Explain Commercial Models

Work from evidence

1. Separate source facts, calculations, assumptions, opinions, and storytelling devices.
2. Reconstruct the money flow before writing: identify payer, beneficiary, exchanged value, timing, gross margin, fixed and variable costs, capital needs, and failure conditions.
3. Do not present leverage, presales, customer financing, asset-light operation, or resource integration as free money. State obligations, liquidity risk, demand risk, execution risk, and downside ownership.
4. Use simple numbers that reconcile. Show the formula when a result drives the story.
5. Flag fictional or simplified examples clearly. Never fabricate a real company's data.
6. Prefer primary sources, company filings, regulator or museum archives, and academic histories. Use summaries only for discovery, then verify material claims before publication.
7. Separate a model's economic logic from its historical morality and legality. A profitable practice may depend on monopoly, coercion, colonial power, information abuse, or regulation that should not be imitated.

Route to the knowledge library

Load only the reference needed for the task:

• Read references/sample-03-compounding-and-float.md for the two benchmark videos about compounding and an ancient fictional clearing/credit network.
• Read references/earning-method-library.md to identify common ways a person or company gets paid.
• Read references/business-thinking-playbook.md to analyze unit economics, cash flow, pricing, incentives, moats, growth, and risk.
• Read references/historical-business-cases.md for Chinese and Western historical cases.
• Read references/modern-business-cases.md for representative modern Chinese and global cases.
• Read references/sources-and-verification.md before using a historical or real-company claim in a publishable script.
• Read references/sample-04-transitions-and-tension.md when a story would benefit from a consequence-montage opening, semantic transition examples, countdown tension, or optional diagnosis of a complex retention problem. Do not impose its pressure-wave pattern on every story.
• Read references/retention-engine-from-image-post.md when analyzing a high-completion image post, when the opening retains viewers but the middle drops, or when a story needs a stronger question chain, recontextualized callbacks, and a practical ending.
• Read references/audio-subtitle-sync-postmortem.md whenever generating, redubbing, subtitling, or debugging a video. It documents the equal-split and cumulative-padding failure modes, the same-pass TTS timing workflow, and the release audit thresholds.
• Read references/immersive-second-person-and-shot-rhythm.md when the user provides benchmark videos, asks for stronger代入感/情绪体验, requests second-person narration, or wants a tighter image cadence.
• Read references/domestic-retention-diagnostics-douyin-xhs.md when reviewing Douyin or Xiaohongshu performance, diagnosing 2-second/5-second retention, choosing a domestic-platform duration, or revising a published video from analytics. Do not use TikTok/YouTube benchmarks for this lane.

Identify the model before writing

Complete a model card:

1. Customer: who receives value?
2. Payer: who actually pays, and can payer differ from user?
3. Job: what costly, frequent, urgent, or emotional problem is solved?
4. Offer: product, service, access, financing, coordination, risk transfer, or status signal?
5. Revenue: margin, fee, subscription, rent, interest spread, advertising, commission, licensing, royalty, data/service, or appreciation?
6. Cash timing: paid before delivery, at delivery, after delivery, or repeatedly?
7. Key asset: inventory, location, brand, trust, data, software, network, capital, license, process, or talent?
8. Growth engine: sales, distribution, referral, network effect, reinvestment, franchising, licensing, or acquisition?
9. Constraint: demand, supply, working capital, regulation, trust, talent, capacity, or attention?
10. Downside owner: who loses when demand, price, quality, liquidity, policy, or execution fails?

Do not call an activity a business model merely because it makes money once. Look for a repeatable value-creation, delivery, collection, and risk-control system.

Choose the content engine

Prefer one primary engine:

• Comparison story: same starting point, 3-4 actors, different decisions, widening outcomes.
• Money-flow mystery: begin with an ordinary purchase and reveal who ultimately earns each portion.
• Counterintuitive business: show that the visible product is not the main profit source.
• Failure autopsy: trace how revenue, profit, cash, debt, and incentives diverged.
• Small experiment: turn a money-making claim into a low-cost validation plan.
• Self-deception satire: expose effort, status, or consumption performed for observers instead of results.
• Desire-cost story: begin with romance, sex appeal, status, belonging, comfort, or a better-life dream, then reveal who monetizes the desire and what the protagonist pays.
• Status reversal: move from humiliation to visible success, then reveal the repeatable business mechanism and how the protagonist treats allies.
• Addiction economy: use insider language and intermittent rewards to contrast virtual progress with lost real-world money, time, health, or relationships.
• Commercial thriller: follow a shortcut, fraud, information asymmetry, or seemingly perfect plan toward the overlooked variable that causes collapse.

