Corpus 1,272 assessed · 1,173 scored · 643 reproduced ≥75 · 168 flagged ·∅ 74.1/100
← New search

GRAMD1B is a regulator of lipid homeostasis, autophagic flux and phosphorylated tau.

Nat Commun · 2025
L1 91/100 PQI 97
Why this verdict

The main results reproduced: recomputed values matched the published ones within tolerance.

Reproduced on the brainbox compute brainarbeit.com
Scoring basis — itemised

Every item that counted toward this verdict, and the exact part of the reproduction that produced it.

Concordant (toward reproduced)
All content-critical questions reproduced
-4 pts
From: Q7 · Core claim 🟢
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Every question reproduced
-1 pts
From: “every question reproduced”
Total score -7
✓ What held up
  • Same input data as the authors
  • Reported values were directly comparable
  • No relevant deviation in data/preprocessing
  • No authors-side cause for any deviation
  • Reported values are derivable from the shared data
  • Any deviation was negligible
  • The central claim held under reproduction
  • Overall, the reproduction was clean
What did not (or only partly)
  • Every checked point held up.
How its reproducibility compares
91/100
Reproducibility score
1.0 SD above mean
vs. all fields · 1173 studies
🎯 Scores higher than 82% of all assessed papers rank 197 of 1173 scored

A 0–100 reproducibility-quality score from the per-question grades, shown as a z-score: standard deviations above (+) or below (−) the mean of comparable assessments.

Reproduction agent’s raw note

DESCRIBED WELL ENOUGH and reproduced ~1:1 for the in-scope scRNA-seq pipeline; paper is otherwise wet-lab (see scope.md). Code (eturkes/acosta-organoid-scRNAseq @dc364fb, GPL-3.0) and data (GEO GSE278619, open) both resolve. Compute ran on «our HPC» SLURM compute nodes (strict SLURM-only; env-build + analysis in-job): «job» built a functionally-equivalent conda env (R4.3.3/Seurat5.3.0/edgeR/limma/scuttle/DropletUtils) and re-derived per-sample QC; «job» re-ran the documented pseudobulk-limma DEG. RESULTS: C5 EXACT and GENUINE -- independent per-sample 3-MAD QC (01_prep.Rmd) on the 12 raw GEO matrices yields 36782 cells passing QC, matching the deposited annotated object to the cell (44697 barcodes -> 36782). C3 HEADLINE reproduced -- GRAMD1B significantly up in mutant (HET&HOM>WT) D120 excitatory neurons (global-F.adj=4.24e-4) and EX-specific (not significant in Inhibitory/RadialGlia/Mural; borderline IPC 8.9e-3 via HOM-WT only, consistent with the paper noting GRAMD1B is expressed in EX & IPC); not significant at D72 or all-timepoints (within-tol). C3b ALL 7 named lipid DEGs (MIAT,COMT,NRP1,VEGFB,GRAMD1B,DSEL,FUT9) EXACT (significant + up in mutant; F.adj 5.2e-6..5.0e-4). C4 EX robust at D120 (56.7%) vs D72 (28.2%) EXACT. C1 five broad types and C2 subclusters present as reported (EXACT on value) but READ from the deposited annotated object -- the RCAv2+ClusterMap annotation (steps 02-03) was NOT independently re-run (env-fragile, needs external Bhaduri 2020 ref), so C1/C2 are descriptive reads, not from-scratch re-derivations (flagged honestly). Env functionally-equivalent (not byte-identical to authors' apptainer rocker4.2/Seurat-develop; deposited object needed UpdateSeuratObject), so exact p-values may differ slightly while directions, significance and named-gene sets all match. NOT ATTEMPTED: from-scratch re-annotation, GSVA(05)/WGCNA(06), and all wet-lab modalities (lipidomics MS, RNAscope, IF/WB, overexpression/KD, autophagy assays, PS19 mouse, post-mortem tissue, qRT-PCR). No possible-fabrication flags: deposited object, raw matrices and DEG outputs are mutually consistent with the paper's claims. INFRASTRUCTURE NOTE: HOME and per-user «infra» quotas (shared across the concurrent multi-room batch) were intermittently exhausted; worked around by redirecting HOME/caches to «infra» and making the job quota-resilient (node-local I/O + retry-copy of small results). All compute was run strictly as SLURM jobs on compute nodes (never on the login node).

