Prediction did not start when the crank turned. It started a day earlier, with a printed sheet of constants, a spanner, and patience. Here is the setup ritual as US Coast Survey operators performed it around 1910 — reconstructed from manuals and logbooks.
What you needed on the bench
- The constants sheet for the port: amplitude and phase (epoch) per constituent, from harmonic analysis like Darwin’s or Doodson’s.
- The machine’s phase-zero reference: a crank position defined as simulated midnight, usually marked with an engraved line.
- A spanner, a magnifier, oil, and the logbook in which every setting was recorded and countersigned.
The steps
1. Wind the machine to zero. Crank gently to the reference epoch and lock the brake. Every phase is measured from here; a careless zero ruins the year.
2. Set amplitudes. For each constituent, loosen its crank-pin clamp, slide the pin to the engraved scale mark matching the port’s amplitude, and re-tighten. Big M2 pins sat far out; minor constituents huddled near the centre. Veterans checked each other’s pins — a slipped clamp was the classic blunder.
3. Set phases. Rotate each individual shaft (via its setting worm, with the main drive disengaged) until its crank points at the constituent’s epoch angle relative to zero. This is fiddly: you are aiming a pointer at a protractor while the shaft wants to drift.
4. Ink and paper. Fit a fresh roll, fill the pen, set the time-scale gearing (one roll-width per fortnight, say), and run a few test hours. The trace should start at the known height for the epoch hour — if not, back to step 1.
5. Run. Engage the drive and crank — or, on later machines, switch on the motor. Log start time, operator names, and any stoppages. A year of hourly heights took days of shifts; the pen was checked like a patient.
Reading the result
The roll came off covered in ink curves; clerks converted them into printed high/low-water tables. The decoding craft has its own guide: Reading an Old Prediction Roll.
What the manual doesn’t say
How long did setup take? A full port setup on a 30+ constituent machine took most of a working day for two people. Re-running the same port later was faster — settings were logged and could be restored.
What went wrong most often? Phase mis-set by one full cycle of a fast constituent (easy on crowded dials), loose amplitude clamps, and pen skips mistaken for real slack water.
Is the ritual worth re-enacting? Absolutely, on any working demonstrator. My museums guide notes which machines visitors may help crank.
One rule for the bench
If you take a single habit from the 1910 operators, make it this: never trust a setting you haven’t verified twice, by two different senses — spanner plus magnifier, dial plus test run. That discipline is what stood between a harbour and a wreck, and it transfers intact to anything you calibrate today. Go on to decode what all that care produced in Reading a Prediction Roll, or compare the instruments that fed the machine in Staff vs. Gauge vs. Predictor.