.claude/skills/video-script-production/references/texture.md

# Texture: personality, rhythm and the human element (Lessons G, H, I)

Added 2026-10-25 after the Moser Baer v5 review. The user's brief: the script is good, but it needs (G) a personality that suits the whole channel, is both **earnest** and **aloof**, and explains with metaphors, similes and parables; (H) a rhythm that distributes hard data moments and easy storytelling moments; (I) a human element: personifying, finding the personal stake of the viewer or of a character we care about. The same review asked why the hook's last line ("So let's start with the city he left") gave the viewer no reason to care.

Read this once per script, at Step 3. The mechanical checks are rules G1 to G4, H1 to H5, I1 to I6 and B8 in `lint.js`; thresholds are in `rules.json` under `persona`, `figures`, `rhythm`, `human` and `hook`.

## How these three fit the pipeline: one plan, one additive pass, then the subtractive pass

Three options were considered.

1. **Three more passes in a row** (a metaphor pass, a rhythm pass, a human pass). Rejected: each rewrites what the last one set. A figure adds words and breaks the rhythm; moving a number to make rhythm moves the person who was standing next to it; adding names where the figures sat changes the load. Three passes mean three rounds of rework and a script that sounds assembled.
2. **Fold everything into the writing step.** Rejected: without a plan, the persona becomes a sprinkle of jokes and the people become decoration.
3. **One plan, written with the chain; one additive pass; then the existing subtractive pass.** Chosen. The three lessons share one unit, the beat as a *texture sandwich* (a person or picture, a claim, a number, a landing), so they are decided together in the chain (Step 3, item 7), written together (Step 4), checked together in one **texture pass** (Step 4a), and then the **voice pass** (Step 4b) takes out whatever sounds written. Order matters: *add, then subtract, then lint.* The texture pass adds words and the voice pass cuts the tells those words may bring.

The plan **is** the chain: `persona`, `figures`, `humans`, `viewer_stakes`, `hero_numbers[].landing_quote` and `cdl.hook_elements` are the plan, and the linter checks that every declared quote is in the text.

## What the evidence supports (and where it does not)

Grades: **A** = replicated or meta-analysed, close to our situation; **B** = reasonable evidence, different setting; **C** = craft knowledge or a contested effect; **D** = do not rely on it.