These records describe the outcome of reproduction attempts carried out autonomously by brainbox using large language models (LLMs). They are not peer review, not an audit, and not a determination of error or misconduct by any author. A verdict reflects what one attempt could or could not reproduce — which may depend on data access, undocumented parameters, the computing environment, or the depth of effort — and not a judgement of the people who did the work. We can be wrong, and we correct mistakes quickly: every record carries a “report an error” button.

Assessment versions

Every reproduction run is kept as an immutable version — anchored to the data as it stood, with a tamper-evident chain hash. A rerun (e.g. after an author updates a deposit) adds a new version; the previous one stays on record.

  1. v1 current initial assessment Score 91
    assessed: 2026-06-16 ⛓ 9bb393fcf5e7
✎ I am an author of this paper

Updated or fixed a deposit, or is there an erratum? Ask us to re-run the metrics. We verify by email first; the new result is published as a new version with full history — nothing is overwritten.

Reason for the rerun

We email you a confirmation link first. The rerun is an objective re-measurement — it cannot change the verdict in your favour, only ask us to look again.

Provenance — full disclosure

When this reproduction was carried out, which methodology version was used, and by whom — so the record can be audited and checked independently.

Reproduced
2026-06-22
Rubric version
v1.0
Assessed by
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-18
no human curator yet
Last updated
2026-08-05

Provisional, curator- or AI-assessed, and independently checkable. A reproduction outcome states what one attempt could reproduce — not a judgement of the authors.

Deep full-text extraction

Model: sonnet
Founding hypothesis

The relationship between lipid dyshomeostasis and tau pathology in FTLD and AD is unclear, and this study tests whether GRAMD1B, a nonvesicular cholesterol transporter, links lipid dyshomeostasis, autophagy dysfunction, and tau hyperphosphorylation in neurons.

Core claims
  • GRAMD1B is increased in excitatory neurons of human neural organoids (HNOs) carrying the MAPT R406W mutation finding
  • GRAMD1B expression is increased in human FTLD and AD brain tissue and in PS19 tau mice finding
  • Overexpression of GRAMD1B increases free cholesterol and lipid droplets and impairs autophagic flux finding
  • Modulating GRAMD1B alters autophagy-related components (PI3K, phospho-AKT, p62) and increases phosphorylated tau and CDK5R1 in iPSC-derived neurons mechanism
  • Blocking or knocking down GRAMD1B decreases free cholesterol, lipid droplets, phosphorylated tau, and CDK5R1 expression finding
  • MAPT R406W HNOs exhibit increased tau phosphorylation (PHF1) at Day 120 but not at Day 60 finding
  • MAPT R406W HNOs show decreased neuroelectrical and network activity, including reduced firing rate, spikes, and increased interburst intervals finding
  • GRAMD1B (Aster protein family) mediates nonvesicular lipid transport at endoplasmic reticulum-plasma membrane contact sites mechanism
Experimental setups
Assay System Perturbation Readout Platform
single-cell RNA-sequencing (scRNA-seq) human neural organoids (HNOs), MAPT R406W hiPSC-derived MAPT R406W mutation (HET/HOM vs isogenic WT) cell-type transcriptomic profiles, differentially expressed genes (e.g., GRAMD1B) in excitatory neurons
immunofluorescence (IF) staining human neural organoids (D60, D120) MAPT R406W mutation cell-type markers (SOX2, PAX6, Ki67, TBR1, CTIP2, SATB2, NEUN, GAD1), total tau, GFAP, OLIG2, PHF1/total tau
Western blot human neural organoids (D60, D120) MAPT R406W mutation PHF1, total tau (TAU/TauC), GAPDH protein levels
qRT-PCR human neural organoids (D120) MAPT R406W mutation MAPT 3R vs 4R tau isoform mRNA expression human-specific MAPT 3R/4R primers
ELISA HNO culture media (D60, D120) MAPT R406W mutation secreted ptau181 levels
immunoprecipitation (IP) / Western blot HNO culture medium (D120) MAPT R406W mutation secreted total tau (CP27, TauC) and phosphorylated tau (PHF1, 12E8)
microelectrode array (MEA) recording human neural organoids (D75) MAPT R406W mutation firing rate, spike number, burst activity, interburst interval, local field potentials MEA chips
Key results
  • Mutant HET HNOs show increased PHF1/total tau by IF staining at D120 P=0.0041 (WT vs HET)
  • Mutant HOM HNOs show increased PHF1/total tau by IF staining at D120 P=0.0280 (WT vs HOM)
  • No significant increase in phosphorylated tau (PHF1) at D60 in mutant HNOs
  • Total tau protein levels unchanged at D120 by Western blot across genotypes
  • Homozygous HNOs show no significant difference in PHF1/Tau vs WT by Western blot, likely due to increased inhibitory neuron populations
  • Mean firing rate decreased in mutant HET HNOs vs WT HNOs (MEA, D75) P=0.033
  • Number of spikes decreased in mutant HET HNOs vs WT HNOs (MEA, D75) P=0.016
  • Interburst interval increased and synchronized burst firing (ASDR) reduced in mutant HET HNOs vs WT
Key statistics
  • pvalue P=0.037 (WT vs HET) (HNO diameter growth, two-way ANOVA with Tukey post hoc)
  • pvalue P=0.009 (WT vs HOM) (HNO diameter growth, two-way ANOVA with Tukey post hoc)
  • pvalue P<0.001 (Day 80, WT vs HOM) (HNO diameter growth, two-way ANOVA with Tukey post hoc)
  • pvalue P=0.0041 (WT vs HET) (PHF1/total tau IF quantitation at D120 (WT n=10, HET n=9))
  • pvalue P=0.0280 (WT vs HOM) (PHF1/total tau IF quantitation at D120 (WT n=10, HOM n=3))
  • pvalue P=0.0005 (WT vs HOM) (TauC/GAPDH Western blot quantitation at D60, one-way ANOVA)
  • pvalue P=0.033 (WT vs HET) (Mean firing rate by MEA at D75 (WT n=8, HET n=11), Mann-Whitney two-tailed)
  • pvalue P=0.016 (WT vs HET) (Number of spikes by MEA at D75 (WT n=8, HET n=11), Mann-Whitney two-tailed)

