artifacts/standard-named

Rendered from markdown source. Open raw source on GitHub.

Consentful Training — Authority and Boundary Matrix

Status: draft Content canon status: unset

| Concept | Decision | Boundary | |---|---|---| | Consentfully trained model | Retain provisionally | Qualified profile, never a binary moral badge | | Training recruitment | Retain | Semantic, human, material, and institutional contributions entering an outer loop | | Training authority | Adopt | Broader than consent; includes license, contract, public mandate, community governance, and other bases | | Source standing | Retain | Interests materially attached to source contribution or representation | | Dataset provenance | Adopt existing foundation | Does not by itself establish permission or legitimacy | | Preference provenance | Adopt | Population, task, policy, alternatives, disagreement, aggregation, and adjudication | | Model constitution | Retain | Broader operational object; distinct from Constitutional AI | | Restriction propagation | Adopt | Restrictions and duties must survive aggregation, training, and derivation | | Withdrawal ladder | Adopt | Collection stop, deletion, future exclusion, retrieval block, unlearning, retraining, retirement | | Unlearning verification | Use established term | Claim only the removal evidence actually demonstrated | | Synthetic-data lineage | Adopt | Generator, source, policy, filtering, proportion, and recursive depth | | Benefit governance | Adopt | Payment alone does not purchase unlimited authority | | Community governance | Retain | Requires legitimate representation and internal minority protection | | Model succession | Retain | Capability transfers with restrictions and obligations | | Training/operation separation | Foundational invariant | Neither profile substitutes for the other |

Governing distinctions

accessible ≠ permitted
permitted ≠ consented
consented ≠ universally legitimate
documented ≠ authorized
public ≠ ownerless
human feedback ≠ human consent
preference aggregation ≠ shared human values
source deletion ≠ model forgetting
unlearning attempt ≠ verified removal
synthetic ≠ lineage-free
open ≠ consentful
closed ≠ protective

Training authority classes

INDIVIDUAL_CONSENT
COLLECTIVE_AUTHORITY
COMMUNITY_GOVERNANCE
CONTRACT
LICENSE
PUBLIC_DOMAIN
PUBLIC_MANDATE
RESEARCH_ETHICS_AUTHORIZATION
FIDUCIARY_STEWARDSHIP
STATUTORY_AUTHORITY
OTHER_LAWFUL_BASIS
UNKNOWN
CONTESTED
EXPIRED
WITHDRAWN

Hard boundary

Consentful operation does not retroactively make unconsented training consentful. Consentful training does not guarantee consentful deployment.