File size: 4,841 Bytes
be94813 1d6efc6 c58afa2 1d6efc6 c58afa2 be94813 1638c7d be94813 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 | ---
tags:
- kernel
library_name: kernels
license: apache-2.0
---
# szl-lambda-gate
**Λ — a governance aggregator as a Hugging Face kernel.** A differentiable, torch.compile-friendly weighted-geometric-mean aggregator with an ADVISORY non-compensatory gate and runtime axiom self-checks, from [SZL Holdings](https://huggingface.co/SZLHOLDINGS).
> Companion to [`szl-governed-norm`](https://huggingface.co/SZLHOLDINGS/szl-governed-norm). Where that kernel makes a normalization *auditable*, this one makes a *governance decision* computable and checkable at the tensor layer.
## Interactive demo
> **ROADMAP** — companion showcase Spaces (`lambda-gate-holo`, `lambda-aggregator-live`) are planned but not yet live. The unified `szl-kernels-live` demo is the planned entry point.
>
> The quickstart above runs fully locally. For a full governed-kernel suite demo, see [szl-kernels](https://huggingface.co/SZLHOLDINGS/szl-kernels). For the live a11oy substrate, see [a11oy Space](https://huggingface.co/spaces/SZLHOLDINGS/a11oy).
## What Λ is — and is NOT (read this first)
Λ is the **weighted geometric mean** over axis scores in [0,1]:
\[ \Lambda(x) = \prod_i x_i^{w_i}, \quad \sum_i w_i = 1, \; w_i > 0, \; x_i \in [0,1] \]
It is a **non-compensatory, ADVISORY** roll-up: any single zeroed (or non-finite) axis drives the whole aggregate to 0 — a conservative "one bad axis fails the gate" signal. **Λ is NOT "proven trust" and NOT a closed theorem.** Its *uniqueness* (that the weighted geometric mean is the only aggregator satisfying the carried axioms) remains **Conjecture 1 — OPEN**. A gate "pass" is an advisory signal, never a guarantee. We label this honestly everywhere.
## Quickstart
```python
import torch
from kernels import get_kernel
# Current `kernels` (>=0.15) requires an explicit revision/version + trust flag for org kernels:
lg = get_kernel("SZLHOLDINGS/szl-lambda-gate", revision="main", trust_remote_code=True)
# (once a tag is published you can pin it, e.g. revision="v0.2.0")
axes = torch.tensor([0.9, 0.8, 0.95]) # axis scores in [0,1]
score = lg.lambda_aggregate(axes) # Λ(x) ∈ [0,1]
res = lg.lambda_gate(axes, threshold=0.5)
print(res.score, res.passed, res.advisory) # advisory is always True
print(lg.selfcheck()) # empirical A1–A4 checks + version
```
## API
| Function | Notes |
|---|---|
| `lambda_aggregate(axes, weights=None)` | Λ over the last dim. Differentiable, batched, torch.compile-friendly. |
| `lambda_gate(axes, weights=None, threshold=0.5)` | Advisory gate → `LambdaGateResult(score, passed, threshold, advisory)`. |
| `lambda_gate_batch(candidates, weights=None, threshold=0.5)` | Score many candidate vectors `(..., N, k)` in one call; returns the advisory pass mask. |
| `selfcheck()` | Empirical A1–A4 axiom checks + adversarial falsification search + version. NOT a uniqueness proof. |
| `is_monotone / is_homogeneous / is_egyptian_exact / is_bounded_by_max` | The four carried axioms as real runtime checks. |
| `yuyay_weights()`, `YUYAY_AXES`, `YUYAY_FLOORS` | Canonical 13-axis Yuyay preset (advisory). |
| layers: `LambdaGate`, `LambdaAggregate` | Pure `nn.Module` for the Kernel Hub layer-mapping mechanism. |
## Carried axioms (verifiable, not a proof)
- **A1 IsMonotone** — Λ is non-decreasing in each axis.
- **A2 IsHomogeneous (deg 1)** — Λ(t·x) = t·Λ(x).
- **A3 IsEgyptianExact** — Λ(c,…,c) = c.
- **A4 IsBounded** — Λ(x) ≤ maxᵢ xᵢ.
`selfcheck()` verifies these empirically on sampled inputs and runs a random falsification search. A clean run is **evidence, not proof** — Λ-uniqueness is Conjecture 1 (open).
## Provenance
Backed by the Lean 4 formalization [szl-holdings/lutar-lean](https://github.com/szl-holdings/lutar-lean) (749 declarations / 14 axioms / 163 tracked sorries), DOI [10.5281/zenodo.20434308](https://doi.org/10.5281/zenodo.20434308). Λ uniqueness = Conjecture 1 (open).
## Honesty
- Pure-Python universal kernel — a correctness reference, not a CUDA speed record. No fabricated benchmarks (50 passing tests).
- Λ is advisory; never "proven trust."
- Prior art honestly attributed: the weighted geometric mean as a less-compensatory composite indicator is established practice (UN HDI 2010, OECD Composite Indicators Handbook 2008); the veto/cut-off idea is ELECTRE. The 13-axis conjunctive form is SZL's own yuyay_v3 gate.
## Compatibility
Python 3.9+, `torch>=2.5`, standard library + torch only.
## License
Apache-2.0. Copyright 2026 SZL Holdings.
---
<sub><b>SZL Holdings</b> · Λ governance aggregator · advisory, not proven trust · <a href="https://a11oy.net">a11oy.net</a> · <a href="https://github.com/szl-holdings">github.com/szl-holdings</a> · <a href="https://huggingface.co/SZLHOLDINGS">huggingface.co/SZLHOLDINGS</a></sub>
|