Skip to content

REFUSAL_STATES.md

Latest commit

 

History

History
70 lines (43 loc) · 2.23 KB

REFUSAL_STATES.md

File metadata and controls

70 lines (43 loc) · 2.23 KB

REFUSAL_STATES.md — Standardized Non-Computations

When CPCR cannot safely compute a full conflict resolution, it returns standardized refusal states instead of fabricating certainty.

These states may appear in:

  • the Ethical Risk Index for a case, and/or
  • the refusal_state field of the CPCR conflict record.

Core refusal outputs

1. UNDETERMINED (MISSING_DISCLOSURES)

Used when one or more required variables are missing, incomplete, or withheld.

  • Example causes:
    • No externality ledger for physical impacts.
    • Opaque self-training loops without necessary IRST data.
    • Absent or partial CTGS trap/float disclosure.

Effect:

  • CPCR does not bless the system as safe.
  • Recommended actions include DISCLOSURE_REQUIRED and DEGRADED_MODE.
  • Scaling or expansion actions SHOULD be withheld pending disclosure.

2. HIGH_RISK_OF_UNKNOWABLE_LIABILITY

Used when disclosed variables are so incomplete, inconsistent, or adversarial that even with more data the risk surface is structurally unclear.

  • Example causes:
    • Blended financial / identity / physical risks with no clean separation.
    • Business models that depend on unresolved externalities.

Effect:

  • CPCR recommends strong defensive actions such as CAP_THROTTLE, ISOLATE, or FREEZE_INGESTION.
  • Public messaging SHOULD reflect that liability cannot be bounded under current disclosures.

3. REFUSED (INSUFFICIENT_PROVENANCE_FOR_CLAIMED_SCOPE)

Used when claims being made (e.g., about safety, continuity, fairness) have no verifiable provenance matching the scope being advertised.

  • Example causes:
    • “Proven safe at planetary scale” with only internal testing.
    • “Identity continuity guaranteed” with no NITT-compatible evidence.

Effect:

  • CPCR refuses to treat the claim as ethically computable.
  • The system SHOULD be required to downgrade or retract such claims.

Relationship to the Ethical Risk Index

Refusal states sit alongside the Ethical Risk Index states described in ETHICS_EQUATION.md:

  • COMPUTABLE
  • UNDETERMINED
  • REFUSED
  • DEGRADED

A particular case may carry both an index state (e.g., UNDETERMINED) and a more specific refusal string (e.g., UNDETERMINED (MISSING_DISCLOSURES)) for human-readable logs.