Chapter 4. The Brief, the Greenlight Gate, and the Loss Cap#

4.1 A brief is a production contract, not a creative introduction#

Many briefs contain a genre, a few reference images, and adjectives — premium, gripping, cinematic. That is enough to start a meeting and not enough to constrain production. The writer does not know which behavior to design for. The asset team does not know which characters deserve heavy investment. The generation team does not know the budget ceiling. The editor does not know whether to protect story clarity or chase a faster cut.

An effective brief is a production contract between everyone on the project. It defines goal, audience, promise, scope, constraints, risks, approvals, evidence and stop conditions. It does not have to answer every creative question in advance, but it must make clear which questions are still open and who has the authority to settle them.

A brief has one more function that is easy to overlook: it stops a project from quietly switching goals under pressure. When pilot click-through disappoints, a team suddenly remembers it always cared more about brand quality. When payment underperforms, at least there were views. If the primary goal and the guardrail metrics were never written down, any result can be narrated as a success — and nothing can ever be learned.

4.2 The twelve things an executable brief contains#

One, the value exchange: how the project creates revenue or business value. Two, the primary audience action. Three, audience desire and viewing context. Four, the story promise. Five, the business model and distribution entrance. Six, format: episode count, duration, aspect ratio, language and versions. Seven, the visual and audio route. Eight, production constraints. Nine, rights and compliance status. Ten, metrics and evidence. Eleven, approvers and change authority. Twelve, stop conditions.

Missing any one of these does not necessarily block concept exploration, but it does block entry into expensive production. Unknowns are allowed to exist. Unknowns must be labelled.

4.3 The opportunity brief for Backlit Takeover#

project_id: MD_BOOK_001
working_title: Backlit Takeover
decision_version: 0.4
value_exchange:
  primary: viewers pay to unlock E007 and beyond
  secondary: produce a recuttable IAA version
primary_audience_action: unlock_after_E006
audience:
  core_desire: retake, with undeniable evidence, the right to interpret that power has seized
  viewing_context: vertical mobile, usually entered muted from an ad, 3-8 minutes per session
  rejection_triggers:
    - the male lead solves the protagonist's core problem for her
    - the rules of evidence change arbitrarily
    - one misunderstanding stretched across many episodes
story_promise: each segment completes a status reversal through a new mechanism of proof, and advances the missing-mother mystery
format:
  episodes: 60
  duration_seconds: 65-85
  aspect_ratio: 9:16
  language: Mandarin
  pilot: E001-E007
visual_route: contemporary urban realism; cool neutral offices, warm gold banquet; proof objects highly legible
audio_route: naturalistic dialogue; restrained narration; cold three-note piano theme; music pulled before key reveals
constraints:
  core_characters_first_10: 5
  major_locations_first_3: 3
  primary_locations_per_episode: 1
  critical_text: all composited in post
  forbidden_spectacle: large vehicle crashes, hundred-person crowds, complex car chases
rights:
  source: original teaching case
  character_likeness: not based on real people
  music: only provably licensed for commercial use, or self-generated assets
success_evidence:
  - a cold viewer can describe the current conflict within 5 seconds
  - E001 completion reaches the test baseline
  - at least one hook family is proven effective
  - E006 boundary conversion and post-E007 retention hold at the same time
approvers:
  commercial: PRODUCER_01
  story: SHOWRUNNER_01
  visual_assets: ART_DIRECTOR_01
  release: PRODUCER_01

Notice what the brief does not say: hit, stunning, industry-leading. None of those can be accepted against. What it does say is when the audience should understand what, how which assets are handled, which expensive spectacle is not being made, and what evidence permits the work to continue.

4.4 Translate adjectives into acceptance conditions#

Clients and creators will naturally speak in feelings. A production system cannot refuse feelings, but it must translate them.

"The characters should feel premium" may mean restrained tailoring, makeup that is not over-styled for social video, unexaggerated movement, no gratuitous slow motion. "Fast pace" may mean a state change in the first three seconds, a goal established within fifteen, and repeated explanation deleted — not every shot cut down to one second. "Consistent characters" means at minimum that facial identity, hair, build, wardrobe stage and signature objects meet a specified match.

Translate using observable behavior, plus condition, plus permitted deviation. For example: across all approved keyframes in E001, Lin Yun retains the identity mole below her left eye, straight collarbone-length hair, and the same grey suit; a mole displaced by a mirrored composition does not pass; the mole being invisible in a wide shot is not a failure, but the face shape and hair silhouette must still match.

That is longer than "maintain character consistency" — and unlike it, it can actually direct a review.