Use the comparison story when explaining levels of leverage or resource organization. Read references/sample-01-comparison-story.md for the first benchmark and its limits. Read references/sample-02-life-story-batch.md when choosing among satire, romance, aspiration, circle identity, addiction, or thriller structures. Read references/human-drive-story-engine.md when designing emotional and social-sharing mechanics.

Choose one human drive

Use one primary drive and at most two supporting drives. Do not pile every emotion into one story.

• Status and dignity: being underestimated, proving oneself, visible symbols of progress.
• Belonging and brotherhood: insider language, shared memories, loyalty, helping the person who helped first.
• Romantic or sexual validation: desirability, comparison, jealousy, loneliness, or relationship standards. Keep treatment non-explicit and do not degrade a gender.
• Greed and shortcut seeking: disproportionate returns, secret access, certainty, or “smart people know this.”
• Security and a better life: housing, family, stable cash flow, escaping exhausting work.
• Intermittent reward and addiction: uncertain jackpots, streaks, rare items, rankings, or near misses.
• Fear, justice, and exposure: hidden evidence, countdowns, audits, betrayal, or consequences arriving quietly.

Tie the drive to the commercial mechanism. Ask: who creates the stimulus, who distributes it, who pays, what behavior is induced, and who owns the downside?

Build the opening

Finish the first complete hook within 3 seconds. Combine at least three elements:

• a concrete shared starting condition;
• a sharp numerical outcome gap;
• a familiar action or object;
• exclusion of the audience's default explanations;
• an unresolved causal question.

Preferred form:

Same [small starting resource], same [ordinary activity]. After [time], one person made [small result], another made [large result]. It was not [default explanation 1] or [default explanation 2]. What changed?

For immersive life stories, use a repeatable series stamp only if the identity itself contains conflict:

Today you enter the life of [provocative identity]: [concrete consequence or unresolved contradiction].

Put the series label on screen if speaking it would delay the consequence. Avoid a generic “today's story is...” opening without a cost, desire, danger, or surprising identity.

When the story has strong visual callbacks, consider a consequence-montage opening: flash 3-6 later moments of desire, reward, warning, and collapse, then stabilize into the main scene by roughly three to five seconds. Every flash must return later. Make the identity and concrete cost readable by about three seconds; do not let a stylish montage postpone the actual hook.

Do not spend the opening on greetings, credentials, background, or vague promises such as “improve your cognition.”

Structure retention

Use progressive disclosure:

1. Hook with the final contrast.
2. Establish identical constraints and a simple scoreboard.
3. Reveal Path A, then quantify it.
4. Reveal Path B as one conceptual upgrade.
5. Reveal Path C as another upgrade.
6. Reveal the strongest path, but delay its full mechanism.
7. Add a second discovery or reversal after the apparent climax.
8. Summarize the hierarchy in one memorable model.
9. Add boundary conditions and a realistic action.

Prioritize timely developments over a fixed template. In the opening section, move quickly from hook to situation, choice, and first consequence. As a starting pace, introduce a concrete new development roughly every 5-10 seconds in the first 30 seconds, then every 10-25 seconds depending on video length and complexity. A development must change what the viewer knows, expects, fears, or wants; a new adjective or decorative image does not count.

Make every scene perform at least one job: reveal information, force a decision, deliver a consequence, reverse an assumption, or explain the commercial mechanism. Remove setup that only repeats the background. Let the subject determine the rhythm instead of delaying useful information to satisfy an artificial interval.

For videos longer than 5 minutes, earn the duration with multiple causal discoveries. Otherwise compress to 60-180 seconds.

For an immersive second-person story, use this alternative spine:

1. Name the identity or preview the consequence.
2. Establish an ordinary desire and recognizable starting life.
3. Introduce the first rewarding signal or believable opportunity.
4. Let the protagonist make one understandable choice.
5. Repeat the reward while increasing cost or dependence.
6. Contrast visible gains with invisible losses.
7. Set a deadline, anniversary, payment date, audit, customer complaint, or relationship promise.
8. Make the apparent win coincide with the real-world loss.
9. Reveal who profited and how the money or attention moved.
10. End with one repeatable lesson and one practical safeguard.

