Review: Congenital Disorders of Glycosylation (CDG) and Deglycosylation (CDDG) groupings

Date: 2026-08-20 · Scope: kb/groupings/Congenital_Disorders_of_Glycosylation.yaml, kb/groupings/Congenital_Disorders_of_Deglycosylation.yaml, and all their member Disease entries · Method: MONDO closure audit (2026-08 release), just check-groupings, just compliance-weighted, per-entry section census, and a PubMed sweep (2023–2026).

1. Headline findings

#	Finding	Severity
1	The CDG grouping listed 7 members while the KB held 14 qualifying CDG entries. Seven curated diseases that satisfy the grouping's own NECESSARY criteria were absent. Fixed in this change.	High
2	The CDDG grouping is MONDO-complete (2/2 descendants curated) but holds the two thinnest entries in the whole glycosylation family.	Medium
3	NGLY1-CDDG is 3 years behind its own literature — no treatments, no clinical trials, no animal models, and none of the NFE2L1/Nrf1, ENGASE, or AAV9 gene-therapy biology that now defines the field.	High
4	MAN2C1-CDDG2 is a single-source entry (one PMID). A second cohort report published 2026-01 is uncited.	High
5	COG1-CDG's evidence is structurally unverifiable: 50 evidence items cite bare `DOI:` references, which `linkml-reference-validator` skips. All three DOIs resolve to PMIDs.	High
6	~~VPS51-PCH-CDG and UGGT1-CDG have no MONDO anchor.~~ Corrected during follow-up: both terms already existed. VPS51 disease is `MONDO:0032831` (PCH type 13, causal gene VPS51, xref OMIM:618606 — the same OMIM the entry was seeded from) and UGGT1-CDG is `MONDO:0980705` (CDG type IIcc, causal gene UGGT1). Both are now bound. No term request was needed.	Medium
7	Four members (ALG1, SLC35A2, NGLY1, MAN2C1) do not declare `conforms_to` against any module, despite a `congenital_disorder_of_glycosylation` module existing that 12 siblings use.	Medium
8	No deglycosylation module exists. The CDDG pair has nowhere to conform to.	Low

2. Membership audit

2.1 CDDG — complete

MONDO:0031376 (congenital disorder of deglycosylation) has exactly two is_a descendants in the 2026-08 MONDO release:

MONDO	Disease	Curated
`MONDO:0800044`	congenital disorder of deglycosylation 1 (NGLY1)	✅
`MONDO:0030770`	congenital disorder of deglycosylation 2 (MAN2C1)	✅

The grouping is MONDO-complete. That is a statement about the small size of the concept, not the depth of the entries. This completeness check is now recorded in the grouping's notes so a future reviewer does not have to re-derive it.

2.2 CDG — was 7/14, now 14/14

MONDO:0015286 has 169 is_a descendants; 23 KB entries bind one as their primary disease_term. Of those, 14 fall inside this grouping's declared N-glycan boundary. Seven were missing:

Added member	Gene	Step	MONDO
ALG1-CDG	ALG1	first β-1,4-mannosylation of the cytosolic-face LLO (CDG-Ik)	`MONDO:0012052`
ALG3-CDG	ALG3	first luminal-face mannosyltransferase (CDG-Id)	`MONDO:0010998`
DPM2-CDG	DPM2	Dol-P-Man synthase subunit — donor synthesis (CDG-Iu)	`MONDO:0014023`
MPI-CDG	MPI	GDP-mannose precursor pool (CDG-Ib) — the treatable CDG	`MONDO:0011257`
PGM1-CDG	PGM1	UDP-Glc/Gal supply — mixed type I/II, galactose-responsive	`MONDO:0013968`
SLC35A2-CDG	SLC35A2	Golgi UDP-galactose transporter (type II), X-linked	`MONDO:0010478`
VPS51-PCH-CDG	VPS51	GARP/EARP retrograde tether → Golgi enzyme positioning	none

All 14 members now return SATISFIED against the NECESSARY criteria with no contradictions (just check-groupings).

