

• A chronograph string over a load lets the interior model learn the burn behavior of your powder in your barrel — the prediction tightens to your reality.
• Your observed drop at distance (your dope) lets the exterior model correct itself to what the bullet actually does downrange — a process called truing.

But — and this is the boundary again — calibration is a deliberate act by you, feeding measured data into a model through a controlled door. It is never the statistics engine silently leaking into the physics. When the referee evaluates whether two loads are genuinely different, it looks only at what you measured. When the predictor tells you where the bullet lands, it never consults your group sizes to make itself look good.

The honest hierarchy

All of this rests on one ranking that the rest of these articles keep returning to:

Measured beats fitted beats estimated.

• A number measured from your own firearm outranks everything.
• A number fitted to published reference data is next best.
• A number estimated from general principles, when nothing better exists, is honest only if it is labeled as such and carries a wide enough interval to admit its uncertainty.

Two engines, kept apart so that each can be honest about what it knows — and a strict rule about which kind of knowledge outranks which. That is the frame. The next four articles fill in the physics and the statistics, one engine at a time.