Confidence matrix
Kong works with most Ghidra-decompilable binaries. Confidence varies by architecture and source language:
| C | C++ | Go | Rust |
|---|
| x86 | High | High | Medium | Medium |
| x86-64 | High | High | Medium | Medium |
| ARM (32-bit) | High | High | Medium | Low |
| AArch64 | High | High | Medium | Low |
| MIPS | Medium | Medium | Low | Low |
| PowerPC | Medium | Medium | Low | Low |
Confidence definitions
High — Kong reliably decompiles, deobfuscates, and recovers names, types, and structure. Expect 80%+ of functions named with high confidence.
Medium — Decompilation is usable but noisier. Expect partial recovery and lower confidence scores. Go and Rust binaries have more complex calling conventions and runtime patterns that reduce accuracy.
Low — Decompilation has significant gaps. Results will be incomplete, noisy, or partially unreadable. Kong will still produce output, but expect many low-confidence results.
Scaling considerations
Binary size correlates with:
- Function count — more functions = longer analysis
- LLM cost — more functions = more API calls
- Time to completion — roughly linear with function count
However, larger binaries tend to produce lower average confidence — there are more utility functions, compiler-generated code, and edge cases that are harder to name accurately.
See LLM Models & Pricing for cost estimates by binary size.Last modified on March 20, 2026
