# Appellate Preference Model Use this reference when comparing a draft appellate document to public Court of Appeal decisions. ## What Counts as a Preference Signal A preference signal is a pattern visible in public judgments or orders. It is not private psychology. Acceptable: - how a judge structures reasons in public judgments, - how a case type is usually framed, - whether recent decisions emphasise text, statutory structure, commercial consequence, discretion, or appellate restraint, - treatment of authority hierarchy, - tolerance or criticism of overlong factual narrative where visible in public reasons. Prohibited: - likely vote, - temperament, - ideology, - private preferences, - personal motivation, - claims from gossip, directories, or model memory. ## Scope Labels Use: - `court_general` - supported across a general Court of Appeal sample. - `case_type_specific` - supported in similar subject matter or procedural posture. - `division_context` - civil, criminal, family, or specialist context. - `judge_specific` - supported by multiple public decisions involving the named judge. - `single_source_lead` - one useful public source; do not generalise. ## Corpus Selection Record: - sources searched, - queries, - date checked, - reason for inclusion, - reason for exclusion, - court level, - date range, - subject-matter fit, - author/panel where available. For judge-specific observations, aim for multiple relevant public decisions. If only one is available, use `single_source_lead`. ## Draft Assessment Dimensions Assess: - issue framing, - appellate error identification, - ground-by-ground structure, - fact-to-law ratio, - authority hierarchy, - statutory or contractual text anchoring, - commercial pragmatism versus black-letter analysis, - tone/restraint, - treatment of adverse points, - remedy/disposition. ## Argument Dependency For key grounds, identify: - proposition needed, - authority relied on, - citation verification status, - proposition support status if checked, - risk if authority/proposition fails. Do not perform citation/proposition checking unless sources are available and the user asks or the skill invokes that workflow separately. ## Output Discipline Every preference finding needs: - dimension, - observation, - scope, - confidence, - supporting decisions, - paragraph references, - short excerpts, - source URLs, - uncertainty/corpus limits. No-source mode may review internal structure only. It must not make source-backed preference findings.