This is what it does
Try it
InfluencerBooster reads a video’s retention notes back to you: paste what your analytics already told you, and it names the second the sharpest loss happens — quoting every figure back in your own words.
Runs in your browser — the product’s own engine, on notes you paste
- What it does: reads the figures your notes contain, shows each one beside the sentence it came from, names the sharpest measured loss, and offers advice as rules of thumb with the evidence attached.
- What to try: edit the notes and read again — delete a figure and watch it move into “not in your notes, so not stated”; press Show an empty paste to see it return an empty curve instead of a decorative one.
- What it runs on: the product’s own engine,
insights.js, copied into the page with its body unchanged (its sha256 is recorded in the file’s header). It is pure by construction — no I/O, no network, no clock — so it runs identically here. - What it does not do yet: there is no hosted service, no account and no automatic ingestion. The optional model pass exists in the product and is deliberately not in this page; nothing you type leaves the tab.
Status
Where this stands
- It is a private tool, not a service. It runs as a single Node process on your own machine and binds the loopback address only; there is no sign-up and nothing to download here.
- 44 tests pass, with no failures. They cover the engine, the memory loop and the HTTP layer: ten named checks each with a falsifier that mutates the engine and requires the check to go red, plus eleven regression tests for defects an adversarial review found in the shipped product.
- The honesty rule is enforced, not just documented. A check fails if any figure reaches the interface that was neither supplied nor computed. A canned figure is exactly what a retention tool must never produce.
- The advice is a second opinion. The cards are rules of thumb about hooks, pacing and endings, drawn from the notes you paste. Nothing here measures whether following one raises retention for anyone.
The numbers
Counts from the suite
44tests passing, zero failures
10named checks, each with a mutation that must make it fail
11regression tests for defects an adversarial review found
4755the port the local console binds on the loopback address
Under the hood
What it is built on
- Stack
- A zero-dependency Node backend with a deterministic, offline analysis engine and a plain HTML, CSS and JavaScript console. The optional model pass is opt-in and needs a key configured on the machine running it.
- Surfaces
- A local console at 127.0.0.1:4755 and an HTTP API for analysis, creator memory, and marking a recommendation applied or recording its outcome.
- Runs where
- Your own machine, as a single Node process, bound to the loopback address only.
- Data
- One JSON file per creator under the memory directory, holding the profile, the recommendations and their outcomes. No database and no account.
- Auth
- None. Every endpoint requires an explicit identifier, which is a filename rather than a security boundary.
- Tested by
- 44 tests: the named checks in-process, a falsifier that mutates the engine and requires the targeted check to go red before restoring the file and verifying its hash, plus regression and HTTP-surface suites.
Plainly
What this is not
- It is not connected to your account. It does not read TikTok or Instagram and it does not pull your analytics; you paste what you already have.
- It is not a prediction engine, and it is not a source of benchmarks. It has no idea what a good number is for your niche, and it does not invent one.
- It is not a counting tool. It will not tell you a total, an average or a percentage that you did not give it, and its own checks assert that.
- It is not uploading your work anywhere by default. Memory is a file on the machine running it, and the optional model pass stays off unless a key is configured there.
- It is not a substitute for watching your own video and deciding what you think is wrong with it.
Questions about InfluencerBooster
What it can tell you from the notes you have, or which figures it can compute from what you paste — ask first.