The Teardown Reads 9 Dimensions

The saved folder keeps growing and nothing comes out of it. 40 reels saved, watched twice, and the only note is that they were good. That is watching. A teardown is logging: 1 spreadsheet row per reel, 1 column per dimension, 9 columns, and about 100 rows before any conclusion is allowed. Row 1 is a URL, a first-frame screenshot and a transcript. Nothing else counts. A reel that was never captured cannot be compared 3 weeks later, and the comparison is the entire product. The sample is built from 3 parts. The top 5 to 10 creators in the niche. Rising creators under 100k followers with 1 breakout. And the format that keeps appearing across both groups. 3 search phrases is the front door - the 3 broadest ones the audience actually types - run through Instagram first, then TikTok and YouTube Shorts for the related formats. 100 rows is the bar, and it is not arbitrary. Under 20 rows it is noise, because a single outlier looks like a pattern when there are only 15 of them, and 1 lucky hit can be copied forever at the cost of the whole file.

The hook column is the most expensive one to fill properly. It takes the first-frame text verbatim - the exact words, the size, the placement - plus the first spoken line, and then it answers 3 questions. Did it register inside 1.5 seconds. Were the visual, the spoken line and the on-screen text all firing at once. Did it carry both topic clarity and a curiosity gap. Structure is the beat arc: hook, context, the first value hit, development, payoff, and the call to action. The 2 things that matter are where the open loops are opened and closed, and whether the value arrives frontloaded inside about 4 seconds or buried at 15. Pacing is logged against the benchmark rather than against taste, which is the only way 100 rows stay comparable. The top-creator benchmark runs about 19 shots inside the first 30 seconds, a change roughly every 1.77 seconds. Then the rhythm curve gets written down as a shape: a fast open, a varied middle, and the beats where cuts cluster. 3 dimensions, 1 column each, and the 3 of them already describe most of what separates a reel that travels from 1 that does not.

Editing is a list of what the cut actually did: hard cut, match cut, whip, zoom, an S-curve speed ramp, a snap zoom, b-roll inserted over a talking head, a text reveal, a sound-design sting, or silence placed exactly where a sting was expected. Captions are logged as burned in or automatic, then the size, then the placement. The winners run 36 to 60 points and above, keyed to the speech, and kept inside the safe zone rather than under the interface where the app covers them. Visuals is where the muted half of the audience is decided. Lighting - a key plus practicals - depth, clutter, single-subject focus, the grade, the motion level, and whether the first frame carries motion, contrast, or a contradiction inside its own text. Half the hook is visual, which is why a transcript-only teardown reads a different niche than the 1 the audience watches with the sound off.

Length is recorded as an exact duration and then placed inside the niche's own length distribution, because a good length does not transfer between niches. The question is which durations cluster at the top of that niche. Storytelling records the point of view, the emotional arc, the relatability to the target viewer, the why behind the post, the stakes, and the personality signal that makes 1 creator recognisable across 50 posts. The call to action records 3 things: the type, the placement and the count. Follow, comment a word for a lead magnet, save, share, link, or nothing at all. Lead generation usually pairs 1 freebie with 1 comment word and a ManyChat reply. 9 columns, fixed before row 1, is what makes row 87 usable. Change a column halfway through and the table stops being a dataset and turns back into a folder of links.

A teardown is not watching reels. It is logging them.

The teardown is not finished when the rows are full. It is finished when the rows have been counted. Move 1 is the tally. Take the top 20 per cent by views and count what repeats: which hooks, which lengths, which calls to action, which pacing signatures, which caption styles. Move 2 is the combo. The hook, structure, pacing and visual bundle the outliers share is the niche's proven format, and it is almost never 1 element. It is 4 things that arrive together. Move 3 is the flops. The bottom 20 per cent carries the most valuable data in the whole table, because a format that fails 15 times tells you more than 1 that worked once. Move 4 is the gap: what every winner misses that your own angle already supplies. Small is fine. Small and unfilled is the whole opportunity, and it is the only row in the report that competitors are not mining.

5 pitfalls show up in almost every teardown that goes nowhere. Under 20 rows is the first. That is noise wearing the costume of research. Judging from the transcript alone is the second, because the first-frame visual and the captions carry half the hook, and a viewer scrolling with the sound off never hears the line the teardown was scored on. Copying a winner instead of finding the gap is the third. A copy competes with the original on the original's terms, and it loses that comparison every time. Ignoring the flops is the fourth, and it is the expensive 1, because the bottom 20 per cent is data almost nobody is reading. The fifth is the scoring question. Would my followers like this is the wrong test. The stranger test is the right 1, and it is the only 1 that predicts reach outside the people who already follow.

The file closes with a written pattern report: the niche, the sample size, and then the winning hook, structure, pacing signature, editing, captions, visuals, length, storytelling and call to action - 9 lines, 1 per dimension, each 1 supported by the count that produced it. Then the gap line: what all of them miss, and the angle that fills it. Then the ideas. 10 to 20 one-line concepts, each built on proven ground with 1 thing nobody else is doing. 20 lines is a schedule. 100 rows is a strategy. This is how the physical builds were done. 200 wood-gasification machines across the UK and Europe. 210 energy systems across Africa and Asia. Companies built in 12 countries, 1 industrial group of EUR 75 million restructured, and a book of 393 pages published on 8 April 2026 - each 1 measured against its own sample before it was repeated. The probability that a teardown of 100 rows produces no usable pattern is close to nil. The failure mode is never the data. It is stopping at 15 rows and calling it research. The next 100 reels are already public. The only thing missing is the table.