# Method note for Part 1: counting protest in India, 2000 to 2026

Draft, 2 October 2026. Everything here is provisional until the ACLED and news-event layers are in.

## What is being counted, and by whom

"Protest" has no official definition in India and no official count. Each source below counts something adjacent to it, and the differences between them are part of the finding.

| Series | Unit | Who records it | What it misses |
|---|---|---|---|
| NCRB rioting cases | a First Information Report under IPC 147 to 151 (from 1 July 2024, BNS 189 to 191) | state police, compiled by the National Crime Records Bureau | every protest that produced no FIR; counts depend on each state's registration practice |
| NCRB riot motives (2014 onward) | the motive the police attributed to a rioting FIR | state police | the same; and NCRB's motive list changed in 2017 (money, land, family, enmity and electricity disputes were added, so "Other" shrank) |
| Internet shutdowns | a suspension order under the Telegraph Act rules or section 144 | SFLC.in from press and user reports; Access Now from the #KeepItOn coalition | orders that were never reported; the two trackers disagree by 10 to 20 a year because they count districts and renewals differently |
| ACLED protest and riot events (2016 onward) | one event per place per day from press monitoring | ACLED's India coders | the same event in several towns is several events; coverage depends on the press |
| ICEWS, GDELT (1995 onward) | machine-coded news sentences | automated | double counting, wrong geocoding; mostly English-language press |
| The curated catalogue | one row per nationally significant protest | this project, from press and official sources | small and local protests by design |

## Rules

1. A series is never spliced silently. Where two NCRB editions give a figure for the same year, the later edition wins and the table name is stored on the row.
2. Rates use World Bank mid-year population for all-India and NCRB's own rate per lakh for states.
3. 2024 NCRB figures sum IPC and BNS cases, as NCRB itself does in the "Total" column. The change of statute is marked on every chart.
4. Crowd sizes in the catalogue keep police, organiser and press figures in separate columns. No figure is ever averaged or invented. "n/a" means none was published.
5. Every claim of foreign funding or amplification is recorded with who made it, what evidence was cited and any rebuttal, in the same row.
6. Reported-event series (ICEWS, GDELT) are shown both raw and divided by total India news volume in the same source, because the growth of media is itself a cause of growth in reported protest.
7. Nothing from the earlier research session on another machine is reused; every figure here was rebuilt from the sources named.

## What the event record shows

- ACLED demonstrations: 10,864 (2016), 17,899 (2019), 15,624 (2022), 21,810 (2024), 20,572 (2025), 17,737 in 38 weeks of 2026 (about 24,300 annualised). Peaceful protests rose from 8,895 to 18,893; violent demonstrations fell from 937 to 706; deaths in demonstrations from 170 to 15.
- Test stated in the plan: the rise after 2022 (32 per cent to 2025; about 55 per cent with 2026 annualised) is not larger than the rise of 2016 to 2019 (65 per cent, partly ACLED's own coverage ramp). Verdict: a real, continuous rise, not a break.

## What the official series show so far

- Registered rioting per 100,000 people peaked in 1981 and has fallen in most years since; 2024 is the lowest rate in the series.
- Within rioting, the motives that resemble protest (communal, political, caste, agrarian, student, "during andolan or morcha", attacks on police) fell from about 11,800 cases in 2017 to about 2,900 in 2024. Agrarian riots peaked in 2016. Most registered riots since 2017 are private disputes.
- Internet shutdowns peaked in 2018 on both trackers and have fallen since, with a second rise in 2023 (Manipur, Haryana) and the lowest figure since 2017 in 2025.
- A falling riot count is consistent with both readings of the premise: fewer violent confrontations, or less registration. It says nothing yet about peaceful demonstrations, which is why the ACLED layer is decisive.

## Known gaps (2 October 2026)

- SFLC.in totals for 2020 and 2021 are not yet confirmed from SFLC's own reports (press figures vary); 2025 SFLC total not yet published in a form found.
- Access Now figures for 2016 to 2021, 2023 and 2024 were entered from memory of the #KeepItOn reports and are flagged "unverified" in the raw file until checked.
- ACLED: the aggregated weekly export for India (2016 to 19 September 2026) was supplied by the editor on 2 October 2026 and is kept out of git under ACLED's terms; the yearly, monthly and state aggregates are committed. GDELT, ICEWS and Mass Mobilization remain blocked; the news-volume normalisation is therefore not done, and the brief says so.
- Labour Bureau disputes series collected 2000-2023 (2004, 2005, 2024 missing; 2020 disputed between sources). MHA communal incidents collected 2004-2017.
- Catalogue: 217 rows, all with two or more sources; 2014 not yet covered; police crowd estimates absent for almost every row before 2023.
- Election calendar: 2025-2026 rows verified; 160 older rows carry dates from memory and need a spot check against ECI.