card(atelier): MEASURED nano silhouette — not a 1.5B retrain

README.md

@@ -3,49 +3,53 @@ license: apache-2.0
 library_name: numpy
 tags:
   - governed-ai
-  - khipu
   - szl-holdings
-  - moons
-  - mlp
-  - silhouette
+  - doctrine-v11
+  - nano
+  - measured
 ---
 
 # Moons-Nano
 
-Two-moons **2→8→2** tanh-softmax SGD. A few hundred floats. **Not 1.5B. Not Qwen. Not a foundation model.**
+Two interlocking moons. A 2-8-2 tanh MLP. The toy that teaches fail-closed before anyone touches 1.5B.
 
-Canonical source: [szl-holdings/szl-khipu](https://github.com/szl-holdings/szl-khipu)  
-Sibling card: [SZLHOLDINGS/szl-khipu](https://huggingface.co/SZLHOLDINGS/szl-khipu)
+**Family.** nano · **Evidence.** MEASURED · **Weights.** numpy · **Params.** 2-8-2
 
-```python
-from szl_khipu.train import moons
+Hub: [SZLHOLDINGS/Moons-Nano](https://huggingface.co/SZLHOLDINGS/Moons-Nano)
 
-weights, ev = moons.train(seed=20260721, steps=400)
-print(ev["acc"], ev["loss"])
-# REPORTED: acc 0.93 · loss ~0.13 on the training moons
-moons.save_npz("moons.npz", weights)
-```
+## The cut
 
-## What it does
+Accuracy is the wrong headline. We publish holdout accuracy AND silhouette AND the decision boundary you can click. Leaders hide the toy. We hang it in the foyer.
 
-- Classic two-moons toy classification. Hidden width 8. Softmax over 2.
-- Trained here on CPU NumPy. Honesty **REPORTED**. Energy **UNAVAILABLE**.
+A model card that is a laboratory: click the plane, watch the gate, read the loss.
 
-## What it is NOT
+### Silhouette → leave → SZL
 
-- **Not SZL-Khipu-1.5B.** Not QLoRA. Not a chat model.
-- **Not sklearn moons as a product claim.** A live silhouette so the estate has a TRAINED tiny MLP that actually ran.
-- **Not proven trust.** Λ uniqueness remains Conjecture 1 OPEN.
-- Energy **UNAVAILABLE**. CUDA **UNAVAILABLE**. Never a fabricated joule.
+| Leader | Take, then tweak |
+|---|---|
+| Anthropic | Small, inspectable, no capability theater. |
+| NVIDIA | MLP instead of cuDNN flex — the kernel idea at 2D. |
+| Unsloth | SGD on 400 points, not QLoRA. Same honesty about scale. |
+
+Nobody else ships this combination. That is the point of a one-of-one.
+
+## Intended use
+
+Onboard the cut: measured numbers, visible boundary.
+
+## Limitations
+
+- Toy. 2D. Noise 0.13. Not a product model.
 
 ## Honesty
 
-| Claim | Label | What-NOT |
-|---|---|---|
-| Weights trained in this package | REPORTED | silhouette, Not 1.5B |
-| acc 0.93 on the training moons | REPORTED | not a published benchmark |
-| Λ | ADVISORY · Conjecture 1 OPEN | never a theorem |
-| Energy | UNAVAILABLE | never a fabricated joule |
-| CUDA | UNAVAILABLE | CPU numpy LIVE |
+| Claim | Label |
+|---|---|
+| This card's numbers | MEASURED |
+| Energy / joules | UNAVAILABLE unless a signed meter says MEASURED |
+| Λ uniqueness | Conjecture 1 OPEN — not a theorem |
+| GGUF as the signed object | FALSE |
+
+Doctrine v11 LOCKED · 749 declarations · 14 axioms · 163 sorries · locked-proven 8.
 
-Doctrine v11 LOCKED · 749/14/163 · locked-proven 8. Apache-2.0. Copyright 2026 SZL Holdings · Stephen P. Lutar Jr. · ORCID [0009-0001-0110-4173](https://orcid.org/0009-0001-0110-4173).
+Apache-2.0. Copyright 2026 SZL Holdings · Stephen P. Lutar Jr. · ORCID [0009-0001-0110-4173](https://orcid.org/0009-0001-0110-4173).