The Machine-Readable Stance

The whole methodology, as data. This file is the canonical otherai:Stance — every principle, refusal, lens, and diagnostic on this site, in a form machines can load and critics can diff.

Why a stance file

The fifth principle requires the method to be published machine-readably, and the reasons are practical, not ceremonial:

• Loadable — developers parameterise audit prompts directly from the file rather than transcribing a webpage.
• Diffable — when the methodology changes, the change is a visible diff in version control, not a silent edit to prose. Accountability for revisions is part of the method.
• Forkable — an organisation adopts the practice by publishing its own stance file; divergence from the canonical file is legitimate and visible, which is what makes adoption checkable. A published stance is a commitment your users can audit you against.
• Comparable — two stance files can be compared mechanically: which refusals were dropped, which lenses added. The ecosystem of forks becomes legible.

The canonical file

Also available raw: stance.json

{
  "@context": "https://schema.org",
  "@type": "CreativeWork",
  "name": "the OtherAI: Preservative AI Stance",
  "version": "1.1",
  "license": "CC BY-SA 4.0",
  "keywords": ["preservative-ai", "epistemic-audit", "omission-detection",
               "suffixscape", "accountability"],
  "principles": ["preserve-provenance", "audit-framing", "surface-omissions",
                 "include-plural-knowledge", "publish-methodology"],
  "refusals": ["no-synthetic-personas", "no-invented-events",
               "no-nonconsented-training", "no-engagement-optimization",
               "no-adversarial-positioning"],
  "lenses": ["indigenous_knowledge", "deep_history", "cross_cultural_wisdom",
             "scientific_evidence", "artistic_perception", "future_modelling",
             "marginalised_voices", "trickster_knowledge"],
  "diagnostics": ["nominalised_evasion", "agency_diffusion",
                  "epistemic_inflation", "temporal_flatness"],
  "audit-endpoint": "/api/v1/audit?lens=all&suffixscape=true",
  "lastUpdated": "2026-06-12"
}

Field reference

principles

The five operational principles, as stable keys. Constant across forks — they define what "Preservative AI" means; a file without all five is a different practice.

refusals

The five refusals. Forks declare which they commit to; omitting one is permitted and public.

lenses

The eight epistemic lenses this stance audits with — including the reflexive trickster_knowledge lens, which audits the stance-holder's own analyses. Forks may add regional or domain lenses; each added key should resolve to a published lens definition.

diagnostics

The Suffixscape patterns in force. Language-specific forks are expected to replace these — English grammar diagnostics do not translate, as the limitations page records.

audit-endpoint

Where outputs can be programmatically audited against this stance. The canonical shape: /api/v1/audit?lens=all&suffixscape=true.

version / lastUpdated

Stance files are versioned artefacts; consumers should pin and diff. History lives in this site's public git record.

Publishing your own

Use the generator for a starting draft, then: place the file at a stable URL (/.well-known/otherai-stance.json is the suggested convention), commit it to version control, link it from your documentation, and — the step that gives it meaning — make your outputs available for audit against it. The licence is CC BY-SA 4.0: fork freely, share alike, attribute the lineage.