Proving LabMeasurements on browser tools, OCR pipelines and AI-assisted developmenthttps://provinglab.dev/2026-08-02T00:00:00ZProving Labhttps://provinglab.dev/https://provinglab.dev/icon-128.pngContent licensed for reading and quoting with attribution.Pages that are gone before you need them: eight moments in a degreehttps://provinglab.dev/notes/pages-gone-before-you-need-them/2026-08-02T00:00:00Z2026-08-02T00:00:00ZEight recurring situations in a degree where something on screen becomes evidence later — the submission receipt, a course room that closes, the version of an examination regulation that applied at enrolment, a statistics query that returns different figures next year. With what not to keep: other people's data, and teaching material beyond your own use. German version on the same page.Which page-saving extensions actually run on Firefox for Android?https://provinglab.dev/measurements/android-capture-extensions/2026-08-02T00:00:00Z2026-08-02T00:00:00Z248 Firefox extensions for saving or capturing a page, checked against the add-ons API on 2 August 2026. 60 declare Android support — including SingleFile with 85,724 users. Declaration is not function, and that distinction is where most claims fall apart.Web citations that vanish: what happens to a source after you cite ithttps://provinglab.dev/measurements/web-citations-that-vanish/2026-08-02T00:00:00Z2026-08-02T00:00:00Z150 web sources from reference lists, checked: 19.3% are gone, and the newest archived copy is a median of 603 days old. What that means for a thesis, and how to secure a source in the moment you cite it. Mit deutscher Fassung.Webpage to PDF for OCR: how much text survives? A measured testhttps://provinglab.dev/measurements/webpage-to-pdf-for-ocr/2026-08-01T00:00:00Z2026-08-01T00:00:00ZA full-page screenshot PDF run through Tesseract and compared against the original article text. 92.6% of the vocabulary survives, and the resolution threshold sits between 72 and 110 dpi. Numbers, method and commands to reproduce it.Six things went wrong building software with an AI assistant in one dayhttps://provinglab.dev/notes/building-with-ai-what-went-wrong/2026-08-01T00:00:00Z2026-08-01T00:00:00ZA single day of AI-assisted development, logged honestly: a wrong assumption about browser engines, a false claim that nearly got published, 22 local paths one commit away from a public repo, and three more. What caught each one — and what would not have.Print to PDF or capture the screen? The same page measured both wayshttps://provinglab.dev/measurements/print-to-pdf-vs-screenshot/2026-08-01T00:00:00Z2026-08-01T00:00:00ZFirefox can already save a page as PDF. So when is a screenshot capture worth it? The same article through both routes: 26 pages against one, 9 sentences cut by page breaks, and a text layer that print keeps and capture loses.Does your PDF extension upload the page? How to check it yourselfhttps://provinglab.dev/measurements/pdf-extension-permissions/2026-08-01T00:00:00Z2026-08-01T00:00:00ZBrowser extensions declare what they may access. Here is how to read that yourself in 30 seconds, what the current PDF extensions request as of 1 August 2026 — and why a permission is not proof of behaviour.An extension is only as dangerous as its permissions allowhttps://provinglab.dev/measurements/extension-permissions-risk/2026-08-01T00:00:00Z2026-08-01T00:00:00ZA year ago Mozilla warned that add-on developer accounts were being phished. That is why permissions matter more than trust: they decide what a compromised extension could do. A practical way to judge the risk before installing.