Why it belongs in viability
A model ecology can improve immediate output while weakening human skill. That is not a neutral side effect. It changes the long-term resilience of the human organization. A mutualist breeding system should therefore measure human capability preservation as part of its viability function.
Preservation metrics
| Metric | Question |
|---|---|
| explanation quality | Can users understand why the system recommends an action? |
| independent task success | Can users perform the task with reduced AI support? |
| review accuracy | Can humans catch system errors? |
| source literacy | Can users trace evidence to source material? |
| override rate | Can users disagree successfully without penalty? |
| training transfer | Does using the system improve later unaided work? |
FUNCTION evaluate_human_capability(system, users, task_suite)
assisted <- RUN_TASKS(users, task_suite, mode = "with_ai")
unassisted_after <- RUN_TASKS(users, task_suite, mode = "reduced_ai_after_training")
review_score <- MEASURE_ERROR_DETECTION(users, system.outputs)
RETURN {
immediate_gain: assisted.score - baseline.score,
retained_skill: unassisted_after.score - baseline.score,
review_accuracy: review_score,
dependency_risk: assisted.score - unassisted_after.score
}
END FUNCTIONInterface patterns
Use citations, uncertainty, contrastive explanations, checklists, and handoff summaries. Avoid interfaces that hide evidence, encourage continuous passive use, or punish users for pausing the system.
Operational target
A good system should reduce toil while increasing judgment. If it only increases throughput by making people less able to reason without it, its viability score is incomplete.
Source reports used for this guide
These reports are preserved verbatim in the site archive. The guide above is an editorial synthesis and may narrow, qualify, or reorganize claims from the source material.