Benchmarks

Two public corpora, 4,207 formula-bearing workbooks, identical code. How we measured this →

97.31%

Accuracy

zero-mismatch recompute rate

4M+

Formulas

compared against Excel's cache

26M/s

Eval speed

single-threaded throughput

736 KB

Engine

compiled to WebAssembly

97.31%zero-mismatch rate

3,254 of 3,344 FUSE workbooks recompute cell-for-cell against Excel's own cached values. 4,033,846 formulas compared. The remaining 90 files concentrate 97% of all mismatches in 10 files.

Findings

Private working models (Enron) consistently show more findings than publicly shared spreadsheets (FUSE). The rate rises monotonically with model size on both corpora.

Enron (private) FUSE (public)

Any high-precision finding, by model size

All formula-bearing

33.9%

20.5%

≥ 100 formulas

52.1%

32.5%

≥ 1,000 formulas

60.3%

41.5%

By detector — substantial models (≥ 1,000 formulas)

Hardcoded plug

43.2%

29.9%

Broken #REF!

22.6%

10.8%

External links

41.1%

9.8%

Performance

2 ms

Median intake

per workbook

19 ms

p90 intake

per workbook

276

Functions

compiled natively

4,207

Workbooks

across two corpora

Go deeper

Reference

276 supported functions

Every function the engine compiles — aggregates, conditionals, lookup, financial, dynamic arrays, and higher-order functions.

Full function list →

Methodology

How we measured it

Corpora, sample seed, detector definitions, limitations, and the commands to reproduce every figure on this page.

Read the methodology →