This is the paragraph that settles arguments in month three. When a shot looks gorgeous but feels off, the review question is not "do we like it" but "does it serve this paragraph." Write what the film is FOR: what it should do to a nine-year-old, and what it should plant that sprouts a week later. Example opener: "A wordless film that makes a kid love two little machines, and through them, the work of healing land." If the paragraph could sit on any nature short's poster, it is not done yet; it should only fit this one.

The age band decides a hundred downstream calls: how dark the storm reads, how long the film can sit in quiet, how scary the low point gets, how much patience the pacing can ask for. Nine-to-twelve plus the parent means real feelings are allowed but cruelty is not, and boredom is the only true sin. If the answer is "everyone," the film obeys no one; pick the person the film must win and let everyone else be a bonus.

Anchors are borrowed feelings, not style references: what each film FEELS like, not what it looks like. The line on each is the deliverable: Wall-E for warmth carried entirely by machines, Planet Earth for scale and reverence, Flow for wordless clarity under pressure. Strike any anchor that does not earn its line, add ones that do. Three is plenty; five is a mood board wearing a trench coat.

The working test: a viewer walks out and says "that was about..." in their own words. If those words are close to the intended feeling, the film worked; if they recite a slogan, it preached. Write the feeling as a sentence a stranger might actually say: "the little ones kept going and the rain came." Then guard it: no character may say it, no narration may explain it, no title card may rescue it.

The most load-bearing field on the page, creatively and legally. Each line must be checkable against a cut by someone with no context: "no drowning imagery" passes or fails on sight. Vague entries ("nothing too sad") get argued; checkable entries get enforced. Aim for five to ten. This list runs at every review, every animatic pass, and every cut, and it is the fastest way to keep the film safe without hovering over it.

Development builds a library on purpose: rig, sets, lexicon, style bible. This field aims that library beyond the film. If a picture book is plausible, the style bible keeps compositions that crop to spreads. If the funding rooms matter, development keeps frames that present at billboard scale. Name the three or four futures worth building toward now; everything else stays a nice idea.

Concrete beats grand. "Accepted at two Oscar-qualifying festivals, a laurels page that opens conservation-partner conversations, the book pitch alive" is a target the plan can steer toward. "It goes viral" is weather. Three sentences, checkable in July 2027: what happened, who saw it, what doors opened.

One sitting, straight through, no research pause. The hedges ("maybe," "we could," "possibly") come out in the trim; the heat stays. This document works when it sounds like a person who knows what they want, because every later argument gets settled by reading it out loud.

The dinner-party test: introduce the robot the way a friend would. "The one who triple-checks the knot and worries the whole way down" is an identity; "brave and curious" is a horoscope. The line has failed if it also fits the other robot: the pair only works because the two lines cannot be swapped.

Two-hander films live or die on this field. Who moves first, who measures twice, who fixes, who breaks the silence after a fight. Write five or six exchanges as behavior: "one rushes the gap, the other measures it, the rushed jump fails, the measured one follows anyway." The comedy comes from the mismatch; the heartbreak comes from the mismatch mattering.

Shared anatomy means identity lives in the deltas: proportion, posture, tempo, wear. One sits a little lower; one carries a dent nobody explains. Movement is casting: decide which unit moves like a sparrow and which like a tugboat, and every animation choice downstream inherits it. Write it physical enough to pose.

Pitch range, phrase length, how fast it answers, what silence means. The two dialects must be tellable apart with eyes closed: that becomes a literal test at sound direction. Note what each unit does vocally under joy, alarm, and stubbornness, and when each one goes quiet, because the quiet is where the acting lives.

Acting arrives through posture and tempo only. Joy might be a higher stance and quicker turns; fear might be folding low and going still. Write one physical tell per core emotion per robot, different for each. If both express fear identically, the pantomime test fails and the pair collapses into one character wearing two shells.

The surface want drives scenes: finish the job, find the water, fix the broken thing. The deep want drives the film: matter to the other one, see the land live. State both, and the fear underneath them. Scene logic falls out of it: a unit that fears being useless volunteers for the dangerous job before being asked.

Directable rules, not personality adjectives. "Always finishes a job, even a pointless one" tells the boards exactly what happens when the storm hits mid-task. "Never harms a living thing" decides an entire beat the moment a creature blocks the path. If a rule cannot generate a scene decision, sharpen it until it can.

The IP field. Three to five signatures a knockoff could not avoid copying: the split-sphere silhouette, a movement tell, a sound signature. These go into the copyright record with the dated pages, which is why they need to exist on paper before any AI touches the designs. Favor marks that survive silhouette; those are the defensible ones.

