Build journal
Each article argues something specific and says what it is still unsure about. There is nothing here written to rank for a phrase.
Why ITISYOU News keeps sources, corroboration, editorial review and revisions as separate signals instead of collapsing them into one confidence score — and why it is still labelled Working rather than launched.
The argument
Most news products answer what is happening. The harder and more useful question, for anyone who looked away for a week, is what moved since you last checked.
Why DEV.ITISYOU attaches evidence to a revision instead of to a page, keeps records of fixes that failed, and refuses to show a confidence score without the counts behind it.
The argument
The problem with a highly upvoted answer is not that it was wrong. It is that it was right, in an environment nobody wrote down, on a version nobody recorded.
I assumed that if I collected almost no personal data, the legal obligations would mostly not apply. That assumption was wrong, and finding out why changed how I build.
The argument
I tried to design my way out of an obligation. Designing well reduces risk; it does not remove accountability, and those are different things.
Every line traced to a confirmed claim. The formatting was parser-safe. It passed its tests. I still could not charge for it, and working out why took longer than building it.
The argument
The generator worked. The documents it produced were correct, and they were still not good enough to sell, which is a harder problem than a bug.
Merging a PDF should not require handing a stranger your document. What it takes to do this work inside the browser, what it costs, and where the approach genuinely does not fit.
The argument
Uploading a file to compress it is a reasonable engineering decision with an unreasonable privacy cost, and for most everyday tasks the upload is no longer necessary.
What actually happened when a trained variational circuit ran on real IBM hardware: the accuracy table, the noise that damaged confidence but not verdicts, and the prediction that turned out wrong.
The argument
A four-qubit circuit matched its classical twin and lost to plain logistic regression. Writing the prediction down first is what turned a disappointing result into a useful one.
How a masters degree, a shift pattern and one small calculator turned into a network of five products — and what the first year actually taught me about shipping.
The argument
The network was not planned. It came from finishing one small thing and discovering how much of the work had nothing to do with the code.
An eighth article was planned, on choosing the right AI model rather than the most expensive one. It is not published because the evidence for it is a social post rather than a project record, and writing it anyway would have made it the one piece of filler on a site that argues it carries none.