Corpus 1,286 assessed · 1,187 scored · 648 reproduced ≥75 · 174 flagged ·∅ 73.9/100
← New search

A cross-species approach to identify transcriptional regulators exemplified for Dnajc22 and Hnf4a.

Sci Rep · 2017
L1 76/100 PQI 92
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.

Supporting (toward a concern)
Minor / cosmetic deviation
+1 pts
From: Q3 · Location of the main deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q2 · Endpoint comparability 🟡
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
Total score -4
✓ What held up
  • Same input data as the authors
  • 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)
  • 🟡Reported values were only indirectly comparable
  • 🟡A deviation arose in the data or preprocessing
How its reproducibility compares
76/100
Reproducibility score
at the mean
vs. all fields · 1187 studies
🎯 Scores higher than 48% of all assessed papers rank 589 of 1187 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

REPRODUCED (described well enough; honest 1:1 on the authors' own code + data). The paper's computational core is co-expression of a gene-of-interest with candidate TFs across tissues; its headline computational claim is that Hnf4a is strongly co-expressed with (a top correlate of) Dnajc22 across species. Re-running the authors' own two R pipelines (repo schultzelab/Dnajc22-regulation @ 72f3d14) on the paper's own data on «our HPC» reproduces this 1:1 in BOTH species: mouse Affymetrix GSE10246 (103 CEL -> RMA -> tissue-mean -> cor) gives r(Dnajc22,Hnf4a)=0.95 with Hnf4a the #1 correlate of all 18,197 genes; human RNA-seq GSE16256/Roadmap (DESeq2 -> tissue-mean -> cor, counts shipped in the repo) gives r(DNAJC22,HNF4A)=0.99 with HNF4A the #1 correlate of all 16,140 genes. The paper says 'one of the highest correlations' / 'strong relationship'; our reproduction shows Hnf4a is in fact THE single highest correlate in both species (if anything the paper understates). A cross-check (DESeq2 re-normalisation from the shipped raw counts reproduces the shipped normalised table's correlation exactly) shows no fabrication in that artifact. NOT attempted (out of scope / 80-20): the SOM 225-gene cluster (shipped SOM code has placeholder bugs: som.input2 undefined, genename='Creld1', hard-coded cluster indices; seed/scale sensitive -> not faithfully runnable as shipped); pcaGoPromoter/iRegulon TF prediction (iRegulon is a Cytoscape GUI; pcaGoPromoter PRIMO depends on the unreproducible SOM membership); WGCNA/BioLayout network; all wet-lab (ChIP-qPCR, luciferase) and fly/zebrafish validation; the exact printed heatmap r cells (inside Supplementary Figure images) were not OCR-compared. «our HPC» notes: bioconda Bioconductor data/annotation packages (genomeinfodbdata, mouse4302.db/cdf, org.Mm.eg.db) have post-link downloads unreachable from compute nodes -> installed source tarballs from the tu-dortmund Bioc mirror; preprocessCore rebuilt --disable-threading to clear the affy::rma pthread_create=22 cgroup bug. These grades are PROVISIONAL; a human auditor decides ground truth.

💻 Code ↗ 🗄 Data: GSE10246

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 76
    assessed: 2026-06-16 ⛓ 1bb4f7947e44
✎ 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
2026-06-16
Rubric version
v1.0
Assessed by
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-16
no human curator yet
Last updated
2026-09-19

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

Transcription factors (TFs) and the genes they regulate are co-expressed across tissues and species, and this principle can be exploited computationally to identify the transcriptional regulator of a gene of interest; the paper tests whether this approach correctly identifies Hnf4a as the TF regulating the uncharacterized gene Dnajc22.

