diff --git a/CHANGELOG.md b/CHANGELOG.md
index 7138dac..0e00657 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -33,7 +33,11 @@ Publishing the GitHub release and tagging `v0.2` is a manual step after merge.
 
 ### Notes
 
-- `evolution_stage` is reported as `human-assessed` and is always `null`. The
-  capability axis is scored by a human in a teardown, not derived from a spec
+- `evolution_stage` is reported as `human-assessed` and is always `null` in
+  `aux-audit`. The capability axis is defined in
+  `schemas/aux-evolution-curve.yaml` (E01 Reactive → E04 Personally
+  Intelligent) and scored by a human in a teardown, not derived from a spec
   ([#5](https://github.com/auxfirst/trustkit/issues/5)).
+- Licensing follows one rule: machine-readable definitions are MIT, prose is
+  CC BY 4.0. That now covers pattern `example.py` files as well as schemas.
 - A spec with no evaluation transcripts is capped at heuristic level 2.
diff --git a/LICENSING.md b/LICENSING.md
index 6062a6d..d595359 100644
--- a/LICENSING.md
+++ b/LICENSING.md
@@ -16,8 +16,9 @@ the prose, which is most of the repository by volume. It does not describe
 | Artifact | License | File |
 |---|---|---|
 | **Schemas** — every `.yaml` under [`schemas/`](schemas/) | **MIT** | [`schemas/LICENSE`](schemas/LICENSE) |
+| **Pattern examples** — `example.py` under [`repos/agentic-ux-patterns/`](repos/agentic-ux-patterns/) | **MIT** | `SPDX-License-Identifier` in each file |
 | **Executable tooling** — `packages/aux-audit`, `action.yml` | **MIT** | [`packages/aux-audit/LICENSE`](packages/aux-audit/LICENSE) |
-| **Prose** — onboarding, vocabulary, pattern write-ups, architecture, this file | **CC BY 4.0** | [`LICENSE`](LICENSE) |
+| **Prose** — onboarding, vocabulary, pattern write-ups and diagrams, architecture, this file | **CC BY 4.0** | [`LICENSE`](LICENSE) |
 | **Teardown content** under `repos/agent-ux-teardowns/teardowns/` (when added) | **CC BY-NC 4.0** | declared in that tree |
 
 ## Why the schemas are MIT
@@ -35,6 +36,7 @@ one click. It is just not a condition of use.
 ## What this means in practice
 
 - **Vendoring `aux-heuristics.yaml` into a closed-source product** — fine, MIT.
+- **Copying a pattern's `example.py` into your agent** — fine, MIT. That is what it is for.
 - **Shipping a tool that bundles the schemas** — fine, MIT, which is why the
   copies under `packages/aux-audit/schemas/` need no separate carve-out.
 - **Quoting the onboarding docs in a blog post** — CC BY 4.0, so credit
diff --git a/README.md b/README.md
index c27b00c..025084b 100644
--- a/README.md
+++ b/README.md
@@ -77,6 +77,7 @@ Or read **[onboarding/README.md](onboarding/README.md)** — the router with all
 │   ├── aux-heuristics.yaml        ← the 10 heuristics
 │   ├── trust-architecture.yaml    ← the 4 trust stages
 │   ├── trust-gap-taxonomy.yaml    ← named failure modes
+│   ├── aux-evolution-curve.yaml   ← the 4 capability stages (teardowns only)
 │   ├── trust-contract.yaml        ← example Trust Contract
 │   ├── agent-spec.schema.yaml     ← audit input format
 │   └── memory-policy.schema.yaml  ← memory governance format
@@ -157,8 +158,8 @@ GitHub renders it behind the *Cite this repository* button as APA or BibTeX.
 
 One rule: **machine-readable definitions are MIT, prose is CC BY 4.0.**
 
-- Schemas (`schemas/*.yaml`) and tooling (`packages/aux-audit`, `action.yml`): **MIT**. Embed them; no attribution condition.
-- Prose (onboarding, vocabulary, pattern write-ups): **CC BY 4.0**.
+- Schemas (`schemas/*.yaml`), pattern examples (`example.py`), and tooling (`packages/aux-audit`, `action.yml`): **MIT**. Embed them; no attribution condition.
+- Prose (onboarding, vocabulary, pattern write-ups and diagrams): **CC BY 4.0**.
 - Teardown content: **CC BY-NC 4.0** (attribution, non-commercial).
 
