What's changed

Release notes for THATSBees, newest first. Currently running v1.8.31, built 2026-10-11.

THATSBees — Changelog — 2026-10-11 (v1.8.31)

New

Watch an external-source refresh as it runs. On the data sources page, a running refresh used to show only Refreshing… until it finished, which for NatureServe can take ten minutes. It now shows which step it is on, a count and progress bar ("Fetching detail records — 1,400 of 1,944"), time elapsed, and roughly how long the step has left. It updates every few seconds, and it is still there if you reload the page or open it in another tab mid-run.
If a step stops advancing for two minutes, the line turns amber and says how long it has been stuck.
If the server restarted during a refresh, the source now says Refresh interrupted and the Refresh button works again. Before, it showed Refreshing… forever.

Fixes

NatureServe Explorer's Refresh button works again. On the data sources page, pressing Refresh for NatureServe Explorer failed straight away with "Cannot find module '../../../natureserve_processor'", and the source showed Last refresh failed. The server was missing the NatureServe loader. It now ships with it, and a refresh fetches the ranks and ESA listings NatureServe has republished since the last run.

THATSBees — Changelog — 2026-10-10 (v1.8.29, v1.8.30)

New

Upload a revised assessment from the species page. National admins now see an Assessment Source section under Conservation. It lists every revision of the species' assessment: the 2024 Google Doc, each regional revision, which one is current, its category, EOO and AOO, and the Drive document it came from.
To load a new revision, choose the regional _assessment.docx, its range-map .pdf, or both. Preview shows what would change (category, criteria, EOO and AOO, old and new side by side) and any ecoregion names that weren't recognised. Nothing is written until you press Load this revision.
The previous revision is kept and marked superseded.
A document identical to one already loaded is skipped.
A document for a different species is refused.
The upload is recorded in the species' history.
This replaces the re-import button on the old /ui species list, which had stopped working.
Find the bees nobody has assessed. The Assessment filter (formerly the "Has Assessment" checkbox) now offers Has assessment or No assessment. The map counts the same way.
SGCN by listing year. The SGCN filter can be narrowed to one year's state lists (2005, 2015 or 2025), on the species list and the map alike.
Red List breakdown of any list. Above the species list, a row of chips counts the listed species in each Red List category, plus how many have no assessment. Click a chip to show only that category.
EOO and AOO columns. With the Assessment filter on Has assessment, or a category chosen, the list shows each species' extent of occurrence and area of occupancy from its current assessment, and can be sorted by either.
How the maps were made. Range & State Occurrence now carries the assessment's own note on its map method, e.g. "Maps were created by generating a species distribution model…".
The API's home is /api. It is now the interactive API documentation, and its "Try it out" calls the server you're on. The documentation now covers the new filters, the assessment upload, and the original read-only /api/species… endpoints, which keep working.

Fixes

The old /ui pages showed superseded assessments as current. Species with a regional revision were listed once per revision, so Andrena amphibola appeared as both Least Concern and Vulnerable. A species' own /ui page could show its 2024 category and range instead of the current ones. The /ui pages are retired, and their links now open the matching THATSBees page. The site's address without /wiki now opens THATSBees too.
Re-importing an assessment failed. The 2024 Google Doc importer had not been able to write since assessment revisions were introduced on 2026-09-09. Every re-import was rolled back. It is now retired: POST /api/import answers with a pointer to the new upload.

v1.8.30

New

One floral table for every source. A species page used to split plant data in two: a Floral Hosts tree for the published host lists, and a separate Floral Associations table for iNaturalist and GBIF records. They are now one Floral Associations table with one row per plant and one column per source, grouped into three tiers: published datasets (Noori, Big-Bee, Bee-Gap, Fowler, BeeFunc), iNaturalist, and GBIF records.
Total, in bold, is the number of records of this bee on the plant across the sources that count records. Host lists that don't count (Bee-Gap, Fowler, BeeFunc) show a tick. Sort by any column; the default is Total.
With a state selected, the plant's own GBIF records in that state appear in a small grey column at the end. As before, plants not recorded in the state are hidden until you ask for them.
Female and male counts, the iNaturalist and GBIF record links, and the wetland ratings carry over from the old sections.
The Sources page has a new Floral data tiers section explaining the tiers and how they overlap.
Habitats and ecoregions on separate tabs. The Habitats section is now Habitats & Ecoregions. One tab holds habitat designations (IUCN habitats from the earlier assessments, and Bee-Gap and BeeFunc habitat classes); the other holds the CEC Level III ecoregions from the regional assessments. Neither is converted into the other. An empty tab says why: for example, a bee assessed only under the earlier scheme has no ecoregions.
Find bees by which scheme they have. The Habitat and Ecoregion filters each offer Has any habitat data or Has any ecoregion data, on the species list and the map.
Shade the map by state rank. The species map's layer menu has a new State S-rank (NatureServe) layer. It colours each state by the bee's NatureServe subnational rank, with a legend, and shows the rank when you hover. With a state selected, that state is outlined. It's on by default for any bee with at least one state rank.

Fixes

Ecoregions listed twice. For the 23 bees with a revised regional assessment, the page listed ecoregions from the superseded revision as well as the current one. Only the current revision's ecoregions show now (migration 067).
Old IUCN habitats looked current. 20 bees whose 2024 assessment was revised under the regional scheme still showed that assessment's IUCN habitats with no qualification. They're still shown, because they're the only habitat designation those bees have, but now marked as coming from the earlier assessment (migration 067).

THATSBees — Changelog — 2026-10-09 (v1.8.27, v1.8.28)

New

Two views of a state: its checklist, or all its records. "Count bees by" now chooses between two different lists, not two ways of counting one list.
Checklist shows only the bees on the state's chosen checklist, using that checklist's names and taxonomy. If Vermont's checklist treats Augochloropsis metallica as a misidentification of A. viridula, the Vermont checklist view says so.
Occurrences shows every bee an outside source records in the state, such as the USGS Bee-Gap database, assessments, or a GBIF occurrence, under the national taxonomy.
A state without a chosen checklist can only be viewed by occurrences. Its Checklist option is greyed out, and on the national map it shows as having no checklist.
Counts are accepted species; lists show every name. Every species count on the site now counts accepted species once. The species list has a new Taxonomic Status column (accepted, synonym, misapplied, or genus-level), and a name that isn't accepted links to the species it stands for. When the list holds more names than the count, its header gives both, e.g. "352 accepted species (365 names listed) on the Vermont checklist".
Clickable phenology chart. In a state or county, clicking a week's bar lists the records behind it under the chart, each linked to GBIF, which makes odd early or late records easy to check. In the national view a click opens GBIF's records for that month.
Licenses on the Sources page. Each contributing dataset now shows the license it is used under. Datasets with no license on record say unknown until it's established.

