No human reviews ZoningPro reports — and no value appears in one unless it survives all five automated gates:
1 · Quote verification (deterministic). The verbatim citation must string-match the ordinance text we ingested. A paraphrase fails.
2 · Entailment. An independent model confirms the quoted passage actually supports the extracted value — a citation that merely mentions the topic fails.
3 · Consensus. At least two independent model runs must agree on the value. Disagreement means we don't publish.
4 · Cross-vendor corroboration. Where licensed national data covers the field, agreement upgrades confidence to “corroborated”; conflict blocks publication and opens a pipeline investigation.
5 · Benchmark gate. A field class (say, “max height, table-formatted codes, Texas”) auto-publishes only while it scores ≥95% on our eval benchmark. Below that, you get the quoted code text instead of a number.
When we don't know, we say so
A field that fails any gate renders as Unresolved — read the source with the governing
section quoted in full and permalinked. Planned-unit developments and negotiated zoning are flagged and linked to their
governing documents — never silently reduced to base-district numbers.
Corrections demote, never edit
A documented error is demoted to “unresolved” across all reports and API responses; the number returns only when the
pipeline clears the case on the benchmark. Every demotion is on the public correction log.
Generated by software, verified by software, cited to the ordinance — never guessed. Informational research;
not an official zoning determination (see legal).