perceptual-listening-harness

name	perceptual-listening-harness
description	Design and execute blinded perceptual listening studies for LocusQ (trial schema, randomization, metrics, statistical gating, reproducibility artifacts) aligned with BL-060 and BL-061 promotion decisions.

Title: Perceptual Listening Harness Skill Document Type: Skill Author: APC Codex Created Date: 2026-03-01 Last Modified Date: 2026-03-01

Perceptual Listening Harness

Use this skill for controlled listening-test design and evidence-driven promotion decisions.

Scope

• Blind protocol design (2x2 and related variants).
• Trial schema and artifact contracts (CSV/TSV/summary markdown).
• Objective metric extraction (MAE, front/back confusion, externalization summary).
• Statistical gate decisions for backlog promotion.

Workflow

1. Freeze protocol before collecting data.
   • Define participant count, scene count, condition matrix, and randomization policy.
2. Enforce blind run discipline.
   • Hide condition labels during response capture.
3. Capture machine-readable trial outputs.
   • Include true angle, response angle, error, condition, and reaction time.
4. Compute and publish gate metrics.
   • Per-condition MAE/confusion/externalization.
   • Statistical significance or predefined threshold outcome.
5. Publish reproducibility packet.
   • Re-run analysis and ensure stable summary outputs.

Required Evidence

• trial_log.csv
• metrics_summary.tsv
• stats_report.md
• gate_decision.md
• reproducibility_check.tsv

Cross-Skill Routing

• Pair with hrtf-rendering-validation-lab for offline/realtime render parity baselines.
• Pair with spatial-audio-engineering for coordinate/scene semantics and renderer assumptions.
• Pair with skill_docs for runbook/index promotion wording and evidence linkage.

References

• references/trial-schema.md
• references/gate-metrics.md
• references/reproducibility-contract.md

Deliverables

• Protocol summary and rationale.
• Explicit gate decision with evidence paths.
• Validation status: tested, partially tested, or not tested.