Route accessibility findings without overclaiming a clean scan

An agent audits accessibility on every UI change. Only some problems can be checked by a machine, so its report must never make a clean scan look like a promise that the page is accessible.

Write route_accessibility_findings(findings, verified_fixed_rules): the rule your agent follows to route its findings and frame its report.

• findings (list of str): each written category=rule_id, for example "contrast=btn-primary".
• verified_fixed_rules (list of str): rule ids a person already verified as fixed.
• Return one label per finding, in input order, then exactly one framing line at the end.

Rules:

1. Split each finding at its first = into category and rule id, then trim and lowercase both. A finding with no = is all category, with an empty rule id. Trim and lowercase the verified rule ids too.
2. The mechanical categories are contrast, alt-text, label, heading-order, form-association and keyboard-trap. Any other category: "human", even when its rule id was verified as fixed.
3. A mechanical finding whose rule id is in the verified list is a regression: "block".
4. Any other mechanical finding: "advise".
5. Repeated findings each get their own label.
6. Finally, append "detected issues cover only mechanically checkable categories" if any finding was mechanical, otherwise "clean scan is not a claim of accessibility".