skills/ai-film-pipeline-master/references/phase-01-story-generation/symptoms-tension.md

# Tension complaints

Part of the symptom router. Open this file when the complaint is about tension. Each complaint below is followed by the cards that repair it, best first; `LOOKUP.md` says which file holds each card.

---

### "There's no tension"

The most common cause is that the audience knows exactly as much as the characters. Tension is mostly an information arrangement, not an intensity setting.

1. Tell the audience about the threat and give them the deadline — the threat alone is half the device
   `tens.bomb_under_table`
2. Build dread rather than reaching for shock; dread lasts, a jolt does not
   `tens.dread_over_shock` · `tens.anticipation_vs_anxiety`
3. Promise the violence long before delivering it
   `tens.promise_of_violence`
4. Demonstrate the threat's rules on a secondary character first
   `tens.rule_demonstration` · `tens.threat_with_rules`

### "The scene is scary but the story isn't"

1. You are spending jolts instead of building a state. A false scare costs credit and lowers the next one
   `tens.false_scare_cost`
2. Run a low continuous dread under everything instead
   `tens.sustained_low_dread`
3. Put one small thing wrong in an ordinary scene and let nobody remark on it
   `tens.wrongness_detail` · `tens.corrupting_the_mundane`
4. Violate a place the audience has been taught is safe
   `tens.safe_space_violation`

### "He's come home and I want the house itself to feel wrong"

1. Reproduce the place correctly in every particular except one, and let nobody in the scene remark on it. Make the deviation a *dimension* rather than an object — the ceiling lower, the corridor longer, the light coming from a side with no window — because an object out of place is a clue and a wrong proportion is a feeling, and the audience will suffer it for minutes before they can name it
   `tens.almost_right_place`
2. Let a minor character be entirely comfortable there. That is what makes the wrongness unshareable
3. Never let anyone measure or verify, and do not explain it inside the same work. Verification converts dread into a puzzle, and a puzzle has a solution
4. The three neighbours, kept apart: breaching a place built as safe, one off note inside an ordinary room, and an object the audience owns being poisoned. This card alters the container and leaves everything in it intact
   `tens.safe_space_violation` · `tens.wrongness_detail` · `tens.corrupting_the_mundane`
5. It needs a template. A strange building the reader has never seen is just a strange building

### "The wedding goes perfectly and it's the worst thing in the film"

1. The ritual proceeds correctly, step by step, to its proper end, while something inside it is badly wrong — and the form is the trap, because stopping would cost more than finishing
   `tens.ritual_that_must_continue`
2. Teach the order of the steps and how many remain, early, so the audience can count. A rite with a date is a countdown supplied by the culture rather than by the plot
   `tens.countdown`
3. Somebody in the room must have the authority to stop it and must be seen deciding not to. Keep the wrongness legible to two or three people present and to nobody else, so the pressure is social rather than supernatural
   `tens.social_containment`
4. The ritual must finish. That is the whole card, and it is what separates it from the two adjacent ones — the single scene where the form is *broken*, and the ceremony used as a structural container with a private conflict per step
   `scene.interrupted_ritual` · `seq.ritual_sequence`
5. It has nothing if the audience does not know the form. Teach the order in an earlier working instance, or choose a ceremony they already carry

### "I need dread without a monster"

1. Tension does not require a villain — a situation, a system or a diagnosis will do
   `tens.tension_without_villain`
2. Social exposure is a real fear and costs nothing to stage
   `tens.social_dread` · `tens.escalating_humiliation`
3. Raise the register instead of the volume
   `tens.polite_threat` · `dial.polite_aggression`
4. The body betraying its owner is the cheapest interior threat there is
   `tens.unreliable_body`

### "The audience already knows what happens"

1. Known outcome is an asset, not a problem — the dread is in watching it arrive
   `tens.known_outcome_dread`
2. This is the correct engine for historical, heritage and true-story work
   `tens.known_outcome_dread` · `struct.freytag`
3. In documentary and history, where the outcome is not yours to withhold at all and the audience arrives already holding it, the engine has to be rebuilt on a different question — that is the complaint below
   `tens.known_outcome_documentary`

### "Everyone knows how this ended two thousand years ago"

**The honest limit, stated first: this does not produce suspense. It produces dread and investment.** Suspense needs a live uncertainty about the outcome and you do not have one. A piece that pretends otherwise — music swelling toward a result the audience learned at fourteen, a narrator asking "would the city hold?" — reads as a cheat, and the audience that knows the most catches it first and stops trusting everything after it.

