Every engagement has two histories. One is the technical record, the architecture notes and metric tables. The other is the operational record, which almost never gets written down: who was awake when the training job finished, and how a team that never once sat in the same room still shipped on a schedule that assumed they would. This is the second history for the work behind VeerNet, the system we built to pull curves off scanned raster well logs for a Texas onshore operator. It is not about the encoder; it is about the map the founding team was scattered across, and what that map did to the schedule.
The temptation, describing a team spread over several countries, is to reach straight for the risks. They are real: handoffs drop, a reviewer and the author of the code they are checking are rarely awake at once, and a thirty-second question sits in a channel for hours. For a small team those costs compound, so the received wisdom is to keep the map tight. We did the opposite on purpose, and it is worth being precise about why it worked and where it would have stopped.
The extra two seats bought coverage, not just capacity
The commercial shape made the geography legible. We offered two tracks: a standard track of 4 FTE over 32 weeks, and an accelerated track of 6 FTE over 16 weeks, which the operator took. On paper the accelerated track just looks like more people over less calendar time, and the naive reading is that six finish in half the time four would. That misses what the extra two seats did, because they were not two more sets of hands on the same shift. They were the seats that made the working day long enough to wrap.
With four people in overlapping hours, the team's real day is one shift long. Work stops when they log off, and the training box, the reviewer, and the open questions go quiet until morning. Add two people at the far end of the clock and a run kicked off at the close of one region's afternoon no longer waits until that region's next morning; someone is already at a desk when it finishes, reads the result, and either starts the next experiment or flags the failure while the first group sleeps. The 16-week window was not reachable by working the same six harder. It was reachable because the calendar day got longer, and the expensive serial parts of the loop, training and review, stopped idling overnight.
The relay only exists if the handoff is written down
Coverage on the clock is necessary but not sufficient, and this is what separates a team that benefits from dispersion from one that merely suffers it. A person awake when the run finishes does nothing if they cannot pick up the thread without a live conversation. The whole advantage collapses the moment the handoff depends on a synchronous chat, because the point of the arrangement is that the two people are precisely not online together. So the discipline that made the relay work was not the map. It was the rule that every handoff left a written trace complete enough for a cold start.
In practice a run never finished into silence. It finished into a short note: what was tried, what changed, what the result was, and the one decision the next person had to make. Whoever picked it up did not have to reconstruct intent from a diff or wait six hours to ask what a parameter was for. That habit converts a spread-out team from people who keep missing each other into a genuine follow-the-sun loop. Without it, more time zones just means more places for context to fall on the floor.
The exhibit below is the tradeoff made tangible. The clock is a single day; drag the time zones the bench is spread across and watch the lit arc, the part of the day with a live owner on the delivery loop, grow while the orange dead band, the stretch with nobody at a keyboard, shrinks. Below a threshold the gap never closes and dispersion is pure risk. Cross it and the relay runs. Keep pushing and a second effect appears, the one nobody puts on a slide.
Past a point, every new zone costs more than it covers
The instrument argues both halves on purpose, because the follow-the-sun story is usually told as if wider is always better, and it is not. Coverage is only the numerator. Every additional zone adds a coordination tax: another seam for context to cross, another set of hours where the answer is wait until they wake up, another standup nobody can attend at a humane hour. For the first few zones that tax is small against the coverage you buy. Past the point where the day is already covered, you keep paying it and buy nothing, and net effective coverage, the hours that move the schedule rather than getting spent re-syncing, turns over and falls.
For our six people the sweet spot was wide enough to close the overnight gap on the training-and-review loop and no wider. We were not trying to be awake every hour, only the two or three that were previously dead, right after a long run finished. Spreading thinner left the gap open; spreading wider bought a few covered hours at the price of an overhead six people could not carry. The number that mattered was never the count of zones. It was whether the specific dead hours on our loop had an owner, and whether the seams between owners stayed cheap.
What this actually asks of a founder
None of this is a case for hiring at random and hoping a relay emerges. The failure mode is easy to reach: a team dispersed enough to lose synchronous overlap but without the handoff discipline to work asynchronously gets the worst of both, the tax with none of the coverage. What made it work was that the geography matched the shape of the work, whose expensive serial steps were long training runs and the reviews that gated them, exactly the tasks that idle overnight and benefit from a second shift. So the founder's job was three decisions: which parts of the loop were serial and idle-prone, where to place the bench so those hours were covered, and how much to invest in the async habits that turn coverage into progress rather than missed messages. Get those right and a map that reads as a risk on a slide becomes the reason the 16-week track was plausible; get one wrong and it is just a slower way to do what a co-located four could have done.
Limitations
This is one team's operational account, not a study. The commercial facts are the real archive numbers: 6 FTE over 16 weeks on the accelerated track against 4 FTE over 32 on the standard one. Everything the instrument draws about coverage as a function of time zones, the per-seat active hours, where the dead band closes, and the shape of the coordination overhead, is illustrative reasoning anchored on that sourced bench, not measured coverage; we did not instrument our own calendar to the hour. The conclusion that dispersion helped is specific to a workload whose bottleneck was long, serial, idle-prone runs. A team whose critical path is synchronous, or one without the discipline to hand work off cold, should expect the opposite sign on the same map.
What the map was actually for
The habit this left is to stop treating a scattered team as a defect to minimise and start treating the clock as a resource to allocate. A spread matched to the serial, overnight-idle parts of the work, backed by handoffs written well enough for a cold start, turns the one property everyone files under risk into the reason an aggressive timeline is reachable at all. The map was never the point. What it bought was a day long enough that the expensive work rarely waited for morning, and a team disciplined enough that when the sun came up somewhere, the work was already moving.
