Promises made. Promises kept?

A neutral, evidence-based look at how Tamil Nadu's ruling parties performed on their promises. Every promise in the manifestos of the parties that formed the government in 2016, 2021 and 2026, checked against official documents: policy notes, budget speeches, Government Orders and audit reports. Nothing here ranks parties or rates governments.

A promise counts as realised when an official step puts it into operation, with the date of the earliest such step.

  • –terms
  • –promises
  • –official documents

How much of each manifesto was realised

Each bar is 100% of a term's promises, split by what the evidence shows. Cool colours are realised (fulfilled, partial or diluted, with a date); warm colours are not. Switch the judge to see how much the answer depends on it.

Judge:

Share realised over the months of tenure

Curves use the date each realised promise was first operationalised, as far as the evidence now shows (Kaplan–Meier). The diamonds are what was documented at each landmark month; the gap between line and diamonds is evidence that appeared only later.

Status counts and who judged what

2 · What goes with faster delivery

Promise characteristics, under both judges

A Cox model estimates which promise characteristics go with faster delivery. A marker right of 1 means faster, left of 1 slower; the line is the 95% interval. Both judges are shown at once, so you can see where they agree.

The same numbers as a table

Is TVK keeping pace?

Projection under historical patterns, built from the two completed terms. It shows what the past would imply, with the assumptions stated below each line; it is not a prediction of what TVK will do. Bands show fitting uncertainty only.

Pace after five months

Like for like: TVK's documented count at month 5 against what the earlier government(s) had documented by month 5, judged by the same judge. 1.0 means the same pace.

How the pace factor is computed

Projection numbers by month

4 · Conclusions

Conclusions

The same evidence was judged by two language models, Gemini and Claude. Some findings are the same under either judge; others move with it.

5 · How far to trust this

Checked against 40 hand labels

    All accuracy tables · all limitations

    The work behind this

    A pipeline that collects official documents, extracts every promise, finds evidence as of any past date, has models judge it under strict checks and validates them against hand labels.