Generation alone is not enough
An LLM that writes a CV will happily produce something that looks polished and quietly fails ATS parsers, drifts off-topic, or invents skills. You only find out when no one replies.
Light judges before heavy ones
MYO runs a cheap text-only judge during drafting to catch structural problems early: missing headline format, broken bridge line, unsupported metrics. Only packages that pass that get the heavier visual judge.
The visual judge inspects the rendered PDF
The heavy judge reads the actual rendered PDF as images, the same view a recruiter sees. Page distribution, line breaks, orphan headings, font consistency, table-like layouts that confuse ATS. If anything fails, Olaf rewrites and the judge re-checks.
You see the verdict before you ship
In the Olaf workspace the rubric is visible: every check is pass, needs revision, or fail. You ship when you choose, with full context.