 GitHub's badge reads CC BY 4.0 because it detects the root **[LICENSE](LICENSE)**, which covers the prose. The schemas are not CC — see **[LICENSING.md](LICENSING.md)** and **[schemas/LICENSE](schemas/LICENSE)**.
diff --git a/docs/schemas.md b/docs/schemas.md
index 193346d..f63d2cc 100644
--- a/docs/schemas.md
+++ b/docs/schemas.md
@@ -12,6 +12,8 @@ the CLI does not duplicate heuristic text, IDs, or the trust-stage mapping.
 | [`schemas/agent-spec.schema.yaml`](../schemas/agent-spec.schema.yaml) | Audit input |
 | [`schemas/memory-policy.schema.yaml`](../schemas/memory-policy.schema.yaml) | Memory governance |
 
+`aux-evolution-curve.yaml` is the one schema `aux-audit` never reads: capability is assessed by a human in a teardown, not derived from a spec.
+
 License: **MIT** ([`schemas/LICENSE`](../schemas/LICENSE)) — embed or vendor them freely. See [`LICENSING.md`](../LICENSING.md) for the full map.
 
 The Evolution Curve is mentioned in
diff --git a/onboarding/appendix-a-vocabulary.md b/onboarding/appendix-a-vocabulary.md
index c7abe00..3aa85e6 100644
--- a/onboarding/appendix-a-vocabulary.md
+++ b/onboarding/appendix-a-vocabulary.md
@@ -33,5 +33,5 @@ Two tables. One for GitHub words (useful to R1 and R4 if they venture there). On
 | **Memory policy** | Declared scopes, retention, and user controls for persistent memory. | [schemas/memory-policy.schema.yaml](../schemas/memory-policy.schema.yaml) |
 | **Pattern** | A named, reusable design move that closes a specific gap. | [repos/agentic-ux-patterns](../repos/agentic-ux-patterns/README.md) |
 | **Teardown** | A structured review of a shipped product: `analysis.md` + `score.json`. | [repos/agent-ux-teardowns](../repos/agent-ux-teardowns/README.md) |
-| **Evolution curve** | The capability axis. Assessed by a human in a teardown, never computed from a spec — a spec says what a product claims, capability is what it does. Stage names are still open. | [trustkit#5](https://github.com/auxfirst/trustkit/issues/5); home is [repos/agent-ux-teardowns](../repos/agent-ux-teardowns) |
+| **Evolution curve** | The capability axis: E01 Reactive → E02 Task-Aware → E03 Context-Aware → E04 Personally Intelligent. Assessed by a human in a teardown, never computed from a spec. Unlike trust, it does not cascade. | [schemas/aux-evolution-curve.yaml](../schemas/aux-evolution-curve.yaml); applied in [repos/agent-ux-teardowns](../repos/agent-ux-teardowns) |
 | **Definition / Executable / Reference** | The three kinds of repo. Every repo is exactly one. | [REPO_STRUCTURE.md](../REPO_STRUCTURE.md) |
diff --git a/repos/agent-ux-teardowns/README.md b/repos/agent-ux-teardowns/README.md
index 4ec6470..8713972 100644
--- a/repos/agent-ux-teardowns/README.md
+++ b/repos/agent-ux-teardowns/README.md
@@ -31,7 +31,7 @@ Every `analysis.md`:
 3. **Where the contract breaks** — named failures from [`trust-gap-classifier`](../trust-gap-classifier).
 4. **Heuristic scorecard** — 10 heuristics, scored 0–3, with evidence.
 5. **Trust stage** — placed on the Trust Architecture ladder.
-6. **Evolution stage** — placed on the capability axis. This is the *only* place the capability axis is assessed: `aux-audit` reads a spec, and a spec states what a product claims rather than what it does. Stage names are still open — see [trustkit#5](https://github.com/auxfirst/trustkit/issues/5) — so `evolution_stage` stays `null` in teardowns until they land.
+6. **Evolution stage** — placed on the capability axis, per [`aux-evolution-curve.yaml`](../../schemas/aux-evolution-curve.yaml). This is the *only* place that axis is assessed: `aux-audit` reads a spec, and a spec states what a product claims rather than what it does. A stage counts only when two distinct scenarios demonstrate it — one clever transcript is an anecdote. Capability is often uneven across surfaces, so record the spread in `evolution_surfaces` rather than averaging it away.
 7. **The fix** — one concrete pattern that would move it up a stage.
 
