Where the numbers come from, including the ones nobody publishes
Every record carries its own provenance. Where two suppliers publish figures that contradict each other, both are kept and labelled — averaging them produces a number nobody published and nobody will stand behind.
The contradictions are the content
Anchored on Onsrud and Amana, whose charts state flute count and RPM. Freud and Vortex publish figures roughly twice as high, which appear to be per-revolution budgets for industrial spindles. Treated as an upper bound, not a target.
AMB are the only vendor who publish both input and output power — 1,050 W in, 600 W out. That ratio of about 57% is applied to the other universal motors, because none of them publish output at all. It is derivation, and it is labelled as derivation.
No manufacturer publishes a speed-droop curve for any trim router. The droop is real and well understood — the dial is good to a few per cent before load, then sags under it, which raises the real chipload, which raises the load again. The magnitude used here is inference, not measurement.
Amana’s own compression-spiral chart is internally inconsistent: its one-flute and two-flute tables do not satisfy Amana’s own feed formula. Compression bits are therefore modelled from geometry rather than from that chart, and the reason is on the record.
Bantam’s published aluminium recipe is roughly 170 times more conservative than their own published maximum removal rate for the same machine. Their beginner recipe targets a safe cut; this model targets a proper chip. Both are defensible, and you are told which one you are looking at.
Soft plastic versus hard plastic is our mapping. Amana and Vortex both use the terms without ever defining which polymers fall in which group.
Nobody publishes static stiffness in N/µm for any hobby machine. A dial indicator on the spindle nose with a known lateral force at the tool tip would settle every rigidity argument in the dataset for the price of an afternoon, and nobody has done it.
What has not been verified
Nothing has been validated against a real cut
The test suites check physical identities and internal consistency. They do not check correctness. No figure this model produces has been compared with what a machine actually did. The test grid exists to close that loop, and the loop is still open. This is the biggest gap in the product and it is not buried in a footnote.
Many feed ceilings are estimates
Carbide 3D, Inventables, OpenBuilds and MillRight publish no maximum feed rate at all. Those machines are flagged in the interface, and there is a field to override with the real number once you know it.
Ferrous metals are excluded deliberately
Steel with a carbide cutter wants 1,500–3,000 RPM on a quarter-inch tool. The lowest spindle floor in the whole dataset is 3,500. Rather than extrapolate past the data, the model declines.
Half of what is on this page says that a named manufacturer’s published figures do not add up. That is the most useful thing Feedbench knows, and it is worth more than a banner. Selling those manufacturers advertising space would quietly end it, so the only adverts here are for our own tools.