| Claim we lean on | Source | Grade | What we take from it |
|---|---|---|---|
| A metaphor persuades a little overall (r about .07) and a lot under good conditions (up to r about .42): one metaphor, novel, with a familiar target, early in the message | Sopory and Dillard 2002, meta-analysis (seen through a summary; moderator detail from a secondary source) | **B** | One governing metaphor, early; not a figure of speech every sentence. More metaphors do not add up |
| A metaphor frames what solution people propose ("crime is a virus" vs "a beast") | Thibodeau and Boroditsky 2011; a later study with a non-metaphor control (Steen, Reijnierse and Burgers) found no framing effect | **C**, contested | Choose the metaphor whose implications you can defend with the evidence, and do not expect it to steer viewers unseen |
| In an analogy only the relations transfer, not the objects; teachers should state the limits | Gentner, structure-mapping; Brown and Salter 2010 | **B** | Name what the picture maps, and say aloud where it stops working (our treadmill: "has a stop button; a bond with a date doesn't") |
| Refutational two-sided messages (state the other side, then answer it) persuade best | Allen 1991 and O'Keefe 1999, meta-analyses (seen through secondary sources) | **B** | The aloof half of the voice: give the case for and the case against in the same breath |
| A competent person who admits a small slip is liked more | Aronson 1966 (pratfall effect); replications mixed | **C** | Honest "I couldn't find" lines are cheap; competence has to be shown first. Not a reason to fake imperfection |
| Presenter persona changes learning or trust in explainer channels | We found no experimental study of narrator persona in explainer videos; the parasocial and credibility papers found concern travel and shopping | **D** | The persona is a design bet, not a finding |
| Narrative exposure moves beliefs, attitudes and intentions a little (r about .17 to .23) | Braddock and Dillard 2016 meta-analysis | **B** | Stories help. But statistics beat stories on beliefs and attitudes and stories beat statistics on intentions (a 2015 meta-analysis), so use both: a number, then a person |
| One named victim raises giving more than a statistic | Identifiable victim effect: Lee and Feeley 2016 found r about .05; a 2023 re-analysis and a 2024 replication found about zero | **D** | Do not design on "one face beats the numbers". Use a person for their *want* and *stake*, not as a tear-jerker |
| "Mirror neurons" explain why stories make us feel with a character | Hickok 2014 (critique): their role in empathy and understanding is unproven | **D** as a mechanism | We call the craft *perspective-taking*, make no neural claim, and do not name mirror neurons in a script |
| Listener brain activity aligns with the speaker's when communication works, and anticipates the speaker | Stephens, Silbert and Hasson 2010, small fMRI study | **B-** | Clear, predictable, concrete narration helps shared understanding. Nothing here about personality |
| Material related to the self is remembered better | Rogers, Kuiper and Kirker 1977; Symons and Johnson 1997 meta-analysis | **A** for memory | "You" lines, the viewer's own memory (the disc) and own decision (your next move) |
| Pacing and arousing content raise attention and encoding up to a point, then overload cuts them | Lang's LC4MP; a 142-article meta-analysis supports its load, motivation and memory domains | **B** (mostly TV and lab) | Alternate demanding stretches with easy ones; measure the stretches |
| A story is anecdote plus reflection; a talk moves between "what is" and "what could be" | Ira Glass (anecdote, bait, reflection); Nancy Duarte (sparkline) | **C**, craft | The hard-moment/easy-moment alternation is the same idea in a measurable form |
| The end and the peak weigh most in remembered experience (peak-end rule) | Fredrickson and Kahneman 1993, from memory; not re-checked this session | **B** | The close is quiet and human |
| Winners in our own corpus | 20 "why X died" teardowns: S-VILLAIN 13/20, S-PROTAGONIST 8/20, S-SELF-REF 8/20, S-SCENE 8/20, S-ANALOGY 9/20, S-PERSONA 6/20. The 8 English transcripts we measured have a median of 0.73 comparison markers per 1000 words (range 0 to 2.3, about one every nine minutes) and a median of zero first-person admissions per 1000 words (v6: 1.7 and 2.4); they swap between number-heavy and quiet stretches about every 70 seconds | counts only | **Winners do not need these things.** Presence is not quality, and 8 videos is a small sample. v6 deliberately differs from them on figures and persona |

So: the strongest supported moves are *one early, apt metaphor*, *two-sided fairness*, *a number followed by a person*, *self-reference* and *alternation*. The weakest are the persona itself, the single-victim effect and anything that names mirror neurons.

## Lesson G: a personality that travels, and figures that explain

### The persona: earnest about facts and people, aloof about verdicts and heroes

The brief asked for two things that look opposed: **earnest** (willing to follow leads, to research, to be honest) and **aloof** (does not care who wins, finds the negative and the positive on every side, unbiased). They stop being opposed once you see that they point at different things.

| | Points at | Sounds like | Never |
|---|---|---|---|
| **Earnest** | the evidence, the gaps, the people | "reading the rating notes and the creditors' claims"; "I couldn't find what it earned"; "I went looking for the figure some videos quote and couldn't find it in any record"; gives a worker his real salary | a claim of effort that was not made; performed humility |
| **Aloof** | verdicts, heroes, sides, the hype | "I'm not going to referee Bengal's politics"; "I don't have a favourite theory, I have the company's own statement"; "Nobody in this story needed to be a villain"; "I don't need it to be a scam, and I won't pretend it is" | outrage; sarcasm at people; aloofness toward the viewer, the facts or the workers |

The test of the combination: *the narrator is never aloof about the person who paid, and never earnest about picking a side.* Aloofness is what lets the script be fair to a founder who may be disliked; earnestness is what stops the fairness from becoming a shrug.

