{"resourceId":"california-poppy-shared-platform-control-claims","versions":[{"version":"external-00b9daec92508b706ff611e9b585b52e453f7e71623d602d8c254b0e5ad511eb","resource":{"id":"california-poppy-shared-platform-control-claims","title":"Poppy describes shared multi-model access without measured productivity evidence","organization":"California Department of Technology","sector":"State government","geography":"California, United States","publishedAt":"Undated page; inspected September 11, 2026 UTC","publicationDate":null,"eventDate":"2025-09-29","sourceName":"CDT: Poppy","sourceLabel":"Official platform guidance and operator claims","sourceUrl":"https://www.cdt.ca.gov/poppy/","evidenceClass":"standards-guidance","outcomeClass":"emerging","topics":["knowledge-work","developers-agents","infrastructure","data-security","accessibility-workforce","operating-model"],"finding":"CDT describes a multi-model employee assistant with reusable workflows and API integration. Its security and productivity statements are operator claims, not independent test results.","sledRelevance":"Direct state shared-service example, newly added to complement earlier California procurement coverage; current rollout completion is unverified.","evidence":"The page reports pilot input from 2,800+ employees across 67 departments. It dates the pilot start to September 29, 2025 and still describes July 2026 rollout as planned. Testimonials provide no measured baseline, quality-adjusted savings, sampling method or controlled evaluation.","architectureImplications":"Interpretation: model switching and API access warrant separate authorization and regression tests. Hosting topology, retention details and connector permissions are not specified.","governanceImplications":"Interpretation: approve each use case and material model change; trusted source branding does not certify answer accuracy.","securityPrivacyImplications":"The page claims no model training and sensitive-information safeguards. Interpretation: validate enforcement and failure handling before supplying protected inputs.","caveats":"Undated and partly forward-looking. No independent security validation or measured service benefit. Event date refers only to pilot launch.","streamIds":["state-government"],"roles":{"sales":"Interpretation: For a state CIO and agency operations leader, the problem is fragmented access to drafting and research tools. Ask which recurring tasks need shared support, which inputs are restricted, and who checks outputs. A bounded engagement could compare one document workflow across approved models with its current process. The value hypothesis is less total effort while preserving completeness. The page's participation count supports relevance, but does not justify savings, staffing reductions or a claim that every department can connect safely. Include budget and workforce stakeholders before sizing adoption. Confirm actual availability and support commitments before proposing expansion.","engineering":"Interpretation: Fit is human-reviewed assistance with controlled integrations. Start with a read-only connector, explicit identity scopes, approved documents and a fixed evaluation set. Treat each provider and model revision as a separate behavior profile; a common interface does not establish equivalent controls. Test source fidelity, omissions, prompt injection and attempted disclosure using authorized synthetic cases. Compare total effort and correction rate with unaided work. Prerequisites include agency data classification and a reviewer who knows the task. The public page cannot settle cloud, on-premises or hybrid suitability; obtain the actual data-flow and retention design before production integration.","delivery":"Interpretation: A platform service owner should coordinate access, support and change notices, while the agency process owner accepts the work product. Inventory the selected workflow, establish its baseline and train staff using approved examples. Dependencies include reviewer capacity, accessible onboarding and incident routing. Proposed acceptance criteria are that every sampled output retains its source trail, restricted test inputs follow the agreed handling policy, and total effort improves without lowering the agreed completeness score. These are local tests, not reported Poppy results. Review at pilot exit and after model changes. Risks include expanding faster than support capacity and counting drafting speed while overlooking verification."},"retrievedAt":"2026-09-11T03:00:45Z","enrichedAt":"2026-09-11T03:02:34Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: test keyboard, screen-reader and low-confidence-user workflows; employee testimonials do not establish accessibility.","procurementImplications":"Interpretation: check export, transition cost and provider terms before accepting a no-lock-in claim.","operatingModelImplications":"Interpretation: distinguish central platform operation from agency responsibility for correct outputs.","updateExplanation":"URL absent from all 182 archive resources and targeted Poppy search. Newly inspected historical platform detail complements the archived Claude agreement; no September change or completed rollout asserted.","sourceVerification":{"openedUrl":"https://www.cdt.ca.gov/poppy/","referenceExcerpt":"Side-by-side model comparison","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}