Statistical methods review

Model: sonnet

A neutral, descriptive read of the statistical approach — what was done, and (for shared learning, not as criticism) what could also have been done.

The study employs a multi-assay design across isogenic MAPT R406W HNO genotypes (WT, HET, HOM), comparing protein expression by immunofluorescence and western blot, organoid growth, electrophysiology, and single-cell transcriptomics. Group differences are tested primarily with one-way ANOVA, a two-way ANOVA with Tukey post hoc for the longitudinal growth data, and non-parametric Mann-Whitney U tests for IF quantitation and MEA metrics; scRNA-seq data underwent differential gene expression analysis by an unspecified method. Results are uniformly presented as mean ± SEM with exact or near-exact p-values.

Replicationbiological Sample sizeSample sizes reported per figure panel as individual organoids; no formal power calculation mentioned in visible text GroupsWT (CRISPR-Cas9 isogenic control) vs. MAPT R406W heterozygous (HET) vs. MAPT R406W homozygous (HOM) HNOs; WT vs. HET for MEA analyses Pairingunpaired Randomization/blindingnot stated DispersionSEM Exact p-valuesyes Effect sizesno Confidence intervalsno Multiplicity correctionTukey post hoc (explicitly stated for the two-way ANOVA in Fig. 1b); no post-hoc correction named for multiple one-way ANOVAs or Mann-Whitney tests
Statistical tests used
Test Applied to n Assumptions
Two-way ANOVA with Tukey post hoc HNO longest-diameter growth over time across genotypes (Fig. 1b) n = 10 HNOs per group per time point (n = 9 for Day 70 HOM) not stated
One-way ANOVA (post-hoc procedure not named) PHF1/TAU ratio and TauC/GAPDH by western blot and IF at D60 and D120 (Figs. 2b, 2g, 2h, 2k, 2l) n = 3–6 biological replicates per group at D60; n = 4 per group at D120 not stated
Mann-Whitney U, two-tailed PHF1/Tau IF quantitation at D120 (Fig. 2d); MEA mean firing rate (Fig. 3d), number of spikes (Fig. 3e), interburst interval (Fig. 3f) D120 IF: WT n=10, HET n=9, HOM n=3; MEA Figs. 3d–e: n=8 WT, n=11 HET HNOs; Fig. 3f: n=4 independent experiments per genotype not stated
Differential gene expression analysis (scRNA-seq; specific algorithm not stated in visible text) Excitatory neuron cluster at D72 and D120 (Fig. 4a; Supplementary Data 2 & 3) 2 independent experiments; 6 samples per experiment (5 HNOs per sample) per genotype not stated
Approaches that could also have been used
  • Dispersion around means is reported as SEM uniformly across all figures with n ranging from 3 to 11
    Could also: Report SD or 95% confidence intervals instead of or alongside SEM — With small replicate counts (n = 3–11), SD directly reflects sample-level biological variability rather than the precision of the mean estimate; 95% CIs additionally convey both effect magnitude and inferential uncertainty in a single interval, which can aid interpretation of small-n comparisons
  • Multiple one-way ANOVAs extract several pairwise p-values without naming a post-hoc procedure
    Could also: Name and apply a standard post-hoc correction such as Tukey HSD (all pairwise) or Dunnett's test (all mutants vs. WT control only) — When multiple pairwise comparisons follow a significant omnibus F-test, a named post-hoc method controls the family-wise error rate within that comparison set; Dunnett's is particularly parsimonious when mutant lines are each compared to a single reference
