Three-Plate Method

Three, not two. The whole method rests on that number.

1. Cut three plates 150 × 150 mm from 10 mm aluminium plate, as flat as you can get them to start.
2. Mark them permanently A, B and C on their edges.
3. Mark a reference corner on each so you always know its orientation.
4. Deburr all edges thoroughly.

Why two plates cannot work. Rub A against B and they will mate beautifully — as a matched convex and concave pair. Every point of contact is satisfied, and both are wrong. Nothing in the two-plate system can detect it, because each plate's only reference is the other one.

You need to SEE which parts touch. That is the measurement.

1. Apply a very thin, even film of engineer's blue or a marker-pen wash to plate A.
2. Lay plate B on it and move it a short distance in a figure of eight.
3. Lift and look at B: the transferred colour marks the HIGH spots.
4. Photograph or sketch the pattern before scraping.

Modern build spec (derived). Engineer's blue is traditional; a permanent marker wiped thin works for a demonstration in aluminium. The film must be thin enough that only real contact transfers — a thick film prints everywhere and tells you nothing.

The sequence is the algorithm. Do not shortcut it.

1. Blue A, rub against B, scrape the high spots off B.
2. Blue A, rub against C, scrape C.
3. Now blue B against C and scrape BOTH.
4. Return to AB and repeat the whole cycle.
5. Scrape lightly and often rather than heavily and rarely.

The BC pairing is the crucial one and the step people skip. A-against-B and A-against-C alone just copies A's errors into both. It is bringing B and C together — two plates that have never met, each matched to A — that exposes the curvature A was hiding, because a convex A would have made both B and C concave, and two concave plates cannot mate.

Flatness here is measured by how the contact is DISTRIBUTED, not by any single number.

1. After each full cycle, blue up a pair and count the contact spots inside a 25 mm square.
2. Record the count and how evenly the spots are spread.
3. Continue until the spots are numerous and evenly distributed across the whole face.
4. Check any pair in any orientation — all three combinations should look alike.

Joseph Whitworth promoted the three-plate method from the 1830s and built his business on it. His argument was blunt: the true plane is the foundation of all accurate work, and until you can make one, nothing else in a workshop can be trusted. He also demonstrated that scraping produced surfaces far flatter than planing, and the surface plates he sold made accurate machine tools possible for everyone else.

This is bootstrapping, and it is rare. Almost every standard is a copy of a better standard held somewhere more precise. Flatness is one of a handful that can be generated from nothing at all, needing no reference, no instrument and no prior standard — only the geometric fact that three surfaces which all mate pairwise must be planes.

Where the same logic reappears: the reversal test used on the angle plate in this batch is the two-object version of the same idea. Three-gauge methods in metrology separate a spindle's error from the artefact's error the same way. All are self-referencing techniques that extract truth from comparison alone.

Honest limits: aluminium is soft and will not hold a scraped surface under use — this build demonstrates the method, it does not produce a working surface plate. Real ones are cast iron, stress-relieved, and scraped over many hours, or granite lapped by a related process. And the method makes a PLANE, not a plane of any particular size or orientation: it gives you flatness, and nothing else.