The evolution loop
Seven stages. No automatic progression.
Observe
- Question
- What happened or changed?
- Useful output
- Source-linked signal
- Decision boundary
- Collecting evidence requires permitted access.
This explorer only reveals explanatory text. It does not run an experiment, approve a change, or execute a process.
A useful outcome is not always adoption.
Adopt within bounds
Evidence supports the exact candidate and its scope; activation still needs separate authority.
Revise and test again
A finding changes the hypothesis or method, not the decision criteria after the fact.
Retain the baseline
A candidate fails or offers no justified improvement.
Defer
Evidence is missing, contradictory, or insufficient for the proposed claim.
Five different meanings of learning.
Human development
Judgment or skill changes; assess practice, transfer, and retention where claimed.
Context and memory
Stored information changes; inspect provenance, permissions, and retrieval, not model weights.
AI-system adaptation
A versioned prompt, retrieval, tool, model selection, or workflow changes; compare outputs and effects.
Model training
Parameters change; require separate data permission, evaluation, and release evidence.
Organizational learning
Shared decisions and practices change; check whether lessons actually affect later work.
The local pilots exercise decision records over declared synthetic measurements, not these learning mechanisms themselves. Read source ↗
Look beyond a single score.
Three depths of change
Improve the current practice; reconsider its assumptions or objective; improve the learning method itself. These are not levels of autonomy.
Four assessment views
AI task performance, human capability, collaboration, and broader outcomes should be considered together without flattening them into one universal score.