Precision: Files, Screws, Flats, and Measurement

Difficulty: hard, slow, and unglamorous. It is also the gate on everything advanced. Prerequisites: iron and steel — especially hardening and tempering — plus woodworking and patience.

Why it matters

Three other guides in this repository stop at the same wall and none of them climb it.

• Mechanical power says that if you want steam, "invest in precision machining first: an accurate lathe and boring mill are the actual prerequisites."
• Woodworking says a lathe with a leadscrew is "how you escape hand-filed threads and enter real precision engineering."
• Navigation and timekeeping says "the escapement is where clockmaking is genuinely hard."
• And printing calls the adjustable type mould "the actual invention," while requiring tolerances nothing else here teaches you to hit or check.

Precision is the difference between a craft and an industry. It's what makes parts interchangeable, machines repeatable, and instruments trustworthy. It is also almost entirely a matter of technique rather than equipment — the whole bootstrap below can be done with files, stones, and patience, starting from nothing more accurate than your own eye.

The bootstrap problem, and its answer

Here is the puzzle: to make an accurate machine you need accurate parts, and to make accurate parts you need an accurate machine. Where does the first accuracy come from?

The answer is that some accurate forms can be generated by comparison and symmetry. Flatness and squareness can be bootstrapped without a prior master. Numerical length, angle, and pitch still need a declared standard and calibration.

The three-plate method — the foundation of all precision

This is the single most beautiful trick in engineering and it deserves to be understood properly.

Take two flat-ish plates and rub them together with abrasive, and they do not become flat. They become matching — one convex, one concave — because a hollow and a dome fit each other perfectly.

So use three — and remove material selectively, rather than merely rubbing. Rough-lap A, B, and C only far enough to bring them close, then repeat this loop:

1. Coat A very thinly with a marking medium (soot and oil, or red ochre), give B a short controlled spotting stroke on it, and scrape the marked high spots from B.
2. Compare B with C in the same way and scrape C.
3. Compare C with A and scrape A.
4. Rotate the upper plate between comparisons and keep each plate supported in a consistent way, so one direction or sag pattern does not become the reference.

No plate is allowed to become the unquestioned master. A concavity that fits one convex partner fails against the third plate; repeated three-way spotting exposes and removes it. Blind abrasive rubbing can round edges or make complementary curves, while marking and hand scraping tell you where to cut. Repeat until the bearing marks are fine and evenly distributed over all three pairings.

You now have three mutually checked surface plates, made from rough stone or cast iron, marking colour, a scraper, and time. A skilled scraper can then fit machine surfaces more closely than an ordinary graduated scale can resolve.

The same logic gives you straight edges by marking and correcting three edges in pairwise rotation, and right angles by comparing a square against its own mirror image — any error reverses sign and doubles visibly.

The master tool: the file

Before lathes and mills, the file is the machine tool. Almost every precise object before 1800 — clock parts, lock mechanisms, gun locks, instruments, type punches — was shaped by a skilled hand with a file.

To make one:

1. Forge a blank of high-carbon steel, annealed soft, and grind or scrape it flat and true against your surface plate.
2. Cut the teeth with a sharp chisel and a light hammer, working across the face at a consistent angle, spacing each cut by eye against the last. The chisel raises a burr that becomes the tooth. For a double-cut file, run a second course across the first at a different angle.
3. Harden and temper — see iron and steel. Heat to non-magnetic, quench, then temper only lightly (pale straw). A file must be harder than everything it cuts, so it stays brittle by design.
4. Protect the teeth during hardening by coating the file in a paste of salt, flour, and ale or urine — this stops decarburization and scaling, and it's the traditional recipe.

Using files well is a real skill. Draw-filing (holding the file crosswise and pushing it along the work) gives a far finer finish than cross-filing. Files cut on the push stroke only — dragging them back under pressure blunts them. Chalk the teeth to stop soft metals clogging them, and clean them with a file card. Fit a sound handle over every pointed tang before use; an exposed tang can be driven into the palm if the file catches.

