yogsoth-ai/de-anthropocentric-research-engine

collect-independent-judgments

Collect judgments/rankings from multiple genuinely independent evaluators or perspectives before exposing any peer outputs, preserving ballots and rationale for later social-choice aggregation.

Vedi sorgente

Come usare

  1. Copia il comando e controlla il pacchetto prima di installarlo nel tuo ambiente agente.
  2. Controlla strumenti, credenziali e ambienti richiesti nel documento originale.
  3. Descrivi input e risultato desiderato; confronta l’output con le fonti.
npx skills add https://github.com/yogsoth-ai/de-anthropocentric-research-engine --skill collect-independent-judgments

Prima di iniziare

  • L’installazione non include abbonamenti, quote API o accesso ai file.
  • Stelle e installazioni indicano attività, non un test indipendente di qualità o sicurezza.
Documento Skill originale

Contenuto dal repository con titoli, esempi, codice, tabelle, link e immagini preservati.

collect-independent-judgments

Purpose

Collect blinded judgments from genuinely independent evaluators or perspectives before exposing peer outputs.

Input contract

yaml
required: [judgment_prompt, evaluator_roster, candidate_set, independence_rules]
optional: [rubric, response_deadline, abstention_policy]
constraints: [evaluators receive no peer judgments; each ballot preserves rationale and provenance]

Procedure

  1. Freeze the candidate set, rubric, prompt, and independence rules before collection.
  2. Obtain one sealed judgment and rationale per evaluator, recording abstentions and unavailable evidence.
  3. Check for shared data, model, prompt, or deliberation leakage before unblinding.
  4. Unblind only after collection, then return ballots, independence flags, and an aggregation-ready dataset.

If independent judgments are complete and their provenance is preserved, consider aggregate-ranking as the next tactic.

Output contract

yaml
produces: [sealed_ballots, independence_audit, rationale_set, aggregation_dataset]
delta_fields: [findings, evidence_updates, uncertainties, decisions, open_questions]

Quality gates

  • A-class gate: declared universe = all commissioned evaluator ballots; numerator = ballots complete, provenance-linked, and leakage-free; batch increment = one sealed ballot; stopping reason = roster exhausted or a declared quorum reached with all exclusions documented; source references = evaluator, prompt, data, and ballot IDs; direction/threshold reason = independence confidence decreases with shared inputs or peer exposure.
  • No aggregation before the leakage check is complete.

Failure and counterexamples

Do not count multiple roles using the same evidence as independent judgments. An abstention is a recorded outcome, not a missing pass.

Provenance map

  • resolved: collective-adjudication

Correlati per attività

dallo stesso repository

Altri Skills

Tutti gli Skills