Fixes

The species list is fast again (v1.8.28). After v1.8.27 a page of the list took about 0.7 s, and 3.5 s with all names showing, because each species' datasets were looked up for every bee that matched, not just the 50 shown. It now takes under 0.1 s.
Vermont's species count matched neither list. It showed 362 while its checklist has 352. The count included every presence record from any source, and counted old and misapplied names as separate species. It now reads 352 in the Checklist view.
"SGCN States" counted listing years, not states. A bee listed by one state in both 2015 and 2025 showed "2 states"; 213 species were affected.
A bee not on a state's SGCN list now says "Not listed as SGCN" instead of a grey "No" or a dash. For the 16 states and territories whose lists include no bees at all, it says so.
Occurrence dots stayed hidden under the vegetation layer. Turning on the vegetation or ecoregion overlay could cover the GBIF dots. Overlays now always sit beneath the occurrences, and the orange points have a dark outline so they stand out against the vegetation colors.
Flight season reads as dates. The functional-traits table showed flight season as day-of-year numbers (125–178). It now shows dates (May 5 – Jun 27), with a note that these are range-wide figures from Smith et al. 2024.

Data

Eleven names restored as accepted species nationally. Earlier checklist imports had written Vermont's and Minnesota's taxonomy onto the national species records. For example, Agapostemon texanus, a valid species across 40+ states, had been marked a misapplied name for A. subtilior everywhere. Each state's checklist keeps its own treatment, and the Checklist view shows it (migration 058).
Unidentified specimens no longer pose as species. The Ostwald trait dataset files specimens not identified to species ("Andrena sp. 2", "Lasioglossum A") under the genus. Our import had gathered each genus's unidentified material into one record, and four of those records had been renamed to the bare genus. They and the measurements are now genus-level records, which are listed under Names → Genus-level and never counted as species. Ceratina floridana had been holding 21 measurements of unidentified Ceratina and now shows only its own. Dialictus guaruvae, a Brazilian bee, had counted as a US species only through the genus's occurrence count, and is now out of US scope. Measurements recorded as "aff." (a related species, not the one named) moved from 16 species to their genus. Two Andrena measurements are now credited to A. osmioides ("osioides" in the source) and one to A. subchalybea (migrations 059, 061). The national count is now 3,971 accepted species.
County records gathered under a genus key move to the genus (v1.8.28). Ceratina floridana had been showing in 14 Vermont counties. Its GBIF county records had been pulled with the Ceratina genus key, so they were every Ceratina in the US. The same was true of Dialictus guaruvae. Those records now belong to the genus-level records (migration 062). Both species get county records again once they are matched to GBIF by name.
A Names filter, everywhere. A new Names filter shows accepted names by default, or all names, synonyms, misapplied names, or genus-level records. It applies to the species list, the map and the home-page stats alike. By default they count accepted species. With another choice they count those names instead, and the labels say "Names" or "Synonyms": for example, Vermont's checklist has 365 names for 352 accepted species. Search finds a species by any of its names (migration 060).
NatureServe's license is recorded: CC BY 4.0 (v1.8.28). NatureServe's data use terms put public NatureServe Explorer data, the only part of NatureServe used here, under CC BY 4.0, with a required citation (migrations 063, 064).
Every dataset records its license (migration 057). New pulls from GBIF and iNaturalist also store each record's own license. Rules for bringing in new data are in docs/DATA_LICENSING.md.

THATSBees — Changelog — 2026-09-30 (v1.8.26)

New

Host plants from GBIF specimen records, pulled on demand. Museum specimens and survey records on GBIF often note the plant a bee was taken on, in their "associated taxa" field: Uvularia grandiflora, "visited flower of: Salix", "Host_id: Cydonia | collected on | Flower". A contributor can now pull these for one bee with Pull from GBIF in the species page's Floral Associations section, or on the edit page's tab. The pull reads every GBIF record of the bee outside iNaturalist (most bees take under a minute; the commonest bumble bees up to about 20), matches each named plant to a GBIF plant taxon, and stores one row per record. Pulling again replaces that bee's previous GBIF pull; it never changes the curated Floral Hosts list.
Floral Associations now counts plants per source. One row per plant, with a column for iNat PIP observations and one for GBIF records. An iNat count opens those observations on iNaturalist; a GBIF count opens the records beneath the row, each with its date, place, the collector's words and a link to the GBIF record. Most specimen labels say only that the bee was taken on or associated with a plant, so the GBIF column also shows how many of its records say the bee visited the flower.
Names GBIF cannot match to a plant are kept as written, grouped last: common names such as "Cherry" or "NE Aster". Blanks ("N/A"), animals, habitats ("mixed forest") and label notes are not stored, and are counted in the pull's report. A value naming several plants ("Solidago and Aster") is kept as written, since which one the bee was on is not recorded.

Data

GBIF occurrence keys now fit. Record identifiers in the interactions table held numbers up to 2.1 billion; GBIF's occurrence keys pass 6 billion. The column is widened (migration 056).
A production database restore keeps production's GBIF pulls, as it already kept its iNaturalist PIP pulls, wherever production's pull of a bee is newer than dev's.

THATSBees — Changelog — 2026-09-29 (v1.8.25)

New

Assessment Range Maps on the species page. Bees with a regional extinction-risk assessment now show the maps produced for it, in a panel after Range & State Occurrence, one tab per map: the extent of occurrence drawn around every record, the occupied 2×2 km area-of-occupancy cells, and the species distribution model's predicted range. Each map shows the figures printed on it: EOO, AOO, record and locality counts, post-1960 records, and the omission threshold the model used. Where no distribution model was built, the SDM tab says why, for example "fewer than 5 post-1960 records" or "did not pass metrics for reliability". The map figures are rounded; the assessment's stated EOO and AOO stay under Range & State Occurrence.

Fixes

Species with a revised assessment showed the old one's headline. For every bee whose 2024 assessment had been superseded by a regional revision, the page's Red List category, criteria, EOO and AOO came from the superseded 2024 assessment while the rest of the page showed the current one. They now all come from the current assessment.

Data

The regional assessments now cover 136 bees, up from 90. All seven regional folders were re-read: 37 bees have their first assessment, 32 have a revised one (23 updating a regional document loaded on 2026-09-10, 9 replacing a 2024 assessment), and 67 documents were unchanged.
Range maps for all 136 assessments: 136 EOO maps, 134 AOO maps and 80 SDM ranges, plus the stated reason for the 56 assessments with no distribution model.
**Lasioglossum batya Gibbs, 2011 added to the species list**, named by the Southeast assessment (GBIF 9040938; Florida, Georgia, South Carolina, Texas).

THATSBees — Changelog — 2026-09-28 (v1.8.16 – v1.8.24)

v1.8.24

New

An Ecoregion filter. It lists the CEC North American Level III ecoregions bees have been recorded in, from the regional extinction-risk assessments, each with the number of US bees recorded there. Only the 90 bees with a regional assessment have ecoregions so far, so choosing one narrows to those bees. It works on the Map and the List, and its ? explains how it differs from Habitat. Habitat and Ecoregion can be used together: a bee must be recorded in both.

