IMSS workers' protests: organic outrage and coordinated amplification

The sample holds 316 posts mentioning «IMSS» over 15 days. The per-post classifier flags 61.4 % as coordinated, 7 % as artificial and 31.6 % as organic. The conversation takes off on 9 October, when protests were held in seven states, and grows until the 13–14 October marches.

There is a genuine organic conversation: 91 accounts sharing wage demands, testimonies and reports from the marches. Coordinated activity is larger: 185 accounts whose retweets concentrate on a small set of commentator and media profiles, and who direct mentions at official accounts.

The two groups never touch. No organic account amplifies a coordinated one, or the reverse; they only overlap in retweeting some of the same reporters covering the protests. These are statistical indications of coordinated amplification riding on real outrage, not proof of authorship or intent.

• From 9 October, volume rises from 2–6 posts a day to 49 (10 Oct) and 65–67 (13–14 Oct).
• Of the 298 accounts that post, 185 are coordinated, 91 organic and 22 artificial. There is no amplification link between organic and non-organic accounts in either direction.
• Of the 250 undefined accounts (they only appear retweeted or mentioned), 167 get more than half of their amplification from coordinated accounts, 57 from organic and 13 from artificial; 13 have no majority.
• 66 % of all amplification weight comes from coordinated accounts. Of the 187 accounts they push, 142 receive amplification from coordinated accounts only.
• Coordination works through retweets of the same sources, not identical fabricated text: only 17.1 % of non-organic content is repeated.
• Artificial accounts appear on both sides of the issue.
• The timing difference between coordinated and organic accounts is not significant (p = 0.14). There are no hourly bursts at this volume.

• Exact search «IMSS» in the local database (X captures), 30 Sep to 14 Oct 2026, Mexico City time, no language filter.
• Each post is classified as organic, coordinated or artificial with seven criteria about the account's behavior.
• Retweet and mention graph laid out with ForceAtlas2; in the 3D view, the vertical axis is time. An account is a "poster" if it posted in the sample, and "undefined" if it only appears retweeted or mentioned; in the graph it takes a faded tone of the category that gives it more than half of what it receives, or grey if none does. Each account's card shows which category the activity it receives comes from.
• Bursts: hours with robust z ≥ 4 (median and MAD) and at least 10 posts.
• Timing: one-sided Mann-Whitney test on each account's mean activity time.
• Everything is reproducible with exportar_analisis.py from consulta.json.

• Topology, not a census: the method identifies the shape of the network (who amplifies whom, whether groups touch, where amplification concentrates), and that shape shows up in a sample; it does not need every account. The sample has 316 posts (0.35 % of the 89,927 captured in the window). Percentages (61.4 % coordinated, etc.) describe this sample; what generalizes is the structure.
• Broad term: «IMSS» includes conversation that is not about the protest, such as service complaints or news.
• Categories come from the classifier: they are behavioral signals, not attributions of authorship, ideology or membership in any group. Undefined accounts are not classified: the share they receive from each category describes who amplifies them, not the accounts themselves.
• No country data: the sample cannot measure where accounts are based, so it can neither confirm nor rule out an international component.
• We do not know who is a worker: it is not known which accounts belong to IMSS workers. The split is between behaviors (organic versus coordinated), not between workers and non-workers.
• Variable capture volume: daily capture rose on 12 Oct, from about 5,750 to about 8,600 posts, so daily shares are given relative to what was captured each day.