Twelve of these fourteen already declared conforms_to against the congenital_disorder_of_glycosylation module — so the module graph knew they were CDG while the grouping did not. That divergence is the cleanest signal that the omission was drift, not a boundary decision.

2.3 Boundary — deliberately narrower than the modern CDG nosology

The 2024–2026 CDG literature treats CDG as spanning N-linked, O-linked, GPI-anchor, and glycosphingolipid defects (>190 genes). This grouping's NECESSARY criteria are N-glycan-specific (GO:0006487 OR GO:0006491). Keep that boundary — the wider concepts are already held by siblings, so nothing is lost:

Concept	Where it lives
GPI-anchor defects (Mabry/PIGV, CHIME/PIGL, PNH, MCAHS2)	`Disorders_of_GPI_Anchor_Biosynthesis`
Precursor-supply / ER-QC multi-pathway (UGDH, UGP2, UGGT1)	`Other_Multiple_Glycosylation_Pathway_Disorders`
Folded-TSR O-fucose/glucose QC (Peters plus, Geleophysic)	`tsr_o_glycosylation_quality_control` module
Dystroglycan O-mannosylation	`Dystroglycanopathy` entry

This is now written into grouping_rationale so the next reviewer sees why a "missing" CDG is not missing. UGGT1-CDG was deliberately not dual-listed here; it is an ER re-glucosylation quality-control defect and the sibling grouping is the better home.

3. Member-by-member completeness

Compliance is the weighted just compliance-weighted score. refs counts distinct literature identifiers, not evidence items.

Member	Lines	Compliance	refs	conforms_to	Tx	Dx	Prev	Trials	Models	Verdict
COG1-CDG	1674	99.6%	1 PMID + 50 DOI + 43 ORPHA	✅	3	1	1	0	0	evidence base unverifiable
ALG12-CDG	1687	99.5%	8	✅	1	3	1	1	0	solid
ALG9-CDG	1624	99.5%	8	✅	0	3	1	1	0	solid; no treatments
COG7-CDG	1436	99.5%	5	✅	1	2	0	1	0	solid
ALG1-CDG	1322	98.9%	18	❌	2	3	1	0	0	wire to module
MGAT2-CDG	1029	98.7%	3	✅	2	2	0	0	0	thin sourcing
SLC35A2-CDG	539	96.6%	10	❌	3	0	0	2	0	thin; fast-moving field
MPDU1-CDG	1367	96.4%	5	✅	3	2	0	0	0	adequate
MPI-CDG	2446	95.8%	14	✅	10	3	2	1	2	best in group
VPS51-PCH-CDG	523	94.7%	3	✅	0	0	0	0	0	thin; no MONDO
NGLY1-CDDG	365	92.9%	5	❌	0	0	0	0	0	badly out of date
MAN2C1-CDDG2	225	91.2%	1	❌	0	0	0	0	0	single-source
UGGT1-CDG*	825	90.5%	—	❌	1	4	0	0	2	no MONDO ID
DPM2-CDG	2115	90.6%	14	✅	5	3	1	0	0	good
PGM1-CDG	2770	90.4%	20	✅	13	3	1	2	0	richest; needs 2026 update
ALG3-CDG	1234	83.4%	10	✅	5	3	1	0	0	lowest score in group

* UGGT1 sits in the sibling grouping, listed here for completeness.

Cross-cutting gaps. Zero datasets: records anywhere in either grouping. Only MPI-CDG and UGGT1-CDG carry any model system, despite a 2025 systematic review of zebrafish CDG models (PMID:40993721) and a new hypomorphic Mpi mouse (PMID:40693465). Only four of sixteen entries carry classifications:.

The compliance score is misleading here. COG1-CDG scores 99.6% — the highest in the group — while resting on one PMID and fifty DOI-cited snippets that no validator ever checks. Compliance measures field coverage, not evidence verifiability.

