AI & ML interests
a11oy.net
Recent Activity
Control before action.
Evidence after.
Governed models and systems for decisions that must survive scrutiny.
Artifact type, evidence, and runtime state remain separate.
One front door. Four focused paths.
Move directly from the mission to its source, boundaries, and current evidence.
Govern the action
Bounded tools and receipts. Provider, signing, readiness, and authorization stay separate.
Open a11oy ->See the mission
Live-feed fusion when dependencies answer; samples stay labeled. Public actuation is SIMULATED—no live weapon command or production authorization.
Open Killinchu ->Inspect the artifact
Weights, data, software, and recipes—with lineage and explicit limits.
View portfolio ->Verify outside the UI
Receipts, replay, proof sources, and current-state readback.
Inspect evidence ->Six stages. One accountable path.
Propose, validate authority, act within bounds, and emit a portable receipt.
- SignalObserve and attribute the request.
- ReasonGround the proposal in disclosed context.
- PolicyAllow, require approval, or refuse.
- ActExecute only inside the permitted bound.
- ReceiptBind inputs, policy, action, and outcome.
- VerifyReplay evidence outside the product UI.
Five artifact classes. No category shortcuts.
A model-shaped repository can contain weights, software, a surrogate, or a recipe. Each class carries a different publication and evaluation contract.
-
Trained weights
Models and adapters
Neural parameters with exact base lineage, files, evaluation, runtime envelope, and autonomy boundary.
Inspect Khipu -> -
Software and substrate
Code-shaped artifacts
Source, manifests, kernels, meters, and governance packages. No trained weights are implied.
Inspect substrate -> -
Surrogates and recipes
Narrow compatibility artifacts
Small structural classifiers or upstream-model recipes, not replacements for the governed runtime.
Inspect a surrogate -> -
Datasets and evidence
Records with admission boundaries
Corpora, receipts, evaluations, manifests, proofs, and snapshots. Presence is not automatic training admission or proof.
Inspect szl-lake -> -
Spaces and demonstrations
Interfaces, not green badges
Consoles, verifiers, and status surfaces. Reachability proves transport availability only.
Open the current README Space ->
Status is linked, not hardcoded.
These endpoints may change without a card edit. A historical mirror, cached screenshot, download count, or green badge is not a current-state source.
Canonical sources
Evidence and runtime are separate axes.
A signature establishes integrity and origin within scope. It does not automatically establish quality, safety, compliance, adoption, profitability, or authorization to operate.
- PROVED
- MEASURED
- REPORTED
- MODELED
- CONJECTURE
- ROADMAP
The dataset SZLHOLDINGS/SZLHOLDINGS is a HISTORICAL profile mirror, not a current organization card, inventory, or runtime source.
Move from model output to accountable action.
Start with the artifact type, follow the evidence, check current state, then reproduce the exact path from source.
buckets 1
spaces 27
a11oy — Command Center
a11oy — governed-AI Command Center, signed receipts
killinchu — Andean Drone Intelligence
counter-UAS edge organ · 13-axis Λ-gate · DSSE receipts
Khipu Loom — Governed AI Estate
Searchable public AI estate with live evidence
Holographic Estate
Live Atlas integrity and brain-trust evidence observatory
SZL Cosmos
Holographic map of the SZL sovereign substrate