Every date on this site is either something we read from a district's own calendar, or an estimate we've clearly labeled as one. Here's exactly how that works.
We collect board-adopted calendars, district PDFs, and school-portal pages: the districts' own sources. We extract the dates (first day, last day, holidays, breaks, half days), then reconcile each district against its federal NCES identifier so the record joins cleanly to Census and enrollment data. For confirmed districts we keep the URL we read it from.
The single most important thing on any page is whether a date is confirmed or a guess. We never hide that in metadata.
Read directly from the district's published calendar. Shown solid, "printed."
Not yet confirmed: estimated from nearby districts or past years. Shown hatched and labeled with the word "estimated."
When we haven't found a district's calendar, each school weekday is estimated from what most calendar-read districts in its state do that day, and marked estimated. Its confidence is measured, not asserted: on districts we held out, days where the state agrees were right 96–99.8% of the time, and days where it splits were right 55–70% of the time. The confidence says which is which.
Where a state has too few calendars to estimate from, we keep the older estimate for most days, but mark out of session the days nearly every school in the country is off — Labor Day, Thanksgiving and the day after, the week of Christmas to New Year's Day, and Memorial Day. Checked against calendar-read districts with their own state left out, those days were right 98 to 99 times in 100.
Every date carries a confidence score derived from how directly we observed it and whether it was corroborated, never a number we simply asserted. Businesses reading the API get that score on every row; to separate confirmed from estimated, filter on source_method. Families don't see a number; they see the plain label instead.
2026-27 is the year we are actively verifying, largest districts first. We publish the real figures:
| School year | Districts | Confirmed |
|---|---|---|
| 2026–27 (current) | 12,023 | 40.0% |
| 2025–26 | 13,744 | not yet audited |
| 2024–25 | 11,621 | not yet audited |
Confirmed = read from the district's own calendar. Everything else is estimated and labeled. For 2026–27, 4,815 of the 12,023 districts with a 2026-27 calendar row have at least one day read from their own published calendar (40.0%), as of 2026-07-26 (S138: data-hub tickets/S138-HI-COLLECT.md, observed_all). District counts: 2026–27 and 2025–26 shells per S138 (data-hub digests/S138-hi-shell-check-r1.md). How each source_method is labeled: the source_method table.
Calendars change — a snow day here, a board revision there. We re-check sources and version the changes, so the archive keeps its history.
For anything that matters — a flight, childcare, a day off work — confirm with your district's official calendar. We're an independent record, not the school. If you spot something wrong, the "this doesn't look right" button on any district page goes straight to us.