Risk LabStrong architectural inferencev1.21.5
In plain English
This page provides local browser worksheets. They help plan reviews; they are not formal safety certifications.
- Why this matters: AI risk can come from the whole arrangement, not one obvious model.
- What to look for: data, memory, routes, adapters, tools, evaluators, updates, and rollback paths.
- Technical version below: the expert terminology remains available and is linked through the glossary.
Edge Browser Runtime Review
Evidence levelStrong architectural inferenceTechnical label: Architectural inference
Use this review page before treating a local browser model stack as reproducible. It is a worksheet, not a certification.
Step 1 — Define the exact local stack
| Field | Record |
|---|---|
| Browser and version | |
.wasm binary hash | |
| Model container hash | |
| Tokenizer hash | |
| Adapter stackA set of adapters loaded together, usually in a defined order. Open glossary definition and load order | |
| Quantization format | |
| SIMD / atomics / threads enabled | |
| Local stores used | |
| Evaluation suite version | |
| Reset ecology command tested |
Step 2 — Review state carriers
- Model weights.
- AdapterA small add-on that changes or specializes model behavior. Open glossary definition deltas.
- Tokenizer tables.
- Prompt package.
- KV cache pages.
- Radix-prefix cache.
- Speculative draft branches.
- IndexedDB / Cache Storage / File System Access handles.
- Worker-shared memory.
- Diagnostics and evaluation sidecars.
Step 3 — Decide release outcome
| Outcome | Use when |
|---|---|
| Promote | All artifacts, state, cache, and reset boundaries are reproducible. |
| Hold | Evaluation is incomplete, but no urgent release need exists. |
| Rebuild | Hashes, compiler settings, or adapter compatibility cannot be verified. |
| No-opThe decision not to change the system. Open glossary definition | The evidence packet is weaker than the existing release. |