634 down. 94 up.

Last week: 132 districts still very high risk. West Singhbhum. Nandurbar. Thirty-eight of them in Madhya Pradesh.

Part 1 → Two nutrition numbers · Part 2 → 132 districts

This week: the film, not the photograph.

April 2023 to August 2026. Same districts, every month, a score we can add. 739 of them have the full 41 months. 729 are usable — we throw out the near-zero 2023 baselines that would fake a miracle.

Of those 729:

• 634 got better.
• 94 got worse.
• 1 did not move.

The dashboard, as a national object, is not stuck. It is also not kind in every direction.

NFHS cannot see this. One wave. The Tracker can. That is why Part 1 bothered you with two thermometers.

The states that fell

Average district score, April 2023 → August 2026:

Slope and level are different sentences. Gujarat can be the best slope in the large-state file and still have Narmada and Chhota Udepur on the 132. Madhya Pradesh can improve and still be the pile.

Clean district falls, among names you can trust: Botad, Panchmahal, Dahod, Tapi. Tarn Taran, Fazilka. Dantewada.

We do not open with Kra Dadi in Arunachal, which drops from a 100% stunting baseline. That is a glitch until someone proves it is a child.

The states that rose

Delhi. Eleven districts in the panel. Ten got worse. Central, North, South East, South, New Delhi, North East. Measurement coverage in the mid-90s. This is not “we finally counted the poor.” This is the capital moving the wrong way on its own dashboard.

Telangana. Twenty-six of 33 districts worse. Six of them are already on the 132: Asifabad, Vikarabad, Adilabad, Narayanpet, Mulugu. Started better than Gujarat in 2023. Not the story people tell about Telangana.

Tripura. Six of eight worse. Six of eight very high risk today. Slope and level agree. Both are bad.

Jharkhand improved on average and still owns West Singhbhum. A district can fall and remain the worst named place in the file.

How to read a slope without lying

A falling score while more children enter the roll can be composition, not height. Enrolment drives do that. Treat a summer collapse as a data event until the same children show up later.

A rising score with 95% measured — Delhi — is harder to explain away as “we found more kids.”

A ministry line that “indicators are improving” can be true in this part and still leave you with last week’s 132.

This week’s only job

Progress is a slope. It is not a press release.

Next — Part 4: Four states hold half the red . Never mind who moved. Who holds the red now. Four states. Half the list.

Data: Part 3 of 5. Balanced panel of 739 districts, April 2023–August 2026. Usable trend = 729 (near-zero 2023 baselines out). Score as in Part 2. Programme data, not NFHS. Also on LinkedIn.