# Research & Development

The report argues that a serious technology site needs visible research and development depth. Runtime.MiRust.com should show that depth without turning speculative ideas into completed-runtime claims.

## R&D pipeline

1. **Intake**: source review bundles, implementation reports, web architecture reports, and improvement reports arrive as source material.
2. **Durable body**: useful bodies are preserved under `/docs/`.
3. **Pointer memory**: `.uai` files record route summaries, checksums, and truth boundaries.
4. **Public synthesis**: PHP-rendered pages explain the parts in human-readable form.
5. **Claim gate**: claims stay bounded to source-backed implementation, manifests, receipts, tests, and browser smokes.

## Active research lanes

| Lane | Current site treatment | Evidence route |
| --- | --- | --- |
| Rust/WASM runtime | Presented as the current implementation boundary. | [Architecture](/architecture/) and `/docs/implementation/` |
| `.slm` model format | Explained as the current file route before broader GGUF import. | [Model Format](/model-format/) |
| Adapters | ADP1, ASP1, and ALR1 explained as package lanes that Rust validates. | [Adapters & Assembly](/adapters-assembly/) |
| Model breeding | Offline compatibility, lineage, operators, receipts, and selector admission. | [Model Breeding](/model-breeding/) |
| Browser proof | Manifest, module-plan, assembly, adapter-family, and drift-test evidence. | [Proof & Testing](/proof-testing/) |
| Agent-readable site | Markdown alternates, `llms.txt`, `llms-full.txt`, docs, and `.uai`. | [Agent-Readable Site](/agent-readiness/) |
| Infrastructure trust | DNS/TLS/security/deploy checks made explicit. | [Infrastructure & Trust](/site-infrastructure/) |

## What should be expanded next

- A concise diagram showing browser, WASM, model file, manifest, module plan, adapter stack, and receipt flow.
- A current-limit table for context tokens, prompt bytes, output bytes, model byte budgets, adapter budgets, and route-count limits.
- A release-evidence page for each deployed package version.
- A benchmark page that separates deterministic runtime proof from trained assistant quality.
- A public issue/decision log distilled from `/docs/source-memory/log.md`.

## Research language rule

Use “research lane,” “design pressure,” “future work,” “receipt route,” and “proof gate” where appropriate. Avoid language that makes exploratory model-breeding concepts sound like uncontrolled live autonomy.