Changes

The Filters panel is in two rows. First: State, County, Count bees by (and which occurrences), then the SGCN, Has Assessment and Has THATSBees Edits checkboxes. Second: Family, Habitat, Ecoregion, Red List Category, Threat, Conservation Action and Contributing Dataset, on one line on screens about 1000px wide or wider.
The Habitat filter offers every habitat vocabulary, grouped as Bee-Gap habitat classes, INPN habitats and IUCN habitats, and only habitats some US bee is recorded in, each with its number of bees.

Fixes

The Habitat filter chose the wrong habitat. Its options carried ids from an older habitat table, which the Map and List read as ids in the current one. Choosing "Forest – Boreal", for example, listed bees recorded in a subtropical high-altitude grassland. Old links with a habitat in them may still point at the wrong habitat.

v1.8.23

Changes

The species page is grouped by topic, with a contents list beside it. Its 25 or so blocks used to run in the order they were built, so distribution turned up in two places and conservation in six. The top of the page keeps the four summary panels, phenology, the occurrence map and Data Quality, followed by the description, identification and distribution. The rest sits in three groups: Life History, Plants & Partners and Conservation, each opening with the matching part of the written account.
"On this page" stays in view in the left column as you scroll. It lists every group and section, highlights the one you are reading, and shows counts such as "Floral hosts 42". Sections with nothing for this bee stay in the list, greyed out, so the list also shows what is not known. Clicking an entry opens that section and scrolls to it. On a phone, the list becomes a strip of group names pinned under the header.
Each group has Expand all / Collapse all.
Sections remember whether you left them open for each species separately. Before, opening Floral Hosts on one bee opened it on every bee.
Links go straight to a section. For example, /wiki/species/2454#s-floral-hosts opens Floral Hosts and scrolls to it, even while the map and charts above are still loading.

Fixes

Clicking a link to part of the page no longer reloads the state-filtered panels. A link to a nesting source number, for example, used to reload the filters, and the link's target was dropped from the address.
Anchored links no longer land under the sticky header, and closed sections no longer show an empty strip below their title.

v1.8.22

Changes

The active filters are back in the header, on a row of their own. v1.8.18 moved them inside the Filters panel, so they were out of sight whenever it was closed. They now sit on one row under the menu bar, as wide as it, each removable with its ×. The filter controls stay in the Filters panel. The row appears only when a filter is set or a species page is open.
A species page names its species at the start of that row and shows the state and county it is scoped to. Filters that only apply to the Map and List (SGCN, Family, Search and the rest) are still listed, faded with a dashed outline, so it is clear they are set but do not affect this page.
The header works on phones. Below about 960px wide it becomes three rows: the wordmark with Login and the menu; search, Filters and Clear across the full width; then the active filters, which scroll sideways if they do not fit. Map, List, Import, Sources and Users move into the ☰ menu. Buttons, filter chips and menu entries are larger to tap, the Filters panel fills the width, and the search box and selects no longer make an iPhone zoom in when tapped.
The species chip has an × that goes back to where you opened the species from: the Map, or the List at the page and sort it was on. If you changed filters on the species page, the new filters come along. Opened from outside the site, it goes to the List.
"Count bees by" and "Contributing Dataset" have a ? beside them that opens a short explanation. The hint line that sat under Contributing Dataset moved into it.

v1.8.21

New

Floral Hosts can be sorted by how much evidence there is. A Sort menu beside the search box orders the list by name (as before), by recorded interactions — how often this bee has been recorded on the plant — or, with a state selected, by the plant's GBIF records in that state. Families and genera move with their plants and show their totals ("Asteraceae · 77 plants · 1,026 interactions"). Where a source gave no count, each of its records counts once. A genus- or family-level record's GBIF count already covers its species, so those totals are not added twice.

v1.8.20

New

The Sources page lists every dataset, all 16. It used to show 10, from an older list that did not know about Noori, NatureServe, iNaturalist PIP or most of the nesting sources. It now reads from top to bottom in increasing detail: headline totals; the records in each part of the database and where they came from; one row per dataset with what it contributes, how many US bees it covers, its record count and its citation; and, folded away at the bottom, the full citation and record breakdown for each dataset.
Every imported record now names its dataset. 34,710 host-plant records from Bee-Gap and Fowler's pollen-specialist lists, and 831 ecoregion records from the regional assessments, were loaded before records were labelled by dataset, so they were counted under none. Fowler now shows 2,070 records instead of none, Bee-Gap 36,996 and the assessments 1,165. 19 more bees now count as having Bee-Gap data. The only records without a dataset are those entered by curators.

Fixes

"inat-pip" appeared as a raw code in the species list's dataset filter. It reads "iNaturalist PIP" again.
Re-importing Fowler's lists replaced Bee-Gap's link on host plants both sources report. It no longer changes another dataset's link. 602 records still carry the Fowler link from earlier imports, because that is the only record that Fowler lists them too.

Database

Migration 052: a datasets table (name, citation and link for each dataset) and dataset_key_of(). The dataset list the species filter uses is built on them and returns exactly the same rows as before.
Migration 053: fills in source_dataset on the rows above, keyed on the citation each row was created with. The duplicate check on habitat records now applies to any record without a source line number.

v1.8.19

Fixes

The map stopped working when no bee in a state matched the filters. Choosing, say, Vermont and then "Has THATSBees Edits" left the previous map in place, and the numbers beside it went back to the whole country's. The state and its counties now show zeros, and the legend says "No bees match these filters." If the map ever cannot load, it now says so instead of leaving an old map up.
The legend repeated ranges when only a few bees matched ("0 – 1, 1 – 1"). It now uses whole-number steps, only as many as there are counts.

v1.8.18

New

Choose a county and the map zooms to it, with its towns. Town boundaries are outlined and named, and pointing at a town names it. Neighbouring counties fade. Towns are the Census county subdivisions: towns and townships in New England, New York and the Midwest, census divisions in much of the South and West. They are there to help you find your way: bees are still counted by county, because that is how GBIF places occurrences.
A quieter header. Only the species search stays in the header. State, county, how bees are counted and which years count are now the first row of the Filters panel, above your active filters. The Filters button shows how many are set, and pointing at it lists them.
The Filters panel opens over the page instead of pushing it down, and closes when you click elsewhere or press Esc.
One control bar on the map. "Count bees by" and "Occurrences" sit in a single line across the top. The place you are looking at is named in the bottom-right corner.