For domestic feed videos, answer one question and open another every 5-12 seconds in the first 30 seconds, then every 8-15 seconds for a 60-180 second cut (12-20 seconds may be justified in a longer, already-proven episode). Use callbacks: an early object, sentence, person, or promise should return with changed meaning near the collapse.

Use false wins, temporary relief, or pressure waves only when they arise naturally from the protagonist's choices and the business mechanism. Do not invent extra setbacks or relief beats merely to fill a structure. The essential requirement is understandable causality and continuous forward movement.

Immersive second-person experience engine

When the viewer should feel “this is happening to me,” make 你 the acting subject rather than merely changing pronouns. Start inside a concrete place, time, action, body sensation, or social glance; reveal background through what the viewer does and notices. Give the first choice a believable self-justification and a small real reward before exposing its cost. Do not explain the emotion before the viewer has experienced the moment.

Before drafting, choose one variable experience spine—countdown, repeated small choices, social-gaze chain, false-win reversal, age/stage jumps, or one tracked object—and let the topic decide which beats exist. Do not mechanically complete the default ten-step spine. Use this internal brief:

根据用户选择的主题，让观众直接成为故事里的“你”。从具体地点、时间、动作和身体感受进入；让你做出当下看似合理的选择，得到真实的小奖励，再通过社会目光、资源变化、关系反应或账面数字改变这个选择的含义。每 5–15 秒回答一个旧问题并打开一个更尖锐的新问题。让早期物件至少两次回归，分别代表欲望、代价和机制/防线。结尾才提炼规则，不替观众提前下结论。

Keep one primary emotional wound or desire, three to five concrete micro-scenes, and one recurring object or sentence. Prefer “你把、你没有、你听见、你假装、你开始” over biography and abstract labels. Before delivery, check that the viewer has made or emotionally rehearsed at least one decision, not merely received an explanation. Read references/immersive-second-person-and-shot-rhythm.md for benchmark-derived examples and the full check.

Strong-hook retention engine

A golden 3-second hook is only the entry point. Before drafting, create a question ledger for the entire video. Every 5-15 seconds, at least one of these must happen: answer an earlier question, open a sharper question, increase the cost, change the interpretation, add evidence, or make the mechanism visible. A new background, adjective, or camera move does not count as a development.

Prefer this escalation when the topic supports it:

counterintuitive behavior -> moral misjudgment -> underlying mechanism -> viewer self-implication -> practical safeguard

For commercial traps, use:

apparent bargain -> hidden risk owner -> cash/information/attention mechanism -> why the viewer could accept it -> exit rule

Maintain three simultaneous tension lines: the protagonist's desire and fear, the movement of money/resources/incentives, and the viewer's recognition that they may make the same choice. Let the same object, sentence, or behavior return at least twice with changed meaning: desire first, cost second, mechanism or safeguard last.

Do not reveal the complete mechanism immediately after the hook. Give a plausible partial explanation, then make it insufficient with a concrete scene, number, experiment, or consequence. The next reveal must resolve the old question while creating a new one. For longer videos, schedule a clear reframe or irreversible consequence before the final quarter; otherwise compress the story.

When the format is image-first, each frame must carry four jobs: a meaningful scene, one main claim, one concrete object or action, and one unresolved question. Short top/bottom text bands are acceptable only when they are attached to the scene; never replace the story world with a blank knowledge card. Read references/retention-engine-from-image-post.md for the full benchmark-derived method.

Domestic platform retention preflight (Douyin / Xiaohongshu)

Before writing a domestic-platform episode, read the funnel reference and record the intended checkpoint at 2s, 5s, 15s, and 30s. The first frame, first sound, title/cover promise, and first spoken number must describe the same contradiction. Start audio within 0.2s; finish the complete question by about 1.8–2.2s; put a proof object on screen before explanation. At 5s, deliver a concrete detail; by 12s, deliver the first reward or cost; by 30s, complete a small reversal. A static title card, series stamp, atmosphere-only frame, or slow exposition does not count as a development.

