Government Trust
This page is maintained by PermitJunkie to explain how our Government & Jurisdiction Intelligence Network (JIN) scores each jurisdiction, where the data comes from, and how you can trust — and audit — every recommendation our platform makes.
Each jurisdiction receives a readiness score derived from four canonical signals:
- Coverage — permit catalog, departments, requirement graph completeness.
- Freshness — days since last verified source sync.
- Source reliability — historical success rate of the underlying data feeds.
- Verification — human or automated confirmation of the underlying facts.
Every jurisdiction we display is built from a combination of:
- Government open-data portals and permit APIs.
- Municipal code and ordinance PDFs indexed with citations.
- Direct department contact metadata (name, phone, email).
- Change events detected by our continuous knowledge crawler.
Sources and reliability are tracked per jurisdiction inside PermitJunkie and cited in every recommendation.
We continuously re-check data sources — most jurisdictions refresh within 30 days. When a source fails, the impacted jurisdiction is flagged as stale or high-risk until re-verified.
New jurisdictions are seeded with low confidence and unverified state; JIN self-heals as sources come online.
- Every AI recommendation shows its evidence, sources, and confidence.
- Evidence snapshots are append-only and cannot be silently edited or deleted.
- Reviewer decisions are replayable — the same inputs always produce the same outputs.
- Enterprise policy decisions are cited on every action they govern.
PermitJunkie provides the government intelligence platform, canonical data schema, verification pipeline, and explainability infrastructure. Application owners, workspace admins, and end-users remain responsible for the accuracy of project-specific information they enter and the final decision to submit any permit.
This page describes controls currently enabled in the PermitJunkie platform. It is not an independent certification and does not by itself constitute regulatory or legal compliance advice.
Every autonomous recommendation surfaced in the product is auditable. When a user accepts, rejects, or overrides a recommendation, or when a predicted outcome is later observed, PermitJunkie writes an append-only learning event tied to the workspace and project. These events are stored in a database table that has no update or delete grants — mutation attempts are blocked by database triggers.
Learning events feed the Adaptive Permit Intelligence Mesh (APIM). APIM re-calibrates confidence per workspace by category — so the more your team engages, the better the system matches your permitting reality. There is no cross-workspace leakage: calibration and mesh queries are strictly workspace-scoped and enforced by Row Level Security.