# Four songs, now four sung campaigns

The owner approved the music in all four first versions. My mistake was treating the rest of the task as a collection of catchy angle demonstrations. Those 32–36-line songs established a premise, named a product, repeated a hook and reached a quick payoff. They did not yet carry the dramatic development of the reference campaigns. Technical audio checks did not catch that conceptual shortfall.

## What I understand about the references

The unit of construction is a **change in a person’s situation**, carried by melody. Verse/chorus labels describe the music; they do not explain why a viewer needs the next line. A useful dramatic chain is: an encounter opens a question → its personal cost becomes concrete → the heroine tries or resists something → another person responds to that resistance → a practical choice becomes possible → an old situation returns → the heroine and supporting people act differently → the opening question receives an answer → the viewer gets a next action.

This is an interpretation of the supplied references, not a universal formula or measured retention finding. Not every reference has a large cast. The garlic explainer mostly uses a sung narrator, escalating time markers and unanswered questions. It contributes momentum and progressive disclosure; the relationship references contribute interaction.

### Why the characters speak when they do

The Target / SEORA reference is the clearest retained transcript example:

- **0:00–1:24:** the encounter with the ex-husband’s new wife opens the wound. A glance, a sympathetic smile, a remembered marriage and a car retreat make the emotional cost specific. The other woman need not give a villain speech: the narrator’s interpretation drives the next action.
- **1:26–1:55:** the heroine meets a familiar mother at her daughter’s pickup and asks about care because she has just described her own failed attempts. The peer asks what she uses before responding. That question–answer dependency gives the advice a reason to exist.
- **1:55–2:07:** the peer hands over a product and gives an instruction; the heroine subsequently acts on it. Receiving advice and following it are separate story events.
- **2:17–2:33:** the daughter notices a change. In that source, the line functions as outside validation after a passage of time. Its efficacy assertions belong to the source ad and are not evidence for goPure.
- **2:35–3:08:** the recital returns the heroine to a related social test. The other woman’s new response matters because we remember the first encounter. The heroine’s final response shifts the story toward her own judgment.
- **3:10 onward:** the CTA follows an emotional resolution. It is not a product list dropped into an unresolved scene.

Evidence: `references/example-reference-transcript.txt`; `references/example3-seora-passport.md`; the retained dense observations and synthesis in `runs/gopure-dna/research/targ/`. Transcript timings are approximate and the retained transcript is not a newly certified verbatim transcription.

The existing **sisterbrazil** analysis maps a climax/flashback opening, isolation and failed attempts, a guide intervention, a proposed explanation, action, a restaurant confrontation, and self-worth before the CTA (approximately 0–30, 30–90, 90–150, 150–240, 240–330, 330–450, 450–481 seconds). The dramatic lesson is that the later social encounter tests an earlier wound. The **svekr** analysis similarly records repeated social pressure, failed attempts, a confidant’s intervention, action and a later response from the original pressure source (approximately 0–30, 30–60, 60–120, 120–150, 150–240, 240–280 seconds). These are retained visual-analysis summaries, not exact new dialogue transcriptions. Their biological mechanisms and transformation claims are not transferable to this product.

The **krov** analysis supplies a different arc: hiding → search and frustration → discovery/explanation → action → social return. It supports making avoidance visible and making return an action. I did not fabricate exact quoted exchanges from that analysis. The **garlic** transcript supplies progressive time markers, consequences and objections before a product reveal, but not an ensemble-dialogue template.

I am preserving these useful functions: a character wants something, a line is prompted by what just happened, the listener has to answer or act, the answer opens a new option, and the eventual payoff recalls an earlier situation. I am discarding imported medical claims, forced humiliation and the idea that a cosmetic product creates family acceptance.

