Resilience Classify

Use this skill to review collateral quality and whole-book custody coverage, adding only evidence-backed metadata.

What these fields drive today: custody eligibility filters in the Selector (shared/lib/selector/) and scoped verdict signals in the DDR depeg-duration resolver (shared/lib/depeg-resolver/). The Selector's retired custody/collateral V8 ranking axes no longer drive ranking or "what to watch". These scalar labels do not feed the current Safety Score methodology. Published safety grades will not move from these labels, so do not promise or expect grade changes.

Read First

• Read resolveCustodyModel() in shared/lib/report-card-policy.ts: authored custody wins; omission resolves to unknown for RWA-backed assets with non-decentralized governance and to onchain for the remaining structural classes. When whole-book evidence cannot support that default, author the reviewed mixed/unknown label explicitly.
• Read shared/lib/methodology-versions/current-version.json before describing scores; use the current Safety Score methodology and let that source file win over remembered versions.
• This skill is only for collateralQuality and custodyModel. Leave governanceQuality alone unless the user explicitly asked for it.

Workflow

1. Read the stablecoin entry in shared/data/stablecoins/coins/*.json (or shared/data/stablecoins/coins.generated.json). For coins with a reserves sidecar, the reserve composition that informs collateralQuality/custodyModel lives in shared/data/stablecoins/domains/reserves/<id>.json, not the base file. Treat the runtime stablecoin re-export as import-only.
2. Review the complete material backing and custody layers. Backing and governance flags alone establish no affirmative custody tier.
3. Flag candidate mismatches when you see:

• off-chain or exchange custody
• bridge-heavy collateral
• delta-neutral or structured strategies
• keywords like CEX, Ceffu, Copper, Fireblocks, bridged

4. Research official docs plus one independent source when the architecture is not obvious.

5. Classify using these questions:

• collateralQuality: what is the riskiest significant backing component?
• custodyModel: does evidence cover every material backing part for the claimed tier? Use mixed for evidenced heterogeneous custody or unknown when whole-book coverage is unestablished.

6. Apply sourced labels in the matching per-coin JSON file, keeping the diff minimal. Explicit mixed/unknown may record the reviewed aggregate outcome; omission follows resolveCustodyModel(), not a universal unknown fallback. Retain partial providers, unknown shares and legal safeguards in the reserves sidecar. For a standalone approved edit, after coupled source edits are complete, converge the aggregate and dependent projections with npm run bootstrap:generated and run npm run check:stablecoin-data. In a coordinated addition, retain the whole-book evidence review above and return evidence, changed source paths, and observed versus unexercised checks in the addition/evidence handoff packet; the orchestrator owns the single Phase 7 generation/check pass after all specialists land.

Tiers

The valid values are COLLATERAL_QUALITY_VALUES / CUSTODY_MODEL_VALUES in shared/types/core.ts; read the source file; do not rely on any list quoted elsewhere.

Decision Rules

• For mixed collateral, classify by the riskiest significant component.
• A predominant verified custodian cannot establish an asset-wide institutional label. Require evidence across every material backing and custody layer; partial coverage stays mixed/unknown.
• Do not turn an undisclosed regulator into proven unregulated custody, or a mixed book into all-onchain custody.

Known Pattern Examples

• Delta-neutral strategies can combine institutional, exchange and issuer-controlled wallet exposures. Review every material layer; do not infer a whole-book custody tier from the mechanism name.