skills/ai-film-pipeline-master/references/phase-01-story-generation/cat-info-3.md

# Information Control, Reveal and Disclosure — Part 3

Category `info`, continued. Parts 1 and 2 are in `cat-info.md` and `cat-info-2.md` — read them first; their 70 cards cover who knows what and when, from dramatic irony and the knowledge ledger through plants, reveals, unreliable narration, secrets, the orders of evidence and the ways a reveal fails. Nothing here repeats them.

**Part 3 of 4.** 12 cards across 5 sections: compelled information; trust as a budget; sourcing, finding and attributing in public; the negative ledger; and one named failure.

These cards close gaps the database reported about itself: each was reached for by a kit or by the intents router and could not be cited because no card existed. Parts 1 and 2 hold everything the database knows about information *the writer chooses to move*. What was missing is the other half — information the writer is **required** to deliver, required to withhold, or required to get right in public, which is most of the information problem in factual and commercial work and almost none of the problem in fiction.

---

<!-- GENERATED:toc — do not edit by hand. Generated in the source package -->

## Contents

- [Compelled information](#compelled-information)
- [Trust as a budget](#trust-as-a-budget)
- [Sourcing, finding and attributing in public](#sourcing-finding-and-attributing-in-public)
- [The negative ledger](#the-negative-ledger)
- [A named failure](#a-named-failure)
- [Coverage](#coverage)
- [Sources](#sources)

<!-- /GENERATED:toc -->

## Compelled information

### The Disclosure Written as Content

**Also called:** the in-voice disclaimer, the honest ad-read, the paid-partnership line, the AI-use line, the native disclosure
**What it is:** A line the piece is legally or contractually obliged to carry — sponsorship, paid partnership, a political payer, a health qualifier, a declaration that something was generated or altered by a machine — written in the piece's own voice, at the moment it applies, so that it functions as information rather than as an interruption.
**Effect on the audience:** Relief, and then trust, in that order. An audience braced for a sales turn is released the moment you name it yourself, and the release is worth more than the concealment ever was. The inverse is the real finding: a disclosure the audience discovers rather than receives retro-actively re-reads everything before it, and they do not go back and re-sort the true parts from the paid ones — they discard the piece.
**Used for and where it works best:** Creator partnerships, UGC and influencer ads, political spots, sponsored segments, long-form direct response, and any factual piece carrying AI-generated or AI-altered material. Four craft rules, and they are enforceable by an audience even where no regulator is watching. **Disclose in the same channel as the claim.** If the endorsement is spoken, the disclosure must be spoken; a caption under a voiced recommendation does not reach the person who is listening with the screen face-down. A platform's own "paid partnership" tag is a supplement, never the disclosure — it is in the platform's voice, and the audience reads platform furniture as furniture. **Put it before the claim it qualifies, not after.** A disclosure after the pitch is a confession; the same words before it are a frame, and cost nothing. **Write it for the least attentive person who will see it** — the one who joined nine seconds in, at 1.5×, with the sound low. If it only works read carefully, it does not work. **Write it in your own grammar.** The whole of this card is that sentence: a compelled line is still a line, and a creator whose disclosure sounds like a lawyer has told the audience that a lawyer is now present. `expo.framing_text_furniture` covers the general behaviour of supers, cards and end-frames; this card is about the specific case where the text is not optional, which changes the craft entirely because you cannot solve it by cutting.
**Best in:** formats: UGC Ad, Advertising, Reels/Shorts, Vlog, Podcast, Carousel | genres: Lifestyle, Advocacy, Political, Journalism, Comedy
**Shelf life:** dated — review 2027-09. This is the fastest-moving card in the category. What must be disclosed, how conspicuously, in which channel and with which words sits in the FTC endorsement guides, the EU AI Act's transparency obligations, election-law payer rules and each platform's own synthetic-content labelling — and all four are being revised. The craft rules above are durable; the compliance floor is not, and it is lower than the craft floor anyway, so writing to the craft rule is also the safest way to pass a rule you have not read.
**Avoid when:** The relationship is so obvious that naming it is a joke at the audience's expense — a brand's own channel does not disclose that it is the brand. Also avoid making the disclosure the hook — where the whole piece is the paid content it is still the first sentence, as in the example below and `rhythm.disclosure_as_content`, but it frames the hook rather than replacing it. A creator who opens every video by announcing a sponsorship has turned an honesty move into a format beat, and the audience now discounts everything they say about anything. Disclose once, plainly, where it applies.
**Example:**
> Bolted on, in someone else's voice, after the pitch:
> "...and honestly I use it every single day. *[on screen, 1.2 seconds, 11pt: #ad — paid partnership]*"
>
> In the creator's own voice, before the claim, in the same channel the claim arrives in:
> "Right — Kayan are paying me for this one, and they're paying me whatever I say, which I made them put in writing because I'm about to be rude about the app.
> The scanner is genuinely the best I've used. The subscription is too expensive and I'd wait for a sale.
> Both of those things are in the ad. That's the deal I wanted."
>
> And the provenance variant, same principle, factual work:
> *Weak:* an end card reading "This programme contains AI-generated imagery."
> *Working:* NARRATOR, over the shot itself — "There is no photograph of the harbour before 1961. What you are looking at was built by a machine from the harbour master's drawings, and the machine got the cranes wrong; we left them in so you can see where it guessed."

