Source Backed Reel Evidence Montage

---
name: source-backed-reel-evidence-montage
description: Use when creating or analyzing AI/tech/news Reels where narration must be supported by article screenshots, highlighted source text, product clips, code/terminal proof, and beat-matched evidence visuals.
---

# Source-Backed Reel Evidence Montage

## Overview
This style turns a fast AI/tech/news narration into a **visual evidence chain**. Every factual phrase gets a matching proof visual: article headline, highlighted sentence, official note, product image, code terminal, company logo, or human/talking-head authority shot.

Learned from 100xEngineers Reel `DYhssDhN-ti` about Claude Mythos allegedly helping crack Apple's Memory Integrity Enforcement. The reusable pattern is not the specific claim; it is the **tight narration-to-proof montage**.

## When to Use
Use this for:
- AI/news explainers that need credibility, not generic B-roll.
- Security, product, startup, or research stories with specific claims.
- Reels where the viewer should feel: “I can see the source while he says it.”
- Scripts with names, dates, companies, numbers, papers, official notes, or demos.

Do **not** use when the video is purely personal storytelling, comedy, lifestyle, or abstract mindset content without sources.

## Core Rule
For every factual sentence, choose one primary proof asset and one visual emphasis:

| Narration phrase type | Proof visual | Emphasis treatment |
|---|---|---|
| Company/model/person name | Logo, founder/person clip, product UI | Quick logo/title card or boxed portrait |
| Big quantified claim | Large serif kinetic text + relevant image | Oversized number/phrase, 0.5–1.0s hold |
| Article/blog claim | Source screenshot with headline/date visible | Crop into article, highlight exact words |
| Technical mechanism | Code, terminal, architecture, UI recording | Zoom into relevant line/pattern, not fake hacker filler |
| Team/startup claim | Founder/team photo, office/location, website | Clean card on light background |
| Result/consequence | Official notes, changelog, patched issue, benchmark | Highlight credited line or bug note |
| Strategic takeaway | Talking head close-up + one proof collage | Slow push-in; captions carry key nouns |

## Shot Structure
A strong 45–60s version follows this rhythm:

1. **Brand/topic cold open (0–2s):** minimal logo or title card on warm/light background.
2. **Hook claim (2–5s):** talking head + bold serif phrase. Immediately show a concrete proof visual before the claim feels like hype.
3. **Entity setup (5–12s):** company logo, executive/product clip, official article headline. Use word-by-word captions for the spoken nouns.
4. **Primary source proof (8–16s):** article/blog screenshot. Headline/date/source must be visible first, then crop closer and highlight the exact phrase being narrated.
5. **Human/startup proof (14–20s):** team photo, company screenshot, founder shot, or location card to make the story concrete.
6. **Technical proof (18–30s):** terminal/code/research screenshot. Use a moving crop or cursor/spotlight to point at bug patterns, exploit output, logs, or research artifacts.
7. **Mechanism beat (25–35s):** visual metaphor or diagram that makes the technical mechanism legible: pattern grid, before/after, hardware/software split, workflow chain.
8. **Consequence beat (35–45s):** official patch notes/changelog/security notes. Highlight the exact credited model/person/company line.
9. **Implication beat (43–50s):** talking-head close-up plus logos/source collage: “what this changes.”
10. **CTA (last 2–4s):** branded follow card; keep it visually distinct from evidence so it does not look like another source.

## Article Highlight System
Article screenshots are the signature move. They must be evidentiary, not decorative.

1. Show the **full source context** briefly: publication/company, date, title, author/source if visible.
2. Crop closer to the relevant paragraph or headline.
3. Highlight only the exact words that support the narration.
4. Use a consistent highlighter style:
   - soft pink/peach rectangle for headline phrase highlights;
   - white rounded text backing for contrast on dark article pages;
   - subtle zoom-in while the highlighted phrase is spoken;
   - no more than 1–2 highlighted phrases per screenshot.
5. Keep source screenshots on screen long enough to read the key phrase, usually 0.8–1.5s.
6. If the source text is too small, recompose it as a source card: headline/date visible + enlarged extracted quote + URL/source label.

## Visual Grammar from the Reference Reel
- **Talking head as anchor:** narrator returns every few seconds to reset trust and emotion.
- **Vertical cards with rounded white borders:** B-roll/product clips appear inside framed cards over dark blurred backgrounds.
- **Mixed real assets:** Apple logo, executive clip, article pages, team photo, terminal windows, laptop/product shots, security notes.
- **Elegant serif emphasis:** big words like “5 years,” “billions of dollars,” “Hardware,” “Software” appear in italic/editorial serif.
- **Kinetic captions:** small white word/phrase captions sit near the bottom, aligned to narration, highlighting one spoken phrase at a time.
- **Proof progression:** logo/title → article headline → paragraph highlight → team/person proof → terminal/code proof → official notes → strategic implication.
- **Motion:** quick cuts every 0.5–1.5s, push-ins on sources, crop jumps from full screenshot to highlighted detail, occasional split-screen/side-by-side comparisons.
- **Background:** dark blurred tech footage or warm off-white cards; evidence cards float above it with drop shadow/border.

