What this site can say, and what it does not say

A programme is a campaign text. It is often vague, sometimes incomplete. It comes out on a different date for each candidate and changes along the way. Translating it into parameters requires reading, choosing, sometimes assuming.

Some candidates have not released a programme. They still appear in the grid, with the mention “No programme released”: their positions are taken from their public statements, from remarks by associates or from their party’s documents, each dated and sourced. They are recorded and costed with the same method as the others; they commit the candidate less than a written programme and may change when it comes out.

We do not hide these choices: we show them. Each page quotes the exact text and its source. When we had to adopt a reading, the page says so. When the text does not allow a costing, the page says why.

They apply a published cost schedule, the same for everyone, to each position. They are static estimates: they count neither changes in behaviour nor effects on employment or growth. They are given with a lower and an upper bound, which reflect the gap between the sources read, not a statistical margin of error.

They do not say:

• what a measure would really cost. A law that is passed would have its own text, its timetable, its exceptions. The real figure would be different.
• what a candidate would do once elected. We read what they write or state. We attribute no intention.
• that a position that is not costed costs nothing. It is not costed because the text gives neither an amount nor a scope, or because the cost schedule does not cover that lever. Its cost is unknown, not zero.
• that a programme is complete. A summary that costs one position in five says nothing about the other four. That is why no total is published while fewer than 60% of the essential levers that can be costed are covered.
• that a programme that costs more is worse. The amounts are static, with no feedback effect on growth, employment or the revenue that economic activity would bring in or lose. Each programme rests on an economic model (demand-led stimulus, lower levies, consolidation of the public accounts…) that the cost schedule does not settle: see models and feedback effects.
• what you will gain or lose. The effects on households concern eight model households, which are reference points and not a sample, and only the measures that the calculation engine can simulate.

• No score, no ranking, no voting advice. The site does not say which programme is best.
• The symbols in the grid (↑, ↓, =) say in which direction a measure goes relative to current law. They do not say whether it is good or bad.
• A high cost is not a flaw, a saving is not a virtue: it is for everyone to judge.
• The online consultation is not a poll: it shows who answered here, not what the country thinks.

Recording hundreds of positions exposes us to errors: a quote attached to the wrong lever, a direction misread, a debatable coefficient. 9 of 770 positions remain to be reviewed.

• Every correction is dated and published in the errata.
• Campaign teams and readers can contest an encoding. A team’s reply is published on the page, together with ours.

• Each page gives the exact quote, its source and its date, with a link to the document.
• The method and the full cost schedule, lever by lever, are public, with extracts from the sources read.
• The positions, the cost schedule and the effects on households can be downloaded in CSV and JSON (files in French).
• The source code will be published when the site opens to the public.