Screw threads

Threads are how you convert rotation into precise linear motion, and everything adjustable depends on them.

The bootstrap sequence:

1. Wrap-and-mark. Wind an even-width strip of paper in a tight, gap-free helix around a reasonably true cylinder and mark along its edge. This gives a rough seed only: tension, strip width, overlap, cylinder error, and marking all affect pitch, so check cumulative lead against an independent length.
2. File it by hand, following that line, with a triangular file. Slow and surprisingly effective — this is how the first screws were made.
3. Make a tap. Cut threads into an annealed tool-steel rod, flute it lengthwise with three or four grooves to give cutting edges and clear the chips, then harden and temper it. You can now cut matching internal threads repeatedly. A die does the same for external threads.
4. The leadscrew lathe — the real leap. Gear the lathe spindle to a long accurate screw that advances the cutting tool as the work rotates. Now the thread you cut is a copy of the leadscrew, and its accuracy is inherited rather than hand-judged.
5. Improve by measuring and correcting. A copied screw inherits the master's systematic lead error; uncorrected generations do not converge on a true pitch. Compare cumulative carriage travel at many positions against an independent length standard, make an error table, and deliberately compensate the tool feed or correction mechanism. Careful finishing may reduce roughness, but copying cannot remove an unknown systematic pitch error.

Standardize your threads. Pick one thread pitch per diameter, make a matched tap and die for each, and stick to it. Interchangeable fasteners are worth an enormous amount and it costs nothing but the discipline. (Joseph Whitworth did exactly this for Britain in 1841, and the chaos before it — where every bolt fit only its own nut — is a good argument for doing it early.)

Boring accurate cylinders

This is what efficient large steam engines eventually waited on. Earlier engines worked with rougher cylinders and forgiving seals, but leakage and an irregular bore limited their size and fuel economy. Wilkinson's 1770s boring machine supplied much more accurate cylinders to Boulton and Watt and made engines practical at larger scale.

The principle: support the boring bar at both ends, and rotate the bar rather than letting it follow the hole. A bar held at one end only will wander into whatever crookedness the rough casting already has and faithfully enlarge it. A bar rigidly supported at both ends, with the cutter fed slowly along it, cuts a straighter, more nearly cylindrical bore instead of merely following the casting. Its accuracy still depends on bearing alignment, bar stiffness, cutter geometry, and a controlled feed.

That support arrangement is the central insight, and it applies equally to cannon, pump barrels, and engine cylinders.

Dividing a circle

Graduated scales are what turn a mechanism into an instrument — sextants, theodolites, clock wheels, gears.

• Exact constructions. Repeated bisection gives powers of two. A chord equal to the radius gives six divisions (and three by taking alternate marks); an exact regular pentagon adds a factor of five. These seed constructions and bisection cover counts of the form 2^a × 3^b × 5^c, where b and c are each zero or one. They do not cover every useful count: 7, 9, 11, and 31 divisions require approximation or a more capable dividing mechanism.
• Step and check. To approximate another count, walk an estimated chord with dividers, note the closure error, and adjust all intervals rather than hiding the error in the last one. Step clockwise and counter-clockwise and reconcile the marks; repeat and verify closure before treating the circle as a master.
• The dividing plate. Once you have one accurately divided circle, use it as a master to index everything else — the pattern for gear-tooth cutting, and the basis of every graduated instrument. Guard it; it's your most valuable object.
• The vernier scale — a genuinely clever idea. A second scale whose divisions are slightly smaller than the main scale's lets you read fractions of a division by seeing which pair of marks aligns. It increases reading resolution, not the instrument's underlying accuracy: graduation, alignment, zero, and calibration errors remain. Vernier published the design in 1631.

Measurement

You cannot hold a tolerance you cannot check.

