Catalogue of Life COL26.10 — pre-publication review

New release 316595 · COL26.10 (RELEASE, project 3, attempt 642, issued 2026-10-14, built 15:42–17:56 on 2026-10-14) compared with 316321 · COL26.9 (RELEASE, attempt 629, issued 2026-09-11). Base release against base release. Release reports: download.checklistbank.org/releases/3/642/.

1. Verdict

2. Red flags

3. Totals and ranks

3.1 Whole-release totals

From GET /dataset/{key}/import?limit=1 (project import attempts 642 and 629).

No shrinkage overall. Type material (+5.29%) is the largest relative move and is not explained by any flagged sector; I did not isolate its source.

3.2 Taxa by rank

Every rank with ≥1,000 taxa, every rank above genus, and every rank that moved.

3.3 Synonyms by rank

3.4 Names at and above order rank (checklist item 7)

The /diff endpoint could not be used (see §8). Instead I enumerated every accepted and provisionally accepted usage at each rank from domain down to order in both releases (/nameusage/search?rank=X&status=accepted&status=provisionally%20accepted, paged): 2,768 names in COL26.9, 2,766 in COL26.10 — equal to the sums of taxaByRankCount over those ranks, so the enumeration is complete. The name sets differ in four places:

No other name at or above order rank changed, and no identifier changed for any of the 2,765 names present in both releases. No newly invented orders.

3.5 Duplicated accepted uninomials, family rank and above

/duplicate/count?category=uninomial&minSize=2&mode=STRICT&rankDifferent=false&status=accepted for every rank from the top of the hierarchy down to family, both releases:

Nothing grew, and the zeros are real rather than an endpoint artefact, since genus returns a non-zero count for both releases. No sign of a sector attaching a duplicated higher taxon. Family duplicates in COL26.10.

4. Issues

Every issue count grew; none shrank. The trend across the base releases of this project is what turns this from "the data got worse" into "the pipeline changed": COL26.8 and COL26.9 report almost no issues at all, and COL26.10 reports more issue types (20) than any release in the window (previous maximum 14).

Earlier base releases of the same series, synonym rank differs: COL25.10 713,148; COL25.11 555,975; COL25.12 515,740; COL26.1 514,940; COL26.2 430,795; COL26.3 421,805; COL26.4 370,917; COL26.5 299,347; COL26.6 299,347. Today's 643,964 is inside the historical range but 2.6× the last validated release.

Because import metrics belong to the project attempt rather than the release copy, I cross-checked the release itself with /dataset/316595/nameusage/search?issue=X&limit=0: it returns exactly the import figure for SYNONYM_RANK_DIFFERS (643,964), MISSING_AUTHORSHIP (63,903), PUBLISHED_BEFORE_GENUS (13,569), RANK_NAME_SUFFIX_CONFLICT (12,826), PARENT_GENUS_MISSING (8,705) and NO_SPECIES_INCLUDED (14,196) — these flags really are in the published index. For COL26.9 the same queries return 1,054 / 0 / 0 / 22 / 0 / 1,035, and two of those do not match its import metrics (22 vs 59, 1,035 vs 2,329) — a further sign that the COL26.9 release copy and its project attempt disagree.

Where the flags sit (facets on the new release): synonym rank differs is concentrated in World Plants (260,447), WCVP (72,194), WoRMS Mollusca (44,876) and Species Fungorum Plus (35,104), and is mostly botanical infraspecific synonyms attached to species (variety 269,400, form 59,471); parent genus missing is 8,705 species spread over ordinary sources (Systema Dipterorum 4,189, WoRMS Mollusca 779, 3i Auchenorrhyncha 724, Tortricid.net 637, GLI 604); missing authorship is mostly genera (33,777). Nothing is concentrated in one broken sector.

5. Identifiers

All counts match the job logs exactly. Nothing here is far above last month — everything except created is far below it, which I attribute to the ID provider now considering 103 past releases including deleted ones instead of 32 public ones (high confidence: the change is quoted verbatim in both digests, and the new superseded.tsv plus the disappearance of 1,176 "ID … reported without name usage in release" warnings come with it). Note that COL26.9's deleted.tsv contains 1,176 rows with empty rank, status and name — the same count as those warnings; COL26.10 has none.

5.1 Deletions by status and rank (all 7,893)

For comparison, COL26.9 deleted 30,300 species synonyms, 4,038 variety synonyms and 976 accepted genera. This month's accepted-genus deletions (86 accepted + 12 provisionally accepted) are roughly a tenth of last month's.