Fixes

The map page fits the window. The legend and the footer were below the bottom of the screen. The page now fits between the header and the footer without scrolling, and the stats beside the map take less room.
The legend covered part of the map. Its title is now two lines ("Listed species", then "with GBIF occurrences since 2000"), so it stays narrow.

v1.8.17

New

Two ways to count bees, always labelled. A toggle on the map, "Count bees by: Checklist | Occurrences", and a matching "Count by" choice in the header whenever a state is chosen, switch every count on the site between them:
Checklist (the default): a state counts the bees on its own species list. A county counts the listed bees that also have GBIF occurrences there, so a county can no longer show more bees than its state.
Occurrences: a state or county counts every bee GBIF holds an occurrence of there, listed or not. GBIF occurrences never add a bee to a state checklist; they are only counted beside it. The banner says which count its numbers are ("Counted by the New Hampshire checklist"), and so do the map legend, the tooltips and the species list heading.
Where the counts differ, the gap is visible. In New Hampshire the checklist has 130 bees; GBIF occurrences since 2000 cover 366. Coos County shows 91 listed bees with occurrences, or 227 bees with any occurrence. Counting by occurrences, the species list gains an "On NH list" column for the whole state, not only for a county, so curators can see which recorded bees their list lacks.
"How bees are counted." The ? beside the map toggle explains both counts, why they differ, and what occurrence maps do and do not show: documented presence, not occupancy; a blank county means no record, not absent; counts follow collecting effort, not abundance.
"Since 2000 | All years" now applies wherever occurrences are counted, including a whole state or the national map when counting by occurrences.

Fixes

With a state, county, counting and years choice all set, the header's filter bar ran under the Login link. It now wraps onto a second line.

v1.8.16

New

Every nesting fact on a species page names its sources. Each term carries small numbers ("Below ground ¹ ² ⁴"), and a numbered source list closes the Nesting section. Each source links to the dataset or paper, and where the source cites primary literature (Rozen 1994, Krombein 1967…) that list is one click away. When several datasets agree, all of them are shown. Before, a term could carry only one source, and a second dataset that agreed was silently dropped.
How a bee nests is now separate from where. "How it nests" says whether the bee digs its own nest, uses an existing cavity, builds a free-standing nest, or has no nest of its own (a brood parasite). "Where it nests" groups sites under Below ground, Ground surface, Above ground and Other animals' nests, with finer sites beneath them: sandy or clay soil, soil banks, animal burrows, leaf litter, grass thatch, snail shells, sound or rotten wood, beetle burrows, pithy or hollow stems, plant galls, rock and walls, human-made structures.
Guesses look different from records. A term that every source inferred from the genus, rather than recorded for this species, has a dashed outline. A term every source marks as doubtful ends in "?". If one source calls a bee a parasite and another says it builds a nest, the page says the sources disagree.
Nest materials, entrance slope, nest aggregation and nest depth now appear in the Nesting section, each with its sources. Nest type, slope and aggregation were in the database but never shown.
Every nesting attribute can be edited. The editor's Nesting tab now edits how the bee nests, where, nest materials, nest aggregation and entrance slope. Each block lists the bee's terms with how many sources report each, and adds a term with an optional qualifier ("optional", "uncertain"…), note and source link. Sites are named by their place in the tree (Above ground › Wood › Rotten wood). Removing a term removes every source's record of it, and the editor asks first when more than one source reports it. Terms you add are credited to you in the species page's source list, and the history view names the attribute and the term.
New nesting attributes need no programming. A new attribute (say, how the nest entrance is finished) is registered with one database migration, and the editor, the species page, the history view and the data loaders all pick it up.
The dataset filter on the species list includes the four new nesting datasets below.

Fixes

747 cleptoparasites had no nesting information. Bee-Gap records them as "Host Nest", which the loader did not recognise. They now show "No nest of its own (brood parasite)". The 227 bees Bee-Gap records as "Multiple" were dropped the same way; they now show "Varied sites". Macrotera opuntiae, the one bee recorded as "Specialized", now shows that it digs its nest in rock (sandstone), from Bee-Gap's own note.
"Cavity" is no longer shown as a place. Bee-Gap's and Ostwald's "cavity" covers above-ground Osmia and below-ground bumble bees alike. It now reads "Uses an existing cavity" under How it nests, and no longer puts bumble bees above ground.
Accented and special characters showed as "�" in about 4,500 places: 4,448 Ostwald measurements (coordinates, localities, dates), 31 study citations (Diekötter, Török, Totland…), 3 Bee-Gap citations (Fandiño, Vélez), 2 vocabulary citations (Müller), 38 ITIS numbers and 5 plant names. Several source files are not pure UTF-8: Ostwald's mix in Mac Roman, Bee-Gap's are Windows-1252. They are now read correctly and the stored text has been repaired. No "�" remains anywhere in the database.
Coordinates read "31°25_00__S". Ostwald's file wrote the minute and second marks as underscores. They now read 31°25′00″S.
Four citations had "_" for a letter the source file could not hold. They now read Dylewski, Ł.; Lepš, J.; Harmon-Threatt; multi-level, taken from each paper's Crossref record.

Data

Bee-Gap nesting reloaded with its citations. 3,861 records for 3,860 bees, each linked to the works Bee-Gap cites for that species (4,310 links). Diller et al. (2020), USGS, doi:10.5066/P9QHQNNS.
BeeFunc nesting reloaded with the primary reference behind each record: 441 records for 117 US bees. Aubouin et al. (2025), doi:10.1038/s41597-025-05626-0.
Bartomeus et al. 2013: nest location, and whether the bee is a parasite, for 183 northeastern US bees. PNAS 110:4656, doi:10.1073/pnas.1218503110; data doi:10.5061/dryad.0nj49 (CC0).
EuPollTrait: how and where 139 Holarctic and introduced US bees nest, and what they build with. Miličić et al. (2026), doi:10.1038/s41597-026-07513-8; data doi:10.5281/zenodo.18032357 (CC BY 4.0).
Cane & Neff 2011, Appendix B: ground nesting and minimum and maximum cell depths for 214 US bees, 430 depth values, each with its primary references. Biological Conservation 144:2631, doi:10.1016/j.biocon.2011.07.019.
Cane, Griswold & Parker 2007, Table 1: nesting sites, nest-plug materials and primary references for 62 North American Osmia. Annals of the Entomological Society of America 100:350.
Two bees added: Lasioglossum mitchelli Gibbs, 2010 (52 US records) and Nomada media Mitchell, 1962 (3 US records), which nesting sources named and the database lacked. Six more names were spellings of bees already here (Osmia cahuila for O. cahuilla, Heriades carinatus for H. carinata, and others) and now match them.
Coverage. 3,739 of the 3,975 US bees now have nesting information (was 2,792). 1,249 have a nest type (was 101), and 539 have two or more independent published sources. Only 237 have a site finer than the top level. The literature says the same: most North American bees have no published nesting observation.

THATSBees — Changelog — 2026-09-26 (v1.8.14, v1.8.15)

v1.8.15

New

"Has Wiki Text" is now "Has THATSBees Edits". It finds every bee edited on the site (wiki text, traits, nesting, habitats, floral hosts, state presence and the rest), not just bees with wiki text. It used to find one bee; it now finds three. iNaturalist flower-visit pulls are imports, not edits, and don't count. Old links still work.
The species list shows what was edited. With the filter on, an Edits column lists each edited section with a count, the last edit date and a link to the bee's history.
One history view, in two places. The species history page and the editor's History tab now show the same table: every edit, newest first, with a Section column naming what changed (Identification, Biology, Traits, Nesting Substrates…). Wiki text changes show the before and after; wiki versions can still be viewed and reverted.

