How Many Rain Days Does a Jobsite Actually Lose?
Construction contracts count working days from the weather record, and the standards that govern the count never say how to read it. What gets read is a daily total — which cannot tell rain at 3am from rain at 10am. We read the same record hour by hour, across 36.6 million station-days at 2,093 American gauges. Measured against the hours crews are actually on site, about half the days the record calls rain days carried no qualifying rain at all.
The standard sends you to the record and stops
A construction contract fixes a number of working days. Weather that stops work can extend it, and disputes turn on how many days the weather should have been expected to take. That number is set before any weather happens — so it comes from history.
Washington State’s Plans Preparation Manual is explicit about where to look, and silent about how to read it:
“If historical working day data are not available, historical rain and temperature data are available from the National Weather Service to develop average working days per month.”WSDOT Plans Preparation Manual M 22-31.10, App. 6 §A6-4(2)
The federal clause governing the resulting claims, FAR 52.249-10, excuses delay from “unusually severe weather” and never defines it. Neither document names a number. We searched and could not find a primary specification that does — so this paper designates no threshold, and reports five.
The same record, read two ways
For each threshold: how often a rain day is registered from the whole calendar day, against how often it is registered from the ten working hours 07:00–17:00. The gap widens as the threshold rises.
Computed on the 17,704,428 station-days from 2,029 stations whose records report to a hundredth of an inch. That classification was built after we found the resolution problem, not before — it looks only at each gauge-year’s own reporting and never at a date, but it is a reading decision made with the results in view. No threshold is designated as the answer; a contract that specifies one may read its own row.
This is not rain waiting until dark
The obvious reading is that rain falls at night. It does not, or not nearly enough to matter. A ten-hour window is 41.7% of the clock, and wet hours land inside it 41.4–42.3% of the time in the cleanest data we have. Weighted by how much rain actually falls, the share sits 1.1 to 1.6 points below the clock share — a real nocturnal lean, and far too small to be the story.
The gap comes from geometry instead. The working window is a slice of the day, and rain arrives in bursts rather than spread evenly across twenty-four hours — so whether a day’s rain lands on the shift is mostly a question of when it happened to fall. If a day’s rain occupied a single hour dropped at random, the window would miss it 58.3% of the time. The measured loss is 49.0%: below that, because real rain lasts more than an hour.
A second finding: the record changed its resolution
Working the numbers turned up something about the instrument. For roughly two decades, most American hourly gauges stopped reporting hundredths of an inch and reported only tenths. The signature is the share of daily totals landing on an exact multiple of 0.10 in — chance would be about 10%.
It is not a flag day — it is a property of individual stations, and roughly half of all station-years are on each side. Where a gauge reports only in tenths, the question “did a hundredth of an inch fall?” and the question “did a tenth?” are the same question. Ask those gauges what share of rain days the working-hours restriction removes, and at 0.01, 0.05 and 0.10 in the answer is 39.2069%, 39.2080% and 39.2088% — one number, not three.
So a working-day table built from 1990s data and one built from 2020s data are comparing two instruments, not two climates. We have not found this stated in the dataset documentation.
Three ways of asking, one answer
Classify every gauge-year by its own reporting resolution — never by date — then ask the same question of different spans. The loss share at a tenth of an inch barely moves.
One rule, applied three ways. The first two spans share no years at all — half a century and the entire coarse era sit between them — and they land 0.03 of a percentage point apart. Pooled across the archive as it happens to be distributed, the figure is 43.6%, but that average is half measurement and half a question the instrument could not answer.
Not every slice agrees. Applying the same rule to 2005–2012 — the years when American gauges were switching back to hundredths — gives 39.6%, well below the rest. Those are the early converters, a small and self-selected group, and rounding does not explain the gap. We report it because it is there, and we do not claim to have explained it.
What this does and does not say
It does not say nature is kinder to construction schedules than the climate record suggests. It says a statistic computed over twenty-four hours answers a different question from the one a contract asks — and that raising the daily threshold cannot fix it, because the correction a jobsite needs varies from 35% to 50% depending on which gauge you stand next to.
This record is precipitation only. It says nothing about days lost to cold, heat, wind or ground conditions — the standards that govern concrete placement and crane operation need temperature and wind, and neither is in this dataset. And it describes the hours at 2,093 gauges, not the hours at any particular site.
Published paper
The full scientific paper, with methods, tables, and references.