Jonas Ekwueme

It never fully stops and that is not the useful thing to measure. Ten years in I still look up date formatting strings and argument order for the less common standard library calls, every single time.

What does change is the shape of the searches. At four months you search for syntax. Later you search for behaviour, things like what happens to this when the connection drops halfway. The day you notice your searches have stopped being how do I write X and become which of these two approaches breaks first, that is the progress marker.

Alternative that often removes the need to copy at all: build the new structure instead of mutating. Something like new = [dict(d, status='done') for d in old] gives you fresh dicts and leaves the original untouched. Code that avoids in place mutation has far fewer of these bugs in the first place.

Separately, while you are in there: fetch does not throw on a 404 or a 500, it only rejects on network failure. So check res.ok and throw yourself, otherwise you will eventually call .json() on an error page and get a confusing parse error instead of a useful message.

My second sentence is definitely about the product. Rewriting the whole sequence around a single question.

Job adverts as a signal is clever, they literally list the painful manual tasks in the responsibilities section.

Also git reflog is the undo history for basically everything. Even after a bad reset you can find the old commit hash there for a good while.

And screenshot everything before you contact them. Statements, the original offer page if you can find it, the date you applied. Offer terms get taken down and then it is your memory against theirs.