Fixes

The history page now lists the current wiki text version; it used to show only earlier ones.
Added and removed nesting substrates now name the substrate instead of showing only "is primary: false".
State presence and SGCN edits are labelled by name in the history instead of by table name.

v1.8.14

New

You can narrow any view to a county. Choose a state, then a county from the new dropdown in the header, or click a county on the map. A bee counts in a county when GBIF holds records of it there. Nothing we curate is recorded by county, so pages say "GBIF records", never "present".
Records since 2000 by default, all years on request. A declining bee's old records can make it look widespread: the rusty patched bumble bee has records in 483 counties, but only 139 since 2000. The header and the map both switch between "Since 2000" and "All years".
The map colours a state's counties by the bees recorded there. Clicking a county shows its species, assessed, SGCN and threatened counts. "Back" steps from county to state to the whole country.
The species list shows each bee's county records, with both counts and a column saying whether the state's own list includes the bee. A bee with county records but no place on the state list is worth a curator's look.
A species page shows the bee's records in the chosen county. The occurrence map, phenology chart and data quality checks follow the county. State presence, SGCN, listing and NatureServe rank stay state-level and are labelled with the state's name.

Data

County records for 3,945 bees, from GBIF occurrence records (presences only, no fossil or living specimens, no geospatial issues). 3,461 bees have records in at least one county. Refreshed after 180 days.
All 3,142 US counties, matched to GBIF's county units. GBIF combines a few neighbours into one unit (the Baltimore, St. Louis, Richmond VA and Franklin VA city/county pairs, and some Alaska areas); both show the combined counts, marked with an asterisk.

THATSBees — Changelog — 2026-09-25 (v1.8.13)

New

You can ask to manage a state. If you keep a state's bee checklist, presence or SGCN data, you can ask a national admin for access to that state. Ask when you sign in, using the new "Request a state to manage" choice on the login page, or later with the Request a state button in your profile. You can add a note, such as your role. A request made on the login page is sent once you follow the link in your email. Your profile lists your pending requests, which you can withdraw, and any that were declined, which you can ask for again.
National admins review state requests on the Users page. Waiting requests appear at the top of the page, with the person's role and note, and the Users link in the header shows how many are waiting. Approving adds the state to that person's assignments. National admins also get an email when a request comes in.

Fixes

The login page said the sign-in link expires in 15 minutes. It lasts 2 hours, as the email says, and the page now says so too.

THATSBees — Changelog — 2026-09-24 (v1.8.12)

Fixes

The occurrence map's light basemap shows the map again, not "API KEY REQUIRED". The site sent its CARTO key under a parameter name CARTO ignores, so every tile came back as CARTO's watermark even with a key configured. The key is now sent the way CARTO reads it. Where no key is configured, tiles load unkeyed as before.
Sign-in emails now come from "THATSBees", not the old "US Bee C-SWG BeeWiki", and the link in them opens over https directly instead of being redirected there.
The phenology chart now uses cleaned records, not raw GBIF counts. For a single state, and for any species with 10,000 or fewer records in scope, each record is checked with the same rules as the Data Quality panel. Absence records, zero counts, fossils and impossible dates are removed. Records that republish the same specimen (same catalog number in another dataset, same occurrenceID) count once. Until now, 545 Vermont Bombus impatiens absence records were charted as sightings.
A state's phenology now includes records without coordinates that name the state. Records whose coordinates fall in the state still count even when the coordinates are imprecise. Checklist taxon keys are queried together, so a record matched under both a name and its synonym counts once.
The note above the chart accounts for every record. It lists what was removed and why, and what could not be placed in a week. Species with more than 10,000 records in scope still chart from GBIF's date counts, with only absences removed, and the note says so.

Configuration

The root .env is now the only configuration file; api/.env is retired and settings are no longer built into the Docker image. See "Configuration" in deploy/DEPLOYMENT_NOTES.md. Before deploying this to an environment that still has api/.env, run ./deploy/env-consolidate.sh (or ... prod).

THATSBees — Changelog — 2026-09-23 (v1.8.10)

New

Wetland status for the plants bees visit. THATSBees now holds the US Army Corps of Engineers' National Wetland Plant List: 27,967 ratings covering 8,113 plants, from "almost always in wetlands" to "almost never". Ratings are kept per region, because a plant can be a wetland plant in the Midwest and an upland plant in the Arid West. 2,306 of the plants our bees visit are rated, and about half the host plants of a typical bee, since the plants bees visit most are the best covered.
A bee's own summary — what share of its rated host plants are wetland plants — is a hint rather than a finding: bees already recorded in wetlands average 40% against 30% for the rest, measured on only 15 bees. The per-plant rating is the solid part.
"Where the host plants grow" now explains itself. A fold-out note under the heading sets out how the ranking is built, how well it does against the bees whose assessments already name their ecoregions, and four things it does not show — that it is not the bee's range, that GBIF records where people collect, that only matched host plants count, and that boundaries are simplified and stop at the border. It stays closed, so the ranking is still the first thing on screen.
Both of those now show on a species page. A new section, "Where the host plants grow", ranks ecoregions by how much of the bee's host flora is recorded in each — "92.1%, 222 of 241" — and says how many of its host plants are wetland plants. It is labelled Inferred and says plainly that it describes where the plants grow, not where the bee has been found. In the floral host tree, each plant now carries its wetland rating: OBL, FACW–FAC and so on, with the wetland ones picked out, and the plain-English meaning on hover.
The plants of every US ecoregion, and what they say about a bee. THATSBees now holds a plant list for each of the 84 CEC Level III ecoregions in the lower 48 — 332,000 plant-in-ecoregion records, built from GBIF the same way the state plant lists are. Because the site already links bees to the plants they visit, the two now join up: for any bee, the ecoregions where its host plants actually grow can be ranked, with the evidence shown as "159 of 213 host plants recorded here".
This is a ranking of candidates, not a statement of where a bee lives, and it is labelled that way wherever it appears. Checked against the 63 bees whose assessments already name their ecoregions, 81% of a bee's host plants are recorded in the ecoregions its assessors named against 35% in the ones they did not, and 62 of those 63 bees show the pattern. An assessor-named ecoregion comes out in the top five for 83% of bees. Taking as many guesses as the assessors made, 41% land — useful for suggesting ecoregions to a curator, not for asserting them.
Flower visits from iNaturalist have their own section on a species page. "Floral Associations" now sits directly above Floral Hosts and holds the plants a bee has been photographed on in VCE's Pollinator Interactions on Plants project, with the number of observations for each and a link to them. It used to sit at the foot of Floral Hosts, where neither the records nor the button that fetches them was easy to find. Floral Hosts is once again just the curated host list.
The edit page has a Floral Associations tab, just before the Floral tab. It shows the same table and carries the "Pull from iNaturalist" button, so an editor can bring a bee's records in, or refresh them, without leaving the edit page.
Floral Hosts is now a taxonomic tree. On a species page and on the edit page's Floral Hosts tab, a bee's host plants are grouped Family > Genus > Species, like the Classification box, instead of one long table (the honey bee had 7,734 rows). Families open on click; a search box finds a plant, genus or family. Each plant shows how the bee uses it, its sources, and its common name. Plants are grouped on GBIF's accepted names, so a synonym and its accepted name are one entry ("recorded as Aster novae-angliae").
Pick a state and Floral Hosts shows only plants recorded there. On a species page the header's state filter applies; the edit tab has its own state picker. Each plant carries its number of GBIF occurrences in the state, and a checkbox brings back the plants not recorded there. The first request for a state builds its plant list from GBIF (about a second); after that it is served from our database and refreshed every 180 days. "Recorded" means at least one GBIF occurrence, garden and herbarium records included, so a count of 1 is weak evidence.
Editors remove floral host records one source at a time. In the edit tab each plant lists its records (relation, source, observation count), each with its own remove button.