**Yields to:** `expo.framing_text_furniture` — The on-screen text is optional, not compelled.  
`info.disclosure_as_content`

---

### The Fact You Are Not Allowed to State

**Also called:** the embargoed fact, the legalled line, the approval clause, the redaction, the bound narrator
**What it is:** Information the piece holds and is barred from delivering — by an NDA, a subject's contractual approval, a court order, a source-protection promise, a legal review, an embargo — and the decision to make the *bar itself* visible rather than letting the hole pass as silence.
**Effect on the audience:** A named bar is dramatic; an unnamed one is a gap they blame on you. An audience told "there is a thing here I am not permitted to say, and here is who is not permitting it" experiences the constraint as part of the story and often as its most alarming fact. The same absence undeclared reads as incompetence, evasion, or a film that ran out of money.
**Used for and where it works best:** Investigative reportage, true crime, sports and music documentary with subject approval, corporate and internal films, memoir involving living people, and any podcast that has been through a legal read. The craft rules are about converting a constraint into content. **Name the bar, its shape and its owner** — not "for legal reasons" but "the family agreed to talk on condition we did not name the hospital, and we agreed." Specificity is the whole transaction. **Say what it costs the piece**, out loud, once: which question you therefore cannot answer. An audience that knows the shape of the hole will not spend the rest of the film wondering whether you noticed it. **Never pretend the bar is an authorial choice.** Dressing an approval clause as artistic restraint is the version audiences catch, and it costs more than the fact would have. **Decide whether the bar is the story.** Often it is — a subject who has approval over a film about themselves is a more interesting fact than most of what the approval removed. Distinguish this sharply from `info.never_disclosed`, which is a designed permanent withhold the author chose and controls; here the author did not choose and does not control, and saying so is the technique. `info.cost_of_telling` covers what disclosure costs a character; this is what non-disclosure costs a piece.
**Best in:** formats: Documentary, Documentary Series, Reportage, True Crime, Podcast, Memoir | genres: Journalism, Crime, Political, Testimony, Workplace, Biography
**Avoid when:** Declaring the bar identifies the thing it protects. A source-protection promise announced too precisely is a source burned by arithmetic, and an audience that can name the withheld fact from the shape of your refusal has been given the fact. Also avoid where the constraint is ordinary and universal — every film has things it cannot say about everyone, and announcing that is self-importance, not transparency.
**Example:**
> The hole left as silence, which reads as a film that did not get the interview:
> NARRATOR: "The company declined to comment. In 2019, the plant closed."
>
> The bar named, shaped, and priced:
> NARRATOR: "We did get the company. Ninety minutes, on camera, with their head of operations.
> Their lawyers and ours spent four months on it, and what came out the other end is a signed agreement that we may use eleven seconds of it.
> So here are the eleven seconds. What they are not, and what I want you to hold onto, is an answer to the only question I asked twice: who signed off the September sample."
> *[Eleven seconds. He is adjusting his collar for nine of them.]*

**Yields to:** `info.anonymity_decision` — Naming the bar would identify the source.  
`info.embargoed_fact`

---

### The Piece With Nothing Filmed

**Also called:** wholly generated work, synthetic-origin declaration, the all-of-it case
**What it is:** The disclosure problem when there is no captured material at all. `info.disclosure_as_content` covers a compelled line attached to a claim, and `expo.disclosed_staging_note` covers the single re-staged shot inside an otherwise filmed piece — both assume a real recording somewhere that the label is measured against. A piece generated end to end has no such anchor, so the per-shot rule collapses: every shot would carry the label, which is the same as none of them carrying it.
**Effect on the audience:** Unlabelled, a wholly generated piece is not received as a technique — it is received as a record, especially on a heritage, historical or documentary channel where the audience's default is that a picture of a place is a picture of that place. The harm is not that they were told something false; it is that they were shown something and drew the ordinary conclusion.
**Used for and where it works best:** Decide it once, at the brief, not at delivery. **The test is what the frame invites the viewer to conclude, not how the frame was made.** A wholly invented world declares nothing — nobody believes a dragon was photographed, and a label on it is noise. A reconstruction of a real place, a real period or a real event declares, because the viewer's default conclusion is wrong and the piece caused it. **Declare once, at the top, in the piece's own voice, and then never again** — this is the exact inverse of `expo.disclosed_staging_note`'s per-shot rule, and the inversion is the card: when everything is generated, the spread-doubt-evenly failure disappears, because the doubt is accurate and even. **Say what it is rather than how it was made.** "No photograph of this room exists — this is how the excavation report describes it" is a claim the piece can defend and a viewer can weigh; "contains AI-generated imagery" is furniture, and the audience reads platform furniture as furniture. **A platform's origin flag is a supplement, never the disclosure**, for the same reason the paid-partnership tag is. And where the piece mixes generated frames with even one real photograph, this card stops applying and `expo.disclosed_staging_note` governs, because the mixture is what makes a per-shot label meaningful again.
**Best in:** formats: Heritage / Historical Documentary, Documentary Feature, Documentary Series, Explainer, Reportage, Museum Installation, Vertical Short-Form (Reels, Shorts, TikTok), Video Essay | genres: Heritage (civilisation-focused), Documentary, Educational, Historical
**Avoid when:** The piece is avowedly fictional and set in an invented world, where a declaration tells the audience something they already know and costs the opening its first beat. Also avoid treating this as a legal instrument: it is a craft rule about what a frame invites a viewer to conclude, and what any regulator or platform actually compels is somebody else's question and changes yearly.
**Shelf life:** dated — review 2027-03. What platforms and regulators require of generated content is moving fast and in different directions in different places; the craft rule above is durable, the compliance floor under it is not, and the floor is checked per project rather than remembered.
**Example:**
> *Weak, and the most common:* an end card reading "This programme contains AI-generated imagery."
>
> *Working, first ten seconds of a heritage piece, in narration:* "Nothing in this film was photographed. The city was destroyed twenty-six centuries ago. What you are about to see is built from what the digs found — and where the digs found nothing, I have said so."
**Yields to:** `expo.disclosed_staging_note` — the moment the piece contains any captured material at all, because a mixed piece needs the label at the shot rather than at the top.
`info.wholly_generated_piece`