1. Stop asking *what happens* and build the piece on a question the record leaves genuinely open: how, at what cost, who decided, and how do we know any of this
   `tens.known_outcome_documentary` · `tens.asserted_tension`
2. **The knowledge gap** is the only free tension non-fiction gets. The people in the material did not know. Write the scene from inside their knowledge rather than yours — the instant the voice says "little did they know", the gap collapses into commentary and the scene is over
   `info.dramatic_irony`
3. **The restored decision.** Find a choice in the record that could plausibly have gone the other way and put the camera on it before it was made. It must be a real option and you must be able to show that it was — the strongest proof is that somebody else in the same situation took it, which is evidence rather than speculation. This is the line between restored contingency and fate as an opponent
   `anta.fate`
4. **The detail that was not in the textbook.** The audience knows the headline and nothing under it, so everything under it is new — but the detail must be load-bearing, changing how the known event reads rather than decorating it. Three per segment, no more
   `expo.heritage_specificity` · `expo.proper_noun_budget`
5. **The process question — how do we know this at all.** Who found this, what survived, what is missing and why. The point where the sources stop is a real cliffhanger about the evidence, and nobody in the audience knows where it is
   `seq.documentary_sequence` · `pov.retrospective_distance` · `end.mirror_ending`
6. The structural move underneath all four: **withhold the known outcome's shape, not its fact.** Everyone knows the empire fell; nobody knows what the last ordinary morning looked like, what was in the storeroom, who was still doing their job. Announce the fact early and without ceremony — pretending to guard it is the commonest cheat in the genre — then spend the film on the shape
   `info.known_ending_structure` · `tens.known_outcome_dread`
7. When the known outcome is a catastrophe the audience is already braced for, they have pre-felt the blow and will not feel it again on cue. Move the weight before or after it, play the event itself flat, short and quiet, and keep the opposition unaimed — inventing a face for a catastrophe will cost you the specialists first and the community second
   `tens.aftermath_not_act` · `anta.ambient_pressure` · `tens.tension_without_villain`
8. Avoid all of it in commemorative work — an anniversary film, a memorial, a community history made with the descendants in the room — where restoring the decision reads as second-guessing the dead. Avoid it where the record is too thin to evidence an alternative, because the restored decision then becomes a dramatised guess in documentary clothes. And do not run it under sixty seconds; the process question needs room to become a question at all

### "The middle of the sequence goes slack"

1. Close the available space, physically or socially
   `tens.closing_space` · `scene.trap_scene`
2. Remove options one at a time rather than adding obstacles
   `tens.shrinking_options` · `scene.escalating_cost`
3. Put a visible limit in the room
   `tens.countdown` · `scene.ticking_scene` · `tens.interruption_clock`
4. Make them wait, and make the waiting cost something
   `tens.waiting_as_weapon` · `tens.delay_by_digression`

### "He's hiding and it isn't tense"

The beat that needs no budget and no new information, and it fails on one of four rules every time.

1. The hiding place must be chosen on screen, from at least two visible options, and chosen *badly for a good reason* — a wardrobe picked because the bed is too low is a decision the audience can argue with, and a decision they can argue with is a decision they are inside
   `tens.hiding_beat`
2. Give the searcher a legitimate reason to be in the room that has nothing to do with searching. A man looking for his keys is far worse than a man looking for you, because the audience cannot predict where keys are
3. Keep the hider's whole body visible to the audience so they can inventory what might betray it, and never end on a clean pass — the searcher takes something or leaves something
   `tens.near_miss_economy`
4. To run it longer than about ninety seconds you need the second card: the pressure moves from the searcher to the hider's own body or possessions. Name the noise-maker before the hiding starts and let it be harmless once, give it a schedule the audience can anticipate, allow the character exactly one intervention and price it — and then *spend* it, because the noise must actually happen
   `tens.sound_you_must_not_make`
5. Not a phone with an obvious silent switch. Audiences have watched that fail for twenty years and will spend the beat thinking about the switch
6. And if the hiding place is genuinely secure, there is no beat. The audience works that out in four seconds and the rest is mime

### "He's in plain sight and I can't get any pressure out of that"

1. Concealment by behaving correctly among people is a different card from hiding, and it needs the crowd to have a *rule he can be seen to break* — a queue order, a dress code, a language, a direction of travel, a rhythm. An anonymous mass is not cover, it is set dressing
   `tens.crowd_as_cover`
2. Give him a task inside the crowd, because a person standing still in a crowd is the one thing everybody looks at
3. Make the searcher hunt a behaviour rather than a face. "He's looking for someone walking too fast" is a sentence that makes the audience police the protagonist's legs for the rest of the sequence
4. The exit must be through the crowd, not away from it — the moment he can simply leave, the crowd is scenery again
   `scene.public_scene` · `anta.the_crowd`
5. It dies if the audience does not know the crowd's rules, and it is deleted outright by a face-recognition premise. You cannot have both

