History of corrections

Council of Paris moments now have titles that say what happened

On place pages, the title of a Council of Paris moment often repeated the reason given in the session rather than what was voted. The titles were rewritten from the fact, in French and English, by a language model, then checked, with a random sample re-read: 1,109 titles across 189 places.

Source of the report: Internal initiative, while re-reading the Théâtre du Châtelet page on a phone.

Grants read from a budget are called “planned”, never “paid”

For some towns, the official list of grants comes from their budget (BP), which gives what is planned for the year, not from their accounts for the past year. These amounts were presented as paid. They are now called “planned in the budget”.

Source of the report: Internal initiative.

The main holder of a joint contract no longer changes from one update to the next

When a contract is awarded to several companies for the same amount, the company shown first could change from one update to the next (in Draguignan: MIDI.TP, then Comptoir Méditerranéen). It is now settled by its SIRET number, so it stays the same, from the national update that follows. And the page of a supplier no longer in a town's list now leads to that town's list of contracts instead of a page not found.

Source of the report: Internal initiative.

Twelve 2024 contracts are no longer counted twice

In 2024 the City publishes the awarded lot in its own list and the national platform (DECP) publishes the framework agreement, under another number: the same contract showed up twice. Rule applied: same amount, same notification day and a word in common in the description, the DECP line is removed. 12 pairs, €433M. The total of 2024 contracts goes from €3.18bn to €2.79bn, on the page, the detail pages and in the assistant's answers.

Source of the report: Internal initiative.

Investment spending on places no longer stops at the end of 2017

The City publishes the same series of programme authorisations in two datasets, split at the change of accounting plan (2009-2017, then from 2018). Only the first was read: on place pages, every operation stopped on 31 December 2017. Both datasets are now read. 24 place pages change, none goes down; the public money linked to places goes from €1,268M to €1,363M.

Source of the report: Internal initiative.

Towns' public contracts were counted seven times

To link each contract to its town, the calculation used a table with one row per town and per year: each contract was thus counted once per year of that table, seven times. There is now one row per town. Recomputed: 12,226 towns, 517,842 contracts.

Source of the report: Internal initiative.

“Paid more than agreed” is no longer presented as an overrun

On the San Francisco pages, a contract whose paid amount is above the agreed amount was flagged as an overrun. On checking the 129 active contracts concerned, 98.8% of the dollars are on rows where the City's register does not add up with itself: the agreed amount is not a ceiling there. The gap is now described as more paid than the recorded base amount, without calling it an overrun.

Source of the report: Internal initiative.

Method page restructuring

The /methode page was condensed into 5 clear sections (sources, transformation, enrichment, editorial choices, limits). The chart-level provenance badge now points to the public BigQuery datasets instead of a local description.

Source of the report: Internal initiative — the previous methodology page listed 12 sections, but an audit showed readers dropped after section 4.

• Page /methode ↗

Removal of "net contributor vs net beneficiary" framing

Panel §06 of /daily-bread showed, for each income level, a "difference" between what you pay and what you benefit from. This framing was both editorially loaded (close to political arguments) and technically questionable (inheritance, capital gains, company exits, hard-to-value public goods are not captured). Reworked into a neutral framing: "here's what you pay, here's how it's spent".

Source of the report: Direct user feedback on the misleading nature of the framing.

• Simulateur de prélèvements ↗