It travels across a channel because it is defined by a **stance**, not by a catchphrase or a topic. The same two lines work on a bank, a cricketer or a dam. A recurring ritual is allowed and encouraged: a short "what I'm sure of, what I'm not" near the end (the Moser Baer script's scam beat does this), because a repeated move is what makes a voice recognisable. The persona's "I" must be true of whoever records the script. Rules:

- **G1.** At least three earnest and three aloof lines, each in at least three beats, the first by 35% of runtime and the last after 65%, so the voice is there from early on and does not fade.
- **G2.** At least two contested calls get both sides in the same beat (`persona.sides`): the case for and the case against. The two-sidedness is the evidence-backed part (Allen 1991).

### Figures: one governing metaphor, a few glosses, a parable

The brief asked to use metaphors, similes and parables "whenever possible". The evidence says the opposite about density, so the policy is **whenever it earns its place, and no more**:

| Kind | What it does | Budget |
|---|---|---|
| **Governing metaphor** | One picture that carries the thesis and returns as a lens (Moser Baer: the treadmill) | Exactly one, introduced by 45% of runtime, returning at least three times, with **its limit stated aloud** once |
| **Gloss by analogy** | Makes a hard word concrete the first time it appears (a bond is a tab you only see at closing time; recasting is asking the landlord to rewrite the lease; liquidation is a house's contents at a forced auction) | At least three, each tied to a declared jargon term and sitting at its gloss |
| **Parable** | A short hypothetical scene that lets the viewer feel a mechanism before the numbers (the tea lane; the restaurant queue) | At least one; 25 to 130 words; opens "say you" or "imagine"; **no figures inside it** |

How to choose the governing metaphor: (1) it comes from the viewer's everyday world; (2) its *entailments* (what follows if the picture is true) match the evidence (a treadmill: effort without progress, the pace set by others, falling if you stop, a faster belt = a faster market; all true of a price war); (3) you can name where it fails, and the failure is itself an insight (a treadmill has a stop button; a bond with a date does not); (4) it is not a cliché the audience has stopped hearing ("perfect storm", "house of cards", "ticking time bomb" are on the stock list). Do not mix worlds in one beat. Do not put a number inside a picture.

- **G3.** Governing metaphor early, returning, with a stated limit; glosses tied to jargon; at least one parable (declared in `chain.figures`).
- **G4.** Budget over the whole script: 4 to 16 comparison markers, at most 2 in any minute, at most 2 stock figures, at most 2 verdict words ("greedy", "reckless", "villain": the script should not pass judgement it cannot prove).

## Lesson H: rhythm, the alternation of hard and easy

A **hard moment** is a claim with figures the viewer must hold. An **easy moment** is a person, a picture, a parable or a plain statement. The viewer's attention is a budget (Lesson A); rhythm is how it is spent and refilled.

The model: inside a beat, *person or scene, then claim, then number, then landing*. The **landing** is the sentence after a hero number that turns it into something felt (a person, the viewer, or a picture) and contains no further figure. Across beats, the heavy beats are followed by a human one.

How we measure it (text only): cut the script into 30-second windows; a window is **hard** if it holds two or more spoken quantities and **quiet** if it holds none.

| | Winners (8 English transcripts) | v5 | v6 |
|---|---|---|---|
| Hard-window share | 9% to 67%, median about 30% | 37% | 26% |
| Quiet-window share | 17% to 74%, median about 46% | 46% | 46% |
| Longest run of hard windows | 1 to 13, median 3 | 3 | 2 |
| Swaps between hard and not-hard per 10 min | 2.6 to 12.2, median about 7 | 9.3 | 9.2 |

Thresholds were first drafted stricter (a run of more than 2 would warn, which would have flagged 4 of the 8 winners) and then loosened to the winners' own range; the rule now warns above 3 and fails above 6. This is a calibration to eight videos, not a law.

- **H1.** At most 3 hard windows in a row (fail above 6). **H2.** Hard windows at most 40%, quiet windows at least 30%. **H3.** No two consecutive beats at 4 or more quantities a minute. **H4.** Every hero number is followed, within 45 words, by a declared landing with no figure and a person, the viewer or a picture in it. **H5.** The close has at most 2 figures and a person or the viewer in it.

