Escapement timing on a machine that spaces twice

Part 01

The complaint and what it turned out not to be

The machine was a mid-sized portable dated by its serial to the first half of 1961, in good cosmetic order, with a note from the owner saying it had started putting gaps in the middle of words. I could reproduce it within a line of typing. Roughly every eighth or ninth character the carriage moved two steps instead of one, and occasionally, less often, two letters would pile up in the same position with no movement at all.

My first assumption was the universal bar, because a sticky universal bar can bounce and trip the escapement twice from one keystroke. I cleaned it, watched it, and found it returning cleanly. My second assumption was a dragging carriage, since a carriage that is being pulled too hard can overrun a step. The draw cord tension was reasonable and the rails were clean. Both of those cost me an evening each and both were wrong, and in hindsight the clue was there from the first line I typed: the fault came and went in a pattern that repeated, and a pattern that repeats points at a wheel rather than at a lever.

The star wheel, or escapement wheel, has a fixed number of teeth. If one tooth is bent, worn or has picked up a burr, the fault appears every time that tooth comes round, and the interval depends on how many characters fit between one pass and the next. Counting the characters between two failures gave me nine, then nine, then eighteen, and that was the answer sitting in plain sight two evenings earlier.

Part 02

How the escapement works, in the words I use to myself

The carriage is always being pulled to the left by the mainspring through the draw cord, and the only thing stopping it is a toothed wheel held by two pawls, which everybody calls dogs. One of them, the rigid dog, is fixed to the rocker and normally holds the wheel still. The other, the loose dog, is mounted on the same rocker but is free to move a little within its own limits, and it is the one that catches the wheel while the rigid dog lets go.

Pressing a key pushes the universal bar, which rocks the whole assembly. The rigid dog swings clear, the wheel begins to turn under spring pressure, and it turns exactly until the loose dog catches the next tooth. Releasing the key rocks the assembly back, the loose dog releases, the rigid dog catches, and the carriage has moved one letter width. The whole handover takes a fraction of the time the typebar spends in the air, which is the point of the design: the letter prints while the carriage is still stationary, and the step happens on the way back.

Timing, in this context, means the overlap between those two catches. If the rigid dog releases before the loose dog is in position, the wheel is free for a moment and the carriage runs on by an extra tooth, which is the double space I was chasing. If the loose dog has not cleared by the time the rigid dog catches, nothing moves at all and the letters pile up. Both symptoms come from the same adjustment being off in opposite directions, which is why the machine showed me both.

Part 03

The adjustment itself, and going slowly

On this machine the loose dog position is set by a single eccentric screw on the rocker, and the whole rocker assembly can also shift on its mounting slots. Both move the same relationship, which makes it very easy to correct a problem at one end and create a worse one at the other. I take the rocker mounting as fixed unless there is evidence somebody has moved it, and I work only on the eccentric.

The useful movement was about a sixth of a turn in total, from clearly wrong one way to clearly wrong the other. I moved it a twelfth of a turn at a time, marking the head with a pencil line so I could count, and after each move I typed the same test line: forty characters of alternating wide and narrow letters, then a full row of the letter i, which is the worst case because the typebar is light and the key is often hit fast. Anything less than a full line proves nothing at all, because the fault only turned up every ninth character.

The burred tooth was a separate job. Under a loupe it had a bright raised edge on one flank, probably from somebody forcing the carriage against the dogs at some point in the last sixty years. I took the burr down with a fine slip stone in maybe twenty strokes, checking after every five, on the principle that I can always remove more metal and can never put it back. The wheel is hardened and it did not want to be stoned, which is the correct behaviour for a wheel that has to survive several million steps.

Part 04

What I got wrong on the third evening

Halfway through the third evening I had the timing very nearly right and decided to improve the feel at the same time by giving the dogs a drop of light oil. The machine immediately got worse, and I spent forty minutes undoing an adjustment that had not been the problem. Oil on the escapement dogs slows the handover, damps the loose dog's return and changes the very timing I had just set. The pivots want oil. The working faces of the dogs want to be clean and dry, and so does the star wheel.

That is now written on a card taped inside the cupboard door above the bench, because it is the sort of mistake that feels like diligence while you are making it. The general version of the rule is that I should never change two things between two tests, and the habit of tidying something up while I am in there counts as changing a thing.

Part 05

How it reads now, and what I would check first next time

The machine went back typing a clean line with no repeats over four full pages of test text, which is around eight thousand characters and about nine hundred passes of the wheel. That is the standard I settled on for escapement work, because a page proves very little when the fault interval is nine characters and the failure rate at the margin can be one pass in a hundred.

Next time I will count the interval first. It takes one line of typing and a moment with a pencil, and it would have saved me two evenings on this job. After that, the order I now work in is: check the universal bar returns freely, lift the carriage and work the rocker by hand while watching the two dogs, look at the wheel teeth under a loupe, and only then reach for the eccentric. Three of those four steps cost nothing but time and none of them can make the machine worse, which is the argument for doing them in that order.

The owner asked whether it would need doing again. I said I did not know, which was the honest answer. Escapement timing does not drift on its own, but a machine that gets carried about, knocked, or has its carriage shoved hard against the end stop can move the rocker on its slots, and there is no adjustment that survives that indefinitely.