artifacts/standard-named

Rendered from markdown source. Open raw source on GitHub.

Model Role, Output Class, and Authority Matrix

Status: candidate pre-production Content canon status: unset

| Model role | Typical outputs | May do | Must not infer | |---|---|---|---| | Extractor | Terms, entities, claims, boundaries | Identify candidate distinctions | That extraction is complete or authoritative | | Structurer | Field map, issue graph, dependency map | Organize reviewed source material | Standing, consent, or public mandate | | Translator | Cross-vocabulary rendering | Offer parallel phrasing | That translation preserves every implication | | Comparator | Difference, tradeoff, uncertainty | Compare under disclosed methods | Which method should govern | | Retriever | Source and context candidates | Return authorized traces | That retrieved content may govern action | | Route generator | Candidate options and portfolios | Expand the option set | Participant authorship or adoption | | Challenger | Counterargument, missing-context prompt | Test assumptions and premature consensus | Adjudicative authority | | Mediator | Turn structure, clarification, bounded summaries | Support process under assigned rules | Ownership of participant positions | | Witness assistant | Draft event and decision records | Prepare reviewable witness artifacts | That a draft is the final witness | | Executor | Tool plan and tool result | Execute only through a passed action gate | Consent, authority, or scope from recommendation alone |

Output-class boundary

extraction
structure
translation
comparison
retrieval result
summary
inference
generated option
generated argument
challenge
clarification question
route recommendation
tool plan
tool result
witness draft

Every material output should preserve:

source inputs
source status
transformations
uncertainty
standing effect: none
authority effect: none
consent effect: none
correction route

Hard distinctions

capability ≠ role
role ≠ permission
permission ≠ authority
authority ≠ legitimacy
summary ≠ source
inference ≠ consent
recommendation ≠ authorization
tool result ≠ interpretation

Governing principle

The model may perform an operation without becoming the authority whose field gives the operation legitimacy.