Fixes

The phenology chart accounts for every record. It now says how many occurrences it plotted and why the rest were left out (year or month only, a long date range, a Jan 1 placeholder, no date). Ranges of up to a week are plotted at their midpoint. Charts for the commonest bees no longer fail, and dates no longer shift a day in western time zones.

Data

Migration 042: plant_taxa (every floral host plant name matched to GBIF — 99.4% of host records resolve) and per-state plant lists from GBIF occurrences. After a new floral host import, run api/beewiki/scripts/plant_taxa_match.js --names.

THATSBees — Changelog — 2026-09-15 (v1.8.4)

New

Contributors can bring a bee's flower visits in from iNaturalist. Floral Hosts on a species page now has a second table, "Flower visits recorded on iNaturalist". It lists every plant the bee was photographed on in VCE's Pollinator Interactions on Plants (PIP) project, with the number of observations and how many were of females and of males. Each count opens those observations on iNaturalist. Anyone signed in as a contributor, editor or admin can press "Pull from iNaturalist" (or "Refresh") to bring them in; it takes a few seconds for most bees and about two minutes for the commonest bumble bees. Only research-grade observations are used. A pull replaces the bee's previous one, so observations deleted or re-identified on iNaturalist drop out, and it never changes the curated floral host list above it.
Plants are listed at the level they were identified. Species and genera come first, then plants identified only to a family or tribe, then names the observer typed that iNaturalist never linked to a plant ("wild bergamot"), shown as typed rather than guessed at. For Bombus bimaculatus that is 4,734 of 4,744 observations; the other 10 named no plant, or named the bee itself.
Where an observer hid the location, so do we. Observations iNaturalist obscures are stored without coordinates.
Species pages link to the bee's flower visits recorded on iNaturalist. The External Links box now has "Flower visits (iNat PIP)", which opens the iNaturalist observations of that bee in VCE's Pollinator Interactions on Plants (PIP) of the NE US project. Each of those observations records the flower the bee was on. The number beside the link is how many observations it opens — 22,070 for Bombus impatiens. The link appears only when the project has observations of the species, and only when iNaturalist knows the species by our name or a synonym of it, so a page never shows another bee's flower visits.

THATSBees — Changelog — 2026-09-14 (v1.8.2)

New

State checklists now sit on top of the data instead of rewriting it. When a state's checklist is imported, every species it lists is recorded as that checklist's own claim, including the ones the data already had. Its presence records are added only where the data didn't already place the species in that state, and they're marked as the checklist's, so removing a checklist removes exactly what it added and nothing else. A checklist's view of which names are synonyms now decides where its own records go, without changing the taxonomy every other state sees.
A checklist adds a species only when it is genuinely new. A listed name is matched against our species by name, by recorded synonym, by GBIF key and by GBIF's own synonym list. A new species is created only when none of those find it and GBIF confirms an accepted bee species under exactly that name. Everything else goes on a review list on the Import page, with the reason for each name. Reading Colorado's published checklist (887 names) this way added 6 species and listed 69 names for review, among them fragments of the document's prose such as "Halictus and".
Editing is scoped to the states people look after. Species data that every state shares — traits, habitats, floral hosts, threats, wiki text, assessments — can now be edited only by national admins. State presence, SGCN status and state checklists can be edited by national admins anywhere, and by contributors, editors and state admins only in the states assigned to them. The edit page shows each person only what they can change, and says why the rest is read-only.

Fixes

SGCN status could not be added or removed. Both actions wrote to database columns that don't exist, so every attempt failed. They work now.
A refused edit looked as if it had saved. When the server turned an edit down, the edit page carried on as though it had succeeded. It now shows the server's reason.
Adding a state by hand overwrote where the record came from. Marking a species present in a state that a dataset already recorded replaced the dataset's attribution. The existing record is now left as it is.
Anyone could re-import an assessment document. The import endpoint accepted requests without signing in. It now requires a national admin.

Data

The species created by February's checklist imports have been re-checked. The Minnesota and Vermont imports created 12 species rows. Checked under the new rule, 4 are genuinely new species and stay. The other 8 duplicate species already held under another name — for example Eucera albata is Tetraloniella albata, and Coelioxys bisoncornuus is Coelioxys bisoncornua — or, in one case ("Nomada mr"), are a fragment of a name. Those 8 are no longer counted, each points at the species it duplicates, and their state records were copied to that species. Nothing was deleted.
Filtering by data source returned nothing, or the wrong thing. The filter panel had two ways to ask where data came from, "Data Source" and "Contributing Dataset", and only the second worked. Data Source was an early shortcut: choosing Fowler or Bee-Gap listed no species at all (they cover 1,091 and 317), and IUCN Assessment just repeated the Has Assessment checkbox. It has been removed. Contributing Dataset is now the one place to filter by source, with a line underneath explaining what it means and a link to Sources, which shows what each dataset contributed.
Species with text written on the site now have their own filter. Data Source's fourth option, BeeWiki, was never about a dataset — it meant species whose description or notes were written by contributors here. That is now a separate "Has Wiki Text" checkbox beside Has Assessment and SGCN.
The dataset dropdown's numbers didn't match what it listed. Each option showed a count across every species in the database, then listed only US species — Fowler read 1,097 and opened a list of 1,091. The counts now describe US species, so the number beside an option is the number of species it opens.
Old links keep working. A bookmarked or shared link that used the removed filter now opens the equivalent new filter, and the address bar updates to match.

THATSBees — Changelog — 2026-09-11 (v1.8.0)

New

The home page's counts now describe US bees, and they follow your filters. The five figures beside the map used to report the whole database and sat frozen no matter what you filtered on. They now count the 3,981 species with confirmed US presence — matching the species list exactly — and every one of them responds to the filter panel. Filter to Apidae and the counts become Apidae's; pick a state and they become that state's. Hovering any figure now says what it counts.
"Contributors" is now "Datasets". The figure was counting registered site accounts, which is not what the word suggests and read as 1. It now counts the 11 published datasets the site's records come from — Noori, NatureServe, Ostwald, Smith, Fowler, the assessments, BeeGap, Big-Bee, Thrift and BeeFunc — and links to Sources, where each one's contribution is broken down. It stays site-wide rather than changing with filters: a dataset does not become a different dataset because one state is selected.

