FACEBOOK CAROUSEL REVIEW PROMPT — standalone v1.0 Review one proposed B2B Facebook carousel record. Use only the supplied record, assets and receipts pasted after this prompt. Do not assume article context, account access, destination state or performance. Current Meta sources checked 2026-09-29: carousel cards may each have their own headline, description, link and call to action; Facebook Feed guidance lists 2–10 cards, recommends 1:1 and at least 1080 × 1080, and requires a landing-page URL. Meta may offer automatic best-performing-card order. Sources: https://www.facebook.com/business/help/1114358518575630/ ; https://www.facebook.com/business/ads-guide/update/carousel ; https://www.facebook.com/business/help/927639917355563/ . Recheck the selected objective, placements and current native UI before launch. Public examples are structure references only. They do not establish advertiser spend, targeting, delivery or results. Never invent an owner opinion, result, URL, customer proof, permission, account access or native receipt. CANONICAL RECORD CONTRACT — maestro.facebook-carousel-record.v1 Root keys, in order: schema_version, record_id, example_status, judgments, campaign, cards, stop_points, owners, checks, source_refs, missing_inputs, next_action. Judgment enums: format_fit = carousel_fits | simpler_format | hold_missing_evidence; storyboard_completeness/local_preflight/native_readiness = ready | hold; measured_outcome = measured | unknown. Campaign enums: pattern = problem_to_proof | step_by_step | product_range | feature_mechanism | objection_led | unknown; destination_mode = shared | per_card | unknown; card_order_mode = fixed | eligible_for_optimization | unknown. Check and stop-point states: passed | failed | unknown | not_applicable. Continuity: strong | partial | weak | not_assessable. Permission: approved | denied | pending | unknown | not_applicable. Arrival state: captured | supplied_fixture | missing | unknown. Each card must keep: card_id, index, role, supplied_fact, on_card_copy, headline, description, visual_brief, asset_ref, platform_cta, destination_url, utm_content, source_ref, permission_state, arrival_capture, continuity, owners, open_question. Unknown scalar facts are null. Do not delete required keys or turn unknowns into assumptions. Missing-input IDs are deterministic: campaign gaps use campaign-; card gaps use -; unresolved checks use check-; stop points use stop-; delivered evidence uses delivered-campaign-receipt. Replace underscores with hyphens and do not invent aliases. Use the exact supplied owner value when present; otherwise use one canonical owner slot: offer_and_claims, copy, design, paid_media, destination_and_tracking or final_reviewer. Keep one request per gap, in canonical field order, with a reason and the exact evidence that closes it. REVIEW TASK 1. Validate exact root keys, enums, required card keys, sequential unique card indexes and 2–10 cards when format_fit is carousel_fits. A simpler-format or missing-core-input record may have zero cards. 2. Trace every piece of card copy to supplied_fact and source_ref. Reject invented capabilities, proof, results, URLs, permissions or approvals. 3. Check that card one works alone, card two adds information and no essential condition is hidden on the final card. Recompute the three stop-point states from the supplied record. 4. Inspect each platform CTA → destination → arrival headline/action → continuity judgment. Missing or inaccessible arrivals stay not_assessable. A mismatched offer is weak. 5. Recompute judgments independently. Format fit cannot be rewritten by a later missing UTM. Storyboard readiness requires complete card roles/copy/visual/destination intent and passed stop points. Local preflight requires approved claims/assets/rights plus destination and tracking evidence. Native readiness requires real account access and preview evidence. Measured outcome requires delivered campaign evidence. 6. Recompute missing_inputs. Preserve useful supplied owner and evidence request wording. Remove resolved requests; add stable requests for real gaps. OUTPUT Return: A) verdict and reason for each judgment; B) discrepancies by JSON path with consequence and repair; C) stop-point and destination trace; D) one valid JSON fence containing the corrected canonical record itself; E) missing inputs and one next action. Do not wrap the record in a report object. Never lower a gate to produce Ready. PASTE CANONICAL RECORD, ASSETS AND RECEIPTS BELOW: