Skip to main content
This layer answers: what kind of video is this, and what is it actually doing to the viewer? It comes in two passes. Classification decides the format — before anything else interprets the video, because a 15-second ad and a 33-minute documentary must not get the same read. Comprehension then watches a storyboard sampled across the whole runtime and reads the video as a piece of communication: the idea it argues, the feeling it sells, the brand move it makes.
Teardown header showing the title and logline: a rapid-fire satirical manifesto, 80 scenes at 80 a minute in 60 seconds
Live example: Coinbase — ‘Break The Cycle’ teardown — the logline under the title.

Classification — the format call

FieldWhat it holds
formatOne of: b2b_ad, ugc_ad, long_form_doc, short_form_hook, explainer, kids, vlog, tutorial, cinematic, product_demo, news_commentary, other
format_confidence / format_evidenceHow sure the call is, and the one-sentence why
matched_templateWhether the structure matches a productized template: pile_up_reveal (chaos → hard cut → calm reveal → proof → CTA), numbers_story, before_after, pov_native — or null
duration_bandmicro (under 15s) | short (under 60s) | mid (under 5min) | long (under 20min) | extended
platform_nativeBest-guess home platforms — linkedin, tiktok, reels, shorts, youtube, ctv…
intentsell | teach | entertain | inform | inspire
primary_audience / language / is_ai_generatedWho it’s for, what language, and whether it reads as AI-generated
The format decides which rubric the Scorecard applies — that’s why classification runs first.

Comprehension — the meaning read

Metrics measure a video; comprehension understands it. A vision model watches 12 frames sampled evenly across the runtime — in order, as a storyboard — and can even overrule the metrics-only format guess (format_correction) when the frames tell a different story. It exists because metrics alone once filed a global brand manifesto as a travel documentary and saw its climactic logo reveal as “a frame”.
FieldWhat it holds
loglineOne vivid sentence: what this video is
premiseThe situation or world it sets up
thesisThe single idea it argues, or the feeling it delivers
sellingWhat it actually sells — usually a feeling or idea, not a feature
emotional_targetThe intended arc, e.g. “awe → belonging → pride”
is_branded / brandWhether there’s a brand, and the brand object: name, appears_at_sec, screen_time_pct, reveal_technique — how restrained the reveal is
genre / sub_genreThe real genre — brand_film, manifesto, ad, doc, music_video, explainer, ugc, narrative_short…
subjects / comparable_toCore themes, and 2–4 well-known reference campaigns it sits beside
signature_techniqueThe one move a director would carry into their own work
director_notesTwo or three sentences on how to rebuild the feeling, plus a list of 3–5 concrete, replicable techniques to reference — patterns, never content
rebuild_caveatsThe gotchas — cost reality, skill required, what the metrics get wrong
format_correctionA better format call than classification’s, when the frames disagree — otherwise null

What to do with it

  • Steer with the thesis, not the topic. Two videos about the same product can argue opposite ideas; matching the thesis is what makes a reference actually comparable to your brief.
  • Study the brand object before your own brand placement. appears_at_sec and screen_time_pct quantify restraint — high-performing brand films routinely hold the logo back for most of the runtime.
  • Use comparable_to to brief stakeholders. “It’s structured like [known campaign]” lands faster in a review than any metric.
  • Trust rebuild_caveats on scope. When the notes say the feeling costs real production, a cheap imitation of the surface won’t inherit the performance.

Scorecard

The format-specific rubric this classification selects.

Script analysis

The skeleton delivering this thesis.

Hooks & re-hooks

How the argument earns its first three seconds.

Cast

The people and world carrying the message.