---

### The Upload Switches: Made for Kids and the AI Label

**Also called:** audience setting, made for kids, the COPPA setting, AI use setting, altered or synthetic content, AI info label
**What it is:** Two settings the platform asks for at upload, each of which changes what the published piece can do. YouTube's audience setting says whether a video is made for kids. YouTube's AI use setting and Meta's AI info label declare realistic material that was generated or altered with AI.
**Effect on the audience:** The viewer never sees the setting; they see what it switches off (no comments, no end screen pointing to the next episode, no notification bell) or they see a label telling them a realistic scene did not happen.
**Used for and where it works best:** Every upload, decided at the brief, because both settings change how the piece has to be built. **Made for kids.** YouTube requires every creator, wherever they are, to set it, and says a video is made for kids when children are its primary audience or when its characters, songs, stories, activities or subject matter show an intent to target children; an animated kids' song meets that on its face. Once set, the video loses autoplay on home, cards and end screens, the video watermark, channel memberships, comments, the donate button, likes and dislikes on YouTube Music, live chat, merchandise and ticketing, the notification bell, personalised advertising, miniplayer playback, Super Chat and Super Stickers, and save to playlist or watch later; a channel set as made for kids also loses posts. So the handoff to the next episode lives inside the song or the last shot, never on an end screen. YouTube says its systems may override the setting and that an inaccurate one can bring action from the FTC or other authorities. **The AI label.** YouTube requires disclosure when AI makes a real person appear to say or do something they did not, alters footage of a real event or place, or generates a realistic scene that did not occur; it lists AI-generated music, and AI-generated footage of a real place, as needing disclosure, and fully animated or fantastical content and minor aesthetic edits as not. The label sits on the player for photorealistic content and in the expanded description for animated or non-photorealistic content; YouTube may apply it itself, including from C2PA metadata, and creators who repeatedly fail to disclose can face removal or suspension from the Partner Program. Meta shows "AI info" when it detects industry-standard AI indicators or when the person discloses, requires disclosure for photorealistic video or realistic-sounding audio that was digitally created or altered, may penalise failure, and may add a more prominent label where the risk of deceiving the public on a matter of importance is high.
**Best in:** formats: Kids Content, Animated Series, Music Video, Vertical Short-Form (Reels, Shorts, TikTok), Heritage / Historical Documentary, Explainer | genres: Children's, Educational, Heritage (civilisation-focused), Documentary
**Avoid when:** The question is what the piece itself says about how it was made; the platform box is a supplement to that line, and `info.wholly_generated_piece` writes it.
**Shelf life:** dated — review 2027-03. Both settings are platform policy and both have changed within two years; the lists above are the help pages as read on 2026-09-24 and are read again per project, not remembered.
**Example:**
> *Brief line for a kids' song channel:* "Audience: made for kids. So: no end screen, no cards, no comment prompt; the next-song tease is sung in the last chorus. AI use: yes (the song is AI-generated music); label expected in the description, the video is fully animated."
>
> *Brief line for a heritage reconstruction:* "AI use: yes (realistic generated footage of a real place). Expect the player label on YouTube and AI info on Meta; the narration's own declaration in the first ten seconds stays."
**Source:** reference — YouTube Help, "Set your channel or video's audience", https://support.google.com/youtube/answer/9527654, checked 2026-09-24; supports the list of features unavailable on made-for-kids videos and channels, the requirement regardless of location, the possible override, and the FTC risk. YouTube Help, "Determining if your content is 'made for kids'", https://support.google.com/youtube/answer/9528076, checked 2026-09-24; supports the primary-audience and intent-to-target tests. YouTube Help, "Disclosing use of GenAI content", https://support.google.com/youtube/answer/14328491, checked 2026-09-24; supports the three disclosure triggers, the listed examples, the label placements, automatic labelling including C2PA, and the penalties. Meta Newsroom, "Labeling AI-Generated Images on Facebook, Instagram and Threads" (2024-02-06) and "Our Approach to Labeling AI-Generated Content and Manipulated Media" (2024-04-05, updated 2025-10-23), https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/ and https://about.fb.com/news/2024/04/metas-approach-to-labeling-ai-generated-content-and-manipulated-media/, checked 2026-09-24; support the disclosure requirement for photorealistic video and realistic-sounding audio, the possible penalties, the AI info label and its triggers, and the more prominent label.
**Yields to:** `info.wholly_generated_piece` — The question is the piece's own declaration, not which box to tick at upload.  
`info.platform_upload_switches`

---

## Trust as a budget

### The Credibility Spend

**Also called:** the trust budget, the campaign-level credibility economy, deposits and withdrawals
**What it is:** Treating an audience's belief as a finite account held across a whole campaign or series rather than inside one piece — deciding in advance which pieces are deposits, which are withdrawals, and where the one claim that needs the whole balance is going to sit.
**Effect on the audience:** Invisible when done, and quietly fatal when not. Audiences do not track credibility consciously; they arrive at piece six already decided about whether pieces one to five were honest, and that decision is made of accumulated small impressions rather than of any single claim. A campaign that spends evenly has nothing left for the week it matters.
**Used for and where it works best:** Political campaigns, multi-part advocacy, documentary series, brand campaigns running longer than a quarter, and any creator's channel, which is the purest case — a channel is one long campaign with the receipts still visible. Three craft rules. **Name the withdrawal before you write anything else.** There is usually exactly one: the ask, the accusation, the price rise, the endorsement. Everything else in the schedule exists to fund it. **Deposits must cost you something visible** — a fact that does not flatter you, a competitor praised, a limitation named — because a deposit that costs nothing is not a deposit, it is a claim. **Never make two withdrawals in a row**, and put the smaller one first; an audience that has just been asked for something is at its least generous, and the second ask is where a campaign discovers its balance was lower than it thought. `tens.credibility_economy` is the card for this *inside a single work*, and it is the one to read for the mechanics of a single spend. This card is the schedule above it, which is where political, advocacy and serial work actually lives and which nothing previously covered.
**Best in:** formats: Documentary Series, Advertising, Series, Podcast, Brand Film, PSA | genres: Political, Advocacy, Journalism, Lifestyle, Crime
**Avoid when:** The work is a single piece with no successor. Budgeting across a campaign that does not exist produces a first piece that holds back its best material for an episode two nobody commissioned. Also avoid in any format where the audience arrives at a random point and leaves — a retail loop, a pre-roll — because there is no account to hold: every viewer is at balance zero and you must earn it inside the piece.
**Example:**
> A four-week advocacy schedule, planned as a ledger rather than as four films:
> **Week 1 — deposit.** The number that is against us. "Forty-one per cent of the people we surveyed said they had never heard of this and did not want to." We open on it. Nobody else in this fight will say that.
> **Week 2 — deposit.** The opposing lawyer, at length, uncut, and not made to look foolish.
> **Week 3 — small withdrawal.** One specific ask: read the eleven-page report. Nothing else. No donation, no signature.
> **Week 4 — the withdrawal the other three paid for.** The accusation, named, with the document on screen.
>
> Reversed — accusation in week one, evidence in week four — the same four films produce a campaign that spent a balance it had not yet opened, and week four arrives as a repeat rather than as proof.

