Reelly’s market is real, its ad data is honest, and its render engine works. The shipping build takes the first 1.5 seconds of every clip, sorted by capture date. Every strategic question asked so far was asked about a product nobody has run.
That is the whole selection logic: sort by timestamp, take the head of each clip. Meanwhile
ios/AutoClipping/Indexing/ is built and benchmarked. Six VLMs evaluated,
a two-model split decided, 3/3 on short-clip golden-moment ground truth and 4/4 on
long, per-clip screenplay beats, roughly four cents a reel. Nothing outside that
folder calls it.
It builds the container and stops there. From the app’s perspective the entire selection layer is dead code.
The left column cost $276 and three weeks. The right column is what decides whether there is a business, and none of it can be bought with ad spend.
The kill criterion in the go/no-go doc is a 30% D7 second-reel rate. Cohort member one has infinite motivation, zero install friction, and wrote every line of the render path. His rate is zero. That reading was free, and it is worse than any number 26 strangers would have produced after a month of TestFlight review and analytics wiring.
A near-identical incumbent runs roughly $7.2M a year on the same one-line pitch. That question is closed and does not need re-asking.
61% of its recent reviews are one to three stars against a 4.78 lifetime average, and the top complaint is the paywall rather than the product. A generous free tier is a real wedge.
One mention across 120 incumbent reviews. Every ad creative that delivered was travel footage, and the winner was a travel montage at $4.11 a signup.
Per-clip screenplay beats and golden-moment windows, benchmarked against ground truth at four cents a reel. CapCut does not know what is in your clips. You do.
Two apps launched in late 2025 are already at $62k and $101k a month. Distribution here is organic TikTok, not App Store charts, so incumbency is not decisive.
At the comp’s monetisation, $10k a month needs about 3,300 downloads a month. That is a solo-founder number, not a venture number. Aim there.
The wedge that won on ad performance is hostile to subscription pricing. Nothing in the plan answers this yet, and it should be answered before any paywall work.
Travel came from creative click-through, not from a person. Ad data tells you which video stopped a scroll. It does not tell you whose problem you solve.
Four cents a reel in a market whose loudest complaint is being charged before seeing value. The obvious wedge and the obvious unit-economics problem are the same decision.
Users name it in reviews. Any positioning that does not answer “why not CapCut” fails, and the answer cannot be transitions or music.
The go/no-go doc says another demand test is procrastination. It was written after running one, three weeks after the 26 emails arrived. The chosen path defers contact again.
Preview and export share one composition. If assembly makes segment lengths content-driven, that invariant is at risk. Finding out in week five is expensive.
You chose to rebuild around content understanding rather than wire what exists and re-decide. The objection stands and is resolved by ordering, not override: assembly logic sits on top of selection. If the moment oracle returns mediocre windows on real handheld footage, no narrative arc above it produces a postable reel.
Roughly 100 clips, not the demo set, through the head-of-clip path exactly as it ships. Export it and watch it. Half an hour, no engineering, and it is the before-shot you will want in week one.
Content-aware ordering replaces sort-by-timestamp. The golden-moment in-point replaces zero. No new UI, no new models, no new research. The renderer already accepts an arbitrary time range, so this changes an argument rather than the engine. Keep the no-API-key fallback intact.
Real account, real people who know you. Deciding not to post counts, and is equally informative. This is a gate, not a proof: a pass proves little, a fail is decisive.
Screenplay beats drive a narrative arc. Templates become story shapes the content selects rather than fixed cuts the user picks. This is the part that answers “why not CapCut,” and it is open work rather than wiring.
Signing, lanes, CI, and the App Store Connect record are all ready. Three sub-day blockers remain: the home screen still says AutoClipping, there is no analytics SDK, and the waitlist API only inserts a row. Hold anyway. Twenty-six first impressions do not come back, and spending them on the head-of-clip build buys a number that means nothing.
Put about 100 clips from one real trip on the device, run the build you have today, and watch the export all the way through.
Not the demo set. Not five samples. The real thing at real scale through the path that ships right now. You need the baseline before you change anything, and you have never actually done it — every export so far was a demo, an ad, or a bug repro. Thirty minutes, zero engineering, and it is the only item here that requires no code at all.
Design doc · ~/.gstack/projects/eugenechantk-reelly/eugenechan-main-design-20260808-001202.md
Sources · go-no-go-2026-08-05.md, format-test-readout-2026-08-06.md,
proving-demand-without-building.md, ios/AutoClipping/{Rendering,Studio,Indexing}
Session 2 · prior project fiftyworkout · signals: real problem, pushback,
domain expertise, taste, agency