The logline discipline: what happens, to whom, and why anyone cares, in one breath. "Two small robots tend a dying valley and hold on for the first rain" is workable. If the sentence needs a second clause to explain the world, the world is doing too much work; if it has no feeling in it, neither will the cut.

Only what the audience must know without being told: the year, what healed, what is still healing. Everything else is texture and belongs in the world notes. The test: a stranger watches the film cold and never feels a missing footnote. If a fact must be known and cannot be shown, it belongs to the narration; if the narration cannot say it naturally, it does not belong in the film.

The counterpoint engine, spelled out: narration states nature's rule in the calm documentary register, and the robots break the rule in the same frame. "The wind crawler conserves energy through the dry months" over two robots frantically building something pointless: that gap is the comedy. Decide what the voice knows, what it never comments on, and how it feels about these two by the last scene. The warmth should arrive so gradually nobody can point to where it started.

A beat is a change of state, not an activity report. "The robots work the ridge" is scenery; "the seed cache floods and the map is suddenly worthless" is a beat. Write each in one or two lines: what was stable, what breaks it, whether it lands as event or feeling, macro or intimate. Mark unknown gaps ??? and keep moving; a boarded hole is fixable, a stalled sheet is not.

The single pivot the film bends around: before it, one kind of story; after it, another. Exactly one beat owns it. If two beats both feel like the turn, the film has two spines and will drift in the edit; if none does, it is a mood reel. Name the beat, then protect it: nothing in pre-production gets to soften it.

Films are remembered backward from the final frame. Describe what is literally in frame and what it should do to the chest, not the plot it resolves. The bar: the image a viewer doodles in a margin a week later. This field deserves more drafting time than any other on the page, because the whole edit gets steered toward earning it.

Story-level keeps and bans, checkable at the animatic: "the first rain lands on screen, not offscreen," "the robots never speak words," "no human appears." This is the beat sheet's contract with the never-list: anything written here is enforceable at every review, so write only lines meant seriously.

Each place is a feeling with geography attached: the meadow that holds the morning light, the machine ruin that reads like a fallen giant, the water line that keeps its distance. What a place FEELS like when the robots enter is the deliverable; names are optional. Set design starts from this list: every place described here becomes an AI design cycle inside the locked style, and an approved set the film can shoot.

The verbs the audience sees but never hears named: catching water, holding soil, seeding shade. Each verb becomes shot ideas: a terrace holds after a slip, a windbreak leans and stands. Write them plain; the film does the translating. Any term only a practitioner would use goes in the lexicon, not on screen.

Vast sells the stakes: one valley in a healing world, the documentary register. Close sells the love: two small machines and their work. Note the two or three moments that must feel enormous and the two or three that must feel like a held breath. Pacing and lensing get built off this map, so disagreements here are cheap and disagreements at the animatic are not.

The whole film waits for the rain, so the waiting needs a look: dry wind, hard light, dust that hangs. Describe the before (heat, stillness), the approach (the light going strange), and the after (everything the film was saving up). Any hard opinions about storms, dusk, or overcast days belong here; light is the cheapest emotion in the pipeline.

The world's own never-list: kinds of imagery, kinds of decay, kinds of danger this place never shows, one checkable line each. The test is the same as the vision doc's: a line either passes or fails on sight of a cut. These merge into the master never-list and run at every review, so write only bans meant seriously.

Anywhere means anywhere: film stills, paintings, photographs, game screens, a phone photo of late light on a wall. No justification required; the pattern across the images is the brief. Ten loves with no explanation beat three loves with paragraphs, because the pattern is the data, and week 1 turns that pattern into the four style directions.

Hates are faster teachers than loves. Three to five images that feel wrong for this film, one line each naming the offense: "too glossy," "feels like a tech demo," "cold." Every hate closes a door the style search would otherwise waste a round exploring. The one-liners matter here in a way they do not for loves.

Five adjectives for how the film should feel to the hand if a frame could be touched: soft, worn, sun-bleached, hand-made, quiet. These words steer material and light choices in every render test. If two words fight each other (pristine, weathered), the fight gets settled now instead of inside week-three frames.

Hard opinions welcome; "surprise me" equally welcome; both are usable answers. What helps most: any color that already belongs to the story (the green the valley is heading toward, the color of drought) and any color that repels. The real palette gets decided on actual frames in weeks 2-3, so instincts recorded here are aim, not law.

Whatever must survive no matter which style direction wins: a silhouette that stays readable, a softness that cannot be traded away, a specific kind of light worth building the look around. Sacred means the style bible writes it down and every later choice defers to it. One or two entries; sacred loses meaning in bulk.