{
 "dataset": {
  "archive": "dandi",
  "id": "000008",
  "label": "DANDI:000008"
 },
 "page": "datasets/dandi/000008/",
 "generated": "2026-10-22T21:23:47+00:00",
 "caution": "A failure to reproduce is not a finding that the paper is wrong. It may be a failure of the attempt.",
 "outcome_counts": {
  "reproduced": 11,
  "partially_reproduced": 4,
  "not_reproduced": 1,
  "blocked": 0
 },
 "outcome_counts_with_outside_data": {
  "reproduced": 9,
  "partially_reproduced": 0,
  "not_reproduced": 0,
  "blocked": 2
 },
 "outcome_counts_elsewhere": {
  "reproduced": 3,
  "partially_reproduced": 0,
  "not_reproduced": 0,
  "blocked": 3
 },
 "outcome_counts_links": {
  "reproduced": 2,
  "partially_reproduced": 0,
  "not_reproduced": 0,
  "blocked": 0
 },
 "attempts": [
  {
   "id": "000008-scala-2021-01",
   "dataset": {
    "archive": "dandi",
    "id": "000008",
    "version": "0.211014.0809",
    "doi": "10.48324/dandi.000008/0.211014.0809"
   },
   "key": "dandi:000008",
   "label": "DANDI:000008",
   "dataset_version": "0.211014.0809",
   "paper": "10.1038/s41586-020-2907-3",
   "paper_title": "Phenotypic variation of transcriptomic cell types in mouse motor cortex",
   "date": "2026-10-20",
   "kind": "agent",
   "accountable": "CatalystNeuro",
   "agent": {
    "harness": "Claude Code (GitHub Actions)",
    "model": "claude-opus-5-5",
    "protocol": "pilot-10"
   },
   "runner": null,
   "tier": "agent_verified",
   "deprecated": "",
   "deprecated_on": null,
   "entry": "000008-scala-2021",
   "entry_title": "Phenotypic variation of transcriptomic cell types in mouse motor cortex (Scala, Kobak et al. 2021)",
   "description": "Tests the findings of Scala, Kobak et al. (2021) on dandiset 000008, which holds the current-clamp recordings of 1,328 Patch-seq neurons of mouse primary motor cortex with each cell's t-type, family, layer and read counts. The paper's own feature extraction is run on every file; the cohort, the stimulus protocol, the features, the kNN family classification, the IT and ET differences, the t-type coverage and depths, and the variability of t-types are tested on the dandiset, and the claims that also need the paper's gene counts, morphology or transcriptomic reference data are tested with those inputs in separate notebooks.",
   "notebooks": [
    {
     "path": "scala_2021_ephys.ipynb",
     "title": "Recordings and metadata of dandiset 000008 (cohort, features, families, t-types)",
     "executed": "notebooks/000008-scala-2021/scala_2021_ephys.html"
    },
    {
     "path": "scala_2021_transcriptomics.ipynb",
     "title": "Dandiset 000008 with the paper's gene counts and 10x reference (RRR, within-family distances)",
     "executed": "notebooks/000008-scala-2021/scala_2021_transcriptomics.html"
    },
    {
     "path": "scala_2021_sequencing.ipynb",
     "title": "Sequencing claims on the paper's count tables and reference data",
     "executed": "notebooks/000008-scala-2021/scala_2021_sequencing.html"
    }
   ],
   "outcomes": [
    {
     "claim": "link-metadata-table",
     "statement": "The dandiset's cells can be matched by cell id to the paper's per-cell metadata table, and the matched records agree on what both describe.",
     "source": "Data availability; the code's data/m1_patchseq_meta_data.csv (mini-atlas at commit 3bf98a524b8e48aa8647e1b9ae8f2351d9c712c9)",
     "quote": "",
     "statistics": [],
     "kind": "link",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "All 1,328 cells of the dandiset match a row of the paper's table by cell id; the table's one other cell, 20190211_sample_10 (L6 CT Gpr139), has no file. For all 1,328 cells the records agree on mouse, slice, genotype, sex, Cre label, targeted and inferred layer, exon, intron and intergenic reads, batch, genes detected, t-type confidence, the ALM/VISp mapping and age, and on t-type and family with the files' PT read as the paper's ET (1,279 agree in the files' spelling). The files' cortical_thickness_4x equals the table's soma depth in 4x units and cortical_thickness_um its cortical thickness in 4x units for all 1,328 cells; the \u00b5m columns are not in the files. The five cells with a t-type that the table excludes are all in the dandiset and are not marked. Numbers are compared exactly, as floats without rounding.",
