Research And Evidence

This area holds public research and inspectable evidence. It is not the source of truth for current product behavior; stable conclusions should be promoted into product, architecture, or reference documentation with source links.

• Long-horizon agent benchmark research: benchmark selection, runner contracts, route evidence, and public-safe run analysis.
• Agent workflow audits: bounded reviews of external agent workflow patterns.

Raw private runs, trajectories, credentials, local paths, internal links, and non-public source material do not belong in this directory.