Thing	Label	Method / N / date / what-NOT
tests (`PYTHONPATH=torch-ext`)	PASS	MEASURED 2026-08-29T15:54:00Z host `betterwithage` Windows-10-10.0.26200-SP0. torch `2.10.0+cu128`. GPU `NVIDIA GeForce RTX 5050 Laptop GPU` arch `Blackwell`. pytest `16 passed in 0.07s`. Failed nodes: `none`. What-NOT: not a leaderboard. torch.compile fullgraph failures on Windows Blackwell (`cl is not found`) are MEASURED, not hidden.
Kernel Hub `get_kernel`	import-LIVE	kernels `0.16.1`. Default: `get_kernel("SZLHOLDINGS/szl-invariants", revision="main", trust_remote_code=True)` → `True`. `backend="cpu"` → `True`. trust_remote_code=False → `ValueError` (SZLHOLDINGS is not a trusted publisher). repo_type=kernel required (kernels 0.16). What-NOT: not a weight load; do not pickle/joblib.load.
formula-tax	ADVISORY	locked-8 `F1 F4 F7 F11 F12 F18 F19 F22`. registry_count=21. Λ geomean `1.0`. uniqueness Conjecture 1 (never a theorem).
I1–I8	catalog	`I1 receipt-chain-continuity; I2 ledger-failure-shape; I3 served-run-has-model; I4 signed-columns-atomic; I5 loop-steps-positive; I6 receipt-ed25519-verify; I7 receipt-columns-consistent; I8 flywheel-lineage`. Executed by `SZLHOLDINGS/szl-invariants`. Statuses never coerced. Λ untouched.
CUDA speedup / tokens/s / joules	UNAVAILABLE	Not claimed. Receipted kernels, not silent CUDA.

GitHub source: szl-holdings/szl-invariants @ 0a1e6bdeaf4d7515bbbf4ca37c3c838a4f293542. Artifacts: BENCH.laptop-blackwell.json, OPERATIONAL.json.

python

from kernels import get_kernel
k = get_kernel("SZLHOLDINGS/szl-invariants", revision="main", trust_remote_code=True)

szl-invariants

Eight falsifiable receipt/ledger invariants, offline. Not a model. No weights.

IS: a pure-Python, stdlib-only governance kernel that replays the same eight checks the a11oy backbone runs at /api/invariants over a ledger JSONL you already hold. Public API: run_invariants, load_jsonl, verify_ed25519, selfcheck. Statuses HOLDS / VIOLATED / KEY_ROTATED / NO_DATA / UNAVAILABLE are first-class.

IS NOT: trained weights or a LoRA. Not a CUDA bench. Passing selfcheck is not an eval leaderboard. Hub model.joblib is QUARANTINED executable serialization — do not joblib.load it. These checks do not prove the export is complete.

Canonical GitHub source: https://github.com/szl-holdings/szl-invariants
Hub package: https://huggingface.co/kernels/SZLHOLDINGS/szl-invariants
This model-type repo is the publish / card mirror. Apache-2.0.

python

<!-- SZL-ATELIER-CUT:v1:START -->
## The cut

Checksums are optional. Ours are the product. If the signature fails, the model is not the model.

Weights that are invalid if unsigned — at load time, not at audit time.

### Silhouette → leave → SZL

| Leader | Take, then tweak |
|---|---|
| Anthropic | Trust, but here we verify. |
| NVIDIA | Supply-chain (signed containers) applied to tensors. |
| Unsloth | After merge, sign. Not during QLoRA. |

Nobody else ships this combination. That is the point of a one-of-one.

## Intended use

Load-time invariant check.

## Limitations

- Proves integrity and origin, never accuracy.

Canonical GitHub: [`szl-holdings/szl-invariants`](https://github.com/szl-holdings/szl-invariants/blob/main/README.md)
<!-- SZL-ATELIER-CUT:v1:END -->

from kernels import get_kernel
inv = get_kernel("SZLHOLDINGS/szl-invariants", revision="main", trust_remote_code=True)
report = inv.run_invariants(inv.load_jsonl("runs_export.jsonl"),
    samples=inv.load_jsonl("training_samples.jsonl"), pubkey="<SPKI base64>")
print(report["summary"])
print(inv.selfcheck())

The eight ids: receipt-chain-continuity, ledger-failure-shape, served-run-has-model, signed-columns-atomic, loop-steps-positive, receipt-ed25519-verify, receipt-columns-consistent, flywheel-lineage. Missing pubkey → #6 UNAVAILABLE; missing samples → #8 UNAVAILABLE.