A propeller in a cup. No gap, so nothing sets the size except how well the beans circulate.
—
One arm on a short handle sustains somewhere between ten and thirty watts. This is the least certain number on the page.
A grinder either has a gap or it does not. Everything below follows from that one fact.
Bean-on-bean restitution is about 0.35, and the energy a bounce keeps goes as the square of that: 0.35² = 0.12, so a bean struck from below rises to about an eighth of the height it would have reached unhindered.
Stack four layers and the fraction of a strike that reaches the top is 0.124 ≈ 0.0002 — over 99% absorbed. The beans nearest the blade are hit again and again while the ones above wait their turn, which is why a blade grinder is set by how deep the bed is and not by how long you hold the button.
Because there is no gap to grind to. The only thing deciding the size is how well the beans circulate past the blade, and that is set by how many beans deep the pile is, not by how many grams are in it. A struck bean has to fight out through the beans above it, and bean-on-bean restitution is about 0.35, so four layers absorb over 99% of a strike.
A grinder either has a gap or it does not, and everything follows from that. A burr grinder sets a gap and nothing larger gets through; a blade grinder is a propeller in a cup, and the size is whatever circulation happens to produce.
An illustration with a calibrated scale. It is good for comparing one setting against another, not for predicting the microns your grinder will produce. Free-flying grains in a slice, verified against the model, and honest about being a picture.
One arm on a short handle sustains somewhere between ten and thirty watts, and that is the least certain number on the page — which is why it is a slider rather than a fact.
A vertical cut through an axisymmetric chamber, so a lid strike at radius r is spread over the whole ring. Bean counts on screen are a slice’s worth: a 75 mm chamber is ten beans wide, not what is in your grinder.
A ten-second grind at a quarter fill was set to land on filter. The spread is physics; the absolute number is a calibration, though it then predicts hand-grinder effort correctly, which it was never fitted to.
Following every fragment to filter grind is about two thousand particles per bean. Each is carried as one shrinking particle sized by the energy it has absorbed: right distribution shape, wrong number of pieces.
At 750 µm in a 40 mm burr set the channel is a hairline. The number is real and the narrowing is real; the width on screen is not to scale.
The energy is Bond’s law and coffee’s specific heat is about 1700 J/kg/K, but the split between grounds, burrs, housing and air is reasoned from what each machine is made of, not measured. It would also climb through a session as the steel warms.
A blunter ceramic edge crushes where a sharp one cuts, and ceramic insulates where steel conducts, so more fines and more heat left in the coffee. Both directions are well attested; the multipliers (×1.5 fines, ×1.35 retained heat) are estimates, and the blade count and pitch effects come from a hazard model rather than any measurement of a real head.
No torque, no burr flex, no popcorning at the intake, no grounds retained between doses, no dulling. Real burrs do all five.
Compared at equal blade-on time over ten seeds, 2.48 steady against 2.55 pulsed with a scatter of 0.27: a quarter of the noise. The heat benefit is real and shown; the evenness one this model cannot see, and tuning it until it agreed would have been dishonest.
Air drag, static cling (which is why grounds climb a real lid), moisture, oil, bean shape, and roast level: a dark roast is more brittle and grinds finer at the same setting.
Releases in which this page changed, newest last. Derived from the archived copy of every release, not from notes written afterwards — so it reflects what actually shipped. Site-wide passes are left out; they are in the full changelog.