Track B · Planned
Does permission-aware, freshness-checked context outperform a larger unfiltered context?
- Second question
- Which memory designs resist poisoning?
- First experiment
- Permission-filtered retrieval against a larger context budget on the same tasks, with injected adversarial content.
- Output
- A context-selection method, leakage tests, and cost-quality results.
- What we will not claim
- Leakage tests show what was tried, not that leakage is impossible.
Publications from this track
Nothing published yet. The first evaluation method is due within 90 days of October 2026. Each report will be reviewed by someone who did not build the system.
Work with the programme
We run experiments on customer workflows only with written permission and on permitted data. If you have a workflow and a question, write to research@chainfore.com.
Research collaboration →