The SIGMENT database was created in the framework of A. Vignoles’ doctoral thesis, which examined ecological dynamics with respect to the evolution of lithic archaeological units during the Middle and Recent Gravettian in France. This work combines typo-technological examinations of archaeological assemblages with Eco-cultural Niche Modeling (ECNM) to test the potential relationships between cultural and environmental changes in this chrono-geographic framework. ECNM employs a variety of methods aimed at estimating and comparing the eco-cultural niches of different archaeological units (i.e., the range of environments occupied by populations using said units). In practice, it consists of defining a mathematical relationship between the geographic distribution of archaeological sites where the archaeological unit is observed and environmental variables that characterize the conditions in which the populations operated. In so doing, it is paramount to ensure the quality of input data in order to avoid situations of “Garbage-in, garbage-out” in which models based on low quality data will inevitably have less interpretive power. The evaluation of occurrence data quality was conducted via a critical review of the literature organized in a relational database: SIGMENT. The aims of this database are to: 1) produce an exhaustive list of sites that can be attributed to the Middle and Recent Gravettian in France (and its margins to a lesser extent); and 2) evaluate the reliability of the occurrence of the archaeological units at each locality via the combination of bibliographic descriptions and direct personal observations. This database allows for a reliable estimation of the geographic distributions of Middle and Recent Gravettian archaeological units that can be employed in large-scale studies. The SIGMENT database is comprised of 21 tables centered around two principal tables: 1) “site”, which contains information about the archaeological locality; and 2) “ensemble” (translated as assemblage), related to a collection of artefacts contained stratigraphically and/or spatially within the site. Each bibliographic reference was scrutinized to extract a variety of information at the scale of the site or the assemblage—examples being taphonomic integrity, presence of other material culture elements and other archaeological cultural complexes, the dates of excavation and diverse data concerning lithic industries (mainly typo-technological data). Finally, we evaluated the presence of each archaeological unit at each locality by applying a qualitative reliability grid comprised of five levels. To ensure the exhaustiveness of this survey, we systematically analyzed the highest possible number of references for each site and made an effort to review the princeps publications and primary data in order to avoid posterior interpretations by subsequent authors. We started the survey with recent critical bibliographies that served to identify older references. In total, 196 references were analyzed. Their publication dates vary between 1910 and 2022, but most sources were published after the 1950s. It covers 166 sites and 254 assemblages, mainly located in France. To enhance the reliability of some sites, we personally reviewed or examined nearly 20 collections from sites located in South-western France: Roque Saint-Christophe, Facteur rock-shelter, Fourneau du Diable, Tourtoirac, Flageolet I, Grand Abri of Laussel, Grotte XVI, La Rochette, Combe-Saunière, Solvieux, Lespaux rock-shelter, and Les Artigaux. As a result of these examinations, we retained 75 sites for the Noaillian, 26 sites for the Rayssian and 15 sites for the Recent Gravettian. The geographic distribution of each of these curated data sets is slightly different from the total number of sites evaluated in the database for each archaeological unit. However, this approach is susceptible to enhance the results of macro-scale modeling procedures, since it allowed for the elimination of sites from the data sets for which the cultural attribution is not certain from the data sets. Using these vetted data sets in archaeological modeling will allow for more robust and reliable models. We followed the FAIR principles defined by Wilkinson et al. (2016). These principals are: Findable (public server, linked to a DOI), Accessible (Creative Commons 4-BY License), Findable (public server, linked to a DOI), Interoperable (data format and SQL language) and Reusable. Reusability of the data is facilitated by the fact that the data model was developed with the open-source software LibreOffice Base 7.5, with a description for each field. The present article also contributes to data reusability by describing the database’s scope and the methodology used to create it. We provide a discussion in order to alert future users to possible biases in these data and their potential implications. The data set is likely biased by the very nature of the methodology used to create it, i.e., a bibliographic review. The review demonstrated that publications yield unequal amounts of information from one site to another, and sometimes even for the same site or assemblage. The critical review of collections served to identify additional factors that may distort the geographic estimation of the Middle and Recent Gravettian archaeological units. Namely, unprecise excavation methodologies used in the early 20th century, and those still in use during the second half of the century sometimes significantly impacted the recovery of discreet fossiles directeurs. Finally, post-excavation and conservation activities also underwent methodological changes during the 20th century, which contributed to masking the presence of certain archaeological units. This discussion serves as a reminder that surveys based solely on the literature may be biased to a certain extent. It is our hope, however, that the integration of historical and archaeological reviews of older collections will allow us to establish better the geographic estimations of archaeological units by integrating formal data quality controls into examinations of archaeological data.