# Defensibility trace

This is ARD's **supplementary defensibility trace**: vocabulary, guardrails, and
stress-testing behind selected commitments, one hop from the
[specification](../SPEC.md), which itself reads standalone.

It is **not the primary warrant tier**. Warrant lives in
[`../evidence/`](../evidence/) — the empirical failure ledger. This directory is
a curated defensibility subset, not the full research substrate: one grounding
artifact per major commitment, each earning its place. It **grows by extension**
as more commitments gain a rationale.

## What's here

- **[COMMITMENTS.md](COMMITMENTS.md)** — the map: each load-bearing spec/catalog commitment tied to the position that grounds it (and an honest list of what is not yet traced).
- **[positions/](positions/)** — research positions: agent-authored syntheses, each carrying its own external grounding (the traditions, papers, and standards it rests on). De-coupled from the originating deployment — decision-record citations and deployment-specific paths stripped — but reasoning and external groundings intact.
- **[references.md](references.md)** — the consolidated bibliography: every external source the positions rest on, de-duplicated into one numbered list. A position's `[handle]{N}` citation resolves to the correspondingly-numbered entry here. Bibliographic only — no verbatim source reproduction.

## What does not travel

- **Raw source fetches** — excluded. Only IP-cleared derived material is committed; per-source-class reproduction limits are respected.
- **Source attestations that merely reproduce external copyrighted material** — the *findings* that defend each commitment are ARD's own (the positions, the control-space model, the failure-shape inventory), so the attestation layer is not reproduced here.
- **Operational substrate** — the deployment's own decision records and working memory stay in the deployment; the research-tier positions carry the public rationale.