### "It's relentless and the audience goes numb"

1. Tension needs release or it stops registering. Place the valve deliberately
   `tens.release_valve`
2. Comedy is a tension instrument, not a break from tension — it resets the ceiling
   `tens.comic_relief_as_tension`
3. Alternate scene and sequel so the audience can absorb
   `scene.scene_and_sequel` · `scene.reaction_dilemma_decision`
4. That is the complaint where the audience is still present and has stopped responding. Where they *leave* — a reader who abandons a draft with no identifiable fault in any single scene — the diagnosis is different and the repair is not a rest. See "Every scene is at ten and the reader gave up" below
   `tens.exhausting_draft`

### "The child has to be scared and then has to sleep"

**The release rule inverts here, and a writer who follows the general cards will get this wrong.** Everywhere else in this section the release beat is shorter than the tension and is cut early. In children's work it runs to completion, because completion is the deliverable. Read this card before applying `tens.trough_placement` or `tens.release_valve` to anything a child will watch.

1. **Softening the threat is not the mechanism and does not work.** A child reads a weak danger in about four seconds, stops paying in, and talks over the scene. Build the danger at full strength and bound it instead — with something the child can see the edge of: a stated time, a place the threat cannot leave, a rule it obeys, an ally within reach
   `tens.kids_bounded_threat`
2. Resolve inside the unit. The peril opens and closes inside the episode, the chapter, the song
3. Give the threat a want rather than an appetite for destruction. A thing that wants the ball, the cake, to be left alone has a stopping condition the child can picture and act against — which is why an adaptive threat is forbidden here and the negotiable and rule-bound forms are the native ones
   `tens.adaptive_threat` · `tens.negotiable_threat` · `tens.threat_with_rules`
4. Put a competent ally inside the frame before the peril starts — visible, named, reachable, and structurally forbidden to solve it. An ally who wins the scene is an unearned reprieve and the child stops paying in for good. The ally exists to make the ceiling visible, not the outcome
   `tens.unearned_reprieve`
5. **Then write the recovery beat**, which is the part adult readers cut as redundant and the entire reason the child sleeps. It runs after the plot has finished, contains one physical action and not only a line, and its length is the only checkable number here: roughly a tenth of the peril's running time, and never under about ten seconds on screen or a full paragraph on the page
   `tens.trough_placement` · `tens.release_valve`
6. The bands differ in checkable ways rather than in degree. **Preschool:** the resolution is shown rather than explained — the shadow revealed to be a coat, in frame — no clock and no dread in advance, the scare and its undoing in one continuous run inside ninety seconds, an ally physically present, no threat to a caregiver, recovery behavioural and social. **Five to eight:** anticipation unlocks, so a short visible clock, the bomb under the table and the hiding beat become available and are why kids' adventure gets good here; the threat may be genuinely real, the ally may be off-screen if named and reachable, and the recovery beat can begin to carry a cognitive line. **Middle grade:** monsters stop working and real dangers start, the bound moves off the threat and onto the arc, peril may run across chapters and the ending may cost something real — but the *emotional* beat must still close inside the unit, and a rescuing adult reads as the book not trusting them
   `tens.countdown` · `tens.bomb_under_table` · `tens.hiding_beat`
7. Whether the world can have jeopardy at all is decided upstream
   `char.species_logic`
8. The opponent side of the same design — the three bounds on the threat itself — is a separate card, and the two are written together
   `anta.bounded_threat` · `anta.preschool_problem`
9. The naming vocabulary the recovery beat borrows one line of, and where the beat finally lands
   `emo.kids_emotion` · `end.child_carry_ending`
10. Two failure directions. **Over-bounded:** you can name, at the moment the threat first appears, the specific thing that will undo it — which reads as being talked down to from about six upward
   `tens.diluted_stakes` · `tens.threat_never_acts`
11. **Left open** is owned completely by its own card, including its invisible half, which is a threat resolved in plot and never acknowledged in feeling. Read that one, do not re-derive it
   `anta.unresolved_kids_threat`
12. And none of it applies to a loss that does not resolve — a death, a separation, an illness at home. Bounding requires a frame that closes, and there is not one; the recovery beat applied to a permanent loss produces the story that says it is fine, which is the single false note children detect fastest

### "The surprise didn't land"

1. A surprise only pays if the material was planted and findable on a second pass
   `tens.earned_surprise` · `info.surprise` · `info.foreshadowing`
2. Surprise is fifteen seconds; suspense is fifteen minutes. Check you wanted the right one
   `tens.bomb_under_table`