Fixes

Filtering never changed the home page's counts. The page requested a web address that did not exist, and the resulting error quietly restored the unfiltered numbers, so the figures appeared to ignore every filter. They now come from an endpoint that exists and agree with the species list for every filter tested.
Assessments was counting superseded revisions. It reported 334 where the rest of the site shows 323, because 11 of those rows are earlier revisions kept as evidence. It now counts current assessments only, as everywhere else.
The map ignored the Contributing Dataset filter. Choosing a dataset changed the species list but left the map and its counts unchanged. The map now honours it on both the national and per-state views.
New changelogs were being created under the old project name. The September rename to THATSBees updated the changelog that existed at the time but missed the template that generates new ones, so every release since was titled "cSWG". The template is fixed and the affected entry corrected. Entries written before the rename keep their original name, which was accurate then.

THATSBees — Changelog — 2026-09-10 (v1.7.0)

New

A menu in the top-right corner, and a profile you can edit. The header's scattered icons are gone. A single menu button now holds Watchlist, Changelog and Logout, and the person icon beside it opens your profile — your name, email, role, institution, assigned states and last sign-in — with your name, username and institution editable in place.
Release notes are readable on the site. The Changelog menu item opens the notes for every release, newest first, alongside the version the site is currently running. Until now these existed only as files in the repository.
The species list can be filtered and sorted by which dataset contributed its data. A new "Contributing Dataset" filter lists the eleven datasets behind the site — Noori (GloBI), NatureServe, Ostwald, Smith, Fowler, the assessments, BeeGap, Big-Bee, Thrift and BeeFunc — each with the number of bees it has something to say about, so you can ask "which species does Thrift's range work cover?" and get the 232 of them. The list itself gains a Datasets column naming those sources per species, and every column heading is now clickable to sort — including by how many datasets describe a species, which surfaces the best-documented bees first.
A species can now carry more than one assessment, and the site always shows the current one. Until now every species had exactly one assessment, enforced by a database constraint. Assessments are now revisions: the new one becomes current, the one it replaces is kept in full and marked superseded, and the species page, map, API and wiki all resolve to the current revision. Nothing from the older assessment is deleted — it remains readable evidence of what was concluded in 2024.
Habitat now shows every scheme that describes a species at once. A species can be described by the IUCN habitat classes from its 2024 assessment and the ecoregions from its 2026 revision, side by side, rather than one replacing the other. Four vocabularies now coexist: IUCN, CEC ecoregions, the French INPN typology, and Bee-Gap.
NatureServe conservation status. Global, national and subnational ranks and the ESA listing are loaded and shown, and the Import page now says when they were last refreshed, so a stale rank is visible rather than silently old.

Data

90 revised bee assessments, covering all seven regional review groups — Midwest 5, Mountain Prairie 12, Northeast 9, Pacific 8, Pacific Southwest 27, Southeast 7, Southwest 22 — across roughly 33 genera. 79 are the first assessment for that species; 11 supersede an assessment loaded in 2024. Assessors: Killingsworth, S. and Martínez-López, O., facilitated and compiled by Chesshire et al.
80 reviewer comments are loaded with the assessments they annotate. Several of these documents are still under review and their reviewers say so — that the rank needs re-evaluating, that occurrences are missing, that an AOO needs revising. Those comments are recorded against the assessment rather than lost when the document was read, so the disagreement stays visible next to the number it disputes.
The CEC North American Level III ecoregion vocabulary, 247 terms across three levels (15 / 50 / 182), from the Commission for Environmental Cooperation's own atlas. This replaces a US-only seed that could not name the Canadian and Mexican regions the assessments actually use. 831 ecoregion claims are recorded across the 90 species.
Quantified decline drivers. For assessments listed under criteria A2c or A3c, the modelled decline percentages that justify the listing — climate change, pesticide loading and land use change, separately, against extent of occurrence and area of occupancy — are now stored as numbers rather than only as prose.
Four species added that the assessments name but the checklist lacked: Anthidium atripoides, Anthophora caudata, Epeolus attenboroughi and Perdita albomaculata, each verified against GBIF. Three were described after the original checklist was built.

Notes

The revised assessments arrive as Word documents rather than Google Docs, in a restructured format. docs/assessment_v2_census.md records what changed, what the format dropped, and what is known to be incomplete in the source documents.

THATSBees — Changelog — 2026-09-03 (v1.4.0)

New

A state's published checklist can now be imported as itself. Twenty states have a bee checklist that GBIF does not carry — Colorado's museum PDF, Virginia's provisional list, the Bees of Wisconsin — and until now the Import page could only file a checklist if GBIF happened to publish it as a dataset. Import Source files the publication itself: it fetches the document, reads the species out of it, and matches them against our database. Virginia's list yields 475 names, 466 of which we already hold; Colorado's yields 957.
Names that match nothing are listed, never created. These are journal PDFs and scanned appendices, and their near-misses are exactly the kind of thing that would quietly become new species: Virginia's nine unmatched names include Nomada lousianae for louisianae and Melissodes fimbriatus for fimbriata. They are reported for review under "see unmatched" and left alone. A name is only counted when the genus we matched agrees with the genus the checklist printed, so a missed heading in a document cannot attach a species to the wrong genus.
Every checklist now says where it came from. Current Assignments has a Type column — GBIF for a checklist dataset, Other for a published document — and GBIF-sourced checklists are named with the acronym, so the source is legible wherever the title travels, including the state records each one writes.
The Import page now lists every state's published bee checklist. Fifteen states have a statewide bee list in print — Utah's 1,167 species, Washington's 565, Colorado's ~1,006 — eight more have partial coverage, and twenty-seven have nothing published. All fifty are now listed with the citation, a link to the list where one is hosted online, and the species counts and caveats each source reports. Filter by coverage, or narrow to one state with the header filter.
Import tries to match a published checklist to a GBIF dataset. Press Import on a state and the site searches GBIF for a checklist dataset of that state's bees. A confident match is filed under Current Assignments with the published citation attached, ready for Ingest Species; Michigan matches Gibbs et al. 2017 and Vermont the Vermont Atlas of Life checklist this way. Most states have no GBIF dataset — that is the normal answer, and the page says so plainly rather than treating it as a failure. Where the search returns something plausible but not certain, the candidates are shown with their bee-species counts for a curator to choose from.
The site is now called THATSBees. The name spells out what it holds: Traits, Habitats, Actions, Threats, Status. The full title reads US Native Bees: Traits, Habitats, Actions, Threats, Status, and the previous name BeeCSWG is gone from the header, footer, home page, sign-in page, page titles and the API docs.
Sign-in emails come from THATSBees. The subject line, the message and the sender name all use the new name — they had still been signed BeeWiki, two names back. The noreply@beewiki.org address is unchanged, so nothing about delivery changes.
Two menu items renamed to say what they do. Data is now Import, which is where GBIF state checklists are brought in. Provenance is now Sources, which is where every dataset behind the site is listed with a link to cite.
BeeWiki remains the project's background identity. The public name is THATSBees, but beewiki.org still serves the site and is now acknowledged in the footer, so arriving by either address makes sense. Everything underneath keeps the BeeWiki name: the domain, the noreply@beewiki.org sender, the stylesheet and script filenames, and the sign-in token issuer. cswg.vtatlasoflife.org is unchanged.