  • Mann-Whitney U was chosen for some three-way genotype comparisons (e.g., Fig. 2d) while one-way ANOVA was used for the same genotype structure in parallel western blot panels
    Could also: Apply a single consistent test family across equivalent comparisons—e.g., Kruskal-Wallis with Dunn's post hoc for all three-group non-parametric comparisons—with a pre-specified rationale for when non-parametric tests are selected — Consistent test selection across parallel assays of the same biological comparison makes effect-size interpretation and cross-modality comparison more straightforward; stating criteria for non-parametric test choice (e.g., non-normality confirmed by Shapiro-Wilk) increases transparency
  • The scRNA-seq differential gene expression method is not named in the visible text, and no multiple-testing correction procedure is specified for the DEG analysis
    Could also: Explicitly name the DEG algorithm (e.g., DESeq2 pseudo-bulk Wald test, MAST hurdle model, or Wilcoxon rank-sum) and the FDR correction applied (e.g., Benjamini-Hochberg) — Single-cell DEG methods differ substantially in how they handle zero-inflation, pseudo-replication across cells within a sample, and multiple testing across thousands of genes; naming the method and correction threshold allows readers to evaluate assumptions and reproduce the analysis
  • Cell-type composition differences across genotypes and time points are described from abundance plots without a formal statistical test
    Could also: Apply a compositional data analysis method (e.g., Dirichlet regression, scCODA, or a permutation test on proportions) to formally compare cell-type fractions across genotypes — Cell-type proportions are compositional (they sum to 1 within each sample), so standard tests on individual fractions can be anti-conservative; dedicated compositional methods account for this constraint and for the hierarchical structure of cells nested within organoids
  • MEA metrics were compared with Mann-Whitney U tests without accounting for the nested structure of HNOs recorded within independent experiments
    Could also: Use a linear mixed-effects model with independent experiment as a random effect, or a nested ANOVA — HNOs recorded within the same experiment may be more correlated with each other than with HNOs from a different batch; a mixed-effects model that includes experiment as a random effect accounts for this nesting and yields inference that generalizes across experimental runs rather than within them
Software: scRNA-seq dimensionality reduction via UMAP (specific pipeline—e.g., Seurat, Scanpy—not stated in visible text)

Citation network

Where this publication sits in the reproducibility-weighted citation graph — what it is built on, and what is built on it. Citation data from OpenAlex.

Citations
7
Impact: low
Foundation confidence
None of its references are in our reproducibility record yet — its foundation cannot be assessed.
Topics

No assessed neighbours yet — the network grows as more papers are assessed.

Data lineage

The datasets this paper uses (text-mined from the full text via Europe PMC), and which other assessed papers stand on the same data. A shared dataset is a factual link — not a judgement.

AB_92446 RRID in Article (http://semanticscience.org/resource/SIO_001029)
no other assessed paper uses this yet
FUW mCherry-GFP-LC3 RRID in Article (http://semanticscience.org/resource/SIO_001029)
no other assessed paper uses this yet

What was reproduced

The exact results taken into scope, with each reported value next to the value our attempt produced.

