User:HearThatWhistleBlow

This account is a bot. It listens to the podcast feeds listed in Category:Podcasts and writes down what came through — one page per episode, using Template:PodcastEpisode.

It makes no editorial judgements, and it is not Magent. That distinction is the reason it exists as a separate account rather than a convenience.

What it does

For every episode in every feed, it records what the feed says: the title, the date it was published, the publisher's page for it, the audio file, and how long it runs. All of that is transcription. Anyone can check it by opening the feed, and anyone can reproduce it by running the importer again.

It also records what the episode appears to be about, as Has topic — which is how an episode turns up on the page of the person, band or festival it concerns. See Template:Firehose.

What it gets wrong

The subjects are guessed from the episode title by pattern matching, and pattern matching on human-written titles is not a solved problem. It has produced "Australia", "Part Two" and "Basic Passing Chords" as though they were people. It sometimes carries a description into the name, so that "Banjo Legend Tony Trischka" arrives as a different subject from Tony Trischka. It cannot tell whether somebody was in the room or was only being discussed, which is why it records a topic rather than a guest — an episode about a musician who died in 2020 is still about them.

Fix it at the source, not on the page. The patterns live on each show's own page, in a guest-patterns block; correct one there and the next run corrects every episode it affects. Editing an episode page by hand works until the importer runs again.

Why its edits are not marked for verification

Most bot edits on PickiPedia are wrapped as proposals and filed for a person to check. This account is exempt, deliberately.

A proposal marker means "an intelligence asserted this and somebody ought to look". That is the right label for Magent, which reads things and forms views. It is the wrong label for a transcription of an RSS feed, and several hundred of them would bury the handful of claims that genuinely wanted a human eye. The provenance is on the page instead, in Has episode url and Has audio url: they say exactly where the page came from, which is a better audit trail than a tag nobody can read through.

The topics are the part that is inferred rather than transcribed, and they are the part to be sceptical of.

Running it

Source is in cryptograss/pickipedia, under tools/:

podcast-episodes.py --json --with-wikitext > episodes.json
podcast-episodes-to-wiki.py --dry-run < episodes.json
podcast-episodes-to-wiki.py --write   < episodes.json

It is idempotent: it compares each page against what is already here and touches only what changed, so it is safe to run after fixing a pattern. It refuses to run under an account that is not in the exempt group, so that machine output cannot be signed with a reviewing identity by accident.

To stop it, revoke its bot password at Special:BotPasswords, or remove it from the exempt group — the importer will decline to write rather than fall back to writing badly.

The name

From the railroad songs, and because a whistle is a dumb signal: it blows because something is coming down the line, not because it formed an opinion about it. It sits alongside Blue Railroad Imports, which does the same kind of work for chain data.