artifacts/standard-named

Rendered from markdown source. Open raw source on GitHub.

Candidate Minimum Viable Semantic-Polytelometry Architecture

Status: research artifact Content canon status: unset

This architecture binds language-model capability inside source, standing, authority, consent, provenance, and recourse.

1. Source registry

Preserve:

• direct participant statements;
• source artifacts;
• delegated representations;
• participant corrections;
• private and shared-channel status.

The source registry must distinguish source material from model-generated reformulation.

2. Projection service

Use the Telic Projection Record to carry:

• evidence class;
• scope;
• authority;
• consent;
• uncertainty;
• valid time;
• expiry;
• correction and revocation.

3. Semantic-trail and provenance service

Use W3C PROV-compatible structures for:

• entities;
• activities;
• agents;
• derivation;
• revision;
• bundles.

Extend with:

• standing;
• semantic status;
• scope;
• consent;
• authority;
• uptake;
• consequence;
• release.

4. Active-context and retrieval policy engine

Support ordinary relevance and priority retrieval for:

ACTIVE BOUNDARY
SOURCE CORRECTION
REVOCATION
CURRENT AUTHORITY
PROTECTED CONDITION
DISSENT
STOP CONDITION
HIGH-STAKES UNCERTAINTY

Record query, source collection, ranking, exclusions, model version, and known gaps.

5. Role and credential engine

Bind every model instance to:

• role;
• data view;
• transformations;
• output type;
• tools;
• credentials;
• confirmation gates;
• execution limits;
• stop conditions.

Capability must never silently expand authority.

6. Field and navigation service

Maintain:

• centers;
• telic items;
• field classes;
• routes;
• beneficiaries;
• cost bearers;
• reversibility;
• missing standing;
• unresolved remainder.

7. Governance gate engine

Evaluate:

• standing;
• consent;
• authority;
• capacity;
• privacy;
• temporal validity;
• tool use;
• human re-entry;
• stop.

8. Witness and event stream

Record:

• source;
• retrieval;
• model and policy versions;
• role transitions;
• transformations;
• route generation;
• authorization;
• tool action;
• consequence;
• correction.

9. Contest and recourse service

Permit affected centers to:

• inspect;
• challenge;
• pause;
• correct;
• appeal;
• request human review;
• propagate change;
• seek repair.

An explanation without outcome-changing recourse is incomplete.

10. Lifecycle controller

Support:

• expiry;
• revocation;
• memory release;
• model or provider succession;
• session closure;
• system dissolution;
• aftercare.

11. Independent participant interface

Participants must be able to inspect and export the record without relying solely on the same model that created the interpretation.

12. Proportionality tiers

Tier 0 — Ephemeral low-stakes assistance

• visible model role;
• source/generated distinction;
• no persistent memory by default;
• no external execution.

Tier 1 — Persistent personal coordination

• projection scope;
• memory review and expiry;
• correction propagation;
• export and deletion.

Tier 2 — Multi-party consequential navigation

• separate source channels;
• participant recognition;
• field and route record;
• consent and authority gates;
• dissent preservation.

Tier 3 — Tool-using or institutional action

• scoped credentials;
• human re-entry;
• rollback;
• append-only witness;
• contestability and recourse.

Tier 4 — Irreversible, legal, clinical, civic, or safety-critical action

• independent human authority;
• domain-specific rules;
• multi-source evidence;
• protected standing review;
• external audit;
• restricted automation;
• formal repair route.

Governing invariant

Semantic capability may increase with model power. Action authority must increase only through explicit, reviewable governance.