4.5 Constraints are not the enemy of creativity#

Because AI suggests anything can be generated, briefs easily lose their budget constraints. But every additional character, location, costume and complex action adds asset production, failed attempts, continuity checks and rework paths.

Express constraints in production units. Do not write only a total budget. Write the maximum number of core characters in the first ten episodes, the number of main locations per episode, the target shot count per minute, the share of complex two-person action, the number of key props, the permitted average number of generation attempts, and the scope of manual repair.

Constraints also generate style. Because Backlit Takeover avoids large action set pieces, power is expressed through seating, doors, documents, screens and changes of eyeline. The low budget did not make the story smaller. It produced a unified visual language in which records contend with power.

4.6 The risk register#

A risk is not "AI might be unstable." Each entry records probability, impact, the earliest stage at which it can be detected, prevention, fallback and owner.

- risk_id: R_VIS_003
  description: face bleed or height drift in two-person confrontation shots
  probability: high
  impact: high
  detect_at: storyboard_and_keyframe
  prevention:
    - reduce complex two-person movement
    - lock a golden reference for each character separately
    - shoot singles and reverses after the master
  fallback: rebuild the scene from single coverage and reaction shots
  owner: ART_DIRECTOR_01

- risk_id: R_STORY_002
  description: audit detail imposes too much comprehension load on the audience
  probability: medium
  impact: high
  detect_at: cold_read_and_pilot
  prevention:
    - introduce only one rule of evidence at a time
    - explain through visible screens and character consequence
  fallback: rewrite the terminology as three visual fields — amount, time, name
  owner: STORY_EDITOR_01

The most valuable field is detect_at. The earlier a risk can be reliably detected, the cheaper the repair. If a two-person shot is unnecessary at the shot-design stage, you should not wait for five failed video generations before breaking it up.

4.7 A greenlight is not one decision — it is a series of small bets#

A project should not go from a one-line premise to permission to make sixty episodes. Greenlights are issued in stages: concept, story, assets, pilot, first ten episodes, and full-season scale-up.

A concept greenlight funds written research and low-fidelity tests. A story greenlight funds the first ten episodes and the asset requirements. An asset greenlight funds golden references and pilot keyframes. A pilot greenlight funds finished material for E001–E003 or up to the first paywall. Only the scale-up greenlight funds batch production.

Each one states the new budget, the proposition it must prove, and the date of the next review. That way continue stops being inertia and becomes a choice backed by evidence.

4.8 A scorecard is not a substitute for judgment#

A scorecard helps a team notice omissions. It cannot produce truth by adding unrelated dimensions together. A project with a very high total still cannot be greenlit if the rights status is unclear or the production cost exceeds the ceiling. Some conditions are thresholds, not bonus points.

greenlight_card:
  gates:
    rights_clear: required
    audience_hypothesis_testable: required
    pilot_within_budget: required
    primary_story_promise_clear: required
  scored_dimensions:
    desire_strength: 0-5
    hook_clarity: 0-5
    differentiation: 0-5
    serial_escalation: 0-5
    visual_producibility: 0-5
    asset_reuse: 0-5
    distribution_fit: 0-5
  decision: conditional_greenlight
  conditions:
    - verify with a rough shot list that the audit proof is comprehensible on a phone
    - run one generation stress test on the two-person meeting scene

Set thresholds against your own team's historical baseline. This book will not pretend there is a magic score that applies to every platform.

4.9 A pilot is not a shrunken episode — it is a combination of experiments#

The pilot scope must cover the project's most dangerous assumptions. Backlit Takeover cannot simply produce the easiest thing, a single close-up of Lin Yun. It has to test office dialogue, two-person identity consistency, on-screen proof objects, key narration, blackout lighting, and the first paywall boundary. Otherwise a successful pilot only proves the team can make easy shots.

A good pilot plan assigns evidence to each risk:

Finish quality can be uneven across a pilot. To test story, early static boards and temporary voice are fine. To test character, you need keyframes made close to the final pipeline. Do not spend budget making irrelevant parts look finished.

4.10 The loss cap and stop conditions#

Stopping is not the last step of failure management. It is part of setting the project up. Without stop conditions, teams keep investing because they have already invested, and use more generation to conceal demand that was never demonstrated.

Stop conditions come in four kinds: market, story, production and rights. A market stop means several genuinely different hook families all fail to build real interest. A story stop means the audience understands the content and does not care about the central question. A production stop means the cost of reaching acceptable quality exceeds the commercial ceiling. A rights stop means key IP, likeness, voice or music cannot be provably licensed.

