.claude/skills/video-script-production/references/review-round-7.md

# Review round 7: the Hinglish v6 annotations, and how they became lessons K, L and M

This is the record of how thirty-five line-level notes on one script turned into three general lessons, fifteen lint rules, two new tools and a v7 of the English and Hinglish scripts. It is written so the same move can be made again: **a reviewer writes notes in the margin; the author finds the viewer problem behind each note, generalises it, checks it against research, applies it to every beat (not only the annotated ones), and makes the checkable part deterministic.** The method is in section 1, every note and what became of it in section 2, what the research found in section 3, and what is still unproven in section 5.

## 0. What was asked (2026-10-25)

The user read `research/the-fall-of-moser-baer/script/tts/TTS-v6-hinglish-gemini-3.1.md` and wrote suggestions in round brackets inside the Hinglish text of beats 0 to 6. Their request, in short: look at all the suggestions, understand where they are coming from (make smart assumptions), generalise, and grandfather the lessons into the production pipeline *and* the decision path that led to this script, and into the rest of the script too; they like the script, so polish, do not restructure; for dense stretches, give a one-line summary of what follows (curiosity, a loop, something to understand), then fill in details, close loops, make people care about the human story and the stakes; produce updated English and Hinglish clips; check any new information claim against research.

Standing rules that applied: no subagents, no commits or pushes; keep the user's direction and change as little as a lesson needs; say plainly what is unproven; record tools and learnings in the same session.

## 1. The method (use it again)

1. **Read all the notes as a set before touching anything.** Notes that look unrelated share a cause. Here: five notes were about nobody being told who someone is (Ratul, "that gate", "he", Xidex, the unnamed fourth escape); six were about words nobody says; four were about a turn that was not spoken.
2. **For each note write four things:** what the reviewer literally wrote; the viewer problem behind it (what a listener who has not read the research would experience); the general lesson (stated without the topic's names); the check (a deterministic rule if the property can be found in text, otherwise a line in the read-aloud checklist).
3. **Where a note makes a claim about the world, research it before applying it.** Four of this round's notes contained claims the sources do not support (section 3). A note is a request for an effect on the viewer, not a source.
4. **Apply the lesson to every beat.** The reviewer annotated beats 0 to 6; the lessons are applied to beats 7 to 13 as well (Ratul tagged in beats 7, 9 and 12; headlines on beats 7, 9, 10 and 12; spoken BUTs at three joins in the later beats).
5. **Make the checkable part deterministic, then test the check twice:** on the new script (it must pass) and on the old script (it must fail where the reviewer found the defect). The new rules were run on the unchanged v6 text and flagged Ratul returning with no tag, Xidex named once, the crore gloss, "whatever was", a buried verb in beat 3, three BUT joins with no spoken contrast, "favourite theory", the lab-voice paragraph and the dense beat 3 with no headline.
6. **Add damage-test mutations** (`selftest.js`) so each new rule is shown to fire when its property is broken.
7. **Do the read-aloud pass on the result.** It found five defects here, two of them created by the changes themselves (a "This pattern" with no antecedent once the echo line moved up into the headline; a pair of "too").
8. **Record**: this file, `history.md` Round 7, `learnings.md`, `tools.md`, the lesson in `principles.md`, the field in `chain-schema.md`, the check in `read-aloud-checklist.md`.

## 2. Every note, what was assumed, and what was done

"Rule" names the lint rule (K, L, M groups in `scripts/lint.js`) or the manual gate. "Applied" says where in v7 (English `SCRIPT-v7.md`; Hinglish `SCRIPT-v7-hinglish.md` follows the English).