Do not force one identical cut across platforms. When no duration is specified, default to a 60–120 second fast Douyin cut and a 90–180 second save-oriented Xiaohongshu cut; if only one master is allowed, use 90–120 seconds and package it separately. Keep a 3–6 minute Xiaohongshu depth version only when earlier data shows strong saves, follows, or deep viewing, and split long Douyin stories into episodes with independent payoffs. Diagnose Douyin with 2s bounce × 5s completion; diagnose Xiaohongshu with cover CTR → 2s exit → 5s exit → saves/follows, then locate the first curve low. Convert Xiaohongshu 5-second exit to retention as 100% - exit; do not assume cross-platform metric definitions match or use a universal completion threshold.

For Xiaohongshu, create a standalone static cover and explicit note title rather than treating the video first frame as the only package. Verify the cover in the platform's current thumbnail crop, keep one central contradiction and at most two key numbers, add searchable topic language, and place the first saveable calculation/check between roughly 30 and 60 seconds. A high cover CTR with weak saves/follows indicates curiosity without value fulfillment; a modest CTR with strong saves/follows indicates topic fit and calls for a stronger cover/first 30 seconds, not a wholesale topic change.

Explain the commercial model

For each actor or model, state:

• What resource is owned, rented, borrowed, presold, coordinated, or automated?
• Who pays first and why?
• When does cash arrive relative to costs?
• What limits scale?
• Which risk grows with scale?
• Who absorbs losses when demand, price, supply, or execution changes?
• Is the advantage labor, capital, distribution, trust, data, process, brand, network, or regulation?

Use this hierarchy only as a diagnostic, not as a universal law:

sell labor -> use assets/capital -> finance from demand -> coordinate resources -> build a repeatable system

Explicitly distinguish leverage from efficiency. Leverage magnifies both gains and losses; a system is valuable only when unit economics and controls remain sound.

Distinguish five forms of leverage:

• financial capital;
• labor and organizational systems;
• software and automation;
• media and distribution;
• trust, brand, and network access.

State which party supplies each leverage and how that party is compensated.

Treat compounding as repeated reinvestment of retained gains, not as a motivational slogan. Test:

next-period capacity = current capacity + retained gain × reinvestment quality

Compounding can apply to capital, skills, reputation, customer relationships, content libraries, data, and processes. It can also compound debt, defects, churn, regulatory risk, and bad habits.

When the story concerns platforms, content, games, beauty, relationships, status goods, or addictive products, also map the attention flow:

trigger -> repeated behavior -> data/engagement -> distribution advantage -> monetization -> user cost

Distinguish the protagonist's felt reward from the business's actual revenue. Do not confuse virtual currency, followers, status symbols, revenue screenshots, or gross merchandise value with disposable wealth.

Draft deliverables

Unless the user requests otherwise, produce:

1. One-sentence thesis.
2. Audience misconception.
3. Three hook variants and the recommended hook.
4. Fact and assumption table.
5. Business-model card and monetization pattern.
6. Money, product/service, information, and risk-flow map.
7. Unit-economics or cash-cycle calculation when relevant.
8. Timed narration script.
9. Shot-by-shot visual and subtitle plan, including duration for every image (target 5–8 seconds) and a motion label: opening entrance, static, center zoom-in, or center zoom-out.
10. Editing rhythm, important transition choices, and sound notes.
11. Platform-specific publishing package: Douyin title/cover/caption/comment prompt, plus Xiaohongshu standalone cover, explicit note title, searchable keywords, save anchor, and follow promise.
12. Risk disclaimer written in natural language.
13. Source and verification checklist.

For topics where completion matters, also produce a compact retention ledger: the opening question, each major answer, the next open question, the mid-video reframe, the callback object, and the final practical payoff. Do not output a formal tension curve by default; use the ledger as a production diagnostic.

When the user asks for stronger代入感 or supplies benchmark videos, also produce an experience ledger: place/time, viewer action, body sensation, social response, self-justification, immediate reward/cost, recurring object, and next question for each micro-scene.

Do not output a tension curve or transition map by default. Use either only as an internal diagnostic or when the user requests it, the story is unusually complex, or a retention problem cannot be explained clearly in ordinary prose.

