The leather log and the schema

I. The shelf

A leather log behind the wheelhouse glass

Behind the glass of a wheelhouse at the Port of Alsea this morning, on the shelf above the radar, there is almost certainly a notebook. It is leather-bound or it is not, it is recent or it is older, but it has columns ruled in it and the columns are filled. Date. High water. Low water. Wind. What came up. Some entries are tidy and some are scrawled and a few are scratched out — but the columns go on, week after week, season after season, the same shape repeated.

The fisherman who keeps it does not call himself a data engineer. He keeps the log because the bar at the mouth of the Alsea is six feet deep on a good day, and the difference between a workable departure and a stranded one is whether he remembers what the tide actually did last Wednesday. The log is operational. It is also, although no one in the wheelhouse uses the word, a dataset.

II. The shape

Two highs, two lows, never quite equal

The Oregon coast runs on what oceanographers call a mixed semidiurnal tide. In any twenty-four-hour stretch, the water comes up twice and falls twice — and the two highs are not the same height, nor are the two lows.¹ The mean range at this latitude sits near five to six feet on an ordinary day, and a spring tide near the new or full moon can pull the range to ten or twelve.¹

Here is the new word for the data scientist reading this: tidal constituent. A real tide is not one wave but a sum of waves, each driven by a different astronomical cycle — the moon's pull, the sun's pull, the tilt of the lunar orbit, the slow precession of the moon's nodes. Sir William Thomson, later Lord Kelvin, formalized the method of resolving tides into these harmonic components in the 1860s, and in 1872 he built the first mechanical machine to sum them.²

The fisherman with the log does not need the math to use the result. He needs only to know what time the water came up yesterday, and the day before, and three years ago this week.

III. The first tables

1853, Coast Survey, in the back of the annual report

The first time the United States published its tide predictions, the year was 1853 and they were not yet a book. They appeared as an appendix to the Coast Survey's annual report.³ The Survey itself admitted the work was “laborious” and the results “not yet in condition to present as scientific data,” but they printed them anyway because the practical value was beyond argument.³

In December of 1866, for the year 1867, the first standalone tide table appeared. The Atlantic and Pacific coasts were split into separate volumes, and the early editions listed only the daily high tides. Low tides were added later. Tidal currents came after that. By 1896, the published tables had widened to ports around the world.³

What was being built, page by patient page, was a structured longitudinal record. A column for the date. A column for the time. A column for the height. Organized by station, indexed by year:

A century and a half before the term time-series database existed, the engineers of the Coast Survey were writing the schema by hand and printing it in books that captains carried into wheelhouses on every American coast.

IV. Old Brass Brains

How tide prediction stopped being a person and became a machine

In 1880, the American mathematician William Ferrel, working at the United States Coast and Geodetic Survey, proposed a mechanical tide-predicting machine of his own design. It was completed in 1882 and entered service the next year.⁴ Each tidal constituent had its own pulley and crank, calibrated to that constituent's period and amplitude. Ferrel's machine summed nineteen constituents and returned direct readings of the times and heights of each high and low water.⁴

A successor was built in 1910 and acquired a nickname inside the Survey that stuck: Old Brass Brains. Together with Ferrel's original, it made the United States' tide predictions from 1885 through 1965 — eighty years of American tide tables out of one room, one machine, one set of brass gears. Digital computation finally replaced it in 1966.³

The point worth holding is that the math being done by Old Brass Brains was the math the fisherman's log had quietly been doing all along, in a different shape. The log was record-keeping. The machine was prediction. Both were running on the same underlying truth: the sea repeats itself, and a faithful record will eventually tell you how.

V. Station 9434939

Waldport, in the modern catalog

The Waldport entry in NOAA's tide-prediction database is Station 9434939.⁵ Open the station page and you see the same three columns the 1867 table printed: date, time, height. The heights are referenced to Mean Lower Low Water, the nineteen-year average of each day's lower low tide. Predictions are accurate within roughly ten minutes on timing and half a foot on height.⁵

The harmonic constituents driving that station are the same constituents Kelvin's machine summed and Old Brass Brains summed. The arithmetic is now done by a server, and the result reaches a fisherman not as a printed book but as a row on a screen — but the loop is unchanged: observe, decompose, sum, predict, verify against the next high water.

A century and a half of patient verification is what gives those numbers the right to be trusted on a six-foot bar.

VI. What scales

The first row of a long, long table

The data scientist who has shipped a hundred dashboards and never wondered who kept records before SQL existed has been quietly outranked. The fisherman with the leather log at the Port of Alsea is not a precursor of the time-series database. He is one of its rows.

A time-series database is what tide-keepers were building. Timestamp, value, station, source. The shape of the schema has not changed in 170 years. What has changed is the depth of the table — from the few thousand observations the 1867 Pacific volume could hold, to the billions of rows NOAA maintains now.

The first row of that table is still a hand on a wheelhouse shelf, opening a leather notebook, ruling a line, writing a number. Patience is what scaled. The numbers came later.