v4.22.1 — doubled citation markers, found by writing to the new guidance

Fixed — doubled citation markers, found by writing to the new guidance

The 4.21/4.22 quality guidance had only been verified as far as "the text
appears in the brief". Closing that loop — authoring the beam lab report
following the reader rubric and the structure discipline, then rendering
it — surfaced a defect the old list-style prose never triggered.

A sentence citing two evidence rows from the same source carries two
separate [CITE:...] markers. Each resolves independently, so the
document rendered "[1] [1]". The existing deduplication only covered ids
inside a single marker, which is why the duplicate survived. Adjacent
identical citations — numeric, author-year, or [Source: ...] — now
collapse to one marker.

The loop itself closed cleanly otherwise: 24 synthesis sentences (topic
and concluding sentences carrying no citation of their own) passed the
gates untouched, so the evidence contract does not stand in the way of
paragraphs built Context → Content → Conclusion. The discussion now runs
result → quantitative comparison → mechanism → verdict, citing the
derived slope, R², and error statistics from 4.21.0.

456 tests and both benchmark --checks pass.