 Every `score.json`:
@@ -41,7 +41,8 @@ Every `score.json`:
   "product": "ChatGPT",
   "version": "4o, 2026-04",
   "trust_stage": "contextual",
-  "evolution_stage": null,
+  "evolution_stage": "aux.E03",
+  "evolution_surfaces": { "web": "aux.E03", "api": "aux.E01" },
   "heuristics": {
     "aux.H01": 2, "aux.H02": 1, "aux.H03": 3,
     "aux.H04": 2, "aux.H05": 1, "aux.H06": 2,
diff --git a/repos/agentic-ux-patterns/README.md b/repos/agentic-ux-patterns/README.md
index 7f4abbd..5c0df8f 100644
--- a/repos/agentic-ux-patterns/README.md
+++ b/repos/agentic-ux-patterns/README.md
@@ -44,4 +44,13 @@ See [`CONTRIBUTING.md`](../../CONTRIBUTING.md). A pattern PR is rejected if any
 
 ## License
 
-CC BY 4.0.
+Same rule as the rest of TrustKit: **if you embed it, MIT; if you read it, CC BY 4.0.**
+
+| File | License |
+|---|---|
+| `example.py` — the thing you copy into your codebase | **MIT** (`SPDX-License-Identifier: MIT` in each file) |
+| `README.md`, `anti-pattern.md`, `diagram.svg` | **CC BY 4.0** |
+
+The examples are labelled copy-paste, so a license that sends an adopter to
+their legal team defeats the point. Attribution is still wanted — see
+[`CITATION.cff`](../../CITATION.cff) — it is just not a condition of use.
diff --git a/repos/agentic-ux-patterns/check-coverage.py b/repos/agentic-ux-patterns/check-coverage.py
index 3300d2a..73dd773 100644
--- a/repos/agentic-ux-patterns/check-coverage.py
+++ b/repos/agentic-ux-patterns/check-coverage.py
@@ -1,4 +1,5 @@
 #!/usr/bin/env python3
+# SPDX-License-Identifier: MIT
 """Enforce the two claims this repo makes about itself.
 
   1. Every named gap in trust-gap-taxonomy.yaml has a pattern that closes it.
diff --git a/repos/agentic-ux-patterns/patterns/behavioral-contract/example.py b/repos/agentic-ux-patterns/patterns/behavioral-contract/example.py
index c8e3da7..d37f1f9 100644
--- a/repos/agentic-ux-patterns/patterns/behavioral-contract/example.py
+++ b/repos/agentic-ux-patterns/patterns/behavioral-contract/example.py
@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: MIT
 """Behavioral Contract — minimal runnable example.
 
 Golden cases assert on promised properties, never on prose. Changing the pin
diff --git a/repos/agentic-ux-patterns/patterns/confidence-cues/example.py b/repos/agentic-ux-patterns/patterns/confidence-cues/example.py
index 2e8b061..1ad985c 100644
--- a/repos/agentic-ux-patterns/patterns/confidence-cues/example.py
+++ b/repos/agentic-ux-patterns/patterns/confidence-cues/example.py
@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: MIT
 """Confidence Cues — minimal runnable example.
 
 The band is computed from evidence, never from phrasing, and the renderer has
diff --git a/repos/agentic-ux-patterns/patterns/escalation-handoff/example.py b/repos/agentic-ux-patterns/patterns/escalation-handoff/example.py
index 0f4fcd0..3093672 100644
--- a/repos/agentic-ux-patterns/patterns/escalation-handoff/example.py
+++ b/repos/agentic-ux-patterns/patterns/escalation-handoff/example.py
@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: MIT
 """Escalation Handoff — minimal runnable example.
 
 A blocked situation produces a route with a context packet and the agent's
diff --git a/repos/agentic-ux-patterns/patterns/intent-handshake/example.py b/repos/agentic-ux-patterns/patterns/intent-handshake/example.py
index 59b0272..bab9476 100644
--- a/repos/agentic-ux-patterns/patterns/intent-handshake/example.py
+++ b/repos/agentic-ux-patterns/patterns/intent-handshake/example.py
@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: MIT
 """Intent Handshake — minimal runnable example.
 
 Run with: python example.py
diff --git a/repos/agentic-ux-patterns/patterns/memory-in-motion/example.py b/repos/agentic-ux-patterns/patterns/memory-in-motion/example.py
index 6b10082..fc77766 100644
--- a/repos/agentic-ux-patterns/patterns/memory-in-motion/example.py
+++ b/repos/agentic-ux-patterns/patterns/memory-in-motion/example.py
@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: MIT
 """Memory in Motion — minimal runnable example.
 
 Run with: python example.py
diff --git a/repos/agentic-ux-patterns/patterns/memory-policy-scoping/example.py b/repos/agentic-ux-patterns/patterns/memory-policy-scoping/example.py
index 3bc7283..9ba89f9 100644
--- a/repos/agentic-ux-patterns/patterns/memory-policy-scoping/example.py
+++ b/repos/agentic-ux-patterns/patterns/memory-policy-scoping/example.py
@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: MIT
 """Memory Policy Scoping — minimal runnable example.
 
