How Paris's Digital Archives Ended Up Full of Ghost Images, And Why It's Finally Being Fixed

Paris's municipal digital infrastructure is carrying a problem officials have known about since at least 2022 but only began systematically addressing this year: tens of thousands of duplicate images embedded across city-run platforms, from the Grand Paris Express project portal to the Ville de Paris open-data repository on data.gouv.fr. The redundant files are not merely a storage nuisance. They have distorted planning records, slowed public-facing databases and, in several documented cases, caused the wrong building facades to appear in official permit documentation.

The timing matters. The city spent much of 2024 and 2025 rushing to catalogue post-Olympics legacy assets, newly renovated quaysides along the Seine between the Pont de Sully and the Pont d'Iéna, the repurposed facilities at Saint-Denis, and the upgraded public realm around the Stade de France, and that documentation push was done at speed, under political pressure, with minimal deduplication protocol. The result was a digital estate that grew fast and dirty.

Where the Accumulation Started

The roots go back further than the Olympic sprint. When the Société du Grand Paris began publishing construction progress imagery for the 68-station metro expansion, now running behind its original 2024 completion targets on several lines, it used an automated ingestion system that did not flag near-duplicate photographs taken at the same site on consecutive days. By the end of 2023, internal audits cited in public procurement documents reviewed by The Daily Paris found that certain station construction folders, including those for the Arcueil-Cachan and Bagneux stops on Line 15 South, contained between three and seven visually identical images tagged with different metadata timestamps.

The Atelier Parisien d'Urbanisme, known as APUR, which maintains detailed cartographic and photographic records of the city's built environment, flagged the problem in a working note circulated to city hall in the spring of 2023. APUR's concern was specific: duplicate images were being indexed by the city's GIS system and, when queries were run for planning consultations in arrondissements like the 13th and 19th, both subject to intensive redevelopment, the duplicates were artificially inflating the apparent density of photographic evidence, making it harder for planners to identify genuine changes over time.

What Deduplication Actually Involves

The fix is not simple deletion. Images that appear identical may carry different legal metadata, licence terms, photographer attribution, rights restrictions, and simply removing one copy can strip a database of a record it is legally required to retain. The Direction des Systèmes et Technologies de l'Information, the city's IT directorate, confirmed in its 2025 annual budget presentation that it had allocated €2.3 million toward a multi-year image governance programme, with deduplication as a core component. That figure covers tooling, staff training and an audit of roughly 1.4 million archived files held across city-managed storage.

The practical work began in January 2026 at the Bibliothèque historique de la Ville de Paris on the Rue de Rivoli, where the oldest digitised holdings were processed first. Staff have been working through collections section by section, using perceptual hashing software to identify near-matches before a human reviewer makes the final call on which version to retain. The process is slow by design: archivists estimate they are clearing around 12,000 files per month, which means the full legacy backlog will not be resolved before late 2027 at current pace.

For residents and researchers who rely on public platforms, particularly those using Paris's open planning portal to track building permit histories in neighbourhoods like Belleville or the Goutte d'Or, the practical advice from APUR is to cross-reference any photographic evidence pulled from the city's database against the cadastre records held on geoportail.gouv.fr, which runs a separate, independently maintained image archive. Until the deduplication programme catches up, that second check remains the only reliable way to confirm that a reference image genuinely corresponds to the address it claims to document.