docs: state that the loader lives in a different repo

The npz weights are real and the card does not overstate them, but there is no config.json and no loader here, so from_pretrained and the inference widget cannot load it and nothing documents the array names or forward pass. Adds a working hf_hub_download snippet, points at szl_khipu in szl-khipu-kernels as the forward pass this was trained against, and labels the repo a test fixture rather than a deployable model until the loader ships alongside the weights.

README.md

@@ -2,13 +2,38 @@
 license: apache-2.0
 library_name: numpy
 tags:
-  - governed-ai
-  - szl-holdings
-  - doctrine-v11
-  - nano
-  - synthetic
+- governed-ai
+- szl-holdings
+- doctrine-v11
+- nano
+- synthetic
+- needs-loader
+- test-fixture
 ---
 
+> ### How to actually load this — the weights alone are not enough
+>
+> `receipt_agent.npz` is a real, honest 4-10-4 MLP produced by a real training run, and the card
+> below does not overstate it. But it is a bare NumPy archive with **no
+> `config.json` and no loader in this repo**, so `from_pretrained` and the Hub
+> inference widget cannot touch it. Nothing here tells you the array names or the
+> forward pass.
+>
+> ```python
+> import numpy as np
+> from huggingface_hub import hf_hub_download
+>
+> path = hf_hub_download("SZLHOLDINGS/ReceiptAgent-Nano", "receipt_agent.npz")
+> w = np.load(path)
+> print(sorted(w.files))   # array names are the de-facto interface
+> ```
+>
+> The forward pass this was trained against lives in the `szl_khipu` package, in
+> [SZLHOLDINGS/szl-khipu-kernels](https://huggingface.co/SZLHOLDINGS/szl-khipu-kernels)
+> — a **different repository**. Until the loader ships alongside the weights (or a
+> `custom_code` handler is added), treat this repo as a **test fixture**, not a
+> deployable model. Evidence status: see TRAINING_RECEIPT.json / BENCH.*.json (SYNTHETIC).
+
 # ReceiptAgent-Nano
 
 ALLOW · DENY · ABSTAIN · ESCALATE. Escalation is a class, not a retry loop.