Governed Kernels & Verifiers
Collection
Kernel software packages and their live/holographic demo surfaces. Cards govern algorithms, hashes, readiness, and proof scope; inclusion is not produ • 19 items • Updated
How to use SZLHOLDINGS/szl-khipu-kernels with Kernels:
# !pip install kernels
from kernels import get_kernel
kernel = get_kernel("SZLHOLDINGS/szl-khipu-kernels")Kernel-Hub card for the SZL KHIPU NumPy silhouettes. This is a kernel suite, not a checkpoint. Do not treat Chaski FIFO as a Hub model card.
CPU numpy path LIVE. CUDA UNAVAILABLE this session — that is honesty, not a missing bench.
from kernels import get_kernel
k = get_kernel(
"SZLHOLDINGS/szl-khipu-kernels",
revision="main",
trust_remote_code=True,
)
# CPU numpy path is LIVE.
# CUDA is UNAVAILABLE this session — that is honesty, not a missing bench.
print(k.CONJECTURE_1)
gate = k.lambda_gate([0.95] * 13)
print(gate["score"], gate["passed"], gate["proven_trust"]) # advisory; Conjecture 1 OPEN
Package fallback (labeled; not a Hub load):
from szl_khipu import (
evaluate_lambda, yarqa_attn, evaluate_greenlight, evaluate_anatomy,
run_chaski, run_ayni, run_shard, run_bay, run_prefix, run_route,
)
gate = evaluate_lambda([0.95] * 13)
print(gate["value"], gate["blocked"], gate["proven_trust"]) # advisory; Conjecture 1 OPEN
import numpy as np
q = kv = np.random.default_rng(7).standard_normal((12, 4))
_out, _probs, leaked = yarqa_attn(q, kv, kv, n_canals=3)
print(leaked) # bound ≤ 1e-9
print(run_prefix(11, hijack=0)["hold"]) # 1
print(run_prefix(11, hijack=1)["hold"]) # 0 · poison after digest
print(run_route(11, tamper=0)["hold"]) # 1
print(run_route(11, tamper=1)["hold"]) # 0 · expert swap
HTTP hologram (same process as the kernels):
GET /api/bench
POST /api/infer {"kind":"tiny_khipu","query":"resolve F18 handle"}
POST /api/prefix {"hijack":1}
POST /api/route {"tamper":1}
GitHub source: szl-holdings/szl-khipu.
| Kernel | Field leader we cut from | Bound | Status |
|---|---|---|---|
| TileReceipt | FlashAttention | residual vs naive | LIVE |
| ScoreMod | FlexAttention | future mass = 0 | LIVE |
| BlockWitness | vLLM PagedAttention | table digest | LIVE |
| YARQA | SageAttention | canal leak = 0 | LIVE |
| PrefixWitness | SGLang RadixAttention | poison after digest → BLOCKED | LIVE |
| RouteWitness | Mixtral / Switch MoE | expert swap → BLOCKED | LIVE |
| GreenLight / Ari | signed assent (dual of Willay) | sorry cannot paint | LIVE |
| Kay Pacha / Anatomy | five-organ substrate | zero HEART fail-closes | LIVE |
| Chaski FIFO | F7 silhouette | reorder/drop → BLOCKED | LIVE · not a Hub checkpoint |
| Ayni | residual bus | skip leak → BLOCKED | LIVE |
| ShardWitness | RS(10,6) GF(257) | need 6 of 10 | LIVE · CHECKED ≠Lean PROVEN |
| Evidence Bay | four rails | Space is not proof | LIVE |
| GovEnvelope | DSSE envelope | STRUCTURAL-ONLY · never a fake key | LIVE |
| Invariants | locked-8 | energy UNAVAILABLE · uniqueness OPEN | LIVE |
| Λ gate | WGM | Conjecture 1 OPEN | ADVISORY |
| Path | Status |
|---|---|
| CPU NumPy | LIVE |
| CUDA / Triton cubin | UNAVAILABLE |
| Energy | UNAVAILABLE — never a fabricated joule |
| proven_trust | false |
| Conjecture 1 | OPEN |
| 1.5B / Qwen / Mixtral / SGLang weights | not scraped, not rehosted, not trained here |
| Not | FlashAttention, SageAttention, FlexAttention, vLLM, Mixtral, SGLang |
Apache-2.0. Copyright 2026 SZL Holdings. Doctrine v11 LOCKED.