{"resourceId":"seattle-civcheck-prescreening-coverage-2026","versions":[{"version":"external-03238b0cd49931a25b24891035d19d38f5b420f3088656833e9028b11538cd0f","resource":{"id":"seattle-civcheck-prescreening-coverage-2026","title":"Seattle permitting evaluation separates check accuracy from eliminating review cycles","organization":"City of Seattle Innovation & Performance and SDCI","sector":"Municipal construction permitting","geography":"Seattle, Washington, United States","publishedAt":"June 17, 2026","publicationDate":"2026-06-17","eventDate":null,"sourceName":"City Innovation Hub","sourceLabel":"Municipal proof-of-value evaluation summary","sourceUrl":"https://innovation-hub.seattle.gov/2026/06/17/ai-construction-permitting-seattle-civcheck-study/","evidenceClass":"government-evaluation","outcomeClass":"mixed","topics":["knowledge-work","developers-agents","governance-procurement","operating-model"],"finding":"Seattle found promising prescreening accuracy but warned that incomplete correction coverage may leave review-cycle counts unchanged.","sledRelevance":"Interpretation: Cities and counties can test intake completeness before committing to complex code interpretation. Smaller departments should budget expert review time and start with one permit class.","evidence":"The city compared CivCheck and staff reviews of real applications and interviewed applicants and staff. It recommends a production pilot focused on completeness; permitting impact was modeled, not demonstrated in production.","architectureImplications":"Interpretation: Keep prescreening separate from authoritative permit decisions. Version local requirements and connect reviewer corrections to evaluation records. Hosting and model details are not established by this source.","governanceImplications":"Interpretation: Define who adjudicates disputed checks and require coverage evidence before expanding scope.","securityPrivacyImplications":"Interpretation: Treat uploaded plans as untrusted inputs; test prompt injection, cross-applicant access and sensitive file retention.","caveats":"Full technical PDF returned 403. The inspected city summary does not provide the application sample size or scoring detail. Small residential projects and Seattle codes limit transfer.","streamIds":["local-government"],"roles":{"sales":"Interpretation: The customer problem is repeated corrections that delay intake. Bring together the permitting director, applicants, frontline reviewers, IT and procurement. Ask which missing items actually cause another cycle and how much reviewer time is available for a pilot. A credible value hypothesis is fewer avoidable resubmissions in one permit category. Offer a bounded baseline and prescreening assessment with jointly agreed quality measures. Do not promise faster issuance or extrapolate check accuracy into housing production. A small department may need shared evaluation support; establish that capacity before qualifying a larger rollout.","engineering":"Interpretation: Fit the tool around existing submission and review systems, preserving staff authority. Prerequisites include current rules, representative historical files, correction labels and an agreed scoring rubric. Use a holdout set and track false alarms, missed corrections and complete-case coverage separately. Record model and configuration versions. Restrict file access, isolate applicants and test malicious document content before any connector receives write privileges. Proposed proof of value: compare assisted and ordinary intake on prospective cases, including net reviewer effort and review cycles. The inaccessible technical report prevents relying on this source for a complete benchmark design.","delivery":"Interpretation: Have the permitting service owner lead rule clarification and evaluator training, with IT responsible for identity and monitoring. Dependencies include reviewer capacity, approved data handling and applicant-support channels. Run a limited intake pilot and review disagreements before increasing volume. Train staff and applicants using realistic incorrect flags. Proposed acceptance: every consequential flag has a traceable requirement and escalation route, no unresolved severe access defects remain, and prospective results report both cycle counts and review effort against baseline. Stop expansion if rework rises. Keep a manual route available during outages and code changes."},"retrievedAt":"2026-09-07T03:06:12Z","enrichedAt":"2026-09-07T03:06:12Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: Test screen-reader access and understandable correction instructions; retain assisted intake and train staff to challenge incorrect flags.","procurementImplications":"Interpretation: Buy a bounded evaluation with exportable test results and clear stop criteria; avoid payment milestones based solely on model accuracy.","operatingModelImplications":"Interpretation: The permitting owner should manage the rule backlog and review-cycle measures with IT operating the integration.","sourceVerification":{"openedUrl":"https://innovation-hub.seattle.gov/2026/06/17/ai-construction-permitting-seattle-civcheck-study/","referenceExcerpt":"we recommend a production pilot for pre-screening application completeness.","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}