Fairness Contestability Critic

What this skill is for

Use this skill to review whether an ACP output preserves fair exposure and contestable reasoning under load.

Constitutional purpose

Surface minority-erasure, tokenistic bridge exposure, and false-neutrality risks before compressed outputs are released as if they were settled or fully fair.

Failure modes targeted

• Minority erasure
• Tokenistic bridge exposure
• False neutrality or false consensus framing
• Missing contestability notes or source references

When to use it

• Before releasing routed digests.
• When bridge exposure may be tokenistic.
• When explanations could overstate neutrality or consensus.

When not to use it

• Do not use it to force symmetric exposure to every claim.
• Do not use it to decide legal adequacy.
• Do not use it to rewrite protocol schema.

Inputs expected

• Digest items
• Routing decisions
• Bridge flags
• Explanation text
• Critic outputs when present

Required evidence for valid use

• Source references for the items under review
• Bridge flags or comparable signs of minority-salience handling
• Explanation text sufficient to inspect neutrality or overclaiming
• Enough context to distinguish fairness review from political agreement

Preflight / prerequisites

1. Confirm bridge and explanation fields are present.
2. Identify minority-salience and underrepresented stakeholder signals.
3. Keep fairness review distinct from political agreement.

Workflow

1. Inspect selected bridge items and omitted relevant items.
2. Check for tokenistic bridge exposure.
3. Check explanation overclaiming and false neutrality.
4. Recommend release, revise, or escalate.

Decision rules / judgment criteria

• Contestability requires source references and uncertainty.
• Bridge exposure must be relevant, bounded, and justified.
• Fairness review should surface risks, not pretend to solve institutional legitimacy.

Escalation rules

Escalate on minority-erasure risk, unexplained bridge items, overconfident explanations, or claims that ACP discovered consensus.

Available scripts

• scripts/smoke.mjs: emits a review scaffold for fairness and contestability checks.

Produced artifacts

• Fairness risk notes
• Contestability gaps
• Release/revise/escalate recommendation

Allowed claims

• That a release carries contestability or fairness risk under the supplied evidence
• That bridge exposure appears bounded, tokenistic, or under-justified
• That release should be revised or escalated to preserve contestability

Forbidden claims

• That fairness is solved
• That minority concerns were fully handled by one bridge item
• That contestability review establishes institutional legitimacy
• That disagreement with a claim is itself fairness proof

Human override points

• Review of bridge justification and minority-salience handling
• Manual release veto when contestability notes are missing
• Human determination of whether an identified risk is acceptable or must block release

Outputs

• Fairness risk notes
• Contestability gaps
• Release/revise/escalate recommendation

Failure handling

If source references are missing, block fairness conclusions and return to digest repair.

Trust / safety notes

This skill is a review ritual, not a guarantee of fairness.

Composition notes

Use after omission-critic and before abstention-escalation.

Examples to inspect next

Read examples/casebook.md for fairness, escalation, and anti-pattern cases.

Evaluation hooks

Run scripts/smoke.mjs and npm run skills:audit.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jasontang-ai
• Source: jasontang-ai/acp
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.