## Lesson I: the human element

The brief: personify; find the personal stake, either the viewer's (implied) or a character's we care about; this "activates mirror neurons" and gives the story an emotional base. We keep the design and drop the neural claim (grade D as a mechanism). What can be built and checked:

| Element | What it is | Moser Baer v6 |
|---|---|---|
| **Protagonist** | A person with a **want** and a **flaw**, named before any pronoun stands for them | Deepak Puri: the landowner's son who wanted no part of the land; the flaw is the playbook that worked three times |
| **Counterpart** | A second person whose words carry the thesis | Ratul Puri: the cost-game line, "never more excited" in 2003, "well-funded" in 2011 |
| **Witness** | The person who pays for the decisions; introduced early and back at the close | Mahesh Sharma: 17 years inside, union president, the pay-cut offer, the shop in Dadri |
| **The viewer** | Spoken to ("you") in the hook, the middle and the close; a mirror ("if you've ever been the one in your family who wanted something different") and a use (the three questions) | The disc in your hand; your own next move |
| **A concrete detail** | One ordinary, specific thing per person (an object, a place, a job title) | "a junior job at the oil company Esso"; "a small electrical shop in Dadri" |
| **A stake line** | What the person stands to lose or has lost | "he had no patience left for it"; "he didn't clear the HR round" |

The failure that triggered the lesson: a pronoun with no antecedent. "So let's start with the city he left" asked the viewer to follow a person they had not met. A hook that names a person has to say who they are, what they want and why a decision of theirs is a puzzle (here: *why walk away from a factory that works?*).

Cautions: a private person named in a national magazine (Sharma) is a judgement call; do not imply facts the sources do not give (the script does not say he stood at the gate that morning); a fair treatment of someone who may be accused of wrongdoing means saying exactly what is alleged and what the courts decided; do not write "villain".

- **I1.** At least two people, one protagonist, one witness; each with a want, a concrete detail and a stake, found in the text. **I2.** The protagonist is named in the hook (name, not a role phrase), by word 160, with a want in the hook or the first story beat. **I3.** A hinge line that uses he/she/his/him needs a name in the same or the previous sentence. **I4.** No stretch over 60 s (fail over 100 s) without a person, a role-name or "you". **I5.** At least three viewer stakes, one in the hook, one in the middle, one near the end, each speaking to "you". **I6.** The protagonist in at least four beats; the witness in at least three, early, and in the close.

## The hook carries weight (rule B8)

v5's hook was 73 seconds with one scene and one thesis, and the user called it thin. A longer hook has to earn its length. It may run to 110 seconds if it carries, and rule B8 checks each: **a named person with a want; the weight of the rise; the weight of the fall; a stake for "you"; at least two moments to look forward to; the learning**. It must still contain no more than one figure, no withheld route term, no greeting, and the click confirmed in the first 30 seconds. The forward moments tease *experiences* (a cocktail party, a press conference with one sentence that aged badly), never the route (solar, the bond, the debt). This is a bet: a longer hook can cost retention if it reads as a menu, and nothing in our data can say. If analytics show a drop in the first two minutes, cut the forward moments first.

## The texture plan and the texture pass

**Step 3, item 7 (the plan).** For every beat, one line: *opening move* (person, scene, viewer or parable) · *claim* · *number and its landing* (if any) · *figure* (governing touch, gloss, parable or none) · *which lines carry the earnest register and which the aloof*. Fill `persona`, `figures`, `humans`, `viewer_stakes`, `landing_quote` and `cdl.hook_elements` from it.

**Step 4a (the texture pass, after the draft).** Read the draft once for each question: (1) Who is in every beat, and what do they want? (2) Where is "you"? (3) What lands after each number? (4) Scan in 30-second windows: more than two number-heavy windows in a row? (5) Does the governing metaphor come back, and is its limit said aloud? (6) Is there an earnest line and an aloof line in each third? (7) Is every "I" true? (8) Does any "he" or "she" at a hinge have its name in the same or the previous sentence? Then run the voice pass (Step 4b) and lint.

## What is unproven

The lessons are design constraints. The linter checks that the structure is declared and present in the text; it cannot judge whether a persona is charming or smug, whether a parable lands, or whether a longer hook holds people. The English winners we measured do not use these devices much. A fresh-reader pass (Step 6) and, with enough comparable scripts, a cohort test are the only ways to find out. Nothing here has been tested on viewers.