**Yields to:** `tens.credibility_economy` — A single work with no successor.  
`info.credibility_spend`

---

### The Concession That Buys Belief

**Also called:** the damaging admission, the flaw named first, the honest weakness, "we're only number two"
**What it is:** A single real weakness admitted by the speaker, placed early and stated flatly, which converts the audience from grading claims to weighing an argument — because a source that will say one true thing against itself is assumed to be telling the truth about the rest.
**Effect on the audience:** Disarmament, and it is close to reflexive. The specific mechanism is that an audience's defence is calibrated against a speaker who is optimising; a speaker who has visibly failed to optimise in one place is no longer being modelled as an optimiser, and every subsequent claim is graded against a lower guard. It is the cheapest credibility in writing and the most commonly faked.
**Used for and where it works best:** Advertising against a stronger competitor, any pitch, advocacy on a contested question, memoir, first-person documentary, and the opening minutes of a piece whose thesis the audience will resist. Four craft rules, and three of them are about honesty because faking this is what kills it. **Concede something the audience can check.** An unverifiable weakness is a claim. **Concede it before they find it**, and preferably before they suspect it; a weakness admitted after it is raised is damage control and reads as such. **Put the concession structurally adjacent to the claim it protects** — the sentence after it, not in a different section — so the audience makes the trade in one breath. And **never concede a flaw that is secretly a boast.** "We care too much", "we're slower because we're thorough", "our only weakness is ambition" are the form's poison; an audience that catches one treats every other honesty signal in the piece as technique. If the flaw does not cost you anything, it is not a concession, and you would be better off with none. This is the *speaker's* version of the move. `anta.right_about_something` is the neighbouring card and is a different technique in a different category: there the *opposition* holds a true position the protagonist cannot answer, which is antagonist design and works on sympathy. Here the speaker concedes against themselves, which is credibility design and works on guard. They are frequently used together and should not be confused.
**Best in:** formats: Advertising, Brand Film, Documentary, Podcast, Memoir, Corporate | genres: Advocacy, Journalism, Political, Lifestyle, Testimony, Biography
**Avoid when:** The audience does not yet trust you enough to survive it. A concession from an unknown speaker is just a bad review; you need enough standing that the admission registers as costly. Also avoid where the conceded weakness is the audience's actual decision criterion — admitting your product is unsafe does not buy you belief about anything else, it ends the conversation. Concede on a second-order axis, and only on one.
**Example:**
> Fake, and instantly recognisable as fake:
> "If we have a weakness, it's that we're obsessive. We probably spend too long getting things right."
>
> Real, checkable, placed next to the claim it pays for:
> "We are the second-biggest firm in this city and we will still be second next year, because the biggest one has four times our people and better offices and we are not going to catch them.
> What we have is that you will get me. Not an associate with my name on the letterhead — me, on the phone, on a Sunday, because there are eleven of us and I know all eleven files."
>
> The first sentence is what makes the second one land. Cut it and the second sentence is what every firm in the city says.

`info.damaging_admission`

---

### The Correction

**Also called:** the retraction, the update, the correction notice, eating it in public
**What it is:** The published repair of a published error, written so that it is believed — which means it is a piece of information with its own disclosure problem, because a correction re-releases the wrong fact in order to kill it.
**Effect on the audience:** A well-written correction raises trust above where it was before the error. This is counter-intuitive, well-attested, and the reason the form is worth learning: audiences assume everyone makes errors and grade publishers on what happens next. A badly written one — quiet, passive, timestamped, ashamed — confirms the worst reading of the original mistake and is often more damaging than the mistake.
**Used for and where it works best:** Journalism and reportage, documentary updates, podcast follow-ups, creator channels, and any advocacy piece whose numbers turned out wrong. The craft rules are specific and each of them is routinely broken. **Correct in the same place, at the same size, in the same voice.** An error in a headline is not corrected in a footnote, and an error stated on camera is not corrected in a caption. **Say four things in order: what we said, what is true, how we know, and what we are doing so it does not recur.** The fourth is optional in a small correction and is the whole of a large one. **Decide deliberately whether to repeat the error** — this is a real editorial split, and the working rule is that you repeat it when it travelled, because readers who only met the wrong version need to recognise it, and you omit it when it did not, because repeating a fact is how a fact survives. **Use the active voice and name the agent.** "Errors were made in our reporting" is the single most trust-destroying sentence available in English; "I got this wrong" costs nothing and buys everything. And **never file it as an update**. Silently editing a live piece and appending a timestamp is a correction the audience is not permitted to see, which is indistinguishable from concealment by anyone who notices. `info.disproved_fact` covers a fact that turns out to be wrong *inside* a story; this is what you publish when it was wrong in yours.
**Best in:** formats: Reportage, Podcast, Documentary Series, Vlog, Essay, Carousel | genres: Journalism, Advocacy, Political, Science
**Avoid when:** The thing is not an error but a disagreement. Correcting a contested interpretation to satisfy a complainant teaches the complainant's side that pressure works and teaches your audience that your published positions are negotiable. Say you stand by it, or say what you got wrong; the hybrid — a correction that concedes tone while insisting on substance — reads as neither and is the commonest form of the failure.
**Example:**
> The version that makes it worse:
> *Updated 14 Sept. An earlier version of this article contained an error which has been corrected.*
>
> The version that is believed, with the wrong fact repeated because it had already travelled:
> **Correction, 14 September.**
> On Tuesday we reported that the contract was worth 4.1 million dinars. It was worth 410,000. We were out by a factor of ten and that number was in our headline for nineteen hours.
> The error was mine: I read a figure off a summary table without opening the contract, and the summary table was in thousands.
> The contract is linked below. From now on the figure in any story of mine comes from the document, not from anyone's summary of it, including the source's.

