An eleven-week records backlog, cut to two

The city's records division carried an eleven-week backlog of public-records requests, and statutory response deadlines were slipping. Each request meant searching email, agendas, and permit systems by hand, printing candidates, and reviewing every page for exempt material before release.

Sunshine-law obligations shaped every option: an automated step that couldn't be explained, logged, and reproduced for a requester — or a judge — was worse than no automation at all.

We scoped inside the division for two weeks, timing real requests. Most hours went to search and first-pass review, not judgment calls: clerks were reading hundreds of pages to find the dozen that mattered, then checking each for exemptions.

The signed scope kept release authority entirely human. Acceptance criteria covered search recall against a test set of past requests, redaction-suggestion precision, and a public-defensibility bar: every automated action logged in plain language a requester could read.

The system searches the city's email archive, document management system, and permit records from a single request description, assembles the candidate set, and suggests redactions with the statutory exemption cited for each one.

Clerks review every candidate document and every suggested redaction before release. The system records what was searched, what was returned, what was withheld, and under which exemption — an audit trail generated as a matter of course, not reconstructed under challenge.

The backlog fell 82% — eleven weeks to two — within the first quarter of operation, and statutory deadlines are being met without added headcount. Clerks spend their time on exemption judgment instead of page-turning.

Requesters see the difference too: acknowledgments with realistic dates, and responses that arrive when promised. The system runs under managed operations, with the exemption library maintained as statutes change.