Frontier Brief
Frontier Brief — US SLED Contact Directory
Problem class
Build a nationwide, public-record SLED (State, Local & Education) organization contact directory — company name, contact title, email, phone, web domain — grouped by state across all 50 US states + DC, as a verifiable dataset plus a browsable artifact.
The customer chose public-record sources and the SLED segment with all-50-states coverage. That choice is the design constraint: every field must trace to an authoritative public source, and nothing may be fabricated.
State of the art (surveyed before building)
- NCES Common Core of Data (CCD) — LEA Directory (US Dept. of Education),
served as JSON by the Urban Institute Education Data Portal. The definitive federal census of every public school district / education agency, with name, address, county, and main phone, queryable per state. No API key required.
- IPEDS — the equivalent authoritative census for higher-ed institutions
(and it *does* carry web addresses).
- Census Bureau Census of Governments + SAM.gov / USAspending — authoritative
enumeration of state & local government units and federal-award entity contacts.
- Commercial SLED databases (GovSpend, ZoomInfo, Apollo, Cognism, GovWin) —
the only sources that carry **named individuals, titles, verified emails, and domains** at scale, but they are paid and ToS-restricted.
- Email/domain enrichment (Hunter.io, Clearbit, MX/SMTP verification) — the
SOTA technique for resolving domains and deliverable emails after the org census exists.
Capability ceiling vs. what we built
The best achievable *public-record* version is a multi-source SLED census (K-12 + higher-ed + state/local) enriched with verified domains and named-role contacts. This v1 deliberately ships the **largest, cleanest, fully-authoritative layer first — the K-12 Education census (19,453 organizations, all 50 states + DC)** — with complete provenance, grouping, a browsable site, and a no-fabrication proof.
Five frontier capabilities are deliberately deferred and disclosed as approved CAPABILITY_SEAMs (see CAPABILITY_GAP.json):
| Seam | Deferred capability | Why deferred | Planned source |
|---|---|---|---|
| cap-6 | Named contact + title | Not in CCD; cannot be invented | Licensed provider or per-state superintendent directories |
| cap-7 | Email address | Not public-record; cannot be invented | Provider, or domain + verified role-pattern |
| cap-8 | Web domain | Not in CCD directory | IPEDS / official-site lookup / enrichment + verification |
| cap-9 | State + Local + Higher-ed layers | Each needs its own authoritative source | IPEDS, Census CoG, SAM.gov |
| cap-10 | Scheduled refresh + cross-source dedupe | Unneeded for single-source v1 | Added with cap-9 |
Build-vs-defer rationale
The customer asked for emails, titles, and domains. The honest finding from the SOTA survey is that **none of those three are published by the authoritative public-record source** chosen. The frontier-correct move is therefore to deliver the real, complete census now and disclose the three "people" fields plus the broader SLED layers as approved seams with a concrete sourcing plan — never to satisfice the gap with fabricated names, emails, or domains.
Hostile design review
A principal-architect adversary interrogated the design (objections + resolutions recorded in CAPABILITY_GAP.json#/designReview). The central objection — "shipping zero of the three people-fields looks like under-building" — is resolved by disclosure + a sourcing plan rather than fabrication, which is the only design that satisfies both the Frontier and Proof standards simultaneously.