### "I need the threat felt without showing it"

1. Put it behind the character and keep the camera or the prose on the face
   `tens.thing_behind_character`
2. Use a recurring sound as the carrier
   `tens.repeating_sound`
3. Reach the audience through a sense other than sight first
   `tens.smell_before_sight`
4. Those are single events kept off-frame in a piece that shows other things. The whole-work commitment — no sighting, no voice, no body, not once, the danger known entirely by what it leaves behind — is a different card and a harder contract
   `tens.inferred_threat`
5. Its four requirements are worth reading before you commit: evidence that is physical, countable and escalating in *category* rather than quantity; one piece of evidence a character explains innocently and is wrong about; never two characters agreeing about it on the page, because agreement converts an inference into a fact; and consequence anyway, or you have written `tens.dread_no_consequence` with better lighting
6. It is off the table if your climax is a confrontation. You cannot confront what you have spent ninety minutes refusing to render, and the substitution will be felt

### "The chase is exciting and weightless"

1. The chase must end differently from how it started — it needs a turn, not a duration
   `tens.chase_turn` · `tens.closing_pursuit`
2. Every obstacle must cost something; a near miss that costs nothing is decoration
   `tens.costly_obstacle` · `tens.near_miss_economy`
3. Let the ground itself apply pressure
   `tens.terrain_pressure`

### "The chase stops and so does the tension"

It should be the hardest point in the chase to look away from. A running audience is watching an outcome; a stopped audience is doing arithmetic about a room, and they cannot stop doing it.

1. This is a **junction card**: it owns the two transitions, motion into stillness and stillness back into motion, and nothing else
   `tens.pursuit_that_pauses`
2. **The stop must be the hunted's decision, not the hunter's failure.** The moment the hunter loses them you have written an escape and the sequence is over. The hunted stops because running has become the worse of two options, and the audience must be able to say why in one sentence
3. **The hunter must be competent and patient, and must prove it inside the pause**: one visible piece of correct reasoning the hunted can hear but not see — the radio turned off, the back of a hand against the kettle, the tap tested for a drip. A hunter who stops because he is lazy or thick downgrades the threat permanently
   `tens.near_miss_economy`
4. **In the pause the instrument changes from speed to inventory.** Chase prose is verbs and elision; pause prose is nouns, in three registers you need all of — what is physically in the room and its distances, what can be heard and how far away it is, and what the body is doing that the character is not choosing
5. **Budget the pause in facts, not seconds.** It dies the instant the audience stops learning, so plan about one new fact per twenty seconds of screen time or per half page. That puts the ceiling near ninety seconds on screen and three or four pages in prose
   `tens.waiting_as_weapon` · `tens.shrinking_options`
6. Break it through one of three doors, and they are not equal. **The hunted acting** is strongest, because it is a choice and it returns agency at the point they have least of it. **The hunter deciding** is second, and honest when the hunted genuinely has nothing left, provided the audience can reconstruct the decision afterwards from something they were shown. **A third thing arriving** is weakest for a structural reason rather than a taste one — the hinge is paid for by neither character, and any arrival that happens to favour the hunted is an unearned reprieve in costume. Its one legitimate use is making things worse for both parties at once
   `tens.unearned_reprieve`
7. Resume at a different level than you stopped at — faster because the hunter now knows, or slower into a stalk. The same speed tells the audience the pause was an intermission
   `seq.stalk_sequence` · `tens.chase_turn` · `tens.closing_pursuit`
8. Do not confuse it with the states on either side: a scene that is already a concealment when it opens and is still one when it closes, and the pressure inside that concealment. Hand the pause a second character and it has *become* one of those
   `tens.hiding_beat` · `tens.sound_you_must_not_make`
9. The public version, where the stop happens among people and the hunted has to keep performing
   `tens.crowd_as_cover`
10. Never as your opening — a cold-open chase is a promise of velocity, and an audience who does not yet know either party cannot value a stillness between them. Not in vertical or short form either, where a stop is a swipe
   `tens.micro_form_tension`
11. And the honest medium limit: in audio drama a pursuit that stops is indistinguishable from a scene that ended. One element must run continuously across the transition — the same rain, the same traffic, the breath from the run still in the room — or the listener loses the geography and spends the pause doing admin
   `tens.room_tone_silence`

### "The threat is vague"

1. Make it specific enough to be described by a character in one sentence
   `tens.specific_threat` · `tens.threat_scaling`
2. Decide whether it has a body or does not — they behave completely differently
   `tens.embodied_vs_formless`
3. A threat that adapts is far worse than one that is strong
   `tens.adaptive_threat`
4. And one you can bargain with is worse again
   `tens.negotiable_threat`
5. Refusing to name it is a choice, not a shortcut
   `tens.refusing_to_name`
6. And check whether the problem is definition at all. Stakes work when the audience is holding one specific image of the bad outcome, not when the piece has told them the stakes are high — a threat can be perfectly defined and still unpicturable
   `tens.picturable_consequence` · `anta.absent_stakes`
7. Two cautions on that. Cosmic horror runs on the opposite principle and belongs to the cards above. And do not over-specify a consequence you intend to avert: an image that precise becomes a promise, and averting it reads as a cheat