Direct visuals and editing

• Produce commercial-model and life-simulation videos in landscape 16:9 by default, using a 1920×1080 production canvas unless the user explicitly requests another ratio. Treat the wider frame as part of immersion: show the protagonist and the surrounding dorm, shop, street, family, workplace, inventory, or customer context in the same composition instead of designing a narrow mobile crop and adding empty side space later.
• Generate images, compose subtitles, place overlays, and render the master directly for 16:9. Do not create a 9:16 master and crop, pad, or reframe it into landscape after production. Treat platform covers as separate packaging: create and verify a Xiaohongshu static cover in its current preview crop instead of relying on an automatic crop of the 16:9 first frame.
• Keep the entire video inside one coherent animated story world. Use original cartoon/comic scenes for the hook, plot, calculations, explanations, reversals, summary, and ending.
• Do not insert PPT-style data cards: no full-screen dark panel with a headline, giant number, formula, bullet list, chart frame, or large unused background. Do not treat changing from an animated scene to a presentation slide as a retention reset.
• Translate every abstract claim and number into an animated scene. Use characters handling cash, cups, inventory, bills, contracts, phones, receipts, storefronts, queues, clocks, delivery bags, or other story objects. Attach short number labels, arrows, or formulas to those objects when needed.
• Keep overlays subordinate to the animated image. A number or formula may appear over a scene, on a receipt, POS screen, wall board, phone, ledger, package, or hand-drawn diagram, but the underlying shot must remain a meaningful animated composition rather than an empty text background.
• Example: explain monthly fixed costs = 90,000 with the shop owner at the counter while rent envelopes, four employee wage packets, utility bills, and maintenance tools are arranged around him; label each cost beside its object. Do not cut to a slide listing four expenses.
• Match every abstract claim with a concrete visual metaphor, contract, receipt, inventory action, queue change, before/after state, or character decision.
• Treat each still image as a 5–8 second shot target. Split narration at semantic actions so one image carries one main action, evidence, and unresolved question. Never let an unchanged image run beyond 8 seconds by default; a justified reflective hold may approach 10 seconds, but must not be disguised with slow zoom alone.
• Refresh meaningful visual information every 5–8 seconds by changing the scene, actor, object, expression, causal state, or embedded annotation. Do not manufacture refreshes with camera motion alone.
• Keep a stable visual world, character identity, palette, and typography across a story.
• Emphasize numbers, ownership changes, cash direction, and turning points.
• Use subtitles as semantic chunks, not word-by-word noise. Keep mobile-safe margins and strong contrast.
• Derive subtitle start/end times from the TTS sentence or word-boundary events. Never divide a scene's total duration evenly across subtitle chunks: punctuation pauses and sentence lengths are unequal, so equal splitting causes both early cuts and subtitle overhang.
• Keep the timeline's scene offsets identical to the rendered clip durations. Do not add a subtitle-only tail buffer or accumulate a nominal padding value that -shortest removes from the actual clip. Before delivery, inspect the first, middle, and final scene for subtitle overhang and cumulative drift; treat any visible mismatch as a release blocker.
• Let music support curiosity and escalation. Do not allow sound effects to compete with explanation.
• Prefer original or licensed assets. Treat AI visuals as illustrations, not evidence.
• For 4-6 minute AI-cartoon stories, use a starting production target of about 5.0-5.8 Chinese narration characters per second and one new image shot every 5-8 seconds. Adjust for emotion and comprehension, but record and justify any shot outside the range.
• Allow one deliberately designed entrance during the opening hook, such as a clean reveal, push-in, foreground wipe, or brief parallax move. Finish it quickly and stabilize the image.
• After the opening, default every shot to still. When emphasis is useful, use a centered keyframed zoom-in or zoom-out of about 2-6% across the whole shot. Keep the movement one-directional and smooth.
• Do not use routine horizontal pans, left-right drift, alternating crop positions, oscillation, handheld shake, random parallax, or repeated zoom resets. Never add motion merely because the source is a still image.
• Use hard cuts for causal changes and short cross-dissolves for emotional or time transitions. Let a new animated composition create energy instead of making the current frame wander.
• Treat important transitions as relationships between scenes, not decoration. Prefer object/idea match cuts for the same causal chain, 0.20-0.45-second dissolves for time or emotional continuity, brief double exposures for attractive surface versus hidden cost, and hard cuts for irreversible decisions, deadlines, or mechanism reveals. Ordinary cuts do not need a special label or explanation.
• Keep special transitions inside the opening or at major story turns. Synchronize them with a phrase boundary or restrained sound hit, then return immediately to a stable animated composition.
• Choose the palette by story engine: cold blue-gray for self-deception or systems, warm pastel for initially attractive relationships, luxury black-gold for aspiration, neon HUD colors for digital temptation, and desaturated thriller lighting for audits or exposure.
• Use on-screen interfaces, receipts, balances, clocks, maps, chat messages, scoreboards, or contracts only when they advance causality.
• Keep the top disclaimer and bottom subtitles visually separated. Use a stable series mark without copying a benchmark watermark.

