/make-content-source-to-units

name: make-content-source-to-units
description: >-
  Step of the make-content playbook: mines one finished asset into scored quotable units and maps a
  different lead unit to each platform. Run make-content instead; not for direct use.
version: 0.1.0

Source to units

Input: the full text or transcript of one finished asset (`../make-content/references/repurposing.md` §1 says how to get it in). Read it completely before mining. Never mine from a summary.

Output table

Source: <title> · <link or media_asset_id> · <length/duration>
| # | Type | Anchor (verbatim, ≤25 words) | Where (timestamp / ¶) | Alone | Charge | Novel | Proof | Total | Best homes |
|---|------|------------------------------|-----------------------|-------|--------|-------|-------|-------|------------|

• **Type:** claim · contrarian take · story · stat · quote · framework/steps · mistake+fix · before/after · objection+answer · audience Q · demo moment.
• **Scores (1–3):**
• **Alone:** makes sense without the rest of the source.
• **Charge:** surprise, tension, humour or stakes.
• **Novel:** says what competitors don't.
• **Proof:** 3 if an approved claim or a real number in the source backs it; 1 if it is opinion only. Opinion is fine for a contrarian take, but it must be the speaker's.
• Mine 15–30 units from a long source and 5–10 from a short one. Drop units scoring ≤6.
• **A short or silent clip** (a 10–30s product video, often with an empty transcript) holds 2–4 units: its caption, its on-screen text, and what the scenes show (`media_analyze` gives `visible_text` and the scene descriptions). Say so plainly: it supports 2–3 versions at most, not "every platform". Ask whether a longer asset exists (a founder interview, a webinar, an article) before stretching a thin source into filler.
• A stat in the source is still a claim. If it is about the brand and not on the approved list, keep the unit but mark "stat needs approval". The claims check will cut or soften it.
• Spoken quotes stay verbatim, attributed to the speaker as named in the source. Never tidy a quote. On-screen text that came from OCR (`visible_text`) is text, not a quote: fix obvious OCR errors and never put it in quote marks.
• **Sources under ~30s:** rank the units instead of applying the ≤6 cut-off, and keep the top 2–4. If fewer than 3 units score 7 or more, recommend a richer asset before writing, naming one from the brand's own posts if you can (e.g. one with speech). Then let the human choose.

Then map

For each platform the brand uses, pick the lead unit by the map in `../make-content/references/repurposing.md` §3. No two versions share a lead unit. When strong units outnumber platforms, a platform can take two surfaces (e.g. carousel + Reel). If the source has fewer strong units than the brand has platforms, say so and make fewer, better versions rather than filler.