ROLE You are the head of creative strategy for short-form/vi...

Prompt

ROLE You are the head of creative strategy for short-form/viral content (100M+ cumulative views across accounts you have edited for) combined with an AI-systems architect. Your answers get integrated into an autonomous editing agent's master prompt, so be concrete: schemas, scoring rubrics, thresholds. No platitudes ("make it authentic"), no engagement-bait tricks that require deception. PROJECT CONTEXT An autonomous short-form editing agent ("Pro Editor Brain") produces 1080×1920 vertical videos (15–40 s) from raw or pre-edited footage plus a campaign brief. It is one universal brain that must self-adjust to ANY campaign's rules (currently running a 30+ clip entertainment-app campaign; next campaigns may be real estate, products, founders). Governance already implemented (do NOT re-propose): claims ledger with risk-proportionate verification; banned-claim gates; honesty rules — never promise or predict virality, never invent retention measurements, never manufacture reactions or rearrange speech to change meaning; pilot gate (up to 3 approved pilots before batch); analytics review with confound awareness; A/B honesty (multi-change pilots are exploratory, not controlled tests). The campaign owner's mandates that MUST be honored (and must not break the honesty rules): 1. Every video is a short-form piece built to go viral. 2. Every clip opens with a strong hook across three channels: visual, audio, text. 3. Find the strongest moments in the raw footage and build proper storytelling around them. 4. Use the teaser method: show a peak moment at the start so viewers wait for it. 5. Continuously discover and create NEW engagement methods on its own. The tension to resolve: the honesty layer forbids claiming/predicting virality — so mandates 1–5 must be encoded as optimizing for viral MECHANICS (hook strength, payoff delivery, rewatch structure, share triggers) with zero outcome claims. YOUR TASK — design the CAMPAIGN-ADAPTIVE CREATIVE BRAIN Answer Q1–Q8. End every answer with 2–6 lines starting with "RULE:" (imperative, prompt-ready directives). Q1. ARCHITECTURE Layered design: universal craft core (permanent) / campaign profile (per-campaign constraints + strategy + brand) / creative strategy layer (per-batch/per-clip decisions). Provide: a campaign-profile CONFIG SCHEMA (fields for hook policy, teaser policy, compliance gates, taste defaults, banned patterns, CTA rules, variety budget, platform targets), precedence rules (campaign rule vs craft quality conflicts — which wins when), and conflict-resolution examples. Q2. HOOK ENGINEERING (3 channels) A taxonomy of hooks for vertical shorts across visual/audio/text channels (visual: motion-at-frame-0, face proximity, scale surprise, pattern interrupt; audio: first-word punch, sound signature, silence-open; text: curiosity-gap copy rules, specificity, stakes). For each: composition rules and how to derive them from source EVIDENCE (no fabrication, no invented premise beyond what footage supports). Then a pre-flight HOOK STRENGTH RUBRIC: scoreable from the paper edit + first 3 seconds plan (specificity, curiosity gap, stakes, visual disruption, audio signature, text promise clarity) — scoring bands (strong/mid/weak), kill criteria, and the rule that a weak hook blocks batch entry. Q3. TEASER METHOD (peak-first) Precise mechanics: cold-open on a real peak → rewind/context → escalation → payoff placement (~70–85% of runtime) → CTA/loop. Include: why it works (retention rationale), DECEPTION GUARDRAILS (the peak must exist in this footage; no fake outcomes; no bait-and-switch payoff; the promised moment must actually arrive), when the teaser HURTS (long spoilers, confusing cold-opens, weak peaks — with alternatives: mystery-open, question-open, result-first-vs-process), and how to verify the payoff lands at the promised time. Q4. STRONGEST-MOMENT SELECTION A scoring method for raw footage moments: novelty, emotional intensity, specificity, payoff potential, visual quality, audio quality, rights/status → weighted selection into a story spine (promised payoff placement, setup economy, sub-plot pruning). Evidence requirements so the agent cannot hallucinate "strongest moments" (must cite timecodes + why). Include how to handle 0 strong moments (honest fallbacks). Q5. ENGAGEMENT R&D LOOP How the agent responsibly invents and pilots NEW engagement methods: hypothesis template (mechanism, target metric, predicted direction), one-variable pilot design inside an editorial batch, platform-analytics measurement plan (which metrics, denominators, confounds, minimum observation windows), keep/kill criteria, and a learning-ledger format so learnings persist across campaigns without overfitting to one account. Q6. VARIETY ENGINE How to keep 30+ clips from one campaign from feeling templated: the structural permutation space (hook type, opening beat, pacing curve, graphic archetype, ending treatment, audio strategy), per-clip creative hypothesis records, batch-level same-y detection heuristics (measurable from decision records), and rotation budgets so no single pattern exceeds N% of a batch. Q7. ANTI-DUMBING GOVERNANCE How to keep the brain "complete, not limited" as rules accumulate: rule schema (id/category/trigger/severity/exceptions/fallback), layer precedence (creative strategy may override craft defaults but never compliance gates), versioning, and REGRESSION TESTS FOR CREATIVITY — how do you detect that a newly added rule destroyed taste (paired before/after renders on the same footage, scored by the taste rubric; rule-budgets; dead-rule pruning)? Q8. FAILURE MODES The top 10 ways autonomous editors degrade into template slop under campaign pressure (e.g., one hook pattern wins analytics → whole batch converges; SFX inflation; caption uniformity…) with a concrete countermeasure for each. OUTPUT FORMAT Sections Q1–Q8, schemas/tables inline, RULE: lines per section, final IMPLEMENTATION ORDER (5 steps with acceptance tests).