# Measurement spec — permit duration The definitions this measurement is collected against. Settled before collection, recorded here so a later pass can tell a deviation from a decision. This job is mostly definitions. A number collected against the wrong definition is worse than no number, because it survives review: it looks like the thing it is not. Every rule below exists because a published comparison has already been ruined by getting it wrong. ## 1. Calendar days, always Where a source publishes business days, convert at the source and record both the original figure and the conversion factor used. Never mix the two in one column. Phoenix reports calendar days and Boise reports business days. Placing those two in one column understates Boise by roughly 40% and is the commonest way a permitting comparison is ruined. ## 2. The primary measure is application to issuance In calendar days, for new residential construction. Entitlement is broken out as its own field where the source distinguishes it, and is **never folded into the primary measure**. ## 3. Applicant hold time is INCLUDED, and broken out where the source splits it This is the one genuinely arguable decision here, so the reasoning is recorded rather than left implicit. New Zealand's 13-working-day median is the authority's own time, with the clock stopped during requests for information. It flatters the elapsed wait a builder actually experiences. The quantity this project cares about is **delivery time, not agency throughput** — a house is not delivered faster because the clock was paused while the file sat with the applicant. Portland OR splits staff time from applicant time and is the best-designed metric found anywhere. Where a source does that, carry both. Washington turns out to be such a source, and the split is not small: across the 48 jurisdiction tables in its 2025 report, elapsed time exceeds active review time in 43. Reporting only the active figure would have described a different quantity and called it a permit duration. ## 4. Product class is split Single-family detached and multifamily, recorded as unsplittable where a source mixes them. A source that cannot separate residential from commercial is **recorded, not estimated**. ## 5. First-review-cycle is not elapsed time Some jurisdictions publish time to first response; others publish end to end. The difference is roughly tenfold. Record which one a figure is, and never place the two in the same column. ## 6. Never compute a total where the components differ in scope The existing state workbook refuses to compute an elapsed total for 223 of its 255 rows for this reason, and that refusal is the most defensible thing in it. This rule is why California and Washington are never added, averaged, or ranked against one another here. They measure different intervals, on different populations, with different clock-start rules. Both are permit durations; neither is the other. ## 7. Statutory deadlines and targets are recorded as such Never converted into duration estimates. Chile is the standing proof: the shortest statutory permit deadline found anywhere in the world sits beside one of the longest observed permit stages. A comparison built on deadlines gets it exactly backwards. Washington's Table 5 is a live instance. It publishes *median review days compared to state deadlines* — a signed difference, not a duration. Read as a duration it would make most of the state's jurisdictions appear to take negative time. --- ## What this spec forced open Recorded per the work order's requirement to name any decision the data reopened. **Right-censoring.** Neither source rule anticipated it. A project that applies and has not yet been permitted is absent from a join, so a naive median over matched projects measures only the ones that finished — and the slow ones from recent years have not finished. The California measure is therefore computed on **application cohorts with at least three years of follow-up** (submitted on or before 2022-12-31), and the censoring is stated on the page rather than corrected for. This is an addition to the spec, not a deviation from it. **Identical application and issuance dates.** A material share of California jurisdictions enter the same date in both tables, producing a zero-day "duration". That is a defect in the record, not a fast permit. It is measured per jurisdiction, published as its own column, and used as a publication screen. See `README.md` for the threshold and why it sits where it does.