**Yields to:** `info.disproved_fact` — The error is inside the story, not your publication.  
`info.the_correction`

---

## Sourcing, finding and attributing in public

### Finding Rather Than Writing

**Also called:** the paper edit, the transcript search, pulling selects, the radio cut, finding the moment
**What it is:** Locating the moment that already exists — in rushes, a transcript, an interview, a livestream, a comment thread — as a distinct discipline from composing one, with its own search strategy, its own failure modes, and its own kind of judgement.
**Effect on the audience:** Nothing you write can produce the specific texture of a person arriving at a thought while being watched. A found moment carries a rhythm that scripted speech does not have and that audiences detect without being able to name, which is why a badly-found documentary with real moments beats a well-written one with bridges.
**Used for and where it works best:** Documentary of every length, interview podcasts, clip programming, reaction and highlight work, oral-history and heritage projects, and any piece assembled from material somebody else already spoke. The craft rules are search rules, and they are not obvious. **Search for turns, not for quotes.** The temptation is to hunt for the sentence that says the thing; what you actually want is the point where the speaker changes position, hesitates, corrects themselves or asks you a question back, and the good sentence is usually the *consequence* of that moment rather than the moment. **The clip almost always starts earlier than you think.** Roughly eight seconds before the quotable line is where the person begins to know what they are about to say, and that is the part an audience reads as truth. **Read the transcript for what the speaker does, not what they say** — a transcript hides the two things that matter most, the pause and the look, so mark timecodes on your first pass and never select from text alone. **Assemble the found material into a shape before writing one word of bridge**, because every bridge written early becomes a bridge defended late. And the discipline that is the whole card: **do not write a line you could have found.** A narrated summary of something a contributor said forty minutes in is a writer choosing their own sentence over a better one that already exists and is free. `rhythm.transcript_clean` covers what to do with the words once you have them; `info.interview_admission` covers the specific moment this search is usually hunting for; `info.doc_reveal_structure` covers where to put what you find.
**Best in:** formats: Documentary, Documentary Series, Interview Podcast, Reels/Shorts, Podcast, Reportage | genres: Journalism, Testimony, Crime, Heritage, Sports, Lifestyle
**Avoid when:** The material genuinely does not contain it. Searching forty hours for a moment that was never shot is the commonest way a documentary edit loses a month, and the honest call — that it must be written, or shot again, or that the film is about its absence — is usually available in week one and taken in week nine. Also avoid in any piece where the contributor has a right to review: a found moment is by definition one the speaker did not intend to give, and there is a consent question in it that a card cannot settle for you.
**Example:**
> Written, because it was easier than searching:
> NARRATOR: "For Samir, the decision to stay was not a simple one."
>
> Found — the same point, from the transcript, starting nine seconds before the quotable line, which is at 41:27:
> > **41:18** SAMIR: The shop was — no, sorry. *(long pause)* You want me to say I thought about it. **41:27** I didn't think about it. I want to tell you I weighed it up like a sensible man and I didn't. I stood in the door and my hands were wet and that was the whole decision.
>
> The sentence at 41:27 is the one a writer would have written. Everything before it is the reason anybody believes it, and no writer would have written "my hands were wet".

**Yields to:** `info.access_as_structure` — The moment was never shot; use the absence.  
`info.finding_the_moment`

---

### The Source on Screen

**Also called:** the attribution super, the citation lower third, the chyron, showing your working
**What it is:** Attribution treated as information design rather than as compliance — deciding, per claim, what the audience is told about where it came from, in what channel, at what moment, and in how much detail.
**Effect on the audience:** The specificity of an attribution is read as the strength of the claim, whether or not it actually is. "Reports suggest" and "a 2024 audit by the Ministry of Health, page 31" are the same assertion carrying different weights, and the difference is entirely in the super. Attribution is therefore not a footnote on the claim; it is part of the claim, and writing it as an afterthought silently downgrades everything it touches.
**Used for and where it works best:** Reportage, documentary, explainer and motion-graphics work, and short-form advocacy, where it matters most because the form has no room to argue and the super is the argument. Four craft rules. **Same frame as the claim, not after it** — an attribution that arrives once the statement has left the screen is an attribution nobody connected to anything. **Name the limit in the super where the source has one.** "Company's own figures", "one witness", "estimated" — a source described honestly in three words is stronger than a source described impressively, because the audience is grading you as well as it. **One super per claim.** A screen carrying four citations reads as a piece defending itself, and the audience's inference is that someone has objected. **Attribute in the channel the claim arrives in**: a spoken assertion needs a spoken attribution, since half the audience is not looking. The deeper design question the card exists for is *which rung each claim is being asserted on* — `info.credibility_ladder` is the card for that, and the super is where the rung becomes visible to the audience. `expo.onscreen_graphics` covers supers and kinetic text as a delivery surface; this is the specific case where the text's job is provenance.
**Best in:** formats: Documentary, Reportage, Reels/Shorts, Motion Graphics, Explainer, Carousel | genres: Journalism, Advocacy, Science, Political, Crime, Historical
**Shelf life:** dated — review 2028-09. Where a citation is expected to sit, how a platform treats on-screen source text, and what provenance an audience now demands of an image are all moving — the labelling of synthetic and AI-altered material is being written into both law and platform policy, and it will change what an attribution super is expected to cover. The craft rules are durable; the conventions around them are not.
**Avoid when:** The claim is the piece's own reporting. Attributing your own work to yourself on screen is odd, and doing it repeatedly reads as insecurity — your reporting is asserted, and the film's standing is the source. Also avoid dense citation in an emotional sequence: a super over testimony converts a person into evidence, which is a choice, and usually the wrong one at that moment.
**Example:**
> Weak attribution weakening a strong claim:
> V/O: "Waiting times have more than doubled."
> *[super: Source: reports]*
>
> Same claim, attributed in both channels, with the limit named in the super:
> V/O: "Waiting times have more than doubled — that's from the ministry's own audit, published in March."
> *[super, in frame with the sentence: Ministry of Health internal audit, Mar 2024, p.31 — figures self-reported by hospitals]*
>
> The last clause of the super costs seven words and is the reason a hostile viewer keeps watching.

