fix(evolution-curve): settle the capability axis as human-assessed - #6

What this PR does

Settles the Evolution Curve with option B from #5: capability is not inferable from an agent spec, so the axis belongs to agent-ux-teardowns where a human assesses it, and is out of scope for aux-audit permanently rather than pending. The reason the two axes kept collapsing is that a spec describes what a product declares about itself, while capability is a claim about what it does — and every field a capability score would naturally read (autonomy, memory.persistent, memory.scopes, tools) is already read by a trust heuristic. No schema turns a declaration into an observation. Alongside that, this fixes four references to the axis that were inconsistent regardless of which option won.

Linked Issue

Closes the aux-audit half of #5. #5 stays open for the stage vocabulary itself — see Out of scope.

Type

• New pattern (must include diagram, runnable example, and anti-pattern)
• New gap (must include reproducible scenario and heuristic / trust-stage ref)
• Teardown (must use agent-ux-teardowns/_template/ and pass schema validation)
• Framework edit (must include rationale, ID migration note if renamed, updated examples)
• Schema fix
• Doc fix / typo / link
• Other (explain)

Framework edit rather than schema fix: no schema is added or changed. What changes is where the capability axis lives and who assesses it. No IDs are introduced or renamed — the aux.E* straw man in #5 is deliberately not implemented here.

The one place B touches the CLI

B was originally described as closing this "without touching the CLI". That was wrong in one respect, and the correction is the substance of the code change:

- evolution_stage_status: "schema-undefined"
+ evolution_stage_status: "human-assessed"

schema-undefined means pending a schema. Under B nothing is pending, so leaving that string would keep the tool advertising work that will never arrive — the same class of defect #5 exists to remove. The computation stays absent, which was the substance of the original claim.

Inconsistencies fixed, independent of A/B/C

agent-ux-teardowns now carries the axis explicitly as step 6 of a teardown, since under B that is its home.

Contribution bar — checklist

• I've read CONTRIBUTING.md.
• If this introduces or renames an ID, I opened a debate Issue first. — no IDs introduced or renamed; the aux.E* proposal stays in [schema] aux-evolution-curve: define the capability axis, or retire it #5, unimplemented.
• If this adds a pattern, it has a diagram, a runnable example, and an anti-pattern. — n/a, no patterns added.
• If this adds a gap, it has a reproducible scenario and references a heuristic or trust stage. — n/a, no gaps added.
• CI passes (schema validation, link check). — verified locally: 34/34 aux-audit tests, check-coverage.py 12/12, yamllint clean on schemas/ and .github/, all relative links resolve, and the teardown example still parses as strict JSON. This branch has had no CI run yet, since validate.yml fires on pull_request and on push to main — opening this PR is the first run.
• No emoji in code or schemas.
• Short sentences. Named things over adjectives.

Out of scope

• The stage names. task-aware and personally-intelligent remain undefined. B settles where the axis lives and who assesses it, not what the stages are called.
• Whether the curve is a ladder at all. The trust ladder is sequential by design because its failure modes cascade; there is no equivalent argument for capability, so the curve may be a per-surface profile rather than an ordinal stage. Undecided, tracked in [schema] aux-evolution-curve: define the capability axis, or retire it #5.
• No schema file is added. Writing schemas/aux-evolution-curve.yaml would mint new vocabulary in the aux. namespace, which CONTRIBUTING requires an Issue to settle first. That is exactly what [schema] aux-evolution-curve: define the capability axis, or retire it #5 is now for.

🤖 Generated with Claude Code

https://claude.ai/code/session_01MhpqWEvwKp7NrmJfdXYq4M

Generated by Claude Code