| # | Beat | The note (short) | Assumed intent | Lesson | Rule | Applied |
|---|---|---|---|---|---|---|
| 1 | B0 | Add pirated movies to the opening list; a little shocking, what most people associate with it | The "you've held one" list is a mirror; it works only if it contains the real, slightly guilty association | **L: mirror**: show what the audience actually did with the thing | manual (checklist 20) | "a blank disc, a pirated movie, a college project" |
| 2 | B0 | Add "paisa hi paisa kamaya" | The rise must be felt in money terms before the fall | **L: prize**: state the winning in the audience's words | L1 | "and made serious money doing it" (profit: BT 2003, Business Standard) |
| 3 | B0 | Start the fall with "Lekin"; both changes deepen contrast and stakes | The turn must be audible | **L: prize, then a contrast word** | L1 | "But one morning in November 2017" |
| 4 | B0 | Add "poore" to emphasise the 17 years | A person's sacrifice should be heard as one | **L: cost**: put an intensifier where someone pays | L2 | "a full 17 years inside" |
| 5 | B0 | "ek hissa" to "hissa tak"; both make it more emotional | Same: the pay cut is a sacrifice, not a figure | **L: cost** | L2 | "part of their own pay" |
| 6 | B0 | "uss cheez se door jaane" is confusing; use a smart English word | The viewer had to decode an abstract phrase | **K: plain, concrete words; nothing to decode** | K8 (vague referents, buried verbs); checklist 21 | "Whenever something stopped working, he moved on to something new" |
| 7 | B0 | "uss gate" is not set up; say it more emotionally; use "pivot" with one definition | A demonstrative needs an antecedent; the thesis should land on the people who pay; a named idea needs one plain definition | **K: antecedent; K3 named idea defined once; L3 thesis names who pays** | K6; checklist 1 | "...part of why those workers were locked out"; pivot defined once in the hook |
| 8 | B0 | The teaser seems too long; hit all points in passing; add one more to give weight | The late hook is a trailer, not a menu | **M: trailer**: a list of fragments, the strongest last | M2 | "a cocktail party..., a press conference that aged badly in half a year, and a national honour" (24 words) |
| 9 | B0 | Personal relevance is important but must be 25% of its length | Keep the "you", cut the explanation | **M: takeaway**: one short sentence | M2 | "By the end you'll have three questions for any company, and for you." (13 words, was 31) |
| 10 | B0 | "ek chalti hui cheez se door chale gaye" is unspeakable; use "first pivot", "said no to a plan that looked successful"; give Mr Puri a character | The last hook line should present a person's first decisive act, in speakable words | **K: speakable; L4 character through a deed** | M2/K6; checklist 22 | "start with the first pivot: the day Deepak Puri walked away from a business that was working" |
| 11 | B1 | "khinchaav" to "feeling" | The common word | **K: plain words (Hinglish: prefer the English word people say)** | checklist 21; `hinglish-adaptation.md` | "you'll know the feeling" |
| 12 | B1 | Define pivot here | The lesson Puri took is the idea; name it where it is stated | **K3** | K6 | defined at first use in the hook; "That was his first pivot" here |
| 13 | B1 | Add that a time-clock company never made much money; "lekin fir" | A ceiling needs the want; the turn is spoken | **L: limit carries the want; L6 spoken BUT** | L3 | "...only goes so far, and he wanted to try something new. ... his next idea" |
| 14 | B2 | The Xidex reference is out of place; it only makes sense if Xidex partnered with Puri | A named entity must be tied to the story | **K: name rent** | K4 | Xidex cut (no documented link) |
| 15 | B2 | Establish the tension and the success: everyone said it was a bad idea; Puri knew this country; he succeeded | Each beat is a small arc: doubt, edge, proof | **L: mini-arc** | checklist 23 | doubt; "more than twenty years making things in this country"; "third-largest maker of floppy disks" |
| 16 | B2 | "Lekin/Fir bhi..." | An explicit turn | **L6** | L3 | "Even so, for years people doubted it would last" |
| 17 | B3 | "unique strategy... strategy of scale" | Name the strategy when it arrives | **K: named idea** | K6 | "So Puri bet on scale: if you can't be first, be the biggest" |
| 18 | B3 | Remove "ek crore matlab dus million" | This audience knows crore | **K: do not gloss what the audience knows** | K7 | gloss removed; `audience.known_terms` |
