GHSA-8jjq-8j9w-m2v6

leftRight (calc.go:14236) tests the length with countUTF16String, which counts a rune above U+FFFF as 2. The RIGHT branch on 14241 then slices []rune(text) at utf8.RuneCountInString(text)-numChars, which counts it as 1. For text with N such runes the two measures are 2N and N, so any numChars between them passes the guard and gives a negative index. The numChars < 0 check at 14216 does not help, since the value that gets through is positive.

What makes it worth reporting is AutoFitColWidth, which evaluates formulas without looking like it does, so normalising an uploaded sheet is enough to reach it. One scoping correction to my own wording there: AutoFitColWidth was added in v2.11.0 and does not exist at v2.10.1, so on v2.10.1 the only reachable path is an explicit CalcCellValue.

A 6,077-byte file containing two U+1D7D9 characters in A1 and  RIGHT(A1,3)  in B1 was created and then opened in a separate program without recovery enabled:

v2.9.1    ok
v2.10.0   ok
v2.10.1   panic: slice bounds out of range [-1:]
v2.11.0   panic: slice bounds out of range [-1:]

So it is a regression, not an old defect. Commit a880146 (2026-01-16) moved the guard to countUTF16String and left the slice on the rune index, and git tag --contains gives v2.10.1 and v2.11.0 only.

RIGHT only. RIGHTB reaches the same function but takes the byte branch at 14225, which is internally consistent, and I probed MID and MIDB from 1 to 5 with no panic. It is the same negative-index family as GHSA-fx5j-qcqg-grpf and GHSA-48hm-4h8j-58fg, though those are shared-string lookups in the reader rather than a unit mismatch in the formula library.