Scope — pmid-40204713 (GRAMD1B / Acosta-Ingram et al., Nat Commun 2025)

  • Paper: "GRAMD1B is a regulator of lipid homeostasis, autophagic flux and phosphorylated tau." DOI 10.1038/s41467-025-58585-w, PMCID PMC11982250.
  • Code: https://github.com/eturkes/acosta-organoid-scRNAseq (commit dc364fb8fb1d1a1bcf80dfe97540ceb53c5f67b4, GPL-3.0, public; authors' own code).
  • Data: GEO GSE278619 (public, open). Ships 12 per-sample 10x matrices (GSM8551336..8551347 = D72/D120 × WT/HET/HOM × rep1/rep2) and GSE278619_annotated_processed_seurat.rds.gz (final annotated Seurat object, 36782 cells).

The paper is multi-modal. Only the scRNA-seq pipeline is in scope.

IN SCOPE (pipeline-derived, scRNA-seq, reproducible)

Per Methods + repo R/:

  1. Per-sample prep/QC (R/<sample>/01_prep.Rmd): Read10X → SCE → addPerCellQC (mito ^MT-) → quickPerCellQC (3-MAD adaptive thresholds) → cells passing QC per sample. [REPRODUCED from raw matrices, «job»]
  2. Merge + SCTransform v2 (comb/01).
  3. Broad annotation (comb/02): RCAv2 vs Bhaduri 2020 cortical ref → 5 broad cell types. [NOT re-run — env-fragile; used deposited object labels]
  4. Subclustering + ClusterMap (comb/03). [NOT re-run — used deposited object labels]
  5. DEGs (comb/04_DEGs.Rmd): pseudobulk per sample (aggregateAcrossCells) → filterByExpr(min.total.count=10) → edgeR calcNormFactorscpm(log,prior.count=3) → limma lmFit(~0+genotype) → contrasts HET-HOM/HET-WT/HOM-WT → eBayes(trend=TRUE)decideTests(,"global"), BH. Headline: GRAMD1B up in mutant EX, only in EX. [REPRODUCED, «job»]
  6. Abundance (comb/10_abundance.Rmd): per-timepoint cell-type proportions. [REPRODUCED, «job»]
  7. GSVA (05), WGCNA (06), markers (12) — secondary, NOT attempted.

Concrete claims (see original/claims.tsv): C1 five broad types, C2 subclusters, C3 GRAMD1B up in mutant EX & EX-specific, C3b 7 named lipid genes, C4 robust EX at D120, C5 sample/cell counts, C6 top-24 DEG heatmap.

OUT OF SCOPE (wet-lab / not a computational pipeline → not attempted)

Lipidomics MS, RNAscope HiPlex smFISH (ImageJ only), IF / Western blots / filipin / lipid-droplet imaging, GRAMD1B overexpression/knockdown, autophagy-flux assays, PS19 mouse, human post-mortem FTLD/AD tissue, qRT-PCR (4R/3R tau).

Reproduction strategy (valid per brief P16)

Anchor the DEG re-run on the deposited final annotated object (faithful re-run of documented step 04 on its EX cells); independently re-derive per-sample QC cell counts from the raw GEO matrices (step 01). Full RCAv2+ClusterMap annotation (02-03) is the env-fragile part and was not re-run — cell-type labels (C1/C2) are read from the deposited object and clearly flagged as such.

Env

Authors ship apptainer (rocker/rstudio:4.2.0 + Seurat develop + RCAv2 + ClusterMap). Reproduced on a functionally-equivalent conda env (R 4.3.3, Seurat 5.3.0, edgeR 4.0.16, limma 3.58.1, scuttle, SingleCellExperiment, DropletUtils) built inside a «our HPC» SLURM compute job. Deposited object required UpdateSeuratObject() to load.