**Yields to:** `expo.onscreen_graphics` — The super's job is delivery, not provenance.  
`info.source_on_screen`

---

### The Allegation You Cannot Verify

**Also called:** the unproven claim, the on-the-record accusation, publishing the thing you can't stand up
**What it is:** A serious claim the piece has good reason to think matters, cannot prove to its own evidentiary standard, and will not omit — published with the *state of the checking* attached, so that the audience receives the claim and its exact weight at the same time.
**Effect on the audience:** Handled well, an audience that grades the claim correctly and trusts the publisher more for having shown its work. Handled badly, an audience that believes the claim entirely, because an unverified allegation delivered without its weight is received as a verified one — insinuation is louder than assertion, not quieter, and every writer who has used "questions remain" to avoid a libel problem has in fact made a stronger accusation than the plain sentence would have.
**Used for and where it works best:** Investigative reportage, true crime, current affairs, advocacy documentary, and any oral-history project handling a communal accusation that no document survives for. Four craft rules, in order. **State the claim plainly, in the shortest sentence you have.** Hedged grammar does not reduce the accusation, it only removes the audience's ability to weigh it. **Name the source's position and their interest** — who says this, how they would know, and what they get from saying it. **Say what you did to check, and what the check produced**, including the checks that came back empty; "we asked for the log and were told it no longer exists" is evidence of a different kind and audiences know how to hold it. **Put the accused's answer in the same breath, not later**, and if there is no answer, say exactly how you asked and how long you waited. Then stop — do not argue the audience towards a verdict you could not reach yourself. This is the factual counterpart of two fiction cards that already exist and should not be conflated with it: `info.hearsay_evidence` is how a story uses unverified information as a *device*, and `info.rumour_chain` is how a story watches one distort in transit. Here the claim is real, somebody is accused, and the writer's problem is weight rather than effect. `info.credibility_ladder`, `info.contradictory_sources` and `info.disputed_history` are the three cards to read alongside it.
**Best in:** formats: Reportage, Documentary, True Crime, Podcast, Documentary Series, Essay | genres: Journalism, Crime, Political, Testimony, Advocacy
**Avoid when:** The claim's only support is its seriousness. "It is too important not to report" is the sentence that precedes most of the form's disasters, and it is not a standard — a claim you cannot weigh cannot be handed to an audience with the weighing attached, and publishing it anyway transfers your problem to a named person who did not agree to receive it. Also avoid the composite: three unverified claims presented together read as corroboration, and are not.
**Example:**
> Insinuation, which is the version that avoids a lawyer and accuses harder:
> "The minister has never explained the four days in March. Questions remain about what he knew, and about why the file was closed so quickly."
>
> The claim, its source, the state of the checking, and the answer, in that order:
> "Two people have told us the minister was in the building on 11 March. One is his former driver, who was dismissed in 2021 and is suing him. The other is a security guard who has nothing to gain and would not go on camera.
> We asked for the entry log. We were told it is deleted after three years, which is true and is also the department's rule.
> The minister's office says he was in Basra that week. They have not sent us anything that shows it, and we gave them nine days.
> That is everything we have. It is two accounts and no document."

**Yields to:** `info.hearsay_evidence` — It is a fiction device, not a real accusation.  
`info.unverified_allegation`

---

## The negative ledger

### What the Audience Must Not Be Able to Work Out Yet

**Also called:** the deducibility audit, the negative ledger, the solvability curve, inference control
**What it is:** The inverse of the knowledge ledger: a running account not of what the audience has been *told* but of what they can now *derive* — and the discipline of checking, at each structural seam, that the derivation you are saving is still out of reach.
**Effect on the audience:** Correctly managed, the sensation of being kept just behind, which is the whole pleasure of a mystery. Mismanaged in one direction the audience solves it in act two and spends act three waiting for the characters to catch up, which they experience as the story being slow rather than as themselves being early. Mismanaged in the other they cannot solve it at all and feel cheated (`info.the_cheat`).
**Used for and where it works best:** Every mystery, thriller and detective story, and any serial with a long-held reveal. The audit is mechanical and should be run at each act break or episode end, and it has four passes. **Derivation, not disclosure.** List what a clever viewer can now *infer* from what they have seen, not what they have been told; the two diverge fast, and it is the gap that gets stories solved early. **Elimination.** Count what is left. If your reveal depends on one of three people and two have been cleared, you have disclosed it without stating it, and this is how most whodunits leak. **The meta-channel**, which is where the leaks nobody audits happen: running time remaining, episode count, cast billing, an actor too well known for the part they appear to be playing, the thumbnail, the trailer, the fact that a character has had no scene alone. Audiences read all of it and writers audit none of it. **Structural symmetry** — a scene built to conceal something has a shape, and a viewer who notices you are avoiding a room knows what is in it. The repair is the useful part: **the fix is almost never to remove the clue.** Removing it breaks `info.fair_play_contract` and costs you the reveal's landing. The fix is to add a competing inference that the same evidence also supports, which is what `info.red_herring` and `info.double_meaning_clue` are for — you are not hiding the deduction, you are giving it a rival. `info.knowledge_ledger` and `info.information_map` are the positive-side cards this one inverts; `info.telegraphing` is what the failure looks like from the outside.
**Best in:** formats: Limited Series, Novel, Feature Film, Series, Vertical Micro-Drama, Podcast | genres: Mystery, Thriller, Crime, Detective, Psychological, Science Fiction
**Avoid when:** The story does not turn on a reveal. Auditing deducibility in a piece whose pleasure is dramatic irony is not just wasted work, it is actively harmful — `info.dramatic_irony` requires the audience to be *ahead*, and a writer running this audit on a tragedy will systematically remove the thing that makes it one. Also avoid running it too early; a first draft that is solvable is fine, and the audit belongs to the pass after the structure is fixed.
**Example:**
> The audit run at the end of episode four of six, written as it is actually used — a list, not a theory:
> **Can be derived:** the letter was written after the funeral (envelope's stamp, ep2) · Hana can drive (ep3, unremarked) · the house has a second staircase (seen in ep1, never used) · whoever made the call knew the dog's name.
> **Still safe:** that Hana was in the house on the 9th.
> **Leaking:** only Hana and her brother knew the dog's name, and the brother has been in Amman since ep1. **One inference short of the reveal, and two episodes early.**
> **Fix — not by cutting the dog's name**, which is the plant the finale lands on: give the name a second holder. Ep5, twelve seconds, the neighbour calls the dog from her gate, correctly, and nobody remarks on it. The deduction now has two ends and the audience must choose, which is the state we want them in.

