Adapter integration guide and contract surface (walkthrough, schema descriptions, docs index; longer-term Adapter trait) #130

There is no code-level adapter contract: src/main.rs is a single 27,586-line file with zero trait declarations; each analyzer is a bespoke run_<tool>_kernel / run_<tool>_case function family plus hand-wired enum variants (Commands::*, partition constants). The conceptual contract at the head of docs/adapters.md is clear (five normalized outcomes, "no adapter may synthesize a tool result", never downgrade failure to not-reached), and schemas/result.schema.json fixes the output shape — but:

• There is no "How to add an adapter" walkthrough anywhere; CONTRIBUTING.md gives it one sentence. The real onboarding path is imitating the (excellent) 1,000-line adapter READMEs, and nothing says so.
• The schemas are nearly description-free: result.schema.json and freeze.schema.json have zero property descriptions; case.schema.json describes 3 of 21 top-level properties. Field semantics (negative_mechanism, witness_checkpoints, expected_analysis_capability) live scattered in docs prose.
• docs/ has no index separating normative contracts from historical amendment logs, and no glossary for insider terms (freeze-bound, recorded-only, wave, stratum, amendment). The published website renders only a handful of pages, so the contract docs are repo-only.

Fix (increments)

1. A step-ordered "Adding an adapter" guide: required inputs, outcome normalization rules, evidence/timing/environment artifacts, validation commands, and the exact src/main.rs touchpoints.
2. description fields across all three schemas (or a schemas/README.md).
3. docs/README.md index + glossary.
4. Longer-term: extract an Adapter trait (or at least a shared runner module) from src/main.rs so an integration is a bounded, reviewable unit.

Found during an adversarial review of the repo (2026-09-02).