Bounded Learning
Protean’s learning systems are designed to adapt without losing control.
The platform can adjust prioritization based on repeated evidence, explanation patterns, failure signals, and experimental feedback. Those adjustments stay inside bounded scoring contracts.
Learning Modes
Protean separates learning behavior by evidence maturity.
Explanation-guided mode uses repeated rationale patterns, warnings, strengths, risks, and failure signals when assay data is absent. This mode can improve prioritization, but it does not create biological truth claims.
Assay-guided mode can use measured outcomes when structured assay data exists. Measured evidence takes priority over explanation-derived signals.
Both modes remain bounded.
Control Principles
- Learning updates remain constrained.
- Scoring weights stay normalized.
- Base assumptions remain recoverable.
- Failure penalties continue to apply.
- Adaptation does not rewrite the engine.
- Candidate reranking remains controlled.
Boundaries
Bounded learning is allowed to adjust prioritization within caps. It is not allowed to:
- Rewrite the base scoring contract.
- Remove failure motif penalties.
- Self-modify code.
- Recursively retrain without limits.
- Convert model explanations into validation claims.
- Run open-ended feedback amplification.
Reranking Discipline
The platform can use learning output to rerank candidates in a controlled way. Reranking is treated as a cycle-level action, not an unlimited recursive process.
That distinction keeps the system adaptive while preserving reproducibility.
Scientific Value
Bounded learning lets the system become more responsive without becoming opaque. The objective is not unbounded self-improvement. The objective is disciplined prioritization that preserves reproducibility.