## Honest audit of the four approved short versions

| Song | What already worked | What was insufficient | What the expanded version changes |
|---|---|---|---|
| Стоп! А шию хто забув? | Immediate recognition, comic hook, strong pop-punk lane, a repeatable mirror/checklist prop. | A witty monologue with a fast ingredient/CTA section. Nobody caused or challenged the forgotten-care habit. The resolution was declared rather than tested. | Family requests create the overload. A friend answers a real objection. Vira asks for a boundary; Taras and Sofiia respond with different actions. The checklist returns with a place for her. |
| Шарф у відпустку | Best single visual metaphor; disco energy and scarf-to-bag payoff are memorable. | The heroine announces freedom before facing meaningful resistance. The wardrobe asks questions, but no actual person answers. Care and clothing freedom risk feeling loosely attached. | An unused ticket and a missed evening establish cost. A daughter unintentionally reinforces hiding; a seamstress asks what Lada actually wants. A kitchen rehearsal precedes the second doorway test. The daughter learns to ask, and the scarf gets its new job. |
| Мамо, де ти на фото? | Strongest emotional question; daughter’s absence-in-the-album discovery and slippers have a real story seed. | The daughter initiates the song and mostly disappears. The rest of the family gets no responsibility or response. The timer solves everything too quickly. | A birthday album gives a concrete occasion. The husband’s habitual praise helps create the exclusion. Mum names her own avoidance, asks him to learn, and the sister and daughter physically help the shared photo happen. He prints the final frame. |
| Баночко, тільки без казок | Clear buyer motive, witty skepticism, strong question-and-answer musical potential, explicit payment process. | A one-person FAQ. The heroine asks and answers herself; trust is asserted without being tested. | A well-meaning friend and an adult son change the method of discussion. Three questions structure the suspense. The parcel clerk makes inspect-before-paying a concrete scene. The skeptic keeps her personality and shares her questions afterward. |

For all four: the owner’s musical approval is explicit. The original instrumental pickups were removed and final vocal openings measured in the prior run. That is evidence of a corrected opening, not evidence of audience retention. A single lead passport was insufficient for this requested ensemble stage; the earlier “supporting cast later” assumption is superseded by the owner’s instruction now.

## The expanded delivery model

Each song has 19 narrative/music blocks and 76 short sung lines. The full campaign JSON maps all 304 lines to the narrative singer, the dramatic speaker(s), the addressee(s), the trigger, action, consequence, emotional state and persuasion function. There are four characters per campaign, each with a want, fear, relationship, change, physical identity, wardrobe, gesture and neutral portrait passport.

The performance contract follows the reference storytelling approach: **the approved female lead sings both narration and introduced quotations**. Four fictional characters do not imply four reliably independent generated singers. Their dialogue is written into the lyrics; no spoken voiceover or video lip-sync is being substituted. New cover jobs use each approved short recording as its own seed and preserve its musical style instructions. A cover can preserve a lane without reproducing every melody or vocal inflection exactly; actual outputs still need review.

The first words remain the original approved hooks. The written opening montages are plans only. No video generation is part of this delivery.

## My editorial judgment before hearing the expanded takes

**Mum in Frame is the strongest complete ensemble story.** Every supporting person contributes to the same physical ending. The husband’s change is especially useful: his earlier praise kept Mum outside the picture; his later printed photo keeps her inside the family memory. The risk is sentimentality and a slower middle, so the timer lesson and slippers must retain humor.

**Forgot Neck has the strongest comic interaction.** The recurrence of family requests creates a genuine test. The cream remains a concrete reminder, while the household changes through a conversation. Its risk is drifting into a household-labor campaign with too little neck-care relevance.

**Scarf Holiday has the strongest prop payoff.** The daughter’s well-meant advice is more interesting than an obvious villain. A small kitchen dance earns the public dance. The product is a companion care ritual; this is the least direct product-necessity argument of the four, and that tradeoff should be tested rather than hidden.

**No Fairy Tales is the most direct purchase narrative.** Its three questions create progress and its postal scene tests a promise. It also has the highest risk of sounding like a sung checklist. Bohdana has a small procedural role, not an elaborate emotional arc; inventing one would add length without helping the story. The friend and son carry the relationship changes.