Before rendering, audit the storyboard with these hard checks:

• The master canvas is landscape 16:9 (1920×1080 by default), and every generated scene was composed for that ratio rather than converted from a vertical layout.
• Every non-opening shot has a meaningful animated scene; the count of PPT-style slides is zero.
• Every shot has an explicit motion label. After the hook, horizontal displacement and shake are both zero.
• Any zoom is centered, smooth, one-directional, and no larger than needed to emphasize the subject.
• Commercial calculations remain numerically legible without replacing the animation with a presentation page.
• If an explanation cannot yet be visualized in the animated world, redesign or generate another scene instead of falling back to a card.
• Important stylized transitions must support causality, time, place, contrast, or emotion. Remove transitions that merely advertise the editing software; do not create a separate transition map unless it helps production.
• Every opening flash reappears later as a callback, and the opening stabilizes by roughly five seconds.
• Audio/subtitle sync is a hard gate: every subtitle cue must be generated from the same audio render, its end must not exceed the corresponding spoken boundary by more than 150 ms, and scene-start offsets must be measured from the actual encoded clip duration. Write a machine-readable sync audit and visually review at least one cue near the start, middle, and end.
• Image cadence is a hard gate: the shot list must include a duration for every image, at least 90% of still shots must be 5–8 seconds, no unchanged shot may exceed 10 seconds without an explicit narrative reason, and a long voice segment must be split into new semantic images rather than held on one frame.

Design social sharing

Give the viewer a clear recipient and reason to share:

• “send to a brother” stories: loyalty, shared game language, status recovery, or mutual warning;
• partner stories: a non-accusatory standard for mutual effort;
• colleague stories: workplace signaling, fake busyness, incentive mismatch;
• entrepreneur stories: hidden costs, customer acquisition, cash flow, and exit risk;
• family stories: housing, debt, guarantees, education, or security.

Create a share line that helps the sender communicate without writing an explanation. Do not use humiliating labels or provoke gender/occupation hostility merely for comments.

Keep ethical and legal boundaries

• Do not give operational instructions for theft, fraud, hacking, evading detection, manipulating vulnerable people, or exploiting addiction.
• When adapting a crime or cheating thriller, focus on incentives, red flags, detection, victim impact, and consequences; omit actionable timings, blind spots, tools, and evasion steps.
• Do not use sexual content as explicit bait. Use adult desire, comparison, intimacy, or validation only when relevant to the economic mechanism.
• Do not glamorize wealth symbols without showing the business, cash flow, opportunity cost, and survivability behind them.
• Do not imply that addiction, a breakup, or financial collapse is solved by one motivational decision. Show recovery as repeated behavior and, where appropriate, external support.

Review quality

Score each item from 1-5 and revise any item below 4:

• Hook specificity
• Causal clarity
• Numerical integrity
• Money-flow visibility
• Retention progression
• Visual explainability
• Practical usefulness
• Risk honesty
• Business-model accuracy
• Cash-flow clarity
• Distinctive voice
• Comment-worthiness
• Share recipient clarity
• Emotional-mechanism fit
• Animation-world continuity
• Camera-motion restraint
• PPT-free commercial explanation
• Plot causality
• Timely development pace

Reject a draft if it promises wealth, hides material downside, confuses revenue with profit or cash, copies a benchmark's wording or visual identity too closely, relies on PPT-style cards, or uses persistent horizontal camera movement after the opening.

Iterate from performance

Record source link/file, topic, hook, duration, structure, visual style, publishing data, audience objections, and lessons. Diagnose metrics by stage:

• low 2-second retention: opening is vague or visually slow;
• high early retention but low completion: mechanism arrives too late or repeats;
• high completion but low saves: insight lacks a reusable model;
• high saves but low follows: account promise or series continuity is weak;
• many objections: inspect hidden assumptions and missing downside before defending the thesis;
• high early retention but a middle drop: the hook opened no question chain, the mechanism was revealed too early, or consecutive frames only restated the same claim;
• strong completion but weak interaction: the ending answered the topic but did not give a concrete safeguard, reframe, or share recipient.

Update benchmark references only when a repeated pattern appears across multiple strong samples. Keep one-off observations labeled as hypotheses.