The Many Clocks

On why nothing here is a race, and what you see instead when you stop counting laps

#

There is a paper schedule on the wall outside Dana’s office, four feet wide, one row per resident and one column per two-week block, running the whole academic year. The program bought scheduling software in 2014 and the software works. Dana maintains the paper anyway, in pencil, and has for nineteen years, because the software will show you a month and will not show you a year, and there are things about a year that are only visible when the whole thing is in front of you at once.

She runs an internal medicine residency at a mid-sized teaching hospital. The people currently in her program applied to medical school between six and nine years ago, made the decision to apply a year or two before that, and will be practicing independently until roughly 2065. The training itself, from the first day of medical school to unsupervised practice, runs eleven years in the shortest path and considerably longer for anyone who subspecializes.

In the same nineteen years she has watched the tools change perhaps eight times in ways that mattered and continuously in ways that did not. The current tools are eighteen months old. Whatever replaces them is being trained now. She is designing, this month, the curriculum for a cohort that will spend most of its working life in a decade nobody can describe, and she has to decide by March.

The pencil is because she erases a lot.

Not a race

#

The standard image for all of this is a race. Who is ahead, who is catching up, whether the labs are outrunning the regulators, whether the workers can keep up, whether one country is pulling away from another. The frame is so common that it has stopped registering as a choice.

A race requires one track, one direction, and a finish line that means the same thing to everyone on it. None of those hold here. Dana is not behind. There is nothing for her to be behind, because the thing she is doing takes eleven years and cannot be made to take four, and no amount of urgency converts a second-year resident into a physician.

The better image is polyrhythm. Several clocks, each keeping honest time at its own rate, none of them wrong, no shared downbeat. What you hear at any moment is a beat pattern, and beat patterns are not failures of synchronization. They are what happens when independent periods run together, and they have their own structure, which repeats, and which you can learn to hear.

Nothing in this is behind. Things are running at different rates, which is a different problem and has different solutions.

The ensemble

#

There are at least seven, and most arguments about the future turn out on inspection to be arguments about which one the two parties are looking at.

Formation time runs about twenty years per depth. It is how long it takes to make a person who knows something: schooling, apprenticeship, the years after credentialing when someone is technically qualified and not yet good. It has not changed measurably in a century and there is no mechanism by which it could. Dana’s eleven years are formation time with a professional wrapper.

Capability time runs in months. What the systems can do moves faster than any prior technology’s capacity has moved, and the interval is still compressing.

Commoditization time runs twelve to eighteen months. This is the gap between a capability being scarce and being free, and it is the least discussed and possibly the most consequential, for reasons below.

Release time is two clocks wearing one name, and they differ in kind rather than rate. A commercial release is an event with a date, a price, terms, and a party who can withdraw it. An open-weights release is irreversible; once the file is out there is no recall, no deprecation, no pricing decision, and no counterparty. Treating these as fast and slow versions of the same thing is the most common error in this whole area.

Deployment time is not a clock at all but a distribution. It runs at wildly different rates in different jurisdictions, sectors, and institutions, and the variance between them is larger than the average tells you anything about. A single number for how fast the world is adopting this is a number about nothing.

A distribution reported as an average is the single most reliable way to say nothing about adoption, and nearly all of the published numbers are that.

Depreciation time governs how long a deployed thing stays worth having. It is the clock that decides whether a capital decision was correct, and it is currently short enough that many institutions are declining to make capital decisions at all, which is itself a decision.

Policy time is the strange one, and it has a property none of the others have. It can run backward. A rule imposed and then withdrawn does not return the world to its prior state, because what the rule destroyed was not a level but a planning horizon, and a reversal is further evidence that the rules move. A migration decision runs on a ten-year window. Volatility destroys the window at any level, and repeal does not rebuild it. Policy time is a ratchet where the other clocks are pendulums.

To that list the arguments about culture add a pair that operate inside a single exchange: what a system does in a second, against the months or years in which anything a person says could conceivably fold back into what a system is. Those two are a matched set and belong to their own argument. They are named here so the ensemble is complete.

Every pair of clocks is a spread

#

Any two clocks running at different rates define a gap. Somebody stands in that gap. What they do there is buy on one clock and sell on the other, and the difference is their income.

