This site measures what political leaders say — and what they do not say. Complete archives, identical rules for every political camp, published data, published corrections. This page is the contract: how we measure, what rules bind us, what vocabulary we use, and every error we have made and corrected.
1. Complete export. We download the complete archive of original tweets and replies from the official public account (no retweets), with a declared cut-off date in each profile.
2. Classification. Each tweet is classified using keyword dictionaries (Spanish and French) with word-boundary delimiters.
3. Crisis scoreboard. A fixed dictionary of 17 humanitarian and international crises — identical for all profiles — produces the raw count per crisis, with the reference scale of each attributed to its sources (UN, IPC, ACLED, ministries with their affiliation declared).
4. False-positive verification. Every count is verified by reading the tweets. No figure is published without this reading. The list of eliminated false positives is public (section 5).
5. Direction reading. Counting is not enough: every mention of a sensitive crisis is read to determine its direction — solidarity with victims, defence of the perpetrator, domestic instrumentalization, neutral coverage.
6. Article with fairness box. Every profile must include the section "What the method compels us to acknowledge": the real commitments, nuances and exceptions of the analyzed leader.
7. Pre-publication audit. Every aggregate figure in the text is recomputed from the dataset; every claim of "collective zero" is verified cell by cell; every condensed sentence is re-verified against the original tweets. The editor carries out a final hostile reading.
We analyze exclusively public figures and their publicly assumed statements — the official account, never attributed anonymous accounts. The method is applied symmetrically: the series includes profiles from the radical left (10) and radical right (2) under the identical dictionary. Corrections apply in both directions: three of the most important increased the counts of the profiled leaders (Obono, Hassan, Bardella). Every profiled leader has a right of reply — we will publish any response linked to their profile. Profiles are living documents: quarterly re-export with a change note.
Register. We use the vocabulary of international wire services (EFE, Reuters, AFP): "West Bank", "settlers". Real terminological disputes — such as the American legislative wave for "Judea and Samaria" (Arkansas 2025, Tennessee 2026) — are documented as objects of analysis, not adjudicated.
Attribute, never adjudicate. Disputed institutional determinations are cited with their source and their contestation: the famine in Gaza is "declared by the IPC system (August 2025), which Israel disputes"; the Gaza death toll is "from the Gaza health ministry — an overall figure that the Israeli military considers approximately accurate, disputing the proportion of combatants"; Hamas's obstruction of aid is "condemned by the UN (July 2026)" with Hamas's denial noted.
Loaded words as data. "Terrorist", "genocide", "hostages", "apartheid" are not this site's editorial vocabulary: they are objects of measurement.
Israel/Palestine · Ukraine · Syria · Yemen · Afghanistan · Iran (repression: 2022 + January 2026 massacres) · Sudan · Mali/Sahel · Nagorno-Karabakh/Armenians · Tigray · Rohingya · Uyghurs · Boko Haram/Chibok · Nicaragua · Cuba · Venezuela · Hong Kong. The list expands when current events require it and every expansion is applied retroactively to all profiles.
"irán" (the Spanish verb "to go") · "crisis"→isis · "ministerio"→amini · "pasado"→asad · "ambassade"→assad · "Amancio Ortega" vs "Daniel Ortega" · "San Isidro" (×4) · "@ieguiguren"→uigur · "#LaMarchaMásGrande"→hamas · "Bahamas"→hamas (×3) · "Moura"→Mourad · "secesionista"/"pensionista"→sionis · "#PatriaYVida" (false negative). Every correction is permanently incorporated into the dictionaries.
Nine audit entries (E1–E9). E8 — the January 2026 Iranian massacres were not covered: detected by the editor, all 12 profiles were reanalyzed. E9 — "sionis" captured "pensionista" and "secesionista": the 10 Israel/Palestine counts were corrected downwards (1-8%). Corrections in favour of profiled leaders: Obono Iran 0→15, Hassan Iran 0→4, Bardella Iran 0→5. Falsifiability test: the supremacist-attack battery was applied to all 12 profiles following an editorial challenge — the comparative result was published, nuancing and sharpening the original finding.
In addition to the permanent profiles, this site publishes 72-hour reaction grids after qualifying events. The format: who said what, verbatim; who said nothing; latency in hours; vocabulary used. A 400-word sober note accompanies each grid.
Qualifying event criteria (pre-defined — not retroactive). An event triggers the grid if it meets at least one of these thresholds, concerning a country or actor already present in the 17-crisis dictionary:
The criteria are published before the next event. No event is selected retroactively. If an event does not meet the threshold, no grid is published — even if it generates intense media coverage.
Grid methodology. The measurement window is 72 hours from the first confirmed report of the event. Recorded: each leader's first tweet on the event (timestamped in hours since the event), the exact vocabulary used (does it use "terrorist"? does it name the perpetrator? does it mention victims?), and silence if no tweet within 72 hours. The same verification and direction rules as in the full profiles apply. The grid is published within 72 hours — in real time, not retroactively.
We do not analyze private individuals or anonymous accounts. We do not adjudicate factual disputes: we attribute them. We do not infer intentions: we measure presences, absences, directions and vocabularies. We do not erase errors: we correct them in public. And we exempt no one: the method that documented the silences of ten left-wing leaders documented, under the same rules, those of two right-wing leaders.