• Dividers and calipers — inside, outside, and hermaphrodite. Simple forged steel, and they let you compare even when you can't read a number.
• Comparison beats measurement. For fitting work, don't measure both parts — set the dividers to one and check the other. Errors in your scale cancel out.
• Go / no-go gauges — two gauges, one that must fit and one that must not. This is how you get interchangeable parts without precise measurement, and it's the practical foundation of mass production.
• Screw micrometer — a fine screw of known pitch turning against a scale. If the thread advances 1 mm per turn and you divide the barrel into 100, its nominal resolution is 0.01 mm. Accuracy still requires checking zero, backlash, frame flex, measuring force, and cumulative leadscrew error against standards.
• Physical standards — keep a master length and a master weight, protected and used only to check working copies. And keep them reproducible from nature where you can: see navigation and timekeeping for the seconds pendulum as a length standard and a volume of water as a mass standard.
• Write it down. Record the tolerances you achieved and the ones you needed. See mathematics and record-keeping.

The order to build in

Each step makes the next possible. Don't skip.

1. Rough files and a scraper made with the best existing tools and abrasives. They need not yet be precision references; they make the first comparisons possible.
2. Surface plates by repeated three-way spotting and selective scraping.
3. Improve the files and scrapers against those plates, then recheck the plates. The bootstrap is iterative, not a finished plate made before its tools.
4. Dividers, calipers, squares checked against themselves by reversal.
5. Hand-cut screws, then taps and dies.
6. A lathe — wooden bed and simple bearings to start.
7. A leadscrew for that lathe, hand-cut, then mapped against a length standard and deliberately corrected.
8. Slide rest — the tool held rigidly in a screw-driven carriage rather than in your hands. This is the change that makes a lathe a machine tool instead of a fancy way of holding wood.
9. A boring machine with a supported bar.
10. Gear cutting off your dividing plate.
11. Micrometers and gauges.

At step 8 you can make clocks and instruments. At step 9 you can make an engine.

Failure modes

Safety

Fit handles to file tangs and secure work before filing, scraping, drilling, or boring. Wear eye protection against chips and brittle tool fragments. Grinding wheels and stones need sound mounts, guards where practicable, controlled speed, and a tool rest; stand clear during startup and stop using a cracked wheel. Heat-treat in a clear, ventilated area with dry tools, face and skin protection, and a quench vessel arranged so splashes or fire cannot trap the worker.

Lathes and boring machines add entanglement, ejection, and stored-energy hazards. Guard accessible belts, gears, shafts, and chucks; tie back hair and avoid loose clothing or gloves near rotation. Positively isolate the drive and wait for all motion to stop before measuring, adjusting, clearing chips, or changing tools.

Sources and limits

• The Science Museum Group's Whitworth surface plates document the three-plate principle and the use of scraping to establish flat reference surfaces.
• NIST distinguishes an instrument's resolution from its measurement uncertainty. More scale divisions do not by themselves make readings accurate.
• OSHA's machine-guarding and hazardous-energy-control guidance covers exposed power transmission and safe isolation before service.
• Self-generating comparison methods can establish geometric form. Numerical dimensions such as length, angle, and screw lead still require a physical or reproducible standard, comparison against it, and an uncertainty appropriate to the work.

Era notes

Roman files, calipers, and screw presses all existed, and Roman craftsmen worked to genuinely fine tolerances — the Antikythera mechanism (~100 BCE) has hand-cut bronze gearing of remarkable quality, made entirely with files and dividers. What the ancient world lacked was not capability but systematic precision: the idea of a transferable standard, interchangeable parts, and machines that make better machines.

The systematic sequence — three-plate flats, leadscrews compared with standards and corrected, and dividing plates indexing their successors — is largely an 18th- and 19th-century realization (Maudslay, Whitworth, Wilkinson). It can begin with steel, abrasive, simple standards, and a long attention span.

Your knowledge advantages, and they are large: the three-plate method, hand scraping against a marked reference, the reversal trick for checking squares and straightedges, leadscrews corrected against an independent standard, supporting a boring bar at both ends, the vernier, and go/no-go gauges. The meta-point is that comparison can generate geometric form and expose error; it does not eliminate the need for standards when a numerical dimension matters.