Pluralism and viewpoints

A dossier groups the events of a same case : the initial fact, then the investigation, tributes, political reactions, and the resulting controversies. Each sub-event keeps its own category, severity, sources, and viewpoint comparison; they are never merged.

After each collection run, new events are compared with events from the last 7 days and open stories: shared locations, people, and organizations, rare words in the title, same department. Plausible matches are then submitted to an AI (DeepSeek, via OpenRouter), which only groups them if they share the same triggering fact — not just the same theme or city. A story only exists from 2 events.

The story title and summary are written by AI from its sub-events, and updated when a sub-event arrives or changes (at most once an hour). The story severity is the highest of its sub-events. After 7 days without activity, the story is closed: it can still be read but takes no new sub-events. An administrator can detach or move an event or rename a story; the automation then respects these choices.

The timeline “Hottest” rates subjects according to activity in the chosen period (6 h, 24 h, 3 days or 7 days). A story counts only once there, with the activity of all its sub-events.

• Activity : each article published during the period counts, with more weight given to recent ones (its weight is divided by e ≈ 2.7 every third of the period);
• Diversity : activity is multiplied by 1 + 0.6 × ln(1 + number of distinct sources), so that a subject covered everywhere ranks ahead of one hammered by a single outlet;
• Severity : multiplier from 0.75 (severity 1) to 1.75 (severity 5).

The displayed score is scaled to 100 for the top subject. The tendance compares the number of articles to that of the previous period of the same duration: up by more than +25%, down by less than −20%, “new” if there was nothing before.

The page 2027 presidential election measures the Place of declared candidates in the media tracked by Acturadar. It measures neither their popularity nor their chances: a much-cited candidate may be cited for a controversy, a little-cited candidate may be very well supported.

Article attachment. An article counts for a candidate when their full name (or a variation held by the administrator) appears in the title, lead paragraph or extracted persons, and that the article's AI analysis confirms it really is about him (his actions, statements, his campaign, a case involving him). A simple citation, a list of names or a homonym do not count. Posts on X do not count: only media articles are measured, and the candidates' official accounts are displayed separately, for information.

Neutrality. Default ranking according to the volume of articles in the period (a measured fact, not a judgment); the visitor can choose alphabetical order or ranking by sponsorships. Same presentation and same colour for everyone, no party colour, no forecast, no "winner" or "loser", no popularity score.

• Articles : articles attached over 7 or 30 days; the tendance compare them to the previous period of the same duration (same thresholds as “Hottest”).
• Outlets by side : distribution according to media positioning (see below); uncategorised media are counted separately. It is the media's side, never that of the article.
• Average tone (-1 to +1) and share of sensationalist headlines (sensationalism score ≥ 0.5): headline reading signals, calculated for each article (see “Headline signals”). A negative tone may reflect the subject (a drama, a case) more than bias.
• Main topics : keywords of the events to which articles belong, weighted by the number of articles.

Polls: strictly separated. They are displayed only if the visitor activates “Show polls”, in a separate block, served by a distinct route, and are never combined with Acturadar indicators. They are automatically collected (web search + AI) with their source, then checked and validated one by one by the administrator before publication: institute, sponsor, field dates, sample, results, link.

Sponsorships. It takes 500 signatures of elected officials (mayors in particular) to appear on the official list. The “Parrainages” column displays the official count from the Conseil constitutionnel as soon as it is published (from early 2027, automatically imported from its open data), dated. Before that, it displays the claimed figure by the candidate or their team, marked “claimed”, with its date and source: it is spotted by the AI in the collected articles and in web pages found by a daily Google search for each candidate, then verified and validated by the administrator. These pages are read via a collection service (ScrapingBee) then summarised by an AI (ScaleDown): only the summary is kept, never the page text; only the latest validated one is shown. A claimed figure cannot be verified and may be inflated. Sponsorships are not an Acturadar indicator : they are shown alongside, for information. Once official counts are published, candidates below the threshold set by the administrator are flagged and may be hidden.