### "The threat never actually does anything"

1. The named failure, and the most common one in horror drafts
   `tens.threat_never_acts` · `tens.credibility_economy`
2. Dread with no consequence attached teaches the audience that nothing will happen
   `tens.dread_no_consequence`
3. Tension cannot be asserted by the narration; it has to be built
   `tens.asserted_tension`

### "I want a dream sequence and I'm told it will cost me"

You will pay, and the price is exact. The moment an audience learns that what they are watching may not be real, every later threat is met with an appraisal instead of a fear — *is this one real?* — and the appraisal is faster than the fear, so it always wins. Attention survives; investment does not, and the discount does not expire.

1. What having one does to everything around it
   `tens.dream_logic_cost`
2. Build the sequence itself — its grammar, its length, what may be in it — on the other card. That one makes it; this one prices it
   `time.dream_sequence`
3. **Pay for the licence up front, once, and never twice.** One declared breach of reality is a genre; two is a policy, and a policy means nothing on screen can cost anything
4. There are exactly three honest ways to keep tension inside unreality, and the first is worth more than the other two together: **make the dream's consequences real in the waking world** — wounds that persist, time that is gone, a person who does not come back. It is the only one that restores real consequence rather than managing the loss of it
5. **Or make the dream obey rules the audience can learn**, demonstrated on a minor character early, so they can do the frightening arithmetic themselves. Fear needs prediction, and a world where anything can happen is a world where nothing in particular is about to
   `tens.rule_demonstration` · `tens.threat_with_rules`
6. **Or announce the unreality before it happens rather than after.** That is hygiene rather than an engine — it prevents the theft, it does not generate stakes — but the audience who was told can be surprised, while the audience who was tricked can only be annoyed
7. Note precisely what *it was all a dream* destroys: not the scene it ends, but every scene before it, converted into a transaction the audience is told they should not have made. They apply the lesson forward, to your next work as well as your next act, which is why a strong final image cannot repair it
8. The neighbouring charges are smaller and different. A single jolt sold as a threat costs one unit of trust at a time and can be paid down by delivering later; this devalues the whole currency at once. A real danger that fails to spend itself is the opposite problem — here the danger spends itself freely and the spending does not register, which is why more violence makes it worse
   `tens.false_scare_cost` · `tens.threat_never_acts` · `tens.unearned_reprieve`
9. The interactive version is the same bill in a different currency: a level revealed afterwards to have been a simulation tells the player their choices were not choices, and no amount of later branching buys that back
   `inter.fail_state_as_story` · `inter.permadeath_writing`
10. **Sustained unreality is a different case and deserves an honest answer.** Surreal work where no waking frame is coming does not build suspense; it trades suspense for dread. Suspense needs *if this, then that*; dread needs only that something is wrong and that the character's condition is true. So hold the emotional stakes literal while the physical logic dissolves — the shame is real, the person who is gone is gone, the debt is owed. Lose that and you have dread with nothing attached
   `tens.dread_no_consequence`
11. And avoid the whole device in work that runs on plot suspense — thriller, procedural, survival, heist

### "Six people are in danger and I don't care about any of them"

1. That is `tens.diluted_stakes`, and one question to a reader settles it: *who were you afraid for?* A number in the answer is the defect; a name is the repair. Tension does not add up across a group, it divides
2. The repair is not to shrink the threat. Put one named person in front of it and keep them there for the whole sequence, so the audience is holding a specific image of the bad outcome rather than a figure
   `tens.picturable_consequence` · `anta.absent_stakes`
3. Where the group genuinely *is* the unit at risk — siege, heist, disaster, expedition, a single-location episode — the craft is concentration rather than distribution: at any given moment exactly one member carries the live risk and the others carry the decision about them
   `tens.ensemble_jeopardy`
4. Rotate the carrier once per sequence and never mid-scene. An audience can transfer its anxiety between people, but only at a seam
5. Generate the pressure from a *division*, because a group only shows its structure when something has to be split — who goes first, who gets the dose, who is told. And the working test: the audience must be able to say at any point which two people in the group disagree. If they cannot, you have a crowd rather than an ensemble
   `rel.group_under_pressure` · `scene.ensemble_scene`
6. The honest exception to the whole diagnosis is work whose subject *is* the mass event — testimony, documentary, memorial — where the number is the point. Even there the answer has the same shape: one named person standing in front of the figure, not instead of it

