Two technicians, two roofs

Four parts of arithmetic have all divided by the same assumption: that a completed survey is a completed survey. It is not. The last variable in this series is the only one that is not really about cost at all, and it moves the number more than any of the others.

Two technicians go out on the same Tuesday. Both do two jobs. Both file on time. Both look identical in every system you have.

One of those surveys goes to design and never comes back. The other comes back eleven days later because the main panel label is legible in none of the six photos taken of it, because the existing array was documented as existing PV present with no microinverter model, or because the imagery went to a lender without an azimuth overlay and was rejected on sight.

Almost every rework we have traced went back to a photo nobody took, not a photo taken wrong.

That distinction matters, because it tells you what kind of problem this is. A photo taken wrong is a skill problem, and skill problems are fixed with training. A photo nobody took is a specification problem, and specification problems are fixed by writing the specification down.

What a redo actually costs

Three costs stack, and only the first one is ever counted.

• The second visit. A full slot, at the fully loaded number from Part 01. Not a wage, a slot. The redo displaced a survey that would have completed.
• The downstream stall. A design team blocked, a permit set that slips, an install date that moves. This is usually the largest of the three and it lands in a different department's budget, which is exactly why nobody adds it up.
• The deals that quietly die. A small share, but each one carries the whole project margin. A homeowner who has now taken two afternoons off for the same site visit is a homeowner listening more carefully to the next company that calls.

Put a number on the gap

Ten points of rework rate is not a rounding error, and it is entirely invisible in the place everybody looks. Both technicians cost the same per hour. Both complete the same number of surveys per day. The gap only exists in a number most installers do not track at all, which is the share of surveys that never come back.

How to close it without hiring anybody

Rework is a spec problem, so write the spec. The categories that produce most of it are stable and short:

Every item there is something a design team has been blocked on. None of them require a better technician. They require a checklist that names the artifact rather than the intent.

Where the series lands

Five parts, one argument. A completed residential site survey costs roughly three times what the wage-divided-by-jobs number says. Utilization, not wage, is what moves it. Headcount is a staircase and demand is not, so a crew sized to the trough with the peak bought is the structure that survives a real calendar. And underneath all of it, the rework rate quietly re-prices everything else.

None of that says build or buy. It says the two options should be compared with the same arithmetic, which is not how the comparison usually gets made. If you run these five models on your own inputs and conclude you should keep your crew, this series did its job.