Fixes

The "Bees only" species counts on the Import page were wrong. GBIF ignores a family filter on a dataset-scoped species search, so every dataset was being credited with six times its total record count as "bee species" — a trilobite monograph looked like a bee checklist. Bee families are now counted from the records themselves, so the number beside each dataset is the number of bee species it actually holds.

Big-Bee (Seltmann et al. 2021)

*The site now records what a bee is found with, not only what it feeds on. The Big-Bee network's interaction database — hand-extracted from 39 papers, some of them from the 1890s — adds 2,695 records about 253 US bees. Alongside flower visits it brings the associations cSWG had nowhere to put: the cuckoo bees that rob a nest, the phorid flies and conopids that parasitise adults, Varroa* mites, the viruses found replicating in honey bees, and the fungi and other insects recorded sharing a nesting site. Each record keeps the paper it came from, the place, the date, the sex and the body part the author reported.
*A new section on species pages, Other Species.* It lists non-plant partners grouped by kind — cleptoparasitism, parasitism, pathogens, predation, co-occurrence — with the number of records and papers behind each and a link to cite. Where two bees in the database are the two sides of one record, it appears on both pages, read from each bee's point of view: the cuckoo bee's page says it robs its host, and the host's page says it is robbed.
*Records a paper reported as not happening are kept and marked.* Forty-one rows are a source's negative finding — it looked and did not find the association. They are shown labelled rather than dropped, and they are excluded from the floral host counts, because counting them as evidence of an association would say the opposite of what the paper found.
438 new bee–plant pairs. The other 1,001 flower-visit relationships in this dataset were already in the database from Noori et al.; those are left as they were, and the Big-Bee record with its own citation now sits behind them.
2,803 specimen measurements of body size. Notes from Nature volunteers measured intertegular distance — the standard proxy for bee body size — from museum specimen images, five people per specimen with the results reconciled. Thirty-one species gain real specimen-level measurements, most of them Hesperapis, a genus with almost no published trait data. They appear under Functional Traits as specimen measurements, kept separate from published species averages as always.
Four things in the source were deliberately not taken at face value. Thirty-four rows citing one 2006 paper carried 31 different DOIs, each made by adding a counter to the real one; they are loaded as the single work they are. Measurements outside 0.5–8 mm are a volunteer's scale-setting slip rather than a bee, and 183 were skipped. A microscope file reporting intertegular distances of 17–20 mm — ten times every published value — is graticule divisions mislabelled as millimetres and was not loaded at all. The American Museum of Natural History expedition was left out because its catalogue numbers match no occurrence record, so there is no way to say which species was measured.
Old names still find their species. LaBerge and Cockerell wrote Tetralonia, Synhalonia and Dialictus for bees the database holds under Melissodes, Anthophora and Lasioglossum. 208 records were attached through the current name, with the name the paper used kept beside them.

cSWG — Changelog — 2026-09-02 (v1.1.1)

New

Every species page now shows what we actually know about the bee, with the measurements behind it. A Functional Traits section lists body size, tongue length and pilosity for 1,500-plus species — the mean, the range, how many measurements it rests on, and how many separate studies contributed. Specimen measurements and published species-level averages are summarised separately and never mixed, because they are not comparable: for one bee the two differ by 13%, and a single averaged number would have matched neither.
A new "Where this data comes from" page. Reachable from Provenance in the top nav, it shows the size of the database, what each published dataset contributed and to which part of the site, and a link to cite for every source. The counts are live, not a snapshot.
Data source badges are now links. The source tags on a species page point at the DOI or the deposit the records came from, so any figure on the page can be followed to its publication in one click.
Bee–plant interactions, at a scale the site has not held before. 157,562 bee–plant associations covering 3,024 bees and 2,716 plant genera. Each records what kind of interaction it is — foraging, flower visiting, larval provisioning, pollination — how many observations support it, and how many of those were made in the United States.
Nesting is described far more fully. New to the site: nest type (does the bee excavate, reuse a cavity, build, or parasitise?), nest slope, and nest aggregation — whether a solitary bee nests alone or packed together with others. Nesting materials went from 1 species to 31.
Flight periods and diet breadth. Recorded flight months for 104 species, flight-season statistics for 1,241, and diet breadth for 1,300-plus — held as claims with the evidence behind each, so a specialist designation from field records and one from a published classification can sit side by side and disagree.
Habitat can now be described in more than one classification. Habitat vocabularies are stored side by side rather than forced into a single scheme, with crosswalks recorded between them. IUCN and the French INPN typology are both loaded; adding another is now a configuration change, not a rewrite.
Software version and schema state are visible. /api/version reports the running version, and the Provenance page warns plainly if the database schema is behind the deployed code.

Fixes

Species records that were the same bee twice have been merged. 164 duplicate species — the same taxon under two spellings, such as Osmia foxi and Osmia foxii — were being counted as separate species and splitting their data between them. They are now one record each, with the discarded spelling kept as a synonym so old exports still resolve.
155 species names were corrected to their accepted form: Andrena cressoni → cressonii, Melissodes bimaculata → bimaculatus, and similar.
Pollen specialisation that had gone missing has been restored. Merging duplicate species had silently dropped 58 specialist designations, leaving two bees — Melissodes perpolitus and M. personatellus — with no record of their specialisation anywhere. Recovered, along with 215 taxonomy and common-name fields.
Bee-Gap's own citations are no longer discarded. All 3,922 species carried one identical blanket citation while the source file's per-species references were read and thrown away. Every species now carries its real references — 15,581 links to 487 published works, averaging four per species.

Data

Five datasets loaded, all filtered to species that occur in the United States.

Ostwald et al. 2024, Toward a Functional Trait Approach to Bee Ecology (Ecology and Evolution 14(10):e70465) — 11,627 body size, tongue and pilosity measurements across 1,535 species.
Noori et al. 2026, A curated and integrated dataset for exploring global bee–plant interactions — 121,953 interaction associations, aggregated from 810,788 observation records of US bees.
Smith et al. 2024, Pollen specialist bee species are accurately predicted from visitation, occurrence and phylogenetic data (Oecologia) — diet breadth, range size and flight season for 1,291 US-native species. Also added 50 US bees the database did not previously hold.
Thrift et al. 2026, Concentrated vulnerabilities in bees: Diet specialists have smaller geographic ranges (Insect Conservation and Diversity) — extent of occurrence and quantitative diet breadth for 232 species.
BeeFunc (Aubouin et al. 2025) — the US-relevant tenth of a French trait database: nesting detail, sociality, phenology and European Red List status for the Holarctic and introduced bees shared between the two faunas.

Species not currently known from the United States are kept, not deleted — a bee can turn up in a new survey or in digitised historical material. They are held in full and excluded from counts and default views, and can be surfaced with one filter.