Stops can also be local. Stopping a visual route, a complex scene or one advertising angle is not stopping the project. A good system decomposes "the project failed" into assumptions that can be replaced.

4.11 Budget by shot risk#

Estimating per episode hides the variance. An episode of twenty stable close-ups and an episode with ten multi-person action shots can cost wildly different amounts to generate. Budget per shot: static attempts, motion attempts, manual repair, sound, compositing and QC.

expected shot cost = still candidates x cost per attempt
                   + motion candidates x cost per attempt
                   + manual repair hours x hourly cost
                   + allocated compositing / audio / review

The metric that matters more is cost per finally usable second, not the price of one generation. A cheap model with a high failure rate and severe identity drift can produce a far higher cost per usable second.

Budgets must also reserve for exploration, rework and the unforeseen. Allocating everything to the first generation means there is no room to learn after the first mistake.

4.12 The decision log and change control#

Briefs change. They must not change silently. Record for each significant revision: who proposed it, why, the evidence, who approved it, the affected assets, the cost and the version.

If E006 changes its core proof from a DNA report to a birth bracelet, the system must find every script, shot list, keyframe, ad and subtitle referencing PROP_DNA_REPORT. Fixing the current cut and leaving later episodes still discussing a DNA report is not a fix.

decision_id: DEC_021
date: 2026-07-04
change: E006 core proof changes from DNA report to birth record plus bracelet
reason: cold viewers considered a single report too easy to forge
evidence: PILOT_INTERVIEW_02
affected:
  - E004_B05
  - E006_ALL
  - PROP_DNA_REPORT
  - AD_SECRET_IDENTITY_A
cost_impact: medium
approved_by: SHOWRUNNER_01
status: proposed

Only an approved decision may change the current fact ledger. Candidates still under exploration must not contaminate locked assets.

4.13 SOP for the greenlight meeting#

Before the meeting, the showrunner circulates the brief, the risk register, the scorecard and the open questions. Do not present them for the first time in the room. In the meeting, confirm commercial goal and rights thresholds first, then walk the evidence for the highest-risk assumptions one by one. The creator explains the story value, the producer explains cost and dependencies, distribution explains the test path, and legal explains the conditions under which nothing may proceed.

The meeting may produce only four outcomes: greenlight, conditional greenlight, return for more evidence, or stop. It may not end with "let's all think about it" or "generally good." A conditional greenlight must state the conditions, the owner, the budget and the stage deadline.

Afterwards the decision enters the log, the brief takes a new version number, and the system queues only the approved scope.

4.14 Fault tree#

Symptom: everyone understands the project differently. The brief has a genre and reference images but no story promise, primary action or exclusions. Have each owner retell the project independently, compare the divergence, and write it back.

Symptom: the project keeps adding characters and locations. Scope was never constrained in asset and shot units, and additions carry no cost analysis. Require every addition to state what it replaces or how much budget it adds.

Symptom: high scores, failed pilot. The scorecard became an opinion averager and the high-risk assumptions were never tested. Check the gate items, and check whether the pilot really covered the hardest scenes.

Symptom: the team keeps generating while knowing the direction is wrong. Either stop conditions were never defined, or the decision-maker is anchored on sunk cost. Return to the original assumptions and compare the future cost of continuing, changing route and stopping — not the money already spent.

Symptom: one small change causes widespread inconsistency. Stable IDs, dependency relationships and a decision log are missing. Freeze new generation, and complete the impact analysis and fact ledger update first.

4.15 Checklist, exercises and deliverables#

Before greenlight, confirm: whether the value exchange and primary action are singular; whether the target audience is described by desire and context; whether the story promise can be retold; whether format and versions are explicit; whether visual, audio and production constraints are observable; whether the rights source has evidence; whether the highest risks are covered by the pilot; who can approve changes; and what must trigger a stop.

Exercise one: translate "make a premium, cinematic, fast-paced female revenge microdrama" into a complete brief. Exercise two: build a risk register for a fantasy battle project, and design a pilot under one minute that still covers identity, action, effects and audio risk. Exercise three: take a project with heavy investment and thin evidence, and write three genuinely executable stop conditions.

Deliverables for this chapter: the opportunity brief, the constraint table, the risk register, the greenlight scorecard, the pilot experiment matrix, the stop conditions, the shot-risk budget, the decision log and the change impact table.

A note on sources#

This chapter absorbs practical signals from Chinese creators about staged confirmation, static before motion, human review, production cost and contract acceptance. Specific budget and metric thresholds must come from your own team's historical data. Without that history, establish a baseline through a small pilot first — never borrow someone else's promotional numbers as the basis for a greenlight.