5.2 Every deleted identifier of an accepted name at genus rank or above (139)

Sector attribution was obtained by resolving each identifier in COL26.9 (/dataset/316321/taxon/{id}) and mapping its sectorKey through /dataset/3/sector/{key}; "fate" is an exact-name search in COL26.10. By sector: GLI Lepidoptera 35, StaphBase Staphylinoidea 27, Systema Dipterorum 24, WoRMS Mollusca 15, IRMNG 7, WOL Odonata 5, 3i Auchenorrhyncha 5, LPSN 4, WoRMS Foraminifera 3, WoRMS Ostracoda / UCD / FishBase / ScaleNet 2 each, StaphBase Histeroidea / StaphBase Hydrophiloidea / SF Psocodea / WoRMS Turbellarians / WoRMS Amphipoda 1 each, and 1 that already had no sector. 102 names are absent from COL26.10, 37 survive under another identifier, 18 have an entry in superseded.tsv.

One row, Merisus [sic] (VPDMW), could not be name-checked because the bracketed epithet breaks the search URL; it is counted among the 102 but should be treated as not checked.

5.3 Created and resurrected higher taxa

Created accepted / provisionally accepted names at genus rank or above: 74 genera, 18 tribes, 11 families, 9 subfamilies, 8 subtribes, 3 superfamilies, 1 infraorder. The new families are mostly fossil cephalopod and amphipod families published in 2025–2026 (Aktjubocheilidae, Badouceratidae, Cadomellidae, Trimeroceratidae, Koekotroididae, Neorchestiidae, Talitriatoridae, Chiyanidae, Agaricicolaceae) plus Cordulegastridae (a re-identification, §2.7) and Dolicchopodidae (a misspelling, §2.6). Resurrected: 6,020 accepted species, 455 accepted genera, 21 accepted families (including Daphniidae from the FADA re-sectorisation and the three Odonata families whose duplicate identifiers were dropped) and 1 accepted order, Unionida.

6. Release job

Both digests describe a job that reached FINISHED and indexed every usage. Neither release logged a single ERROR line.

Steps that did not run. COL26.9's timeline has no "Remove copied issues and validate release", no TreeCleanerAndValidator duration and no "flagged with issues" line; COL26.10 has all three ("743271 usages out of 5466019 flagged with issues during validation of release 316595" = 13.60%). That is the largest difference between the two jobs and the explanation for §4. Nothing else was skipped or processed nothing that should not have been: both jobs report "Removed 0 sectors without data", both copied 0 treatments and 0 medias (as every release in this series does), both removed 0 orphan names / relations / type materials / references, and both archived their complete source list (158 datasets now, 159 logos last month). This job copied 653 sectors, 249,756 editorial decisions, 7,086,748 verbatim sources, and triggered the portal build (HTTP 201).

Durations other than the ID provider and the new validation step are either improvements or noise; the −86% on the text-tree step and the 33× difference in NameUsageProcessor "loaded taxa" (299,454 → 10,018,067) look like caching and I/O differences on the host and I would not read anything into them. Both jobs removed ~7.0–7.1 million orphan verbatim sources from the project before copying, which is normal for this pipeline.

7. Per-source findings

158 sources in COL26.10, 160 in COL26.9. Source metrics come from /dataset/{key}/source/{sourceKey}/metrics for both releases; all 318 calls returned 200. Side-by-side in the UI: sourcemetrics?releaseKey=316321.

7.1 IRMNG (2007) — retired content lost

Absent from COL26.10. In COL26.9 it held one sector, 67368 (Protozoa → Eukaryota, attempt 11, version 2025-07-11), 1,544 usages / 1,224 taxa / 320 synonyms; the search index resolved 1,529 of them, a 15-usage discrepancy with the sector file that I cannot explain. Its content was the higher classification of Protozoa: kingdom Protozoa, 3 subkingdoms, 14 phyla, 55 classes, 98 orders, 270 families, 1,034 genera. 333 identifiers were deleted; the remainder kept their identifiers and now have no sector. Confirmed losses: order Protostelida, families Protosteliidae and Acramoebidae, genera Protostelium, Planoprotostelium, Microglomus, Acramoeba, and 13 accepted unranked names (Guembelia, Loeschia, Xastrum, Chasmostoma, Ramesella, Lymphohaematocytozoon, Conradinema, Eneidamonas, Metacoronympha, Teixeiramonas, Arcuothrix, Pleuromastigella, Nodamastix). Genus Orbulites survives with a new identifier (VPXLT → 8FVQB).

