Lost first, noticed last

A 38-foot gillnetter, hull turned to the sky

The boat comes out of the water at the Port of Toledo yard, lifted on a marine straddle hoist rated to 660 tons and set down on blocks in the open air.¹ For the first time in three years, the skipper sees his own bottom.

It is not a clean sight. Below the waterline the hull wears a crust — chalk-white barnacle plates, the rock-hard tubes of calcareous worms, a beard of weed dragging off the keel. The yard smells of copper and bottom paint. A pressure washer throws a silver arc against the planking and the growth sloughs away in wet grey sheets.

He knew the boat had gone soft. Slower onto plane, hungrier at the fuel dock, half a knot down on the same throttle. He just did not know how much he had been carrying until the water let go of it.

Three years is a long time to leave it. Most skippers come out more often. This one had reasons, the way everyone has reasons, and every reason was cheaper than the lift — right up until it was not.

The water moves in, in order

A clean hull does not stay clean. Put a new boat in the sea and within a day it wears a thin coat of bacteria — a slime film, slick under the thumb.² That film is the welcome mat.

Weeks later the larvae arrive. Barnacles and tubeworms read the slime, decide the surface is alive enough to live on, and cement themselves down for good; the soft fouling makes way for the hard.² Biofouling is only the sea moving in — the bacteria first, then the builders that follow them.

Everything in salt water grows what it must grow, whether or not you have time for it. The hull is real estate the ocean reached before you did, and it never once asked permission.

The tax you cannot see

Growth is not only ugly. It is expensive. A fouled hull is a rough hull, and roughness fights the water across every foot of it.

A naval study put a number on the fight. Heavy calcareous fouling — the barnacle kind — can raise the power a ship needs to hold cruising speed by as much as 86%.³ A slime film alone, with nothing hard on it yet, already bleeds speed and fuel.³

None of it arrives as an alarm. It comes a percent at a time — a little more diesel each month, a run that finishes a little later than it used to. Performance is what you lose first and notice last.

The copper is a guess made in advance

The defense is bottom paint, and most bottom paint runs on copper. Cuprous oxide leaches slowly off the hull into the thin skin of water against it, and that release is what keeps larvae from settling; copper coatings sit on roughly 90% of the vessels protected against fouling at all.⁴

But notice when the choice gets made. The skipper picks the coating, how thick, how soon he will be back — in spring, before a single barnacle has settled. He is deciding what the water will do before the water has done anything.

Statistics has a name for a belief you commit to before the evidence arrives: a prior. In Bayesian terms it is the starting point — what you assume about an unknown ahead of the data.⁵ The copper is a prior, painted on: a bet, placed in advance, on what three years of this one bay will grow.

And like every prior, it is only a starting point. The water gets a vote.

Then the evidence comes back up the lift

Three years on, the boat is on the blocks and the hull is telling you something. The crust is evidence — hard data on how wrong, or right, the spring guess turned out to be. A heavy beard says the prior ran optimistic; a near-clean bottom says it held.

A good skipper reads the hull and adjusts: more copper next time, a harder coating, a haul-out at two years instead of three. That corrected belief — the prior moved by what the water actually did — is the posterior.⁵ Do it for thirty seasons and the judgment that looks like instinct is just a prior the sea has corrected so many times it has run out of new ways to be wrong.

A model is built to run the same loop. It ships with priors — assumptions about what normal looks like — and every day of production data is fresh evidence about whether those assumptions still hold. Let the evidence move the belief, and the model stays honest.

The model that stopped updating

Now the one that didn't. An engineer has a model in production. It shipped clean and fast and right, six months ago, and nobody has hauled it out since. The evidence kept arriving — it always does — and nothing was done with it.

The world the model learned has shifted underneath it. The field calls the slippage concept drift: the distribution of the incoming data changes, and a model trained on the old water keeps steering by it.⁶ Put it in Bayesian terms and it is plainer still — the prior never got updated. The belief stayed frozen at launch while the evidence piled up unread.

A prior nobody refreshes is old copper: a defense tuned for last year's ocean, still leaching against threats that have already changed shape. The loss came on gradually and the noticing came all at once — the quarter the numbers finally dipped far enough that someone looked. The fouling on the hull and the frozen prior in the model are the same debt under two names. Both are the cost of not looking.

The cure is the same shape

The cure on the Yaquina is plain. Pull the boat, blast the growth, scrape down to a sound surface, lay on fresh copper, set her back in the water.¹ It is an open yard — they hand you the hoist and the hose, and the scraping is yours. It happens on a schedule, every fall, because the sea keeps no other appointment.

The cure at the desk has the same shape. Pull the model, look hard at what the world is doing now, scrape off the assumptions that have gone stale, retrain on the new water, set it back. Measure it again next season. Nobody claps for either job.

The water never stops growing what it grows. The only choice is whether you haul out on a schedule or wait for the morning the boat will not make the bar. The skipper books the lift. The model deserves the same scraper — and the same honest hand on it.