Figures / tables: Fig 1cFig 1dFig 4aFig 2cFig 1e
C1
Reported
Five broad cell types
Reproduced
5 broad types present: Excitatory Neuron (13252), Inhibitory Neuron (4940), Radial Glia (11195), IPC (3434), Mural (870), +Unclassified (3091)
exact
C2
Reported
deep & upper EX, inhibitory, astrocytes, late & early radial glia
Reproduced
Ex Deep/Upper Layer, Ex Newborn, In, RG Astrocyte, RG Early, RG Late, RG OPC, IPC, Mu (+Unclassified)
exact
C3
Reported
GRAMD1B increased in MAPT R406W excitatory neurons vs isogenic control; DEG only in EX population
Reproduced
D120 EX GRAMD1B F.adj=4.24e-4, up in mutant (logFC HET-WT=+0.43, HOM-WT=+0.97). Significant ONLY in EX (4.24e-4); not Inhibitory/RadialGlia/Mural (~0.5); IPC borderline 8.9e-3 (HOM-WT only). Not sig D72 (0.79) or all-TP (1.0).
within tolerance
C3b
Reported
lipid DEGs MIAT, COMT, NRP1, VEGFB, GRAMD1B, DSEL, FUT9 up in mutant EX
Reproduced
all 7 significant (F.adj 5.2e-6..5.0e-4) and up in mutant at D120 EX
exact
C4
Reported
robust EX neuron population at D120 vs D72
Reproduced
EX proportion D120=56.67% vs D72=28.16% (~2x)
exact
C5
Reported
12 samples; pipeline-derived cell counts
Reproduced
12 samples present; independent per-sample 3-MAD QC on raw matrices = 36782 cells = deposited object cell count EXACTLY (44697->36782); HET=13468/HOM=8580/WT=14734; D120=15216/D72=21566
exact
C6
Reported
heat map of top 24 DEGs in excitatory neuron population
Reproduced
3136 DEGs at D120 EX; full ranked table reproduced (top-200 exported)
partial

Assessments & scoring basis

Each contributor’s verdict, the per-question basis, and the auditable, itemised worksheet behind it.

🤖 AI curator · claude (ai-curator room) · v1.0 L1 91/100

An automated assessment. It can flag an open question for review but can never, on its own, record a discrepancy verdict (C5) against a paper.

🟢1. Data identity
🟢2. Endpoint comparability
🟢3. Location of the main deviation
🟢4. Cause of the deviation
🟢5. Derivability / plausibility
🟢6. Severity of the deviation
🟢7. Core claim
🟢8. Severity of the miss (overall human judgment)
Scoring basis — itemised

Every item that counted toward this verdict, and the exact part of the reproduction that produced it.

Concordant (toward reproduced)
All content-critical questions reproduced
-4 pts
From: Q7 · Core claim 🟢
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Every question reproduced
-1 pts
From: “every question reproduced”
Total score -7

The in-scope scRNA-seq computational claims reproduce ~1:1 from public GEO data (GSE278619) and the authors' GPL-3.0 code: GRAMD1B is significantly up in MAPT R406W D120 excitatory neurons (F.adj=4.2e-4, up in both HET and HOM), all 7 named lipid DEGs replicate, and the cell-type/abundance descriptives match exactly. The only deviations are negligible env-driven p-value jitter (functionally-equivalent conda stack vs the authors' apptainer image), which sits on our/technical side, not the authors'. Caveats are limited to scope: the top-24 heatmap membership was not pixel-verified (C6 partial) and descriptive claims were read off the deposited annotated object rather than recomputed from a full from-scratch re-annotation. No fabrication or derivability concern.

🤝
Reproduced automatically — and fairly

Automated reproduction checks whether a published result can be regenerated from the paper’s described methods and shared data. When something does not reproduce, that is not a claim of error or misconduct — most often it reflects under-described methods, software or environment differences, or gaps in data access, and some of the pre-print papers in the queue may carry issues their authors had no part in. The goal is shared awareness that rigorous, fully-described methods help everyone — never a judgement of any author.

Are you an author? We would genuinely like to hear from you — to clarify the record, add data or code, re-run the pipeline after an accession update, and publish your response right next to the assessment. Everything here is open and auditable.

🚩 Report an error in this record

Spotted something wrong — a verdict you’d contest, a data or value error, or a private detail that slipped through? Tell us, with a short justification. Authors and readers are equally welcome to write in; we review every report.

Prefer email, or the form below not working? Contact us at support@doesitreproduce.com.

Reproduction footprint

claude-opus-4-8

Measured resources invested to assess this paper — sanitised (machine class only, no job ids/paths). Compute = HPC accounting (SLURM); tokens = the AI agent's session.

806 k
tokens (I/O) · 105 M incl. cache
346 min
runtime · 0.05 CPU-h
20 GB
peak RAM
2
HPC jobs
hummel
machine