Core claims
  • Hnf4a is a major transcriptional regulator of Dnajc22 finding
  • A combined co-expression (SOM clustering) and TF binding site prediction approach can identify candidate regulatory TFs for a gene of interest method
  • Hnf4a/HNF4A directly binds the Dnajc22/DNAJC22 locus in rat, mouse, and human finding
  • The H4 motif (+50 bp) is the functionally relevant Hnf4a binding site in the murine Dnajc22 promoter finding
  • Hnf4a-mediated regulation of the Dnajc22 ortholog is conserved in Drosophila, zebrafish, and human finding
  • An independent network approach (WGCNA) also predicts Hnf4a as a regulator of Dnajc22, corroborating the SOM-based approach method
  • Dnajc22 is the vertebrate ortholog of Drosophila wurst, previously shown essential for tracheal development resource
  • The described cross-species in silico plus wet-lab workflow is a generalizable strategy for identifying TF-GOI regulatory relationships method
Experimental setups
Assay System Perturbation Readout Platform
co-expression clustering (self-organizing map, hierarchical clustering) murine tissue atlas transcriptomes (22 organs, 14 tissues, 16 cell types) none TF-GOI co-expression cluster membership GSE10246
co-expression clustering (self-organizing map) human Roadmap Epigenomics transcriptomes (6 organs, 5 tissues, 2 cell types) none TF-GOI co-expression cluster membership GSE16256
transcription factor binding site prediction in silico murine Dnajc22 promoter locus none predicted HNF4 binding sites/motifs TFBIND, PROMO, MATCH, iRegulon, pcaGoPromoter
reanalysis of published microarray/transcriptome datasets from Hnf4a loss- and gain-of-function experiments mouse liver-specific Hnf4a KO; human hepatocellular carcinoma HNF4A siRNA knockdown; HEK293 HNF4A overexpression; HCT116 HNF4A overexpression; rat INS-1 cells HNF4A/HNF1B/HNF6 expression; human renal carcinoma tissue Hnf4a knockout/knockdown/overexpression Dnajc22/DNAJC22 transcript levels GEO microarray datasets
ChIP-seq reanalysis rat kidney, mouse intestine, human HCT116 colon carcinoma cells none Hnf4a binding peaks at Dnajc22 locus MACS2 peak calling
ChIP-qPCR murine kidney cortex none Hnf4a binding enrichment at Dnajc22 promoter fragment (vs Apoc3 positive and Hprt1 negative controls)
luciferase reporter assay murine M-1 kidney cell line heterologous HNF4A overexpression; site-directed mutagenesis of predicted binding sites (ΔH1-H4) luciferase activity
quantitative real-time PCR Drosophila melanogaster Hnf4 mutant larvae; zebrafish embryos injected with human HNF4A; HEK293 cells overexpressing HNF4A Hnf4a loss-of-function (fly); HNF4A gain-of-function (zebrafish, HEK293) wurst/dnajc22/DNAJC22 transcript levels
Key results
  • Dnajc22 shows predominant expression in liver, kidneys, and intestine
  • Liver-specific Hnf4a knockout reduces Dnajc22 transcript to background levels
  • HNF4A knockdown by siRNA decreases DNAJC22 levels; HNF4A overexpression in HEK293 and HCT116 cells induces DNAJC22 expression
  • Hnf4a ChIP-seq/ChIP-qPCR shows binding peaks near the Dnajc22 transcription start site in rat, mouse, and human, confirmed by ChIP-qPCR in murine kidney cortex
  • Heterologous HNF4A expression increases luciferase activity of the Dnajc22 promoter construct relative to control 10.45-fold, P = 0.02, n = 6
  • Mutation of the H4 binding site abolishes HNF4A-induced luciferase activity, reducing it to background levels; mutation of H1-H3 does not P = 0.0003, n = 4
  • H4 motif shows the highest cross-species conservation among the four predicted binding sites
  • Hnf4a-Dnajc22 co-regulation is confirmed in Drosophila (wurst), zebrafish, and human HEK293 cells
Key statistics
  • fold_change 10.45 fold (luciferase activity increase with WT Dnajc22 promoter construct after HNF4A expression)
  • pvalue P = 0.02 (significance of luciferase increase for WT Dnajc22 promoter construct)
  • pvalue P = 0.0003 (significance of luciferase activity reduction for ΔH4 mutant construct)
  • count 225 genes (number of genes in the Dnajc22-containing SOM cluster)
  • count 19 genes (genes in Dnajc22-containing subcluster after hierarchical clustering, 5 of which were TFs)
  • count 104 transcriptomes (mouse gene atlas transcriptomes analyzed (of 182 total in GSE10246))
  • count 25 transcriptomes (human Roadmap Epigenomics transcriptomes analyzed from GSE16256)
  • other 49 species (number of species with single orthologs of wurst/Dnajc22 per Ensembl)

