Skip to content
AletheIA Docs
Esc
↑↓navigate↵open⌘Jpreview
On this page

Governed Skill Optimization

Optimization is evidence, not authority. Governance outranks optimization.

This document explains why AletheIA treats skill optimization as a governed activity and where the boundary sits between producing evidence about a proposed skill change and deciding to make that change.

It is conceptual background. The normative rules live in skill-evolution-validation-contract.md.

The problem

As a skill library grows, proposals to evolve a skill can be accepted because they seem reasonable rather than because they were tested. Three failure modes follow:

  1. A skill improves in one scenario and silently regresses in another.
  2. A change looks cosmetic but shifts decision behavior.
  3. A knowledge-aware skill blurs skill evolution with misuse of a governed source.

Optimization techniques that treat a natural-language skill document as an improvable artifact make these failures worse if their output is trusted directly. The risk is not the experimentation — it is letting an experiment’s result act as authority.

The stance

AletheIA governs the decision; it does not run the optimization. An experiment may generate evidence, a comparison, or a recommendation. It may never produce a canonical change. The strongest thing an experiment can do is create a proposal, which then re-enters the normal governed path: human review and a pull request.

This mirrors how AletheIA already treats its own evolution (see self-application.md) and how it separates what a skill needs from which source satisfies it (see knowledge-governance-layer.md).

Where it sits

AletheIA                      contracts, gates, policies, evaluation depth, human decision
Knowledge Governance Layer    sources, authority, sensitivity, scope, restrictions, packs
Adaptive Skills               skills, modules, triggers, verification, handoff signals
Evolution Layer               observations, proposals, reviews, outcomes
Skill Evolution Validation    validation cases, experiments, regression, comparative evidence

The execution side of this layer — validation cases, experiments, the regression check, and the structural validator — lives in the Adaptive Skills repository, beside the existing Evolution Layer. AletheIA holds only the contract: the conditions an experiment and its evidence must satisfy to be considered governed. AletheIA governs the decision; Adaptive Skills runs the experiment.

The authorized path

Observation
  → Validation Case
  → Skill Evolution Experiment
  → Regression Check
  → Proposal
  → Human Review
  → Pull Request
  → Canonical Skill Update

Everything up to and including the experiment is evidence. Everything after the proposal is human authority. The two never collapse into a single automated step.

What this is not

  • Not an optimizer runtime, a benchmark engine, or a training loop.
  • Not an auto-writeback into any skill document.
  • Not a scoring system whose number becomes a decision.
  • Not a license to copy confidential, restricted, or regulated content into experiment artifacts — synthetic-first applies, and the restricted-knowledge-usage-policy.md still binds.

Good outcomes that are not “a change”

Confirming that a skill already behaves correctly under a new case (reinforced) or that a candidate is not worth adopting (no-change) are successful results. They prevent churn and are evidence in their own right. Few strong cases beat many weak ones.

Was this page helpful?