Limitations

Known gaps, each of them deliberate rather than pending:

• INDIRECT is unsupported. A reference computed at evaluation time cannot be seen by the static dependency extractor, so the recalculation order would be wrong -- see Topological recalc.

• xlsx export of formulas is EXCEL mode only. PYTHON and HYBRID syntax (**, list comprehensions, py.*) is not strict Excel, so those modes export values.

• 3D range refs (Sheet1:Sheet3!A1:B2) are unsupported and return nan. Workaround: expand them manually with +.

• Cross-sheet ranges (Sheet1!A1:Sheet2!B5) are rejected at parse time. Excel does not support them either.

• Grid size is 256 columns by 1024 rows per sheet. xlsx import drops cells outside it and warns.

• xlsx cell styles (fonts, fills, borders, column widths) are neither read nor written. Number formats are, but only the date ones -- enough to keep dates dates through a round trip; a currency or percent format from Excel still arrives as a bare number.

The Excel function coverage audit tracks the function library itself against Microsoft's documented set, including which absences are architectural and which are merely unimplemented.

For what the desktop frontend does not do yet -- a separate question from engine limitations -- see Desktop app.

No longer limitations

This page once listed LAMBDA and its higher-order helpers (MAP, REDUCE, SCAN, BYROW, BYCOL, MAKEARRAY), OFFSET, and the packing of dynamic-array results into their origin cell. All three have since shipped: lambdas are a first-class value type, OFFSET returns a real reference, and array results spill into neighbouring cells. tests/test_docs_conformance.py now fails if a function named on this page is one the evaluator actually resolves.

It also said xlsx dates were neither read nor written. They now are: a date's number format is read, rendered, written back, and understood by COUNTIF-style criteria -- see Dates.