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 paper combines computational co-expression clustering (self-organizing maps, WGCNA, Pearson correlation across tissue transcriptome atlases) with experimental wet-lab validation to identify Hnf4a as a transcriptional regulator of Dnajc22. Significance testing in wet-lab experiments relied on one-sided t-test variants (unpaired, one-sample, and paired) applied separately to each dataset or assay. Results are reported with selective exact p-values and one fold-change estimate; no correction for multiple comparisons across the many independent tests is described.

Replicationmixed Sample sizeSample sizes stated per figure caption; range from n=2 to n=6 for wet-lab experiments; 104 murine and 25 human transcriptomes for computational analyses; no formal power calculation described GroupsHnf4a/HNF4A perturbation (knockout, siRNA knockdown, or overexpression) versus matched controls, across six independent GEO dataset re-analyses and three cross-species qPCR validation experiments Pairingmixed Randomization/blindingnot stated Dispersionnone Exact p-valuesyes Effect sizesyes Confidence intervalsno Multiplicity correctionnone stated
Statistical tests used
Test Applied to n Assumptions
one-sided t-test (unpaired) Re-analysis of six GEO transcriptome datasets for Hnf4a perturbation effects on Dnajc22 expression (Fig. 2a–f) varies by dataset: n=3, 4, 2, 3, 2, 4 not stated
one-sided one-sample t-test ChIP-qPCR enrichment of Hnf4a at the Dnajc22 locus in murine kidney cortex (Fig. 3d) not stated
one-sided paired t-test Luciferase reporter assays in M-1 cells with heterologous HNF4A expression, wild-type and single-site mutant constructs (Fig. 4b, 4d) n=6 (Fig. 4b); n=4 (Fig. 4d) not stated
one-sided paired t-test qPCR validation of Dnajc22 co-regulation with Hnf4a in Drosophila, zebrafish, and HEK293 cells (Fig. 5a–c) three independent experiments per panel not stated
Pearson's correlation coefficient Co-expression matrix of Dnajc22-related family members and predicted TFs across murine tissues (Supplementary Fig. 2b) 104 murine transcriptomes not stated
Approaches that could also have been used
  • All inferential tests throughout the wet-lab validation were one-sided (one-sided t-test, one-sided one-sample t-test, one-sided paired t-test)
    Could also: Two-sided versions of the same t-tests could also have been applied — Two-sided tests do not require a prior directional hypothesis to be specified; they are the conventional default in many reporting guidelines and remain valid even when the direction of an effect is anticipated, while guarding against undocumented directional assumptions
  • Multiple separate significance tests were performed across six independent GEO re-analysis comparisons and three cross-species qPCR experiments with no stated multiplicity adjustment
    Could also: A Benjamini-Hochberg FDR correction or Bonferroni adjustment could also have been applied across this family of tests — Adjusting for the number of comparisons controls the expected rate of false discoveries; this is commonly described when many tests are performed within a single study, and its absence or explicit justification is often requested by reviewers
  • Some wet-lab groups had very small sample sizes (n=2), and parametric t-tests were used throughout
    Could also: Non-parametric alternatives such as the Wilcoxon signed-rank test (for paired designs) or exact permutation tests could also have been applied — At n=2 the normality assumption underlying t-tests cannot be assessed empirically; non-parametric and permutation-based approaches make no distributional assumptions, though they also have very limited power at such small n
  • Pearson's correlation coefficients were used to build the co-expression matrix across tissues
    Could also: Spearman's rank correlation could also have been used — Spearman correlation is robust to outliers and to monotone non-linear relationships and is widely used for microarray and RNA-seq data, which can be right-skewed or contain extreme values that disproportionately influence Pearson coefficients
  • No dispersion measure (SD, SEM, or CI) is reported or depicted for any of the wet-lab quantitative results
    Could also: Reporting SD or 95% CI alongside point estimates could also have been standard practice — Dispersion measures communicate data variability and estimation precision, which are especially informative at small n; many journals' statistical reporting checklists explicitly request them alongside means and p-values
  • SOM clustering was used as the primary co-expression grouping method, with WGCNA presented only as a supplementary confirmatory analysis
    Could also: WGCNA or hierarchical clustering with a defined linkage criterion could also have served as the primary framework, with SOM as a cross-validation — Different clustering algorithms make different assumptions about cluster topology and gene-module boundaries; presenting two methods with equivalent status and comparing their outputs can strengthen the case that identified co-expression relationships are robust to the choice of algorithm
Software: iRegulon · pcaGoPromoter · WGCNA · MACS2 · TFBIND · PROMO · MATCH · HNF4 Binding Site Scanner