4. Latest publications (2023–2026) and what each would change

4.1 NGLY1 — the largest gap in either grouping

The entry cites 5 references, the newest from 2019. Since then:

PMID	Year	What it establishes	Would populate
`41468431`	2026	NFE2L1/Nrf1 sequence editing. NGLY1 converts glycosylated Asn→Asp in Nrf1; editing of Asn574 is required for HCF-C1/OGT binding, chromatin binding, and proteasome-subunit gene expression; other sites drive CREBBP/EP300 coactivator recruitment.	A whole missing `pathophysiology` arm: deglycosylation failure → Nrf1 activation failure → proteasome bounce-back failure
`41721346`	2026	Aging Ngly1⁻/⁻ rat. ~50% dead or euthanized by 9–10 months; ~92% ↓ rotarod latency, ~82% ↓ rearing; widespread neuroinflammation; loss of peripheral axons and spinal motor neurons.	`animal_models` with `modeled_mechanisms` + readouts; supports a progressive neurodegeneration node the entry lacks
`41176936`	2025	AAV9-hNGLY1 ICV in Ngly1⁻/⁻ rats suppresses non-epileptic convulsions but fails to correct EEG abnormalities or sleep fragmentation.	`treatments` + a `FAILS_TO_RECAPITULATE`/partial-rescue model link; an honest negative result
`40687377`	2025	IND-enabling preclinical pharmacology/safety for an AAV9 NGLY1 gene-therapy trial.	`treatments`, `clinical_trials`
`42114141`	2026	Prospective natural history, 15 participants (~10% of the known population): widening developmental gap, GlcNAc-Asn (GNA) elevated in all.	`progression`, `biochemical` (the entry has aspartylglycosamine but not the GNA endpoint framing), trial-endpoint context
`42361657`	2026	Multicenter cohort, 15 patients / 11 families. Abnormal EEG 10/15, seizures 9/15, hyperkinetic movement 8/15, feeding difficulty 9/15, scoliosis 7/13, hypo-/alacrima 6/15, transaminase elevation 12/14 (transient in 9/12), low total cholesterol 7/10 and low HDL 5/10; two novel variants.	Frequency bands for existing phenotypes; new phenotypes: scoliosis, feeding difficulty, auditory neuropathy, hypolipidemia
`40773511`	2025	Natural SEL1L variants rescue an NGLY1-deficiency model and modify ERAD function.	Genetic modifier / `discussions`
`28426790`	2017	Engase deletion partially rescues Ngly1⁻/⁻ mouse lethality.	The canonical modifier axis — absent from the entry
`40643555`	2025	Structure–function of NGLY1 pathogenic variants.	`genetic` variant-level detail
`41229635`	2025	Tofacitinib improved motor symptoms in parkinsonism with a heterozygous NGLY1 variant + autoimmunity (see also `42052850` on NGLY1 in immune function).	`treatments` — but note this is a heterozygote, not classical CDDG; curate as a discussion, not a disease treatment
`41917400`, `41096971`, `40730667`, `40602583`	2025–26	Four reviews of NGLY1 biology and therapeutic strategy.	Orientation for the rebuild

The mouse/rat divergence (mouse null embryonically lethal on C57BL/6, rat viable but progressively degenerating) is a textbook HUMAN_MODEL_MISMATCH discussion that the entry does not carry.

4.2 MAN2C1

PMID	Year	Note
`41623318`	2026	Second report. Novel pathogenic MAN2C1 variant in trans with a 15q24.1q24.3 microdeletion — the only patient described since the founding cohort. Directly addresses this entry's single-source problem.
`37486637`	2023	Exome sequencing in polymicrogyria — independent support for the polymicrogyria phenotype the entry cites from one paper.

