Competition Ops Agent

Competition Ops Agent is a local evidence desk for participants reviewing changing competition rules. It keeps proposed facts beside exact source passages, preserves conflicting deadlines instead of choosing one, marks missing evidence as unknown, and exports a review packet. A genuine Meta MuseSpark 1.3 Standard structured-output call supplies the recorded synthetic extraction example. The deterministic import and quote checks are separate from that model run. The demo exposes a failure: the model shortened "USD 1,000 total" to "USD 1,000". Its full quote and a qualifier warning remain visible. A self-reported local review does not authorize an entry, terms acceptance or spending. The public MIT source has 152 passing tests. This is a prototype with one measured synthetic model case, not a benchmark or autonomous competition-entry service. The evidence desk, import/review/export flow, quote and conflict checks, synthetic fixtures, test suite, bounded Meta transport and recorded-response UI shown here were newly built for this event during October 27-November 4, 2026.