| 19 | B3 | "ke baad" to "just piche" | The idiom | Hinglish idiom | `hinglish-adaptation.md` | "ke just piche" |
| 20 | B3 | "was running one of the two biggest" to "ban gaye the"; simple, clear | The verb belongs next to its subject | **K: one idea per sentence** | K8 | "Many people had doubted Puri... Now Puri's company was one of the two biggest disc makers on Earth." |
| 21 | B3 | Europe anti-dumping cases: a few lines; analyse and fact-check | A result whose opponent is unexplained is noise | **K: explain the opponent in one clause** + research | ledger | "Europe brought five anti-dumping cases against it, accusing it of selling too cheap, and by mid-2003 it had won all five" |
| 22 | B3 | Who is Ratul? Constantly appearing new figures must be solved in the pipeline | A person needs a tag when they arrive and when they return | **K: cast tags** | K1, K2, K3 | "Ratul Puri, Deepak's son and a company director"; "Ratul, Deepak's son" at each return |
| 23 | B3 | "strategy of scale samajh gaye the" | Call the named idea back when the others copy it | **K: callback** | K6 (reuse) | "had worked out the same strategy: scale" |
| 24 | B4 | Make the mechanism line clear and simpler; use English if it is clearer | The core mechanism was in one heavy clause | **K1** | K8 | "every maker keeps cutting its price, and each cut drags the price down for all of them" |
| 25 | B4 | "bhari" to "busy"; "chalte" to "bhaagte"; "treadmill-of-competition" | Plain words; the metaphor's verb | **K1; G (keep the verb of the metaphor)** | `hinglish-adaptation.md` | "busier than ever"; "Puri kept running anyway" |
| 26 | B4 | "aur sabse kam cost wo laga sakta hai jo sabse bada ho" | Short cause to effect | **K1** | K8 | "when prices fall, you cut your costs. And the biggest maker can cut them the most" |
| 27 | B4 | Replace the whole "if this is right you would expect" paragraph with "panting on a treadmill... but signs were starting to show" | The narrator is announcing an experiment | **M: no lab voice**: show the sign, do not announce the test | M3 | "A company panting on a treadmill can look healthy right up to the moment it doesn't. But the first signs were starting to show." |
| 28 | B5 | "kisi ne bhi gayab..." | The subject of "didn't blame" was unclear | **K: restate the subject** | checklist 1 | "The company didn't blame a missing customer" |
| 29 | B5 | Add "unki kismat solidly unke haath mein thi" | Re-establish the height before the fall point | **L: restate the peak** | checklist 24 | "while the company was still a champion, with its fate in its own hands" |
| 30 | B5 | Replace "I have no favourite theory" with "ignored the real reason and focused on other things"; reason about why and generalise | A neutrality disclaimer about a mechanism the record supports gives the viewer nothing to change their mind about | **L: aloof about people, plain about mechanisms** | L4 | "But that story skips something... I'll take the company at its word until someone shows me different." |
| 31 | B5 | "firse pivot" | Reuse the named idea | **K3** | K6 | "or pivot again... had already pivoted three times" |
| 32 | B5 | "Deepak Puri pehle (hi)" | Emphasis on "already" | **K1** | - | "had already pivoted three times" |
| 33 | B6 | "free" to "free movie cds" | Say the thing, not its category | **K: concrete nouns** | checklist 21 | "the pirated copy, which was nearly free" |
| 34 | B6 | "thodi success bhi mili" | Be fair to a partial win | **L9/G fairness** | G2 | "so it did gain real ground" (BT 2008: largest player) |
| 35 | B6 | "Doosra rasta... sab sunne wale ek second ke liye dang reh gaye" | End the beat on the feeling, in the viewer | **L: cliffhanger lands in the viewer** | L3, checklist 25 | "But the other exit was so far from discs that you'd do a double take." (the viewer, not "everyone", so no unsourced reaction is claimed) |
| G1 | all | The script is liked: polish, no structural change | Minimal change | standing rule | - | 14 beats, the chain, the numbers and the treadmill unchanged |
| G2 | all | When dense: a one-line summary of what follows, then details | Headline first | **M1** | M1 | headlines opening beats 3, 7, 9, 10, 12 |
| G3 | all | Check any new info claim with research | | `research-protocol.md` | ledger | section 3 |
| G4 | all | Generalise into the pipeline | | this file and the references | check.js | - |

