# Independent AI pre-review request — Module 4
This is not independent GMP consultant approval, certification or release authorization.

Review ValiStera Academy Module 4: Installation Qualification using the preview linked in draft PR #7, https://github.com/AkashVali02/valistera-website/pull/7.

First establish the exact served source commit from the review-page footer and assets/academy/module-4/review/review-build.json. Compare it with PR head and the media-generation workflow's final saved commit. If identities differ, state the mismatch; do not assume the workflow triggering SHA is the final delivery SHA.

Review all ten lessons (4.1–4.10), forty visuals/descriptions, narration/transcripts/captions, worked examples, workbook/templates, ten lesson checks, twenty assessment questions/keys, sources and rights/privacy records. Consult official references, not a superseded consultation draft. Distinguish regulatory provisions from fictional teaching assumptions and practical recommendations.

Pay particular attention to: IQ versus OQ/PQ boundaries; controlled criteria and actual observations; installed-material traceability; utility quality versus connection identity; calibration applicability and decision-rule limitations; source-data retention and corrections; supplier-test reuse; deviations/change control; activity-specific conditional progression; and the difference between 85% practice scoring and 90% example coverage.

Where tools permit, play every video, inspect scene transitions/captions, click the native timeline at 25% and 75%, confirm one seek bar, operate disclosures/captions by keyboard, test 320/390/768/1280-width layouts and 16/20 versus 17/20 scoring, storage opt-in/clear and media retry. Inspect the blocked removal path (4.3 part 3), signed error (4.6 part 3), and coverage (4.10 part 2) against narration. Do not claim tests based solely on reading another tool's results.

Return one consolidated report: verdict; exact commit/materials reviewed; Critical/Major/Minor findings with location, evidence, impact and proposed correction; independent assessment-key audit; direct tests versus reviewed evidence; untested limitations; consultant and owner release gates. State explicitly if any media could not be heard or images could not be viewed. Never infer GMP approval from an AI review.

R2: begin with CLAUDE-FINDINGS-RESPONSE and PINPOINT-VERIFICATION. Open the public exact-commit evidence link on the review page; require its tested_commit to match the footer and manifest. Historical R1 evidence is not R2 proof. Distinguish direct tests from reading automated evidence.