4.3 CDG members

PMID	Year	Relevance
`41807832`	2026	Cryo-EM structures of ALG3/9/12 reconstituting all four mannosylation steps; explains donor selection (Dol-P-Man over Dol-P-Glc) and gives "mechanistic explanations for enzyme dysfunction in CDGs". Directly upgrades the molecular node in three members at once.
`42269412`	2026	Recurrent HLH in COG deficiency (COG6/COG7): 3 patients, plus ~40% of published COG patients had unexplained febrile episodes; corticosteroids mitigated neurological impact. A clinically actionable phenotype+treatment absent from COG7-CDG.
`42511652`	2026	Comprehensive MS N-glycan profiling for CDG type II subtyping — `diagnosis` for MGAT2/COG1/COG7/SLC35A2.
`41718976`	2026	Review: COG complex in Golgi trafficking and glycosylation.
`41867720`	2026	SLC35A2: disrupted O-GalNAc glycosylation as mechanism and biomarker. Mouse forebrain knockout shows a specific O-GalNAc defect (other galactosylated glycoconjugates intact); truncated O-GalNAc glycans on ECM molecules; variant burden correlates with truncation in human epilepsy tissue. Reframes SLC35A2 disease from generic hypogalactosylation to a specific O-glycan lesion. (Preprint — flag as such.)
`41373710`	2025	Review: SLC35A2-related brain disorders — genetics, pathophysiology, therapeutics.
`42542984`, `42216953`, `42168328`	2026	MOGHE (mild MCD with oligodendroglial hyperplasia) — the somatic-mosaic brain-limited arm of SLC35A2, including Y-chromosome mosaicism and snRNA-seq oligodendrocyte dysregulation. The SLC35A2 entry has no `has_subtypes` for the germline vs somatic-mosaic split.
`41723528` / `40631269`	2026 / 2025	PGM1 cardiomyopathy is glycosylation-independent: patient iPSC-cardiomyocytes show Z-disk depletion via a predicted PGM1–LDB3 (ZASP/Cypher) interaction plus mitochondrial protein loss — explaining why galactose corrects glycosylation but not cardiac disease. A major mechanistic addition.
`40242152`	2025	Creatine supplementation outcomes in PGM1-CDG.
`41018607`	2025	LC-MS hexose-phosphate resolution → new PGM1-CDG pathophysiology.
`41099230`, `41306474`	2025–26	PGM1-CDG misdiagnosed as Laron syndrome; ASD closure for refractory heart failure.
`42376639`, `41172867`	2025–26	Two pregnancy-management reports in MPI-CDG, including oral D-mannose through pregnancy.
`40962549`	2025	D-mannose treatment outcomes in 5 children with MPI-CDG.
`40693465`	2025	Hypomorphic Mpi mouse — an in vivo tool for global N-glycosylation deficiency.
`39984963`, `38876156`, `38717015`	2024–25	ALG12-CDG: novel intronic variant with low mRNA; Duane syndrome association; expanded prenatal phenotype incl. bilateral multicystic kidneys.
`38831602`, `36755425`	2023–25	MPDU1: erythrokeratodermia variabilis presentation; severe ciliopathy-like phenotype.
`40902550`	2025	Review: genetic disorders of dolichol synthesis and utilization — covers DOLK and MPDU1.
`38597022`	2024	ALG3-CDG: deficient glycan extension and ER stress — a second mechanistic arm.
`40743674`	2025	Multi-omics across six CDG (incl. NGLY1-CDDG and PGM1-CDG): shared disruption of autophagy, vesicle trafficking, and mitochondrial function; EMUDRA-predicted repurposable drug classes. The strongest argument yet for shared module-level biology.
`40993721`	2025	Systematic review of zebrafish CDG models.
`40868218`, `41713138`	2025–26	Clinical glycomics/glycoproteomics for CDG; albumin as a glycoprotein biomarker.
`41554664`	2026	CDG due to defective membrane transporters: update — covers SLC35A2.
`40119203`, `39236565`	2024–25	Diagnostic and therapeutic approaches in CDG; treatment overview.

Two literature items to not curate as-is. 42537239 ("COG7 links Golgi integrity to stress signaling and senescence") is an Arabidopsis thaliana study — interesting for the module, not evidence for human COG7-CDG. 41229635 (tofacitinib) concerns a heterozygous NGLY1 carrier with autoimmune disease, not biallelic NGLY1-CDDG.