 Ancestors are readable, siblings never are, and out-of-scope hits are withheld
diff --git a/repos/agentic-ux-patterns/patterns/user-aligned-objective/example.py b/repos/agentic-ux-patterns/patterns/user-aligned-objective/example.py
index 9671024..e74f24c 100644
--- a/repos/agentic-ux-patterns/patterns/user-aligned-objective/example.py
+++ b/repos/agentic-ux-patterns/patterns/user-aligned-objective/example.py
@@ -1,3 +1,4 @@
+# SPDX-License-Identifier: MIT
 """User-Aligned Objective — minimal runnable example.
 
 User value and platform value are scored separately and never summed. The user's
diff --git a/repos/aux-frameworks/README.md b/repos/aux-frameworks/README.md
index 493df46..b191e27 100644
--- a/repos/aux-frameworks/README.md
+++ b/repos/aux-frameworks/README.md
@@ -14,6 +14,8 @@
   /trust-architecture       # 4-stage trust maturity ladder
     schema.yaml
     stages.yaml
+  /aux-evolution-curve      # 4-stage capability axis (human-assessed)
+    schema.yaml
   /aux-audit                # audit rules derived from the above
     schema.yaml
     rules.yaml
diff --git a/schemas/aux-evolution-curve.yaml b/schemas/aux-evolution-curve.yaml
new file mode 100644
index 0000000..633bce7
--- /dev/null
+++ b/schemas/aux-evolution-curve.yaml
@@ -0,0 +1,88 @@
+# Evolution Curve — capability maturity axis
+# Version: 0.1.0
+# License: MIT (see schemas/LICENSE)
+#
+# The second axis, independent of trust. Assessed by a human in a teardown,
+# never computed from an agent spec: a spec states what a product claims,
+# capability is a claim about what it does. See trustkit#5.
+#
+# Referenced by: agent-ux-teardowns
+# NOT referenced by aux-audit — the CLI reports evolution_stage: null by design.
+
+version: "0.1.0"
+name: "Evolution Curve"
+source: "auxfirst"
+axis: capability
+
+principle: >
+  Capability is what the agent can attempt. Trust is whether it should be
+  allowed to. The two move independently, and the dangerous corner of the
+  grid is high capability with low trust — not low capability.
+
+# Unlike the Trust Architecture, this is not a ladder.
+# Trust cascades: one advocacy violation collapses trust back to functional.
+# Capability does not. A product that fails at E04 has not become E01; it has
+# demonstrated E04 unreliably. Nothing is lost, so nothing snaps back.
+cascades: false
+
+stages:
+  - id: aux.E01
+    order: 1
+    name: "Reactive"
+    question: "Does it answer what it is asked, one turn at a time?"
+    observed_when:
+      - "responds to the request in front of it"
+      - "no tool use, or a single tool"
+      - "each turn is independent of the last"
+    not_evidence:
+      - "a long answer — length is not capability"
+
+  - id: aux.E02
+    order: 2
+    name: "Task-Aware"
+    question: "Can it carry a multi-step task to completion?"
+    observed_when:
+      - "holds a goal across several turns without being re-prompted"
+      - "composes two or more tools toward one outcome"
+      - "recovers from a failed step instead of stopping"
+    not_evidence:
+      - "a numbered plan it does not then follow"
+
+  - id: aux.E03
+    order: 3
+    name: "Context-Aware"
+    question: "Does it use the situation, not just the request?"
+    observed_when:
+      - "reads state the user did not hand it"
+      - "acts differently in the same request under different circumstances"
+      - "operates across more than one surface with continuity"
+    not_evidence:
+      - "reciting stored preferences back — that is E02 with memory"
+
+  - id: aux.E04
+    order: 4
+    name: "Personally Intelligent"
+    question: "Does it adapt to this particular user over time?"
+    observed_when:
+      - "behaviour differs between two users beyond their stored preferences"
+      - "the difference grows with use rather than being configured"
+    not_evidence:
+      - "a settings page"
+      - "a preference the user typed once and the agent replays"
+
+assessment:
+  by: human
+  in: "repos/agent-ux-teardowns"
+  # A product sits at one stage: the highest it reaches *reliably*.
+  scoring: ordinal
+  reliability_threshold: 2
+  reliability_rule: >
+    A stage counts only when the teardown demonstrates it in at least two
+    distinct scenarios. One clever transcript is an anecdote; the curve
+    describes what a user can rely on.
+  # Capability is frequently uneven across surfaces. Record it rather than
+  # averaging it away — the spread is often the finding.
+  per_surface: optional
+  evidence_required: >
+    Every stage claim cites a reproducible scenario, per CONTRIBUTING. No
+    stage is assigned from a product's marketing or its own documentation.