**Yields to:** `info.dramatic_irony` — The piece runs on irony, not on a reveal.  
`info.deducibility_audit`

---

## A named failure

### The Bolted-On Disclaimer

**Also called:** the speed-read legal, the two-frame super, the end-card confession, compliance as furniture
**This is a named failure, not a recommendation.**
**What it is:** A compelled line — legal qualifier, sponsorship, payer identification, AI-use, a health caveat — treated as not-writing: dropped at the end, at minimum size or maximum speed, in a register borrowed from whoever demanded it, on the theory that the obligation is satisfied by presence rather than by reception.
**Effect on the audience:** They notice. That is the entire finding and it is the opposite of what the form assumes. A disclosure formatted to be missed communicates one thing perfectly — that the piece would have preferred you not to know — and it tells them that at the last moment, which is where an ending lives, so it is also the note the whole piece ends on. Worse, it retro-actively re-labels the preceding sixty seconds as an advertisement in the one second when the audience is deciding what they just watched.
**Used for and where it works best:** It does not — but it is worth knowing where it costs most, because it is not where compliance teams think. It is cheapest in broadcast advertising, where audiences already expect the furniture and discount it. It is ruinous in creator and UGC work, where the entire product is the impression that a person is telling you something, and where a bolted-on tag exposes not the sponsorship but the *performance of not being sponsored* that preceded it. It is nearly as bad in factual work, where a machine-generation or reconstruction line dumped in the end credits invites the audience to re-read every image they have just trusted.
**Best in:** formats: — | genres: —
**Avoid when:** Not applicable. The repair is `info.disclosure_as_content` and it is four moves: put the line **before** the claim it qualifies instead of after; deliver it **in the channel the claim arrived in**, spoken if spoken; **write it in the piece's own grammar** rather than the lawyer's; and **let it cost you something small and visible**, because a disclosure that is allowed to be slightly awkward is the one the audience believes. If the compelled wording is fixed and cannot be rewritten, the move is to say the true thing in your voice *and then* carry the fixed wording — the audience has already been told, so the furniture is now confirmation rather than revelation, and it lands as a formality instead of as a reveal.
**Example:**
> Bolted on — and note that everything before the last line is engineered to sound unsponsored, which is what the last line destroys:
> "I've been using this for about three months now and I genuinely don't think I'd go back. Anyway — link's below, that code still works, love you all, bye.
> *[0.8s, 9pt, bottom left, over the outro music: Contains paid promotion]*"
>
> Repaired, same obligation, same words available to the compliance team, placed where it works:
> "This one's an ad — Halab paid me, they saw this script, and they asked me to cut one line and I cut it, which I'll tell you about at the end.
> Three months in, I don't think I'd go back. Here's why, and here's the bit they didn't want. *[super, held, legible: Paid promotion — Halab]*"

**Yields to:** `info.disclosure_as_content` — The stated repair for the failure.  
`info.bolted_on_disclaimer`

---

## Coverage

**Two items on the brief were already written, and are not written again.**

**The open secret has a card**, and a good one: `info.open_secret` in `cat-info-2.md` treats it exactly as the brief asked — as an information-control problem rather than a subtext one, with the maintenance cost of the pretence as its craft rule and the requirement that the audience be told early and flatly. Nothing useful remained to add. The slot was spent instead on `info.embargoed_fact` — the fact the piece holds and is *barred* from stating, by an NDA, an approval clause, a court order or a source promise — which `formats-doc.md` recorded as a gap under "the approval clause" and which nothing in the category covered. It is the true opposite of `info.never_disclosed`: there the author chose the hole and controls it; here the author did not and does not, and saying so out loud is the whole technique.

**The rumour has two cards**, `info.rumour_chain` (motivated distortion through a community, with the original shown first so the audience can measure the drift) and `info.hearsay_evidence` (the cheap plant that can be promoted later). Between them the fiction side is complete. What was genuinely missing is the factual side, which is a different problem entirely — not a rumour moving through a story but an allegation moving through a publication, where somebody real is accused and the writer's job is to attach the *weight* rather than to manage the effect. That is `info.unverified_allegation`, and its central finding is one the brief did not anticipate: hedged grammar accuses harder than plain statement. "Questions remain" is a stronger allegation than the sentence it avoids, which means the careful-sounding version is the reckless one.

**The concession is not `anta.right_about_something`, and the difference is load-bearing.** That card is antagonist design: the *opposition* holds a true position the protagonist cannot answer, and it works by complicating sympathy. `info.damaging_admission` is the *speaker* conceding against themselves, and it works by lowering guard — the audience stops modelling you as an optimiser. Different actor, different mechanism, different category, frequently used in the same piece. Both cards now say so.

**What else the research showed.**

**The disclosure card is four kits' most-wanted for a reason the kits did not state.** the commissioned-advert material (`symptoms-commissioned-and-applied-work.md` here, and `../phase-03-shooting-script/ads-trailers.md`) and [`formats-social.md`](formats-social.md) both record it, and social noted that it "bites harder here". The reason is now in the card: in creator work the product *is* the impression that a person is telling you something, so a bolted-on tag does not expose the sponsorship, it exposes the performance of not being sponsored that preceded it. That is why the failure card exists as a separate entry rather than as a paragraph — the defect is a distinct, diagnosable thing with its own repair, and `formats-social.md` will want to rank it as a standing diagnostic in the creator-partnership kit.

**The AI-use line did not need its own card.** It behaves identically to every other compelled line — same channel, before the claim, in your own grammar — and the only thing that differs is which authority compels it. It is written as a variant inside `info.disclosure_as_content`, with the shelf-life field carrying the volatility, and `expo.disclosed_staging_note` in `cat-expo-3.md` holds the per-shot factual case where the declaration is about a *picture* rather than about a payment. Splitting it would have produced three cards with one craft rule between them.