### "I need paranoia rather than danger"

1. Make the ally genuinely possible to doubt
   `tens.suspect_ally` · `tens.isolation_by_doubt`
2. Make the place itself untrustworthy
   `tens.untrustworthy_environment`
3. A lie that has to keep growing supplies its own escalation
   `tens.escalating_lie`
4. Where the doubt is about being observed rather than about who to trust, the complaint and the cards are separate — see "They can see everything and I can't write around it" below
   `tens.being_watched`

### "They can see everything and I can't write around it"

1. Start with the named failure, because it is usually what has happened: `tens.total_surveillance` — every camera working, every database joined, every device live — after which concealment is impossible and every escape the plot needs is an exception the writer grants. The repair is one page of worldbuilding nobody will ever see on screen: who operates the system, how many of them there are, and what they are doing at three in the morning
2. Specify the *gaps* as carefully as the reach, and note that the gaps are almost always human rather than technical — latency, jurisdiction, headcount, retention periods, the analyst with forty screens and a shift ending at six. A surveillance state with a published staffing level is frightening; one with infinite staff is weather
   `tens.threat_with_rules` · `anta.institutional`
3. The working engine is not the watching but the *impossibility of checking*. Confirm the observation exactly once, early and undramatically — one letter that quotes something said in a private room — then never confirm it again, keep the instrument visible and inert, and never show the watchers' room, because the moment you cut to three bored men at a desk the tower has an occupancy figure
   `tens.being_watched`
4. Then stage what it actually costs, which is not the arrest but the thing the character decides in advance not to do. Make the suppression a physical action with an object — a draft deleted, a book turned spine-inwards, a number not saved, a flag folded rather than binned — show the unsuppressed version first so there is a baseline, and give it a witness who does not remark on it
   `tens.self_censorship`
5. Never let that character explain why. The instant they articulate the fear it becomes a speech and the scene starts arguing instead of frightening
   `sub.silence_as_content`
6. And decide which engine you are on: once the audience is *certain* there is a watcher, you are running the bomb under the table, with the fear in the audience rather than in the character. You cannot run that and retreat to this
   `tens.bomb_under_table`

### "How much should I show?"

1. Implication usually beats depiction, and the aftermath usually beats the act
   `tens.implication_over_depiction` · `tens.aftermath_not_act`
2. Reach through a different sense
   `tens.sound_not_sight`
3. But explicitness is sometimes the honest choice, and the card says when
   `tens.honest_explicitness`
4. Check you are producing fear and not merely disgust — they are different responses
   `tens.disgust_not_fear`

### "I want dread about something real"

1. Scale and indifference, rather than malice
   `tens.scale_dread` · `tens.indifferent_force`
2. Dread about a condition rather than a monster — illness, debt, age, a system
   `tens.real_condition_dread`

### "The comedy and the tension are fighting"

1. A laugh placed correctly raises the ceiling rather than releasing the pressure
   `tens.laugh_raises_ceiling` · `tens.comic_relief_as_tension`
2. Placed early it spends the scene
   `tens.premature_laugh`
3. Farce runs on pressure too, and the mechanism is the same
   `tens.farce_pressure`

### "The tension doesn't hold across the whole thing"

1. Map it. Most drafts have never had their tension curve drawn
   `tens.tension_map` · `tens.trough_placement`
2. Place the false victory deliberately
   `tens.false_victory` · `scene.granted_wish`
3. Across a season, the pressure needs its own structure
   `tens.season_long_pressure`

### "The climax is over and there are twenty minutes left"

1. The fix is upstream of the problem and cannot be applied after it: plant the second question no later than the scene *before* the climax, and make it small and answerable — who tells the mother, where the dog is, whether the letter was posted
   `tens.post_climax_hold`
2. Change the kind of tension rather than the amount. Before the climax the audience is tense about an outcome; after it they should be tense about something they now know has to happen
   `struct.freytag_falling`
3. Set the length by the piece's emotional weight rather than by the size of the climax. Thrillers land in two scenes, family drama may want ten minutes, and getting that backwards is why big films end badly
4. Never introduce a new threat after the climax. That is the commonest attempt at this and it produces a different fault
   `end.too_many_endings` · `end.held_final_beat`
5. In short form there is no falling action to hold — the climax should be the last thing that happens. And some stories genuinely are complete at the peak, where the honest move is brevity rather than manufactured residue

