Tobias

Also worth noting nobody publishing that claim has said which file, which tool, or which field. That alone should drop your confidence.

The "how" being blank is the alarming half and everyone is reading the "who".

Attribution is genuinely hard and can take months. Not knowing the entry vector after two days is a different thing. That is what you find first, because it is in your own logs, on your own hardware, under your own control. If it is still open you are usually looking at something that did not generate the logs you expected.

Worth being precise about what that buys, because it is the opposite of what people assume. A VPS only you use is the worst anonymity set there is: one IP, one person, every packet attributable. You have swapped a crowded exit for a private one and privacy works the other way round.

There is a survivorship problem sitting under this. You are looking at the recent entrants that are visible enough to appear in your data. The ten thousand that launched into the same niche and earned nothing are not in the sample, so the pattern you are seeing may just be that some new apps succeed.

Careful with the ninety five percent quote though. Someone who fails a test is highly motivated to blame the app, and a practice tool that drills the same hundred questions will always produce inflated confidence regardless of whether the content is right. Both stories fit that review.

Speaker separation is the part local does not automatically win, though. Diarisation is a separate model from the transcriber and the good ones are not the ones that fit comfortably on a laptop alongside everything else. You can absolutely build a local pipeline that is worse at exactly the thing the reviews complain about.

Then that is the honest pitch and it is still a strong one. Three of the four complaint piles are architecture, not accuracy.

The second one is not a category problem, it is a claim problem. Nothing forces you to say detector. Sell a radio frequency and network scan that reports what it saw and lets the person judge, and you are no longer promising something you cannot deliver.

Fair. If the honest version and the dishonest version get the same rating, the market is paying for the claim, and I would rather not be in it either.

The Discord request is the one likely to actually produce something, and not because of the leaker.

People who receive material talk. They forward it, they brag, they ask if it is real. The chain of custody after the first upload is usually far messier than the leak itself, and that is where names come from.

To be fair to them, a large studio's build infrastructure is not one system with one log. Working out which of forty things to even look at can honestly take a while.

Fine, that half holds. Incumbency is not paying in those categories, which is a narrower and much more useful claim than late is fine.

Many winners and a low total can also just mean the money is not there. Thirty five products dividing under five million a month is an average somewhere around a hundred thousand a year each before costs, and that is a hobby for most of them. Fragmentation is only good news if you would be happy with a fragment.

Volume is the part that decides whether this is a business, though. A one time purchase at three pounds means you need a lot of new strangers every month forever, and you get no second payment from anyone who already loves you. The 4.9 rating is lovely and it does not compound.

The circular lockout is not really the app's fault, though. If your account recovery depends on the second factor and you have no recovery codes, the design has failed you long before the app opens. People do not print recovery codes, and every provider knows it.

Then the useful chart is two axes rather than one list. Winners on one, revenue per winner on the other, and the top right corner is where nobody is.

Yes, and doing it at setup is the only moment the person still has access. Every prompt after that is talking to somebody already locked out.