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.

GSE1589 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE16256 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE2700 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE29084 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE3126 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE62891 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE781 GEO in Methods (http://purl.org/orb/Methods)
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-28642491

A cross-species approach to identify transcriptional regulators exemplified for Dnajc22 and Hnf4a. Aschenbrenner, Bassler, Brondolin, Bonaguro, Carrera, Klee, Ulas, Schultze, Hoch. Sci Rep 2017; 7:4365. PMCID PMC5481429. DOI 10.1038/s41598-017-04370-9.

  • Code: https://github.com/LIMES-immunogenomics/Dnajc22-regulation (redirects to schultzelab/Dnajc22-regulation, default branch master, HEAD commit 72f3d1463e7605a6ec27a843c9d1b3f9d47909c7, pushed 2017-04-11, public, not archived, no license file). Authors' OWN analysis code.
  • Data:
    • Mouse: GEO GSE10246 (Lattin/GNF mouse gene-atlas, Affymetrix Mouse430 2.0). Paper used 104 of 182 transcriptomes (22 organs, 14 tissues, 16 cell types). Sample→tissue map shipped in repo (affy_scripts/sample_info.txt, samples_*-dataset.txt).
    • Human: RNA-seq HTSeq count tables for 13 tissues (×1–2 replicates, 26 files) shipped inside the repo (RNA-Seq_scripts/expression-files/ *_counted.txt) plus the authors' DESeq-normalised table (RNA-Seq_scripts/DESeq-norm_ex_tbl.txt). No external download needed.

Nature of the paper

Methods/proof-of-concept paper: a generic in-silico workflow (co-expression by self-organizing maps + sequence-based TF-binding prediction) to nominate a transcriptional regulator, exemplified on Dnajc22 → Hnf4a, then validated with wet-lab (ChIP-qPCR, luciferase) and cross-species (fly, zebrafish, human). The computational core is two R pipelines (mouse affy + human RNA-seq) that: RMA/DESeq-normalise → background-filter → unique-gene → SOM-clusterpcaGoPromoter PRIMO TF prediction → Pearson correlation matrix of the GOI vs candidate TFs/family members.

In scope (pipeline-derived, attempted)

The repo's two R scripts are the pipeline. The cleanest pinnable, low-hanging quantitative output is the Pearson correlation between Dnajc22/DNAJC22 and Hnf4a/HNF4A across tissues — the paper's central quantitative co-expression claim ("One of the highest correlations was observed between Hnf4a and Dnajc22", Results; Suppl. Fig. 2b mouse / Suppl. Fig. 4 human). cor() of R is named explicitly in Methods ("Pearson correlation coefficient matrix").

  • DP-human (primary, self-contained — data shipped in repo): run RNASeq_pipeline.R logic headless on the shipped HTSeq counts (DESeq2 size-factor normalise → floor → ≥200 background filter → tissue-mean → log10 → cor). Report r(DNAJC22, HNF4A) across human tissues and HNF4A's rank among DNAJC22's correlations. Cross-check against shipped DESeq-norm_ex_tbl.txt. Maps to Suppl. Fig. 4.
  • DP-mouse (primary): download GSE10246 CEL for the 104 paper samples on a «our HPC» compute node → affy RMA → background filter (>6) → unique-gene → tissue-mean → cor. Report r(Dnajc22, Hnf4a) and Hnf4a's rank among Dnajc22's correlations within the 16-gene TF/family panel of the script. Maps to Fig. 1 / Suppl. Fig. 2b.
  • DP-tf (secondary, attempt-if-time): pcaGoPromoter PRIMO on the Dnajc22-associated gene set → is Hnf4a among the top predicted TFs (paper Fig. 1d: pcaGoPromoter → Hnf4a, Hnf1b, Nr1h2, Hnf1a)?