7.2 Coleoptera type genera (290628) — retired content unsectored

Absent from COL26.10. In COL26.9: sector 60392 (Coleoptera → Insecta, attempt 19, version 2.2 of 2025-05-28), 3,398 usages / 2,496 taxa / 902 synonyms. 908 identifiers deleted, ~2,490 survive without a sector. Only 6 of the 139 higher-taxon deletions (accepted subgenera) come from here — the rest of its deletions are synonyms, which is why the synonym-by-rank table moves and the taxon-by-rank table does not.

7.3 StaphBase (1204) — new version, the largest real content change

Version Aug 2022 → Sep 2026, attempt 20 → 24. Usages 137,007 → 152,118 (+11.03%); taxa 93,043 → 95,948; synonyms 43,964 → 56,170 (+27.8%). Flagged sectors: 60397 Staphylinoidea 120,513 → 134,017 (+11.2%, 85,524 taxa / 48,493 synonyms) and 60396 Hydrophiloidea 7,263 → 8,087 (+11.3%); the third sector (60395 Histeroidea) grew by the residual +783. A four-year version jump that adds 12,206 synonyms is a legitimate new edition, and it is the source of 29 of the 139 higher-taxon deletions (Leptusa folded into Euryusa; tribes Catopocerini, Glacicavicolini, Hameedini, Nephanini, Valdini; several Leiodidae subtribes). No sign of failure.

7.4 ICTV MSL (1014) — +46.7% with an unchanged version string

See §2.5. Attempt 31 → 33, version 41.v1 in both; the increase is 10,591 synonyms where there were none (10,120 of them species), taxa flat (+1). Applied decisions 1 → 14 and ignored "name informal" 1 → 49 show the sector was actively re-curated. The release uses attempt 33 while the source's latest import is attempt 35 (33,342 usages) — expected, since sources keep importing after a release is cut.

7.5 Systema Dipterorum (1101) — new version 7.2 → 7.3

Attempt 47 → 51; usages 296,549 → 302,913 (+2.15%); taxa 175,760 → 178,879; synonyms 120,789 → 124,034. 24 of the 139 higher-taxon deletions, several of them rank changes rather than losses: Orygmatinae, Oniromyiinae, Plagioneurinae, Pleurocerinellini and Prosochaetini now exist only as family-rank synonyms, which is also where much of the +12,767 rank name suffix conflict sits (1,924 flags in this source). It removed the misspelt families Pipinculidae and Platrystomatidae and introduced the misspelt Dolicchopodidae. It also carries 4,189 of the 8,705 parent genus missing flags (provisionally accepted species sitting directly under Chironomidae and similar) — unchanged data, newly flagged: 9JHXY Ablabesmyia aurea has parent 84X Chironomidae in both releases.

7.6 FADA Cladocera (1138) — re-sectorised, no content change

Attempt 8 → 10, version Jan 2013 in both; 1,874 usages in both; sectors 17 → 86. One sector removed (1982, "Not assigned" → Cladoceromorpha, 1,254 usages) and 70 created, one per genus (Ceriodaphnia, Daphnia, Megafenestra, Scapholeberis, Simocephalus under Daphniidae; 65 more under Cladoceromorpha), summing to exactly 1,254 usages. Benign, but it is the kind of change that inflates resurrected-identifier counts.

7.7 SF Miscellaneous Insecta (315175) — +3.9%, one sector +29.4%

Attempt 8 → 9 (Sep → Oct 2026); usages 4,909 → 5,099. Sector 66719 (Zygentoma → Insecta) 582 → 753 (+29.4%), i.e. +171 of the source's +190 — consistent, a small source adding a block of silverfish names. No reason to suspect duplication.

7.8 The ten largest sources

No large source lost material content, and no source contributed zero usages except the two retired ones. 44 sources were re-imported (attempt changed) with identical usage counts — mostly small WoRMS groups whose version date moved 2026-09-01 → 2026-10-01, but also WoRMS Bryozoa (25,225 usages across attempts 89→94), Taxapad Ichneumonoidea (60,347, static version "May 2009"), Geometridae (52,696) and Tortricid.net (25,488). From the metrics alone I cannot distinguish "the source genuinely did not change" from "the re-import produced nothing new", so I flag it only as something a source manager could confirm. I deliberately did not use latestAttempt/latestUsagesCount as a staleness signal: they describe the whole source dataset as it stands now, which for sectored sources such as WCVP (1,449,468) is much larger than the sector's contribution (303,875).

