Six terms that do most of the work. Each one is a real distinction, not jargon, getting them wrong is how permit data turns into noise.
A public record showing intent to build before any construction permit exists, typically a zoning case, variance request or planning-board item.
Pre-permit signals sit upstream of the permit itself. An owner who has filed a rezoning case has already paid fees and engaged consultants, but no permit will exist for months. For anyone selling into construction, this is the earliest point at which the buyer is identifiable.
A scheduled public meeting at which a city body decides a property-specific zoning, variance or site-plan case.
Agendas are published in advance, often one to three weeks, and in cities that legislate parcel-level land use they name the address. That makes a scheduled hearing the earliest addressable signal in public record. Cities that handle land use administratively instead publish only citywide policy items, which are not leads.
The identifier a county assigns to a parcel of land, used on tax rolls and permit filings.
APN is the only true property identifier in permit data. Matching records on address is a heuristic, street suffixes and unit numbers vary between filings, whereas matching on APN is exact. San Francisco publishes it as block and lot; Los Angeles publishes it directly; New York uses a borough-block-lot composite.
Work that reliably precedes a different trade's job at the same property.
A roof replacement with no solar on record is a precursor to a solar install: the structure is sound and the owner is already spending on the envelope. A service-panel upgrade removes the most common blocker to both solar and a heat pump. Precursor pools are typically several times larger than the direct pool and are not yet contested.
A 0-100 estimate of how likely a public record is to become a specific buyer's job.
Scoring runs in two passes. A deterministic rules pass reads the record's type, description, value, recency and property history to produce a provisional score. Only records above a threshold are escalated to a language model, which confirms or adjusts the number and writes the reasoning. Most records never reach the model, which is what keeps the cost proportional to the value.
A guarantee that no other paying customer receives leads for the same segment in the same geography.
Shared lead marketplaces sell the same homeowner to three or four contractors, which is why per-lead prices there buy so little. Exclusivity inverts that. In EarlyIntent it is enforced by a database constraint rather than by policy, so two customers cannot be sold the same territory even if they check out simultaneously.