**The negative ledger's real leaks are outside the text.** Writing `info.deducibility_audit` turned up the thing the brief's framing would have missed: the channel that actually gives reveals away is not the prose but the meta-channel — running time remaining, episode count, cast billing, an actor too famous for the part, a character who has had no scene alone. Audiences read all of it and writers audit none of it, and the audit in the card lists it as its own pass for that reason. The second finding is the repair: the fix for a leaking deduction is almost never to cut the clue, because cutting it breaks `info.fair_play_contract` and costs you the landing. The fix is a rival inference the same evidence also supports.

**Not written, because it lives elsewhere.** The permanent designed withhold is `info.never_disclosed`. The orders of evidence are `info.credibility_ladder`. Trust as a budget *inside one work* is `tens.credibility_economy`, which `info.credibility_spend` sits above rather than replaces. A fact that turns out to be wrong inside a story is `info.disproved_fact`, and a story's own rewriting of its past is `info.the_retcon`; neither is `info.the_correction`, which is what you publish when the error was yours. Transcript language is `rhythm.transcript_clean`.

**Still missing after this file**, named in prose above with no id because no card exists: **rights-led structure**, designing a cut around what can be licensed, which `formats-doc.md` records and which three kits needed; **re-consent as a scene**, the late conversation with a contributor filmed inside the film, recorded by the vérité, diary and family kits; **writing the reply as part of the post**, where the published object includes the author's own replies, recorded by `formats-social.md`; and **the platform contract that expires** — the discipline of separating the part of a format that is craft from the part that is a current spec, which every dated card in this database is a symptom of and which nothing yet teaches.

Named failures in this file: `info.bolted_on_disclaimer`.

*Build notes from when this file was drafted — targets, running totals and what remained to be written — are kept verbatim in the [coverage history](../tools/history/coverage-history.md). They were true of the draft, not of the file, so they no longer sit here as if they were measurements.*

## Sources

Verified 2026-09-17.

- [FTC's Endorsement Guides: What People Are Asking — Federal Trade Commission](https://www.ftc.gov/business-guidance/resources/ftcs-endorsement-guides-what-people-are-asking)
- [16 CFR Part 255 — Guides Concerning Use of Endorsements and Testimonials in Advertising (eCFR)](https://www.ecfr.gov/current/title-16/chapter-I/subchapter-B/part-255)
- [2026 FTC Influencer Disclosure Rules — Launchpoint](https://www.launchpointhq.com/blog/ftc-influencer-disclosure-guide)
- [FTC AI Disclosure Rules for Creators, 2026 Guide — Promise Legal](https://blog.promise.legal/startup-central/ftc-ai-disclosure-rules-creators-2026/)
- [Article 50: Transparency Obligations for Providers and Deployers of Certain AI Systems — EU Artificial Intelligence Act](https://artificialintelligenceact.eu/article/50/)
- [The EU AI Act's Transparency Rules: A Practical Guide to Article 50](https://artificialintelligenceact.eu/transparency-rules-article-50/)
- [Code of Practice on Transparency of AI-generated Content — European Commission](https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content)
- [Advertising and Disclaimers — FEC.gov](https://www.fec.gov/help-candidates-and-committees/advertising-and-disclaimers/)
- [FEC Adopts New Rules for Sponsorship Disclaimers for Online Political Advertising — Broadcast Law Blog](https://www.broadcastlawblog.com/2022/12/articles/federal-election-commission-adopts-new-rules-for-sponsorship-disclaimers-for-online-political-advertising-and-to-consider-rules-for-political-marketing-through-social-media-influencers/)
- [Best Practices: Corrections — Ethics and Journalism, NYU](https://ethicsandjournalism.org/resources/best-practices/best-practices-corrections/)
- [Make corrections a building block of trust: a step-by-step guide — Trusting News](https://medium.com/trusting-news/make-corrections-a-building-block-of-trust-with-our-step-by-step-guide-3c7e96645a5c)
- [Be loud about mistakes — and how you correct them — Trusting News](https://trustingnews.org/be-loud-about-mistakes-and-how-you-correct-them/)
- [How to correct website and social media errors effectively — American Press Institute](https://americanpressinstitute.org/digital-corrections/)
- [Corrections policies in journalism — Britannica](https://www.britannica.com/topic/corrections-policy)
- [Explain Yourself: How Corrections and Epistemic Explanations Impact the Public's Trust in Science Reporting — Journalism Practice](https://doi.org/10.1080/17512786.2026.2693188)
- [Attribute All Sources — *Be Credible*, University of Kansas open textbook](https://opentext.ku.edu/becredible/chapter/attribute-all-sources/)
- [Lower Thirds, Chyrons, Banners: What Are They? — NBCU Academy](https://nbcuacademy.com/chyrons-lower-thirds/)
- [What is a Chyron — Definition, Examples and How to Write Them — StudioBinder](https://www.studiobinder.com/blog/what-is-a-chyron-definition/)
- [Elevating the Paper Edit for Documentary-style Filmmaking — Reduct](https://reduct.video/blog/paper-edits-for-documentary-filmmakers/)
- [How to Build String-Outs and Selects for Documentary Editing — Aspect](https://aspect.inc/blog/post-production/how-to-build-string-outs-and-selects-for-documentary-editing)
- [Stages of Documentary Post-Production — New Doc Editing](https://newdocediting.com/stages-of-documentary-post-production/)
- [Cutting with Curiosity: Inside a Documentary Editor's Workflow — RedShark News](https://www.redsharknews.com/cutting-with-curiosity-inside-a-documentary-editors-workflow)
- [Ofcom Broadcasting Code, Section Two: Harm and Offence](https://www.ofcom.org.uk/tv-radio-and-on-demand/broadcast-standards/section-two-harm-offence)
- [Avis "We're only Number 2" Ad Campaign — Swipefile](https://swipefile.com/avis-were-only-number-2-ad-campaign)
- [Avis: "We're Only No. 2. We Try Harder." — CrazyAds](https://www.crazyads.org/campaigns/avis-were-number-two)
- [Self-deprecation — Wikipedia](https://en.wikipedia.org/wiki/Self-deprecation)
- [Direct Response Copywriting: Classic Examples and Tips — Crazy Egg](https://www.crazyegg.com/blog/direct-response-copywriting/)