Ministry tagging, 9 October

Half of the errors are state stories sent to the Centre

of the faulty articles are purely state, district or municipal matters that no Union ministry has any role in — a mayor handing out relief, a block officer building soak pits, a panchayat laying a pipe, a district tender dispute. Each one lands on a central ministry's desk anyway.

One tag causes most of the damage

Of the articles with a wrong tag, the Ministry of Jal Shakti alone accounts for of them. It attaches to any article containing water, river, flood, drought, canal, pipeline or drain, whether or not a Union ministry has any role.

What is being missed

The system over-tags far more than it under-tags, but the gaps are real and they repeat.

Why it happens

• Keyword matching with no sense of context. A single word triggers a tag. Real examples from today: an actor's obituary, a Navratri temple decoration piece, two IAS officers flying to a conference, and a frog-shaped swimming robot were all tagged to the water ministry.
• State and local stories routed to the Centre. Municipal flooding, district administration, state police cases and state election matters have no Union ministry, but roughly a third of the pool carries one anyway.
• Portfolio bleed. One minister holding six portfolios means every statement he makes picks up Space, Atomic Energy and Personnel regardless of subject. This is the main source of spurious Space and Atomic Energy tags.
• Nobody owns the leftovers. Each journalist reviews only their own ministry's tag. A story can be opened, correctly fixed for one desk, and left carrying four wrong tags belonging to other desks.
• The same story, tagged differently. Syndicated copy running in several papers gets different tag sets in each, so the error is not even consistent.

What was looked at

• Today's full edit log was read first: tag changes across articles by people (IST day, still in progress).
• of those articles still carry at least one ministry that an earlier audit found is wrong almost every time it appears. 300 were drawn from that group and read in full.
• Three reviewers worked separately, 100 articles each, without seeing one another's verdicts.

How each was judged

• Each reviewer read the headline and opening text, decided which Union ministries the story is genuinely about, then compared that to the approved tags.
• A purely state, district or municipal story was given no Union ministry rather than being stretched to fit one.
• Severity runs 1 to 5. Critical (4+) means a ministry desk is actively misled on a substantive national story. Moderate (3) means a clearly wrong tag worth fixing. Below that is minor or arguable.

Limits worth knowing

• Reviewers are AI models reading the stored English text. Treat each item as a strong lead, not a final ruling.
• The day is not over, so this is a partial day and the counts will grow.
• The 300 were deliberately drawn from the highest-risk group, so the 81% error rate describes that group, not the whole day's output.
• Several near-identical syndicated stories appear more than once; the same underlying error repeats across copies.