8. Everything checked and found fine (and what I could not check)

• Fine — Totals: every count grew or held; no shrinkage anywhere (§3.1).
• Fine — Ranks above genus: family +0.12%, order −0.11%, class 0.00%, phylum 0.00%, kingdom 0.00%, and ~24 further high ranks identical in both releases (§3.2).
• Fine — Duplicates above family: 0 at every rank from domain to family in both releases, with genus (1,884 → 1,831) as a control that the endpoint answers for releases (§3.5).
• Fine — Names at/above order: only the four changes in §3.4; no identifier changed for any of the 2,765 shared names; no newly invented orders.
• Fine — Identifier stability: deleted −82.8%, unstable names −99.5%, 5,424,551 identifiers reused. No count exceeds last month's except created (+22.1%).
• Fine — Release job: 0 ERROR lines, 2 WARN lines, "Removed 0 sectors without data", Elasticsearch reports exactly 5,466,019 usages and 0 bare names, text tree written with 5,466,019 lines, portal build triggered (201), notification e-mail sent, job FINISHED.
• Fine — Sector file: no ZERO and no DECREASED rows at all; all three REMOVED rows, all 70 NEW rows and all four INCREASED rows were traced to a named cause (§2, §7).
• Fine — Metadata: both are RELEASE origin from project 3, confidence 5; size equals usagesCount in both.
• Not checked — Checklist item 7 via the API. GET /dataset/316595/diff/316321?minRank=order&authorship=false&synonyms=false never returned: one attempt ran 295 s without output, four later attempts were refused with HTTP 429 "You can only run one diff at a time" while an earlier one was still running, and a final 240 s attempt also returned nothing. §3.4 is a reconstruction from per-rank name lists, not the endpoint's output; it covers accepted names only (as the requested diff would) and cannot show authorship-only changes.
• Not checked — base-deleted.tsv does not apply (both are base releases), and COL26.9 produced no superseded.tsv, so this month's 2,332 superseded identifiers have no baseline.
• Not checked — the 15-usage gap between the sector file's prevUsagesCount for sector 67368 (1,544) and the search index (1,529), and the ~15-usage residual when reconciling retired-source survivors (3,686) against the growth in unattributed usages (+3,701).
• Not checked — the cause of +5.29% type material, −15.02% "infraspecific name" synonyms, and why synonym rank differs is 643,964 rather than the 244,969 of the last validated release.
• Not checked — sectors that moved less than the threshold used to build sector-metrics.json: the file has 77 rows while the release has 653 sectors, so most sectors were not inspected individually. The API cannot report sector metrics for a release.
• Not checked — one name in §5.2 (Merisus [sic], VPDMW); and whether the 44 re-imported-but-identical sources are genuinely unchanged at source.
• Not applicable — media and treatment counts are 0 in both releases, as in every release of this project in the window; bare names are 0 in both, so there is nothing to compare.

9. Method

Pair under review: 316595 (COL26.10, RELEASE, project 3, attempt 642) against 316321 (COL26.9, RELEASE, attempt 629) — base against base. Trend context from the base releases 316115, 315777, 315448, 315149, 314909, 314455, 314216, 313722, 313452, 313062 and 312563, via /dataset/{key}/import?limit=1 for each. Every figure above comes from one of the calls or files listed here; none is estimated or extrapolated.

API (https://api.checklistbank.org, bearer token, read-only; no write request was attempted): /dataset/{key}; /dataset/{key}/import?limit=1; /dataset/{key}/source; /dataset/{key}/source/{sourceKey}/metrics (158 + 160 calls); /dataset/{key}/duplicate/count with category=uninomial&minSize=2&mode=STRICT&rankDifferent=false&status=accepted for twelve ranks in both releases plus genus/species/subgenus as controls; /dataset/{key}/nameusage/search with issue, rank, status, sectorKey, sectorDatasetKey, TAXON_ID, facet and q filters; /dataset/{key}/tree and /dataset/{key}/tree/{id}/children; /dataset/{key}/taxon/{id} (all 139 higher-taxon deletions resolved in COL26.9, plus spot checks in both releases); /dataset/3/sector/{key}; /dataset/{key}/sector; /vocab/issue. /dataset/316595/diff/316321 was attempted six times and never returned (§8).

Mounted files: /mnt/session/uploads/sector-metrics.json (77 rows: 70 NEW, 3 REMOVED, 4 INCREASED, 0 ZERO, 0 DECREASED); job-log-digest.md and previous-job-log-digest.md; id-reports/ (created 10,332, deleted 7,893, resurrected 31,136, superseded 2,332, unstable 571 lines) and previous-id-reports/ (created 8,465, deleted 45,984, resurrected 66,237, unstable 105,263 lines). Counted and filtered with wc, cut, sort | uniq -c, awk, comm and grep; no ID report was read whole.

This review is published alongside the release reports at https://download.checklistbank.org/releases/3/642/.