### "Tension in a medium that isn't film"

1. Audio has no offscreen — everything is offscreen, which changes the whole approach
   `tens.audio_no_offscreen`
2. Prose has no soundtrack; rhythm does that work
   `tens.prose_no_soundtrack` · `rhythm.short_after_long`
3. On stage, the blackout is the instrument
   `tens.stage_blackout`
4. In interactive work the audience controls the pace, which removes your main tool
   `tens.interactive_tension` · `inter.pacing_without_control`
5. In audio there is also no silence. A true drop reads as a fault — the listener's first interpretation is that the file has stopped — so a dread pause runs on an audible bed of nothing, and the scare is a *subtraction* from it: a room going from three sounds to two is worse than anything you can add. Write the pause with a duration and a content, because "beat" will be performed as half a second by every reader of your script
   `tens.room_tone_silence`
6. In print comics the page turn is the only hard cut you fully control. Hook in the bottom-right panel of a right-hand page, payoff on the turn, and the pre-turn panel shows a character seeing something the reader cannot — but count your pagination first, because interior advertising destroys the odd/even guarantee and a turn planned for page 11 that lands on page 12 is simply gone
   `tens.page_turn` · `beat.panel_gutter`
7. In theatre the interval is a passage of the play that the audience writes. Go into it on a question with exactly two answers, break *before* the consequence of your biggest beat rather than after it, and answer nothing in the first three minutes back
   `tens.the_interval` · `tens.stage_blackout`
8. And in prose the deeper problem is not the soundtrack but the clock, which belongs to the reader. That has its own complaint below
   `tens.reader_owns_the_clock`

### "The subtitle gives away the line before the actor says it"

**Which subtitle card you want, decided here.** This one is for a **reveal, punchline or turn being spent early by the cue**. `rhythm.subtitle_beat` is legibility — can they read it in the time it is up, and does the break fall at a grammatical joint. `dial.subtitle_constraint` is line construction, writing the original knowing it will be read. `open.subtitled_open` is the first minute. Same 42-character box, different questions, and a cue can pass all three of those perfectly while failing this one completely.

1. The audience receives the turn twice — once silently, once late — and the second arrival is a formality. The laugh comes a beat early and dies alone, the reveal lands on an audience that has already adjusted, and the performance looks slow. They rarely blame the subtitle; they blame the actor, the edit, or the writing
   `tens.subtitle_preempt`
2. Assume pre-emption on anything that ships with subtitles, because the arithmetic makes it unavoidable. A seven-word English line is roughly 38 characters, read at the adult reading-speed ceiling in about 2.2 seconds and spoken in about 2.8 — add the lead-in that legitimately pulls a cue back to a shot change and the last word of the cue sits in the eye a full second before it reaches the ear. On a full two-line cue the gap is larger
3. **Four repairs, in order of cost.** Move the reveal to the end of the spoken line, so the subtitle physically cannot outrun it — a punch in the first three words is already gone
4. **Split the cue**, so the turn opens the second cue rather than closing the first. This is the cheapest fix and the one that survives translation
5. **Carry the turn on something a subtitle cannot pre-empt** — an action, an object entering frame, a face, a held silence
6. **And write the original knowing it will be read**
   `dial.subtitle_constraint` · `rhythm.subtitle_beat` · `open.subtitled_open`
7. **The case worth the most is the one where you write both channels** — an English voiceover with burned Arabic subtitles, same author. A translator inherits your word order and can only re-time; you can re-order the *information*, push the named fact later in the English so the Arabic cue has room, break the Arabic cue at your own breath rather than at a line-fill, and end both lines on the same noun so the two channels land together
   `rhythm.arabic_rtl_rhythm`
8. Do not mangle a good sentence to solve a spotting bug. If the fix is pushing a cue in-time to the frame of the punch word or splitting at the breath, a subtitler does it in ten seconds and no rewrite does it better. And do not pre-shrink for a translation that is not coming
9. In interview documentary, unscripted comedy and archive the line already exists and your only lever is the edit point. Know which part is yours
10. The same damage caused by the writing rather than the timing, and reveal order generally
   `comedy.audience_gets_there_first` · `info.staged_reveal`
11. The dubbing versions of the same problem, and what the unsaid and the claim underneath do across languages
   `rhythm.lip_sync_adaptation` · `lyric.singable_translation` · `sub.across_languages` · `comedy.translation_resistance` · `theme.translation_survival`
12. And what a subtitle does to a person rather than to a beat
   `char.smaller_in_this_language`

### "The reader can flip ahead and I can't stop them"

On a screen the author owns time; on a page the reader does, and every suspense device has to survive an audience with that much power. This is the first thing to read if you are a screen writer on a first prose draft, because the draft is almost certainly built on timing you no longer own.

