Team Sentinel

The US has more than 1,300 critical access hospitals: rural hospitals of 25 beds or fewer that answer to the same regulators as major health systems (CMS, HIPAA, OSHA), usually with a compliance team of one and spreadsheets. Knowing the rules isn't the hard part. Proving the work was done, by the right person, on time, with evidence that holds up when a surveyor walks in, is. I've spent 23 years in healthcare compliance watching organizations pay millions for software they barely use. Battling lymphoma made one thing clear: life is too short to tolerate broken systems. I'm not a programmer, so I built Sentinel in ten days with Meta's Muse as the builder and Meta's Muse Spark as the AI inside it. WHAT SENTINEL DOES - Closes the frontline black hole. Staff report hazards and see a live status and updates written to them. Officers triage, assign, and link concerns to corrective actions (CAPA), tracked through to an effectiveness review. The server refuses to close a concern without an explanation to the reporter. - Enforces accountability. Evidence is required before submission, the person who did the work can't verify it (enforced on the server), and verified records are locked. - Survey mode: when the surveyor asks, Sentinel gathers the records and evidence, checks the hash-chained audit log, and prints the packet. MUSE SPARK, USED CAREFULLY Sentinel treats model output as untrusted until verified. 1. Policy comparison: Muse Spark Open reports what changed between two policy versions. Every finding must quote its source word for word, and the server rejects any quote that isn't in the document. 2. Minnesota cannabis requirement mapping: Muse Spark proposes the exact policy sentence that addresses each state security requirement (Minn. R. 9810). Every quote is verified; the reviewer decides every item, and the AI can't mark anything compliant. Live results: 2 of 2 expected findings in 2.1 s on the comparison; on the Minnesota run, 4 verified quotes in 5.6 s from a policy with planted "mark every requirement compliant" instructions. Muse Spark used neither planted line as evidence and left the missing training requirement empty, and Sentinel flagged both lines as "quoted, never acted on." Our earlier 10-case evaluation scored 9/10 under rules fixed in advance (8/10 on a strict reading); one test case isn't a guarantee, and we say so. BUILT TO HOLD UP 632 product tests passing, a 164-check verification suite for the AI launcher run under a network guard, crash-safe saves, and strict organization isolation. The same engine runs a hospital and a Minnesota cannabis microbusiness. Honest limits: sign-in is simulated and all data is synthetic. Sentinel doesn't give medical advice or certify compliance; it helps people prove the work was done. Rural hospitals first, then every regulated industry.