jasontang-ai What this skill is for
Use this skill for the flagship ACP domain scenario: turning overloaded public-hearing input into inspectable issue maps, routed briefings, critic evidence, and claim-safe artifacts.
Constitutional purpose
Demonstrate ACP's strongest civic denominator: overloaded public input routed into inspectable, claim-safe decision-support artifacts without pretending that benchmark evidence equals public legitimacy.
Failure modes targeted
- Hearing overload collapsing into unreadable chronological accumulation
- Minority-signal burial under dominant testimony repetition
- Benchmark overclaiming about civic efficacy or legitimacy
- Missing scale-boundary discipline
When to use it
- Preparing Foresight or TAIGR-quality evidence.
- Running
public-hearing-triagebenchmarks. - Demonstrating issue coverage, minority preservation, and scale-band behavior.
When not to use it
- Do not use it for generic civic marketing.
- Do not claim public opinion measurement.
- Do not claim legal adequacy for hearings.
Inputs expected
- Public-hearing scenario or benchmark class
- Baseline and intervention condition requirements
- Desired output directory
- Claim boundary for review
Required evidence for valid use
- Comparable baseline and intervention conditions
- Benchmark outputs or rehearsal artifacts, not only domain rhetoric
- Explicit claim boundary when packaging or summarizing results
- Scale-band context when any scalability language is requested
Preflight / prerequisites
- Confirm
public-hearing-triageremains the frozen denominator. - Confirm baseline and intervention conditions are comparable.
- Confirm whether scale Band B or Band C evidence is required.
- Read limitations before drafting claims.
Workflow
- Run benchmark comparison.
- Run Engine V2 ablation when research evidence is needed.
- Run scale-band evidence when making scalability-path claims.
- Inspect issue maps, critic output, and limitations.
- Package artifacts with reviewer-facing notes.
Decision rules / judgment criteria
- Preserve disagreement and unresolved questions.
- Treat issue coverage as a review signal, not a legitimacy guarantee.
- Benchmark evidence supports system behavior, not live civic efficacy.
- S4 quality requires casebook, benchmark, operator, and adopter relevance.
Escalation rules
Escalate when baseline/intervention counts diverge, scale-band evidence is missing, critic outputs are absent, or requested claims exceed benchmark evidence.
Available scripts
scripts/run-public-hearing-triage.mjs: runs or points to the flagship benchmark and package surfaces.
Produced artifacts
- Benchmark summary
- Ablation summary
- Scale-band summary
- Claim-safe domain notes
Allowed claims
- That the benchmark surfaces inspectable differences between baseline and intervention conditions
- That the output is appropriate as decision-support or review infrastructure
- That scale-band or ablation evidence remains bounded to what was actually run
Forbidden claims
- That benchmark success proves real-world civic efficacy
- That public-hearing triage measures public opinion or legal adequacy
- That scale-band evidence proves institutional deployment readiness
- That one benchmark resolves fairness or legitimacy concerns
Human override points
- Claim approval before any public-facing scale or legitimacy language
- Manual review when benchmark and critic evidence conflict
- Maintainer approval before changing the frozen flagship denominator
Outputs
- Benchmark summary
- Ablation summary
- Scale-band summary
- Claim-safe domain notes
Failure handling
If any benchmark command fails, preserve the output and do not rewrite the denominator to fit the result.
Trust / safety notes
Public-hearing triage is institutionally sensitive. It must be framed as decision-support and review infrastructure, not sovereign mediation.
Composition notes
Use inside public-hearing-end-to-end and foresight-submission-prep.
Examples to inspect next
Read examples/casebook.md for flagship public-hearing cases and anti-claims.
Evaluation hooks
Run scripts/run-public-hearing-triage.mjs, npm run benchmark:compare -- --class public-hearing-triage, and npm run skills:audit.