Adaptive Perceptual Reading
Read what changed. Verify what matters.
APR is a research architecture for deciding when an agent should observe, what it should read, how deeply it should read, and when fresh evidence is sufficient to stop.
Research statusv0.10 release candidate — public research infrastructure
Perceive → Verify → Recover
APR maintains a closed loop: perceive evidence, update bounded belief, identify unmet needs, gate action, observe the outcome, verify it, and recover when reality disagrees.
- Observe Acquire only the evidence a live need justifies.
- Budget Spend finite attention where information gain matters.
- Verify Require post-action evidence before declaring success.
- Recover Retry, replan, compensate, or stop under explicit policy.
Current boundary
The repository is a research MVP. This static site explains evidence already present in the repository; it does not host an agent, call a model Provider, or expose desktop control.
Start with evidence
Explore the runtime guide, inspect the bounded offline lab, or read the papers. Every technical page links to an immutable candidate source reference.
Source evidence
Immutable links pinned to the candidate source reference.