5. Knowledge gaps worth recording as `discussions`

NGLY1: which substrate explains the phenotype? Nrf1 sequence editing is now mechanistically resolved (41468431), but no one has shown that Nrf1/proteasome failure accounts for alacrima, the movement disorder, or the neuropathy. KNOWLEDGE_GAP.
NGLY1: mouse vs rat vs human. Ngly1⁻/⁻ mice are embryonically lethal on C57BL/6 and rescued by Engase deletion; rats survive and degenerate progressively; humans have a ~13-year median lifespan. HUMAN_MODEL_MISMATCH.
NGLY1: gene therapy's partial rescue. AAV9-hNGLY1 suppressed convulsions without correcting EEG or sleep (41176936). Curating gene therapy as "effective" would overstate the evidence — record the dissociation.
PGM1: two independent disease mechanisms. Galactose corrects glycosylation; cardiomyopathy persists via a Z-disk/mitochondrial axis (41723528). This is a competing-mechanistic_hypotheses shape, not one chain.
SLC35A2: germline hypogalactosylation vs somatic-mosaic O-GalNAc/MOGHE. Arguably two diseases sharing a gene. Needs has_subtypes before it needs more phenotypes.
COG deficiency and immune dysregulation. Why does a Golgi trafficking defect produce HLH in ~40% of patients (42269412)? Unexplained.
MAN2C1: is free-oligosaccharide accumulation causal? The founding cohort shows accumulation and a phenotype; nothing connects them. The entry's own INDIRECT_UNKNOWN_INTERMEDIATES edge already admits this — it should be an explicit KNOWLEDGE_GAP with proposed experiments.
Is there a shared CDG therapeutic axis? 40743674 finds autophagy, vesicle trafficking, and mitochondrial dysfunction shared across six CDG. If it holds, it belongs in the congenital_disorder_of_glycosylation module, not in six disorder entries.

6. Recommended follow-up work, in priority order

#	Action	Why now
1	Rebuild NGLY1-CDDG from the 11 references in §4.1 — pathophysiology (Nrf1 arm), treatments, clinical trials, animal models, progression, two discussions.	Biggest single gap; the field moved and the entry did not
2	Add the 2026 MAN2C1 report (`41623318`) so CDDG2 is no longer single-source.	Cheap; removes a structural fragility
3	Convert COG1-CDG's 50 `DOI:` references to PMIDs — `10.1111/cge.13980`→`PMID:33960418`, `10.1186/s12887-021-02922-7`→`PMID:34625039`, `10.1073/pnas.0507685103`→`PMID:16537452` — then re-run `just validate-references`.	50 evidence snippets currently bypass validation entirely
4	Add `conforms_to: congenital_disorder_of_glycosylation#…` to ALG1-CDG and SLC35A2-CDG.	Two-line fix; they already satisfy the criteria
5	Add COG HLH phenotype + corticosteroid prophylaxis to COG7-CDG (`42269412`).	Clinically actionable
6	Split SLC35A2-CDG into germline and somatic-mosaic/MOGHE subtypes.	Three 2026 papers on the mosaic arm alone
7	Add the glycosylation-independent cardiomyopathy arm to PGM1-CDG (`41723528`).	Changes what "treated" means for the group's flagship treatable disease
8	Raise MONDO term requests for VPS51-related PCH-CDG and UGGT1-CDG.	Both entries are stuck without an ontology anchor
9	Consider a `cytosolic_deglycosylation` module so the CDDG pair has a conformance target.	The two members share a substrate pool and currently share nothing structural
10	Add `datasets:` to MPI-CDG and PGM1-CDG (the two with real cohort/omics data).	Zero dataset coverage across both groupings

6a. Follow-up: all ten recommendations applied (2026-08-20)

Sections 4 through 6 were written as recommendations. They were subsequently carried out in full. What follows is what was actually done, including the two places where doing the work changed the finding.