## 3. What the research found (and where it parted from the notes)

| Claim in or behind a note | Finding | Source | Effect on v7 |
|---|---|---|---|
| Xidex partnered with, or supplied, Puri | Business Today (22 Jun 2003) mentions Xidex only as a company the party's talk concerned, which "later became a billion-dollar corporation". No partnership | BT 2003 (re-read) | Xidex cut; the party stays as the source of the idea |
| Puri studied this country and its companies for his whole life | No source says so. Sources do say: trading from 1964, manufacturing two years later, the floppy plant in 1987-88, more than twenty years | BT 2011, BT 2003 | Use "more than twenty years making things in this country" |
| Floppy success | BT 2003: "the world's third largest manufacturer of magnetic recordable media (micro floppy disks)" and "an unqualified success story" | BT 2003 (re-read) | "by 2003, Business Today was calling Moser Baer the world's third-largest maker of floppy disks" |
| "Paisa hi paisa" | Net profit ₹138.5 crore (2000-01), ₹245.3 crore (next year); ₹109.78 crore (year to March 2007) | BT 2003; Business Standard | "made serious money"; no figure in the hook (rule A3) |
| "Won all five anti-dumping cases" | BT 2003: five cases, "emerged victorious in all". Commission Decision 2003/371/EC of 22 May 2003 terminated the CD-R proceeding, "no dumping for Moser Baer India Ltd". **But** a parallel anti-subsidy case ended in a countervailing duty (Reg. 960/2003; BT: 7.3 per cent proposed); Moser Baer appealed to the Court of Justice (2009) | BT 2003; EUR-Lex decision text via mirror; search summary for the appeal | The script claims only the anti-dumping wins |
| "Sales doubled between 2002 and 2004" | Outlook: ₹770 crore to ₹1,580 crore. BT 2003: net sales ₹333.9 crore (2000-01) to ₹1,069.5 crore (2001-02). The bases differ | both pages re-read | "more than doubled in the early 2000s" |
| Ratul is "bete/nephew" | Son of Deepak Puri; executive director in 2003. Nephew of Kamal Nath (not used) | BT 2003 | "Deepak's son and a company director" |
| "araam se" (calmly) | Nothing in any source about manner | - | dropped |
| "Pivot" for the Calcutta-to-Delhi move | A move of place is a looser use than a change of product | - | flagged in the hand-off; defined broadly |
| Signalling a structure helps comprehension | Alpizar, Adesope and Wong (2020): 29 experiments, 2,726 people, d = 0.38 for learning outcomes (abstract); Schneider et al. (2018) report larger effects (secondary summary, unverified). Advance organisers: positive on average with wide variation (a 1992 update; earlier vote-counts disagreed). These are learning studies with instructional media, not entertainment explainers | search summaries | Lesson M's grade: B- for the mechanism, C for its use in a story video |

## 4. What was built

- **Three lessons** in `principles.md`: K (nothing to decode: cast, terms, antecedents, plain words), L (the turns: prize, contrast, cost, character, mini-arc, spoken BUT, plain mechanisms, cliffhanger in the viewer), M (density: headline first, a trailer hook, no lab voice).
- **Fifteen lint rules**: K1 to K8, L1 to L4, M1 to M3 (70 rules in all). **Fourteen damage-test mutations**. Chain fields: `humans[].tags`, `humans[].cost_quote`, `cdl.hook_elements.turn_quote/tease_quote/takeaway_quote`, `beats[].headline_quote`, `counted_sets`, `concepts`, `audience.known_terms`, `not_cast`, `one_off_names` (see `chain-schema.md`).
- **Manual gates** added to `read-aloud-checklist.md` (items 20 to 26): the mirror, decoding, the mini-arc, restating the peak, the cliffhanger.
- **Two tools**: `scripts/hinglish-parity.js` (a language version keeps the facts and the structure of the English) and `scripts/tts-gemini.js` (the per-beat Gemini TTS prompts, built from the script and a small spec, with the words guarded). The generator that made the v6 prompt files had been a throwaway scratch script and was lost; it is now part of the skill.
- **`references/hinglish-adaptation.md`**: how to turn a finished English script into Hinglish.
- **v7** of the English and Hinglish scripts, the TTS files (3.1 and 3.8, English and Hinglish), narration text for both.

## 5. What is unproven

- No viewer has been tested; no blind reader has read v7 (the reader pass needs subagents the user has not authorised). The rules show that the declared structure is in the text.
- The fifteen rules come from one review of one script by one reviewer. Their thresholds are design budgets (`rules.json`, `clarity`, `turns`, `density`). The `max_new_names_per_beat` budget would not have flagged v6's busiest beat (it had four first mentions by the heuristic); the dense-beat headline rule did. The name heuristic reads capitalised words inside sentences, so it misses a first mention that opens a sentence and it treats a publication as a person unless `not_cast` lists it.
- The winners we measured were not checked for these properties. Whether tagged cast, spoken contrasts or headline-first beats appear in winners is open (a cohort rubric v4 with E to M checks would answer it).
- The Hinglish has had no native read-aloud pass and its spoken length is unmeasured; the parity tool checks structure and facts, not how it sounds.