Somewhere in every tide office stood a ledger no machine ever touched: years of inked gauge records waiting to become constants. Predictors get the glory — constants in, future out — but the constants had to come from somewhere: years of inked gauge records reduced by hand arithmetic (see Kelvin to Doodson). A few engineers asked: could a machine run the problem backwards? The tide analyser was their answer — and it is the rarest beast in the archive.
The reverse problem
Analysis means finding, for each constituent, the amplitude and phase best fitting the observed record — essentially a Fourier decomposition done by least squares. Human computers did it with Darwin’s schedules or Doodson’s full machinery: weeks of ledger work per port-year. An analyser mechanised the integration: feed the marigram trace through a follower, let geared integrators accumulate each constituent’s contribution over the record, read off the constants on dials.
Why so few?
Three reasons analysers stayed rare while predictors multiplied to forty. First, analysis runs once per port (constants change slowly), while prediction runs yearly forever — the economics favoured predictors. Second, analysis demands higher precision than prediction: small errors in constants corrupt every future year, so only the most refined mechanisms qualified. Third, Doodson’s least-squares methods kept improving faster than brass could follow; by the time a good analyser design matured, the analysis office had a new procedure. The Bidston team evaluated analyser schemes repeatedly and kept its human computers — machines assisted, but the mathematics stayed on paper until digital computers arrived (see Why the Machines Died).
What survives
One or two analyser mechanisms survive in museum stores, usually mislabelled as predictors — check the follower arm (it traces an input curve) versus the pen carriage (which draws output). My museums guide notes where to ask. The vocabulary distinction is pinned in the glossary.
What I had to look up twice
Could an analyser replace Darwin’s computers? Partly — they excelled at the heavy integration but still needed human judgment for datum shifts, storm-surge gaps, and faulty gauge periods. The dream of push-button analysis waited for software.
Are analysers related to planimeters and integrators? Directly — they descend from the same family of wheel-and-disc integrators as Lord Kelvin’s integrator and the famous differential analysers. A predictor sums known waves; an analyser weighs unknown ones. Same blood, opposite direction.
Why should I care? Because the analyser story explains the predictor’s shape: prediction mechanised easily because analysis stayed human. The division of labour defined ninety years of practice — setup day on one side, ledger weeks on the other.
The road not taken
The analysers lost — economics, precision, and Doodson’s ever-improving paper methods saw to that — but losing designs teach what winners hide. The analyser story explains why predictors look the way they do: prediction mechanised easily precisely because analysis stayed human. Keep that division of labour in mind when you read how Kelvin’s prototype worked, and spare a thought for the human computers whose ledger weeks the reverse machines never managed to replace.