Adoption Guide: how to use, adapt, or critique

Five audiences, each with concrete first steps that work today. Adoption does not require permission, partnership, or even agreement — critique is an adoption path.

1. For developers

• Load stance files into system prompts. The canonical stance JSON is designed to be ingested: its principles, refusals, lenses, and diagnostics keys map directly onto audit instructions. Start from the prompt template and parameterise it from the file.
• Add ?audit=true parameters to APIs. Any endpoint that returns generated or summarised content can offer a parallel audited mode: same response, plus framing analysis, omission tags, and Suffixscape results (the stance file's declared endpoint shape is /api/v1/audit?lens=all&suffixscape=true).
• Replace accuracy-only metrics with CMR scoring. Wire completeness / multiplicity / responsibility into your evaluation harness alongside whatever you already measure; the CMR gauge defines the axes.
• Publish limitation ledgers quarterly. A versioned, dated, public record of what your system structurally omits — the engineering counterpart of this site's limitations page.

2. For researchers

• Fork the methodology for regional adaptation. The licence (CC BY-SA 4.0) and the machine-readable stance exist so forks need no negotiation. Regional forks have authority over their lenses, not just their translations.
• Publish critiques with source links. A critique that anchors to the specific text it contests is itself a Preservative AI artefact; this site's versioning makes every claim addressable.
• Propose new lenses or diagnostic patterns. The eight lenses and four diagnostics are a version, not a canon. A proposal needs: the question the lens asks, the class of absence it detects that existing lenses miss, and a scoring rubric.
• Co-author audits with marginalised scholars — as named, compensated co-authors with editorial authority, per the governance model, not as consulted acknowledgees.

3. For journalists & educators

• Use Suffixscape on corporate and government communications. The four diagnostics were built for exactly this material; the live demo needs nothing but a paste. "Who determined?" is a question, and questions are the trade.
• Preserve original sources; flag synthetic summarisation. Link primaries prominently; when AI-generated synthesis appears in the workflow, label it and keep the path back to the original intact (Principle 1 as newsroom policy).
• Teach students to audit framing, not just consume content. The classroom version of the practice: give students the auditor prompt and a week of press releases. Framing literacy — seeing the construction, not just the claim — is the durable skill; the tools merely scale it.

4. For communities & advocates

• Submit omission reports via public channels. An omission report — "this coverage/system/dataset consistently lacks our perspective, here is the evidence" — enters the audit record with standing. You hold exactly the knowledge the Marginalised Voices lens cannot generate from outside.
• Claim right-of-reply for audited entities. If you or your community are the subject of an audit, the reply channel is yours by design, and the reply attaches permanently.
• Adapt lenses for local epistemic contexts. The lens questions are starting points; your context may need different ones, and the fork is yours to make without asking.
• Demand machine-readable stance files from AI providers. "Publish your stance file" is a concrete, checkable demand — unlike "be ethical", it has a deliverable, a format, and a public way to verify compliance against output.

5. For AI labs

• Publish otherai:Stance files. Declare your principles, refusals, lens coverage, and audit endpoint in the open format — the generator produces a starting draft. A stance file is a commitment your users can audit you against; that is its cost, and its entire value.
• Enable audit endpoints. Allow outputs to be programmatically audited — yours and third parties' — rather than requiring screenshot forensics.
• Compensate community reviewers. Structural inclusion of marginalised perspectives is skilled labour. Unpaid "feedback opportunities" reproduce the extraction under audit.
• Treat omission detection as a core safety metric. The frontier framing, stated plainly: a system that reliably erases perspectives is unsafe in a way no current red-team finds. Put omission rates on the model card next to refusal rates.

Adoption is forkable too

These five paths are the ones the framework's authors could see — which, per its own positionality limitation, means the list has gaps. Archivists, librarians, lawyers, clinicians, local governments: the on-ramp is the same three questions everywhere (anchor or replace? audit or amplify? surface or smooth?), and the Invitation is open.