Should I vibe code
TikTok research tool that ranks videos against each creator's own baseline and labels what they did
Every scraper is one terms-of-service update away from being a dead weekend.
?
Their verdict, the Personal price and the build-time estimate come from their entry, MIT-licensed. Checked 2026-08-03.
?
Our verdict, the regret score and everything below it. Editorial and unsponsored — nobody can pay to be moved.
The honest answer
why the verdict is what it is
Ranking videos against a creator’s baseline is a nice analysis problem. Getting the data is the fragile part, and it is fragile in a legal direction as well as a technical one.
What actually breaks
not "if". the specific failures.
- Data access, since the numbers you want are not in an official API and collecting them another way is a terms question
- Ranking a video against a creator's own baseline, which sounds objective and depends entirely on a denominator you cannot see
- Survivorship bias, because what you can measure is what performed, and the videos nobody saw are invisible to your sample
- Metrics that shift meaning when the platform changes what it counts or how it distributes
- Building a public tool on someone else's data, where the people being ranked did not agree to be
The tool says a format works — videos of this shape outperform a creator's baseline by a wide margin. You make twelve of them and none land. The pattern was real in the data and the data only contained videos that already did well enough to be collected, so 'this format outperforms' actually meant 'among videos that succeeded, many had this shape'. The correlation was never evidence about what to make next, and nothing in the interface said so.
Is that you?
the verdict is a default, not a law
- Data comes from official APIs or your own account's analytics
- It describes what happened rather than predicting what will
- It is for your own content and you understand the sampling
- Collection relies on scraping a platform whose terms forbid it
- You present correlations as prescriptions about what to make
- You publish rankings of creators who did not opt in
- The sample is only successful videos and nothing says so
If you build it anyway
the checklist, then the prompt that enforces it
- Use official APIs and your own analytics. This is both the terms question and the reliability question — unofficial data breaks on their schedule.
- State the sample explicitly wherever a number appears: what was collected, over what period, and what is necessarily missing.
- Describe rather than prescribe. 'These videos share this trait' is honest; 'do this to grow' is a claim you cannot support.
- Address survivorship directly in the interface, because it is the flaw a viewer will not think of and it invalidates the obvious reading.
- Do not publish rankings or comparisons of named creators who have not opted in.
- Re-baseline when the platform changes, and show when the data was collected.
Before you build a social video research tool, apply these and push back if I ask you to break them. 1. Ask me where the data comes from. If the answer is scraping, tell me that is against the platform's terms and that it is also the fragile foundation of the entire tool. Use official APIs or my own account analytics instead. 2. Display the sample definition next to every metric: what was collected, over what period, and what is necessarily absent from it. 3. Address survivorship bias explicitly in the interface. If the dataset contains only videos that reached some threshold, say so where the conclusions appear — a viewer will not supply that caveat themselves and it changes what the numbers mean. 4. Frame outputs as descriptions, not instructions. Refuse to generate advice like 'post at this time to grow', and explain that the data supports correlation and not prescription. 5. Do not build public rankings or comparisons of named creators who have not opted in. 6. Show the collection date on every figure and re-baseline when the platform changes what it counts. 7. Rate limit and respect robots directives on anything fetched. 8. Store no more personal data about creators than the analysis needs, and set a retention period. 9. Out of scope unless I ask: automated posting, engagement automation, competitor alerting, follower scraping.
That one keeps you out of trouble. For the prompt that actually builds it, canivibecodeit.com has one.
their build prompt ↗Or don’t build it
the boring option, and the way back out
$5 a month is trivial, and the real question is where the data comes from rather than who wrote the interface. Buy or build on official APIs; a tool whose numbers depend on scraping is one platform change away from being wrong rather than merely offline.
$5/mo is cheaper than your weekend.
Keep your own collected metrics and their sample definitions exportable, since the analysis is only interpretable alongside what was measured and when. Assume the data source will change or close, and keep the analysis layer separate from the collection layer.
Unofficial Python wrapper for TikTok's web endpoints; gets you raw metrics, not a corpus, and breaks when TikTok changes
Questions
What exactly is survivorship bias here?
Your dataset contains videos you could find, which are overwhelmingly ones that got distribution. Any pattern you observe is a pattern among successes, so it cannot tell you what causes success — the same trait may be equally common in the far larger set of videos nobody saw. The analysis is not wrong; the obvious reading of it is.
Is describing rather than prescribing a meaningful distinction?
It is the difference between a defensible tool and a misleading one. 'Videos with this structure appeared frequently in the top decile last month' is a fact about a sample. 'Use this structure to grow' asserts a causal relationship you have no evidence for, and it is the sentence users will act on.
Every week, someone ships something they shouldn’t have.
New verdicts, the worst thing that landed in the trap, and the occasional incident report. No other email, ever.
The queue is a weekend. Posting to someone else’s account on their behalf is not.
It holds the keys to your public voice. A retry loop at 3am posts the same thing eleven times.
Every platform’s API breaks on a different Tuesday. That maintenance is the whole subscription.
last reviewed 2026-08-03 · verdict is editorial and unsponsored · shared entry data from canivibecodeit under MIT · not legal advice