On this page
Every article carries a byline. If the byline is wrong, readers get a wrong idea of who chose the angle and who answers for the text. That is why AndroidLab runs a check before publication that compares the declared byline with the one the site rules calculate from category, tags, title, slug and excerpt.
If the two disagree, publication stops. This is not an abstract idea: it has already happened, and the case is a good example of the limits of this kind of automation.
How the check decides
The check is a script that reads the article metadata and scores three possible bylines: Kaffeine, Michele and Chiara. A keyword found in the tags is worth 8 points, in the title 6, in the slug 4 and in the excerpt 3. The body counts little: at most 2 points, so an isolated mention cannot overturn the angle declared by the title.
Laboratory sections are a separate case. AI Lab is always signed by Kaffeine, while RetroLab and Diario di Laboratorio are always signed by Michele. No calculation happens there: the byline is fixed.
Five tests, nothing published
On 1 November 2026 we ran the script by hand with test metadata and a minimal body, without creating any post. Here are the outcomes.
- Title about Samsung Game Booster, news tags, declared byline Michele: blocked, because the calculation returned Kaffeine. Exit code 78.
- Same metadata, declared byline Kaffeine: approved.
- AI Lab category with words such as Amiga and ssh in the title and excerpt, declared byline Kaffeine: approved, because the fixed section wins over vocabulary.
- Tags and title about privacy and updates: blocked as ambiguous, because Chiara and Kaffeine scored the same (14 to 14). The check does not pick at random, it asks for a review.
- No declared byline on a general publication: rejected. The byline is always chosen by whoever publishes.
The real case of 29 October
The editorial ledger keeps a note from 29 October 2026: a piece about Samsung Game Booster and a penalty reported by the press, for example by Yonhap, had been prepared with the byline Michele. The check calculated Kaffeine and stopped before upload. No post was created, no byline was changed, no quota was consumed. The note asked for a review of the angle or of the rule, not for tweaking the metadata to make the numbers agree.
What it does not do
- It does not judge whether the article is good or true. It only checks that byline and metadata are consistent.
- It uses hand-written word lists, so a missing term carries no weight. A score is therefore not a truth about the content.
- It does not decide for a human in ambiguous cases: it sends them to review.
- It does not replace reading the article: a text with the right byline can still rest on weak sources.
Why it is still useful
A simple check that stops with a clear error costs little and prevents a silent kind of mistake. The practical rule is this: a block should not be bypassed by changing the data that triggers it, but by working out whether the rule or the angle is wrong. The exit code 78 seen in the test is explained in the Python sys.exit documentation, which describes how a program reports its outcome.
In short
- The declared byline is compared with the calculated one before upload.
- AI Lab, RetroLab and Diario have fixed bylines; other news uses a metadata score.
- A real case on 29 October 2026 was stopped without creating a post.
- The check verifies consistency, not quality or truth.
Another limit of automatic checks is described in AI Lab: the guard checks form, not truth.