#	Recommendation	Outcome
1	Rebuild NGLY1-CDDG	Done. 365 → 1,614 lines. Two new pathophysiology nodes (ENGase bypass / GlcNAc-Asn accumulation; failed NFE2L1 sequence editing), 9 new phenotypes with frequency bands from the 2026 cohort, the GNA biomarker as a structured `biochemical` readout, `prevalence`, `progression`, 2 treatments (GS-100 gene therapy with `target_mechanisms`; GlcNAc for alacrima), 3 clinical trials, 3 animal models with `modeled_mechanisms` and readouts, the SEL1L modifier and the recurrent Arg401* allele, 2 GEO datasets, and 3 discussions.
2	Add the 2026 MAN2C1 report	Done. `PMID:41623318` and `PMID:37486637` added; the entry is no longer single-source. A `KNOWLEDGE_GAP` on free-oligosaccharide causality and a note on the 15q24 compound genotype were added with it.
3	Convert COG1-CDG's DOI references	Done. All 50 `DOI:` references converted to `PMID:33960418`, `PMID:34625039`, `PMID:16537452`. On first validation 101 of 103 snippets verified — the curator's DOI-cited quotes had been accurate all along, just unverifiable. The 2 failures were one quote spanning a bracketed HGVS span (`[Arg889Profs*12]`), which the validator strips; it was shortened to a verbatim span that does not cross the bracket. Now 103/103.
4	`conforms_to` for ALG1 and SLC35A2	Done. ALG1 at three nodes, SLC35A2 at two (the terminal node was left unconformed — SLC35A2 disease is brain-predominant, not multisystem glycoprotein dysfunction).
5	COG7 HLH arm	Done. New pathophysiology node, 2 phenotypes (hemophagocytosis; recurrent unexplained fever at FREQUENT, from the ~40% literature figure), corticosteroid prophylaxis as a treatment with `target_mechanisms`, and a `KNOWLEDGE_GAP` on the unexplained glycosylation-to-immune-dysregulation link.
6	Split SLC35A2 into subtypes	Done. `Germline` and `Somatic MOGHE`. Four phenotypes that depend on body-wide transporter deficiency (transferrin profile, skeletal, dysmorphic, failure to thrive) are tagged to `Germline`; shared phenotypes are deliberately left untagged. A `KNOWLEDGE_GAP` records the O-GalNAc reframing as an unreplicated preprint rather than curating it as mechanism.
7	PGM1 cardiomyopathy arm	Done. New `Z-Disk Destabilization via Loss of PGM1-LDB3 Interaction` node; the existing `galactose_resistant_cardiomyopathy` hypothesis extended with arm (c) and its human-cardiomyocyte evidence; the patient iPSC-cardiomyocyte model added as an `experimental_models` entry with two mechanism links and three readouts.
8	MONDO term requests	Not needed — the finding was wrong. Both terms already exist. `MONDO:0032831` (pontocerebellar hypoplasia type 13) xrefs `OMIM:618606`, the same OMIM this entry was seeded from, and records VPS51 as its causal gene; `MONDO:0980705` (CDG type IIcc) records UGGT1. Both are now bound rather than requested. The original finding repeated each entry's own stale note; the earlier VPS51 audit had rejected PCH1A correctly but not gone on through the numbered PCH series. One real upstream gap survives: `MONDO:0032831` sits under pontocerebellar hypoplasia, not under CDG, so VPS51-PCH-CDG is a grouping member without being a descendant of the mapped class. That is a much smaller MONDO ask than a new term.
9	`cytosolic_deglycosylation` module	Done. Five nodes, registered in `CLAUDE.md`, with both CDDG entries wired as conformers. The central node is named Cytosolic Glycan Catabolite Dysregulation rather than for free oligosaccharides: MAN2C1 accumulates free oligosaccharides but NGLY1 accumulates a glycoasparagine, and naming it for the pool lets both arms attach without either overstating its catabolite. The NFE2L1 sequence-editing branch is fenced off as NGLY1-only.
10	Datasets for MPI-CDG and PGM1-CDG	Partly done, and the negative result is the finding. Neither disease has a relevant public dataset: `just discover-datasets` returned only `GENE_ONLY` gene-symbol collisions (yeast PGM1 deletion compendia; a Burkitt lymphoma methylation series and an unrelated liver-fibrogenesis "MPI MT" series), and direct GEO searches for `PGM1-CDG`, `PGM1 deficiency`, `MPI-CDG`, and `MPI deficiency glycosylation` all return zero series. Adding any of them would be Named Entity Confusion reached through dataset search. Both entries now record the search and its date in `notes` and keep `datasets: []`. Four genuinely relevant datasets were found and added elsewhere in the same groupings — `GSE301626` and `GSE295078` to NGLY1, `GSE318030` and `GSE284073` to SLC35A2 — all four resolved by `just verify-datasets`.