Out of scope (not attempted, why)

  • SOM subcluster size = 225 genes (mouse): the shipped SOM block has placeholder bugs (som.input2 is undefined; genename = "Creld1" instead of Dnajc22; hard-coded cluster indices \b3\b,\b7\b). Result is highly sensitive to seed/scaling/training-subset and the exact (unshipped) variable state. The hard ~20% → documented, not chased.
  • iRegulon predictions (Hnf4a, Hnf1b, Gata1, Cdx2): iRegulon is a Cytoscape GUI plugin, not in the repo, not scriptable headless → external.
  • WGCNA / BioLayout Express 3D network (464 nodes, 5,834 edges): BioLayout is a GUI tool; not shipped → external.
  • TFBIND / PROMO / MATCH binding-site screens; ChIP-qPCR, luciferase, fly/zebrafish wet-lab validation → manual/we
Figures / tables: Fig. 2bFig. 4aFig. 1dTable
corr_mouse_Dnajc22_Hnf4a
Reported
One of the highest correlations was observed between Hnf4a and Dnajc22 (murine GSE10246 atlas; Suppl. Fig. 2b; r printed only inside the heatmap)
Reproduced
Pearson r(Dnajc22,Hnf4a)=0.95 (panel) / 0.949 (genome-wide); Hnf4a is the #1 most-correlated gene with Dnajc22 of all 18,197 background-filtered genes. 103 GSE10246 CEL, affy RMA, 52 tissues.
within tolerance
corr_human_DNAJC22_HNF4A
Reported
strong transcriptional relationship between DNAJC22 and HNF4A; same SOM cluster (human GSE16256/Roadmap; Suppl. Fig. 4)
Reproduced
Pearson r(DNAJC22,HNF4A)=0.99 (panel) / 0.985 (genome-wide); HNF4A is the #1 correlate of DNAJC22 of all 16,140 genes. Shipped DESeq-norm table AND independent DESeq2 re-normalisation from shipped counts agree exactly.
within tolerance
gse10246_samples_used
Reported
104 of 182 GSE10246 transcriptomes; 22 organs + 14 tissues + 16 cell types
Reproduced
103 CEL (the pipeline's samples_Fig1-dataset.txt has 103 rows) -> 52 tissue groups (=22+14+16)
within tolerance
mouse_SOM_cluster_size
Reported
SOM cluster containing Dnajc22 = 225 genes (Suppl. Table 1)
Reproduced
NOT ATTEMPTED (out of scope: shipped SOM code has placeholder bugs and is seed/scale sensitive)
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 76/100

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.

Supporting (toward a concern)
Minor / cosmetic deviation
+1 pts
From: Q3 · Location of the main deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q2 · Endpoint comparability 🟡
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
Total score -4

The paper's central computational claim — a strong cross-species co-expression between Dnajc22/DNAJC22 and Hnf4a/HNF4A — reproduces 1:1 on the authors' own pipelines and public/shipped data (mouse r=0.95, human r=0.99, with Hnf4a/HNF4A the single top correlate genome-wide in both species). The only deviations are on the periphery and are minor or on the authors' side: the operative r is never printed numerically (heatmap-only, hence an indirect endpoint, q2 yellow), a trivial 103-vs-104 sample off-by-one (q3 yellow), and a buggy shipped SOM code block that leaves the secondary 225-gene cluster / pcaGoPromoter claims unverified. A DESeq2 re-normalisation from the shipped raw counts matches the shipped normalised table, so there is no fabrication signal; severity is negligible and the core conclusion holds fully.

🤝
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 [email protected].

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.

442.2 k
tokens (I/O) · 69.1 M incl. cache
70 min
runtime · 0.09 CPU-h
2 GB
peak RAM
2
HPC jobs
hummel
machine