That is not a metaphor for arbitrage. It is what arbitrage is. The consultant who reads the capability clock and sells into an organization running on deployment time is holding a clock spread. The firm that hires against formation time and bills against capability time is holding a clock spread. The trader in a market that has not repriced is holding one between information time and settlement time, which is the classical case and the reason the word exists. Every spread the arbitrage essays catalogue turns out, on inspection, to be two clocks disagreeing and somebody standing between them.

A spread is not a gap in knowledge. It is a gap in rate, and it closes when the clocks resynchronize rather than when anybody learns anything.

Which is why the advice everyone gives about closing spreads mostly fails. You cannot educate your way out of a rate difference. Telling an institution running on deployment time to move faster is telling it to change a clock that is set by procurement cycles, liability exposure, staff formation, and the number of people who have to agree. The spread closes when one of those changes or when the fast clock slows, and not before.

Work one through. A staffing firm places contract engineers. Its cost side is formation time: it recruits people who took years to become employable and it pays what that scarcity costs. Its revenue side is capability time: clients want a skill that became valuable eleven months ago and did not exist as a category two years before that. The firm is no smarter than either party. It is standing in the gap between how long a person takes to make and how quickly a demand appeared, and the entire margin is the width of that gap.

Now run the commoditization clock forward. The skill becomes ordinary, the premium collapses, and the firm’s margin goes with it, not because anyone competed it away but because the two clocks it was straddling came back into line. The people it placed are still good at the thing. The gap is gone.

It also explains who captures the value while it lasts. Whoever can stand in both times at once, which is a rare position and a temporary one, because the spread that made it valuable is the spread that will close.

Every arbitrage position in this economy has an expiry date and it is set by a clock the holder does not control.

The one that may govern the rest

#

If the ensemble has a master clock, the candidate is commoditization.

Everything else on the list is financed. Formation is paid for by institutions that need revenue. Deployment is paid for by organizations making a return calculation. Capability itself is paid for by people who expect to charge for what it produces. Commoditization time sets the interval during which anything can be charged for, which means it sets the ceiling on what can be spent on all the others.

If a capability is scarce for eighteen months and free thereafter, then eighteen months is the whole window in which the position it creates is worth financing, and every institution downstream is making twenty-year decisions inside an eighteen-month revenue horizon. That mismatch is the reason so much of what is being built runs on a schedule that makes no sense against the thing it is supposedly for.

Dana’s problem is the same problem from the other end. She has to commit an eleven-year pipeline against a competence that may be free before her current interns finish.

What Dana cannot do

#

The formation clock is the one that will not move, and every scenario turns on that.

If the fast clocks slow, the ensemble comes back into some relation and Dana’s problem becomes an ordinary curriculum problem, hard but familiar. If they do not slow, then no cohort will ever complete its formation inside a stable regime, and every professional entering practice will be a permanent migrant in their own field, and the idea of a generation formed under given conditions stops describing anything. If the likeliest thing happens, which is that the surface stabilizes while the depths keep churning, then the mismatch stops being visible and starts compounding, and it will be legible only to somebody looking back from far enough away.

None of those is a forecast and the essay will not pretend to pick. What holds across all three is the shape of Dana’s constraint: she is the slow clock, she cannot become a fast one, and the decision is due in March regardless.

We wonder whether anyone standing inside a beat pattern can hear it as a pattern, or whether the recurring structure is only ever audible to someone who was not keeping any of the time.

The paper on the wall is a year wide because a year is the unit the program actually runs on, and no interface anyone has built will render one. Dana has looked. The software has a year view and it fits on a screen by making each day too small to hold a name.

So she keeps the pencil, and every August she draws a new grid, and the old one comes down and goes into a drawer with eighteen others. She has never thrown one away. If you laid them out end to end you would have nineteen years of a single institution’s actual time, at the only resolution that shows both a shift and a career, which is a thing that exists nowhere else and that nobody has asked her for.

This essay supplies the clock ensemble as a retroactive key to several standing arguments: the simultaneity problem, the two civilizations, the asymmetric transition, and Margaret’s world. Its central weld, that every arbitrage spread is a clock differential, reframes the Arbitrage series rather than extending it. The two tempos that operate inside a single exchange are named here for completeness and argued in The Two Tempos. Policy time carries the corpus’s open question about policy as a first-order forcing function, and Pax Artifica raises a further complication about whether specification rather than policy is the layer where the durable decisions are made.