{"resourceId":"pennsylvania-compass-document-quality-pilot","versions":[{"version":"external-47d3e3a511883e3a96d296a170171ead006b174a424d8402d58bc51724d65b95","resource":{"id":"pennsylvania-compass-document-quality-pilot","title":"Pennsylvania reports document-quality gains while keeping eligibility decisions with staff","organization":"Pennsylvania Office of Administration and Department of Human Services","sector":"State public-benefits administration","geography":"Pennsylvania, United States","publishedAt":"December 15, 2025","publicationDate":"2025-12-15","eventDate":"2025-12-15","sourceName":"Pennsylvania Office of Administration","sourceLabel":"Government pilot results and launch announcement","sourceUrl":"https://www.pa.gov/agencies/oa/newsroom/shapiro-administration-launches-new-tools-to-improve-customer-ex","evidenceClass":"government-evaluation","outcomeClass":"effective","topics":["knowledge-work","infrastructure","data-security","governance-procurement","accessibility-workforce","operating-model"],"finding":"Pennsylvania reports an 80% reduction in illegible or incorrect documents and over 700 staff hours saved in its COMPASS document-processing pilot.","sledRelevance":"A state benefits workflow with a bounded assistive role. Effective refers only to operator-reported document-quality results, not verified eligibility outcomes.","evidence":"The release describes approximately 12,000 documents screened and concerns identified in 25% of cases. Piloting began in October 2025; December 15 is the announced launch. The tool screens image quality and relevance, not eligibility.","architectureImplications":"Interpretation: insert quality feedback at upload while keeping case records authoritative; offer a human route when scanning fails. Specific model and hosting topology are unspecified.","governanceImplications":"Interpretation: preserve eligibility authority with authorized staff and monitor whether document flags delay legitimate applications.","securityPrivacyImplications":"Interpretation: isolate uploaded benefits documents, restrict retention and vendor access, and avoid exposing document content in telemetry.","caveats":"Operator-reported evaluation in a launch release. No control group, baseline counts, error intervals, time-accounting method or subgroup analysis. Document totals are not participant counts. No independent replication.","streamIds":["state-government"],"roles":{"sales":"Interpretation: Explore unreadable-document rework with benefits operations, caseworkers, digital service leaders and resident advocates. Ask how often uploads require resubmission, how much staff time this consumes, and which residents face the greatest barriers. A bounded engagement could measure the existing upload journey and trial quality feedback on representative documents. The value hypothesis is fewer avoidable resubmissions. Do not promise the reported percentage or staff-hour benefit elsewhere, and do not describe this as automated eligibility determination. Require evidence of improved applicant experience as well as reduced internal work before proposing expansion.","engineering":"Interpretation: Prototype an upload-quality service behind the existing application interface with human override and an accessible alternative. Require a labeled, approved document set spanning devices, formats and difficult images. Separate quality assessment from extraction and eligibility logic; a flag must not silently become a denial. Validate false rejection and missed-defect rates, latency and retention behavior against a manual review baseline. Trace retries and service failures. Hosting, model selection and vendor interfaces need local discovery. These checks address the missing accuracy and measurement detail without assuming that the reported aggregate improvement proves safe performance.","delivery":"Interpretation: Have benefits operations own the workflow, caseworker supervisors own exceptions and IT own integration and monitoring. Dependencies include document permissions, usability testing, staff training and a usable fallback. Pilot with explicit review checkpoints and capture applicant completion, repeat uploads, handling time and false flags. Proposed acceptance criteria: every flagged document has a documented recovery path, no quality flag alone determines eligibility, and measured errors remain within thresholds agreed before testing. Compare outcomes across accessibility and device scenarios before rollout. Risks include applicant abandonment, biased quality thresholds, sensitive-document exposure and overstated savings that omit support work."},"retrievedAt":"2026-09-07T03:01:14Z","enrichedAt":"2026-09-07T03:03:44Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: accommodate low-quality cameras, assistive technology, language needs and assisted submission. Track shifted work as well as staff savings.","procurementImplications":"Interpretation: request document-level validation, integration pricing, retention terms and reversible exit before copying this workflow.","operatingModelImplications":"Interpretation: benefits operations owns exception handling; IT owns uptime and interfaces; quality reviewers monitor false flags.","updateExplanation":"Unarchived primary source found through the assessment. Distinct from archived Pennsylvania ChatGPT workforce pilot: this covers COMPASS upload-quality processing. Historical evidence, not a new September launch.","sourceVerification":{"openedUrl":"https://www.pa.gov/agencies/oa/newsroom/shapiro-administration-launches-new-tools-to-improve-customer-ex","referenceExcerpt":"The tool will not determine eligibility but will support DHS staff and improve accuracy and efficiency in case processing.","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}