The 19 blocks are an authoring grid, not proof that all four need identical screen timing. Longer duration is not automatically more compelling. Subsequent listening checks must reject lost dialogue, rushed explanations, flattened emotions and missing endings, and report remaining uncertainty honestly. Audience retention, conversion, biological “dopamine,” and native-speaker approval of the expanded versions remain unmeasured.

## What the actual recordings changed in my judgment

Twelve raw takes were generated through six explicit song requests. The first two Forgot Neck requests were not accepted: the first repeated a completed story block, and the second fixed that sequence but still triggered a brand-pronunciation objection. The third request put the actual English brand `Go Pure` into the lyrics. Take 1 then cleared the audio review for both sequence and brand; take 2 still repeated story material and was rejected. This is why a valid script alone is insufficient.

| Selected expanded song | Final duration | Script lines above the ASR match threshold | Actual listening assessment and remaining issue |
|---|---:|---:|---|
| Forgot Neck, version 3 / take 1 | 3:40.5 | 73/76 | Complete linear story and sung CTA; the review hears the corrected brand. Three English-brand lines are still transliterated inconsistently by Ukrainian ASR. Native pronunciation remains an owner-listening check. |
| Scarf Holiday, version 1 / take 1 | 4:05.7 | 74/76 | Complete four-character story and disco lane. An extra final chorus reprise is retained; it repeats the musical payoff, not a block of story actions. One brand reading and the alignment across that reprise need interpretation. The reviewer also calls the ingredient diction compressed; its approximate timestamps and final-word identification are unreliable, so that is a listening flag rather than a certified defect. |
| Mum in Frame, version 1 / take 2 | 3:47.1 | 76/76 | The cleanest line-completeness evidence and clearest ensemble payoff. The audio review favors its character entrance and complete ending. This strengthens my first-place editorial judgment; it does not establish audience performance. |
| No Fairy Tales, version 1 / take 2 | 3:59.2 | 75/76 | The complete purchase story with the more concise ending. One brand phrase remains ambiguous in ASR. Its quieter initial syllable needed a documented onset-threshold resolution so it would not be cut off. |

Across the selected masters, **298 of 304 scripted lines** cross the independent ASR threshold. The six exceptions have retained line-specific reviews: five brand-name readings and one alignment spanning the additional Scarf chorus. These are not six silently omitted lines. Equally, a threshold match is not proof of perfect diction. The raw transcription, disputed text and audio-review responses are retained for inspection.

The musical lanes are preserved through the approved seeds and actual comparison reviews, with a measurable qualification: first-minute tempo estimates are approximately **172 versus 161 BPM** for Forgot Neck, **129 versus 123 BPM** for Scarf Holiday, and **103 versus 103 BPM** for both acoustic stories. Thus the two higher-energy expansions run somewhat faster by this estimate. Chroma profiles are very close, but neither chroma nor a genre label proves an identical melody or singer. The original recordings remain available beside the expanded ones for direct comparison.

My recommendation remains **Mum in Frame for the most complete emotional campaign**, **Forgot Neck for comic energy**, **Scarf Holiday for its visual callback and groove**, and **No Fairy Tales for a clear purchase test**. I would validate the middle sections most critically: the care explanations in the first three, and the repeated question format in the skeptic story. All four have more causal development now; none should be declared a winning ad without the owner's listening judgment and an audience test.

Two Scarf review attempts returned a provider error sentence instead of an evaluation. Those outputs were rejected and retained. A third review used one candidate and a different available audio-review model. Broad model genre labels, exact timestamps and “perfect” language were not accepted as measured facts. One final-opening check was started before the mastering receipt existed, failed without changing audio, and was rerun after mastering completed.

Final mastered hook checks pass for all four: isolated vocal activity begins at 0.00, 0.04, 0.03 and 0.04 seconds respectively. Independent three-second ASR recognizes each first word. Short clipped phrases cause later-word transcription mistakes, so full-song alignment and audio review remain the evidence for the complete hook. There is no fade-in or instrumental lead-in added during mastering.