Twenty percent of the Brazilian Amazon has now been deforested, and deforestation rates are increasing. This compels us to evaluate the conservation potential of human-modified landscapes, yet the ecological value of regenerating and fragmented Amazonian forests remains poorly understood. To date, most faunal studies in disturbed forests have examined metrics derived from presence or abundance. Although valuable, these data cannot tell us how old-growth species are using disturbed forests. In this study, we complement presence data with breeding observations to assess more comprehensively the habitat quality of disturbed forests in central Amazonia. We synthesized breeding evidence from standardized passive mist-netting, mixed-species flock-following, and opportunistic nest-searching across a full annual cycle in 30-35-year-old secondary forest, 10-ha fragments, and undisturbed forest. We then used multi-state occupancy models to estimate the number of undetected breeding species in each forest type, which illustrated that fewer species bred in secondary forest (-43%) and 10-ha fragments (-17%) than in undisturbed forest. Both of these losses are larger than the associated decrease in species richness alone (-17% and -10%, respectively). Notably, we confirmed breeding by only three terrestrial and near-ground insectivores in fragments and secondary forest combined (of the 9 species found in undisturbed forest). Disturbed forests also supported fewer breeding individuals (-35-50%) and, in secondary forest, fewer successful breeding attempts (-24%). Encouragingly, however, some forest-dependent birds are breeding and producing fledglings in disturbed forests, including representatives from almost every guild. This was especially apparent for mixed-species flocks and army-ant followers, two guilds that have historically been considered vulnerable to anthropogenic disturbance. Therefore, despite a loss of breeding habitat in disturbed forests, these data suggest that landscapes composed of regenerating forest and small fragments have conservation potential for forest bird populations.

The archived dataset is the full encounter history for the main analysis in this paper (a single-season multi-state occupancy model). This encounter history includes 15 months (encounter periods) of data for 166 species, with each species*month denoted by one of three states: not detected (0), detected (1), and showcasing evidence of breeding when detected (2). (For the latter state, we considered a suite of physical and behavioral cues as evidence of breeding [Table 1].) Detection and breeding data were combined from standardized passive mist-netting and observational data collected while following mixed-species flocks. For more details on these two methods, please see the appropriate sections from the Methods: "Bird capture data" and "Flock-following observational data," respectively. Only the highest value was retained for each month. For example, if a given species was captured (detected = 1) in September 2015 and, while flock-following, was also seen feeding a recently fledged young (breeding = 2), then only a "2" would be displayed for that month.

For every species, there are six rows of encounter history data, one for each of the six research plots (or "Replicates"). These six replicates fit into three discrete forest types (or "Treatments"): secondary forest, 10-ha fragments, and primary forest. (For more details on these forest types, please see the "Forest types" section in the Methods.) Therefore, because each species contains six rows of data, there are a total of 996 rows (6 replicates*166 species = 996 replicates*species). Species taxonomy follows the South American Classification Committee (https://www.museum.lsu.edu/~Remsen/SACCBaseline10.htm), circa September 2019. Finally, the notation for the 15 months is given by the last two digits of the year, followed by the first three letters of that month. So, for instance, September 2015 is indicated by "15-Sep".