## Production Workflow
1. **Fact table first:** split script into factual claims. For each claim, list source URL, exact quote, asset type, and confidence.
2. **Asset gather:** prefer official company blogs, security notes, papers, docs, product pages, public demos, founder/team posts. Avoid random Google Images unless used only as non-factual texture.
3. **Shot table:** create columns: timestamp, narration, proof asset, crop/highlight phrase, on-screen caption, motion/SFX.
4. **Screenshot/source-card preparation:** capture full-page or viewport screenshots, then create 2 crops per source: context crop and highlight crop. If live screenshots are blocked or too small, build a readable source card with source/date/title plus enlarged quoted excerpts.
5. **Edit beat matching:** reveal the proof asset as the noun/claim is spoken, not after the sentence ends.
6. **Draft-video path:** when no talking-head footage exists yet, create a faceless evidence-montage MP4 with generated VO, source cards, highlighted quotes, diagrams/charts, and CTA. Treat it as a motion/storyboard that can later accept talking-head cuts. See `references/faceless-evidence-montage-video-production.md`.
7. **Mobile readability QA:** inspect a contact sheet. If a highlighted phrase is unreadable on a phone, enlarge or remake as a source card.
8. **Claim audit:** no unsupported numbers, dates, names, or causal claims. If unsupported, soften wording or remove. Label company-reported metrics clearly.

## Template Shot Table
| Time | Narration | Visual | Highlight / caption | Source |
|---|---|---|---|---|
| 0.0–2.0 | “X is so powerful…” | Logo/title card + talking head | Big serif hook phrase | N/A or brand asset |
| 2.0–5.0 | “it cracked Y” | Terminal/demo clip | caption: “working exploit” | demo/source URL |
| 5.0–9.0 | “spent five years…” | executive/product clip + big number | `5 years`, `billions` | article/company source |
| 9.0–14.0 | “This is official system Z” | official blog headline | highlight exact title words | official blog URL |
| 14.0–20.0 | “a four-person startup…” | team photo / company page | `four-person startup` | team/company URL |
| 20.0–30.0 | “spotted vulnerable patterns…” | code/research screenshot + pattern grid | `bug patterns`, `prior research` | paper/repo/changelog |
| 30.0–42.0 | “patched/credited…” | security notes screenshot | highlight credited line | official notes URL |
| 42.0–50.0 | “companies must rethink…” | talking head + source collage | key implication words | prior sources |

## Viral second-hand report discipline
When a Reel is based on a viral business/AI headline whose primary source is second-hand, the credibility move is to **hook with the claim, then qualify it quickly, then extract the durable operator lesson**.

Use this pattern:
1. Identify the primary source and attribution chain: direct company disclosure, official document, named source, anonymous source, consultant claim, or secondary pickup.
2. If the claim is unnamed/second-hand, say so in the first 5–8 seconds: `reported`, `Axios says an AI consultant claimed`, `unnamed company`, `not independently confirmed`.
3. Do not treat syndications or rewrites as independent corroboration. Use them as context only.
4. Run a quick magnitude sanity check for huge numbers when public pricing/benchmarks exist, and present the math as `rough public-rate context`, not proof.
5. Convert the headline into a safer lesson: e.g. metered compute, budget caps, access controls, routing, ROI-per-workflow, or governance failure.

Reference: `references/claude-500m-enterprise-ai-spend.md` captures the May 2026 Claude `$500M in one month` research pattern, safe wording, token-cost sanity math, and evidence visuals.

## Common Mistakes
- Using article screenshots as background texture without readable source/date/title.
- Highlighting generic words instead of the exact claim-supporting phrase.
- Showing terminal/code clips that are unrelated or unverifiable.
- Letting logos imply relationships or endorsements that the source does not state.
- Making every caption large; only key nouns/numbers should pop.
- Claiming a model, exploit, benchmark, or research result without source proof.
- Turning second-hand viral reports into definitive claims; use attribution and caveats early.
- Treating secondary pickups/syndications as independent corroboration of the same primary report.
- Using generic robot/neural-network B-roll when a source screenshot, product UI, or code proof exists.

## Verification Checklist
Before delivering a script/storyboard/video:
- [ ] Each factual sentence maps to a proof asset.
- [ ] Each article/source screenshot shows context first, then a highlighted detail; source cards may substitute when screenshots are blocked or unreadable.
- [ ] All highlighted phrases are readable in a 9:16 mobile contact sheet.
- [ ] Numbers, dates, names, and causal claims have source URLs.
- [ ] Company-reported metrics are labeled as company-reported or `Source says`, not framed as independently verified.
- [ ] Any speculative implication is phrased as implication, not fact.
- [ ] Code/terminal visuals are relevant or clearly labeled as illustrative.
- [ ] If delivering an MP4, ffprobe confirms 1080x1920 video + audio and ffmpeg decode QA passes.
- [ ] If no talking-head footage is included, the output is described as a faceless/evidence-montage draft, with talking-head insertion offered as a follow-up.
- [ ] The final CTA is separated from evidence/proof visuals.

## Baseline Failure This Skill Prevents
A default Reel plan often says “add screenshots, logos, code clips, and fast cuts” but fails to specify which claim each asset proves. This skill forces a claim → source → crop → highlight → motion chain so the visuals become evidence, not filler.