{"resourceId":"bellevue-govstream-permitting-backtest-2026","versions":[{"version":"external-03238b0cd49931a25b24891035d19d38f5b420f3088656833e9028b11538cd0f","resource":{"id":"bellevue-govstream-permitting-backtest-2026","title":"Bellevue permitting case study reports historical back-test gains and estimated staff savings","organization":"City of Bellevue and Govstream.ai","sector":"Municipal permitting and staff knowledge work","geography":"Bellevue, Washington, United States","publishedAt":"August 4, 2026, as dated by the city’s case-study link","publicationDate":"2026-08-04","eventDate":null,"sourceName":"City of Bellevue","sourceLabel":"City-hosted case study with vendor-attributed evaluation pages","sourceUrl":"https://bellevuewa.gov/sites/default/files/media/pdf_document/2026/dsd-08042026-govstreamai-case-study-early-results.pdf","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["knowledge-work","developers-agents","data-security","governance-procurement","operating-model"],"finding":"A city-hosted vendor case study reports improved intake completeness in historical replay, alongside estimated time savings from staff assistants.","sledRelevance":"Interpretation: Municipal intake and inquiry assistance are plausible bounded use cases, but another jurisdiction must validate its own rules and applicant mix.","evidence":"Back-testing 276 building permits reported completeness rising from a 22% historical baseline to 67%, with 96% of required documents identified. Separately, staff estimated 152 hours saved during one month between June and July 2026.","architectureImplications":"Interpretation: Map code, GIS and permit-record permissions into retrieval; keep draft responses and form suggestions distinct from authoritative changes. Assess cloud, hybrid or on-premises options through actual supplier documentation; the case study does not establish deployment topology.","governanceImplications":"Interpretation: Maintain separate scorecards for offline testing, live service quality and estimated versus observed labor effects.","securityPrivacyImplications":"Interpretation: Verify supplier data-use restrictions, access controls, log retention and deletion; source citations alone do not prevent leakage or malicious instructions.","caveats":"Vendor/operator evidence; no independent validation, held-out test description or net cost analysis. Back-test completeness is not observed live applicant behavior. The efficiency target is not an achieved result.","streamIds":["local-government"],"roles":{"sales":"Interpretation: Engage development services, GIS, customer support, IT security and finance around inquiry workload and incomplete submissions. Ask whether the bottleneck is missing documents, ambiguous rules or staff capacity. A bounded engagement could reproduce a local historical test and design a prospective pilot before purchasing broad automation. The value hypothesis is less preventable rework with acceptable review cost. Keep customer commitments conditional on local evidence, including net time after edits, licensing and maintenance. Do not present the replay results as a live improvement, guarantee savings or assume a different municipality can reuse Bellevue’s configured rules.","engineering":"Interpretation: Build a versioned mapping between permit types, property attributes and required documents. Approve each retrieval source and test whether staff-only records can leak into applicant responses. Preserve source links and validate the answer rather than trusting the citation. Obtain a held-out dataset before further tuning, then run a prospective comparison that measures complete applications, missing requirements and false flags. Record latency and human edit time. Do not enable autonomous permit approval. Request architecture, retention and subprocessor evidence from the supplier because the case study cannot establish suitability for a particular hosting or security boundary.","delivery":"Interpretation: Plan data cleanup, rule configuration, staff training and accessible applicant guidance as separate work packages. The development-services owner accepts workflow outcomes; IT and GIS maintain sources; procurement and privacy approve handling and support obligations. Gate external release on a reviewed prospective protocol and demonstrated escalation. Proposed acceptance: eligible cases have a recorded baseline and outcome, staff validation time is counted, severe data-access defects are resolved, and support can restore the prior workflow. Monitor changes by permit category to detect uneven performance. Risks include tuning to historical cases, stale property data and underestimated maintenance effort."},"retrievedAt":"2026-09-07T03:06:12Z","enrichedAt":"2026-09-07T03:06:12Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: Include novice applicants, language needs and assistive technology in prospective validation; count the staff work needed to verify drafts.","procurementImplications":"Interpretation: Request the test-set selection and tuning history, prospective validation rights, export terms and a full operating-cost proposal.","operatingModelImplications":"Interpretation: Assign one business owner to maintain rules and one platform owner to handle integrations and change control.","sourceVerification":{"openedUrl":"https://bellevuewa.gov/sites/default/files/media/pdf_document/2026/dsd-08042026-govstreamai-case-study-early-results.pdf","referenceExcerpt":"Early results based on a back test","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}