Two findings changed on contact with the work. Recommendation 8 dissolved: the ontology gap did not exist, and the review had propagated the entries' own stale notes instead of checking MONDO. Recommendation 10 inverted: the gap is real but unfixable for those two diseases, and the honest output is a recorded negative search rather than a filled field.

Validation of the applied work. just validate-disorders over all ten changed disorder files — the exact command CI runs — passes: 10 files, all validations passed, 728/728 snippets verified. Across the ten disorder files plus the new module, just count-verified-snippets reports 744/744. just validate and just validate-terms pass individually on every changed file; just validate-grouping passes on both groupings; just check-duplicate-keys passes over all 4,249 YAML files; just check-title-snippets reports no new title-quoting snippets; just verify-datasets resolves all four new accessions; every conforms_to reference in the repository resolves to a real module node.

7. What this change actually did

Changed:

kb/groupings/Congenital_Disorders_of_Glycosylation.yaml — added 7 members with differentiating_mechanisms; rewrote grouping_rationale to state the N-glycan boundary and name the sibling groupings that hold the rest; corrected the MONDO consistency note from "All 7 listed members" to the verified 13-of-14 with the VPS51 exception explained; extended notes with the type I/II/mixed axis and the treatability axis.
kb/groupings/Congenital_Disorders_of_Deglycosylation.yaml — recorded the MONDO completeness check and the shared free-oligosaccharide-pool rationale in notes.
cache/hgnc/terms.csv — one row (hgnc:7216 MPI) added by the term validator.

Not changed: no kb/disorders/ entry was edited. Everything in §4–§6 is a recommendation, not an applied change — those need the full fetch-and-verify evidence workflow, not a review pass.

Validation run:

just validate-grouping kb/groupings/Congenital_Disorders_of_Glycosylation.yaml    ✅
just validate-grouping kb/groupings/Congenital_Disorders_of_Deglycosylation.yaml  ✅
just check-groupings   kb/groupings/Congenital_Disorders_of_Glycosylation.yaml    ✅ 14/14 SATISFIED, 0 contradictions
just validate-terms    kb/groupings/Congenital_Disorders_of_Glycosylation.yaml    ✅
just check-duplicate-keys (both files)                                            ✅
just check-term-cache-integrity                                                   ✅
just validate-history-all                                                         ✅ 5825 records
just normalize-cache                                                              ✅

The grouping member foreign-key check was run as a direct script applying the same rule as tests/test_data.py::test_grouping_member_foreign_keys (every members[].member of type DISEASE/SUBTYPE must resolve to a Disease.name or has_subtypes[].name): 5,030 names indexed, 14 CDG members and 2 CDDG members, zero failures. The pytest node itself was not run to completion here — building its _disease_names() index over kb/disorders/ takes several minutes in this environment — so CI remains the authoritative run of that test.

Literature identifiers in this report were resolved through the PubMed E-utilities API and titles/abstracts read directly; they have not been fetched into references_cache/ or snippet-verified, because none of them is cited as evidence in a KB entry yet. Any curator acting on §4 must run just fetch-reference and just count-verified-snippets before committing a snippet.