Comparison guide
SuperCompress vs truncation
Truncation keeps the beginning and end, drops the middle. If answer evidence sits in the middle, truncation removes it.
Why truncation is risky
Truncation achieves only ~25% oracle recall at a 65% compression budget. That means 75% of the time, the line containing the answer gets dropped.
Benchmark comparison
| Metric | Truncation | SuperCompress |
|---|---|---|
| Oracle recall | 25% | 100% |
| Query awareness | None | Full |
Frequently asked questions
When is truncation acceptable?
When relevant context is always at the start or end.
Can I combine both?
Yes. Use compression first, then hard truncation as a safety guardrail.
Build with less context
Put compression in front of your next LLM call.
Use the hosted API or run SuperCompress locally. Keep the evidence, drop the token waste, and measure the savings before it reaches your model.