1. Put the chapter break inside the unresolved moment. The break is the one pause you control, and a reader who has to turn a page to find out is a reader you have held
   `tens.reader_owns_the_clock`
2. Never run a wait longer than about a page. Film can hold eleven seconds of a man reading a file; prose cannot, because the reader simply reads faster and arrives at the end of your silence in four seconds looking for the next event
3. Keep the answer further away than the eye can reach — not on the same spread as the question, because the eye reaches the bottom right of the recto before the mind has finished the top left
4. Assume every reader skims description and none skim dialogue, so the load-bearing piece of dread goes in a line somebody says out loud
5. The sentence-level half of the same medium difference, and the longer wave
   `tens.prose_no_soundtrack` · `rhythm.book_level_pace`
6. All of it inverts for audiobook, where a listener can neither skim nor easily reread, and a structure built to defeat skimming lands as padding when narrated
   `rhythm.listener_no_rewind`

### "The short piece has no grip"

1. The first five seconds must establish the threat, not the world
   `tens.first_five_seconds` · `open.first_second_hook`
2. End on unresolved pressure rather than on an answer
   `tens.suspended_ending` · `dial.unanswered_line` · `scene.early_exit`
3. Under thirty seconds you cannot build stakes at all, so stop trying. Borrow them from a situation the audience already carries, and spend the entire runtime on escalation rather than on setup
   `tens.micro_form_tension` · `expo.short_form_frontload`
4. Which fails the moment the situation has no cultural shorthand. A borrowed clock the audience does not own is worse than no clock, because the piece then spends its whole runtime explaining and arrives nowhere

### "There's no deadline in this story and it still has to feel urgent"

Three different clocks that are not a countdown, and they do not combine — an audience holds one piece of arithmetic and drops the second.

1. Start with the one that needs no figure: a terminus the audience feels precisely and cannot measure — before someone finds out, before she changes her mind, before the money runs out, before the child asks again
   `tens.soft_deadline`
2. Or make the clock a person. Somebody with standing will come back and ask the same question, and the character knows roughly when. The requirement is that the asker can act on the answer; without power it is nagging, and the audience sides against them
   `tens.human_deadline`
3. Or charge rather than chase: a price that rises while nothing happens — interest, infection, a story spreading, a body in a room, a child getting older
   `tens.cost_of_waiting`
4. Never run two askers, and never run a meter beside a countdown. And an indefinite wait with no terminus at all is not a soft deadline; it is a different instrument doing a different job
   `tens.sustained_low_dread`
5. Heist, disaster and most action want a visible number, and a soft deadline there reads as a writer avoiding the work
   `tens.countdown`
6. The named failure this complaint most often turns out to be is `tens.announced_deadline_dropped` — a deadline stated out loud in act one and then neither met, missed, extended nor mentioned again. The repair is mechanical and takes an hour: list every stated time figure in the draft and write beside each one the page on which it is paid. Any figure with no page is deleted or paid
   `tens.asserted_tension` · `end.unpaid_setup`

### "The danger lifts and I can't say who lifted it"

1. That is `tens.unearned_reprieve`, and the diagnosis is two questions rather than the loose charge of deus ex machina. *Who solved it* — if the answer is not a character present in the scene, it is unearned. *What did it cost* — and note that foreshadowing does not rescue this, because plausibility buys credibility and only cost buys weight. A planted, entirely free deliverance is still free
2. The repair is mechanical: give the rescue to somebody already in the room and charge them for it. The person who saves her loses the job, the alibi or the relationship that saving her costs
3. The legitimate cousin is worth knowing, because it is free to use at any point: coincidence that makes things *worse*. Only coincidence in the protagonist's favour is a debt
   `end.deus_ex_machina` · `tens.false_victory`
4. A run of these produces a different complaint entirely — a threat the audience has learned will never do anything
   `tens.threat_never_acts` · `tens.near_miss_economy`

### "Every scene is at ten and the reader gave up"

The fault most often misread as pacing and then made worse, because the standard wrong response is to cut the quiet scenes for speed. Two checks, in this order, and both are procedures rather than repairs.

1. The **trough audit**. Draw the curve and look for the floor: if no scene in twenty pages drops below a six, the problem is structural, and the placement rule and the shape of the release beat are held on their own cards
   `tens.exhausting_draft` · `tens.tension_map` · `tens.trough_placement` · `tens.release_valve`
2. The **hope audit**, which is the one almost everybody skips. Write down what the reader currently *wants* to happen, as distinct from what they fear. If that list is empty, the draft is running on anxiety alone — and anxiety alone is exhausting and burns out, while anticipation can run for hours
   `tens.anticipation_vs_anxiety`
3. So the repair is not a rest, it is a want. Give the reader one outcome to root for and the identical amount of pressure becomes bearable, and sharper, because now it threatens something rather than merely continuing
4. Distinct from "It's relentless and the audience goes numb", which is an audience still present and no longer responding. This one is an audience that left
5. The honest exception is a short piece deliberately built as a single unbroken pressure — under about twenty minutes or forty pages — where there is no accumulated fatigue to relieve and a trough cut into it would simply be a hole
   `tens.micro_form_tension`

---