     "deviations": [],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    },
    {
     "claim": "link-ephys-features",
     "statement": "The electrophysiological features extracted from the dandiset's files with the paper's code match the paper's published per-cell feature table.",
     "source": "Methods, Extraction of electrophysiological features; the code's data/m1_patchseq_ephys_features.csv and preprocess-ephys-files.ipynb (mini-atlas at commit 3bf98a524b8e48aa8647e1b9ae8f2351d9c712c9)",
     "quote": "",
     "statistics": [],
     "kind": "link",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "Every file matches a published row; 1,320 of 1,328 cells (99.4%) agree to a relative 1e-4 on all 28 features other than the rheobase. Three cells (20190128_sample_3, 20180215_sample_1, 20190222_sample_5) differ in 22 to 25 features, so their files hold other traces than those the published features were computed from; four differ in 5 or 6 features of the highest-firing trace (Fano factors, CVs, spike frequency adaptation, wildness, or the spike count), and one in the time constant only. The rheobase differs for 41 cells (-40 to 20 pA); it is fitted by RANSAC with no seed in the paper's code, and for 4 of the first 8 such cells one of seeds 0 to 9 gives the published value.",
     "deviations": [
      "The rheobase is computed with the global numpy seed set to 0 before each cell; the paper's code sets no seed."
     ],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    },
    {
     "claim": "cohort-cells-mice",
     "statement": "1,329 cells from 266 mice passed the initial quality control and were sequenced; the paper's recordings come from these cells.",
     "source": "Results, Patch-seq of mouse primary motor cortex, paragraph 1; Methods, Animals",
     "quote": "In total, we performed whole-cell recordings from more than 2,000 cells, of which 1,329 cells (from 266 mice) passed initial quality control.",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "The dandiset has files for 1,328 distinct cells from 266 mice, which matches reading (b) and the mouse count. Three pairs of files hold identical voltage matrices, so the files hold 1,325 distinct recordings, but the six cells of these pairs are distinct cells (different read counts, genes detected and depths, four of the six on different slices). The criterion counts distinct recorded cells, so the duplicated recordings are a fault of the deposit, recorded as a finding, not missing cells. Changed from partially_reproduced after verification round 1, which counted the cells this way.",
     "deviations": [],
     "disagreed": true,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "cohort-ages",
     "statement": "The mice were 35 to 245 days old, with a median of 75 days and an interquartile range of 64 to 100 days.",
     "source": "Methods, Animals; Extended Data Figure 2a",
     "quote": "Experiments on adult male and female mice ( n = 266; median age 75 days, interquartile range 64\u2013100, full range 35\u2013245 days, Extended Data Fig. 2a )",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "One age per mouse (no mouse has two). Over 266 mice the median is 75 days, the quartiles 64 and 100 days, the range 35 to 245 days, under the linear, lower, higher and nearest quantile rules alike.",
     "deviations": [],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "cohort-genotypes",
     "statement": "The 266 mice were wild-type C57Bl/6 (27), Viaat-Cre/Ai9 (24), Sst-Cre/Ai9 (75), Vip-Cre/Ai9 (46), Pvalb-Cre/Ai9 (76), Npy-Cre/Ai9 (2), Vipr2-Cre/Ai9 (7), Slc17a8-Cre/Ai9 (6), Gnb4-Cre/Ai9 (1) and Slc17a8-iCre/Ai9 (2).",
     "source": "Methods, Animals",
     "quote": "were performed on wild-type C57Bl/6 ( n = 27), Viaat -Cre/Ai9 (vesicular inhibitory amino acid transporter, encoded by the Slc32a1 gene, n = 24), Sst -Cre/Ai9 (somatostatin, n = 75), Vip -Cre/Ai9 (vasoactive intestinal polypeptide, n = 46), Pvalb -Cre/Ai9 (parvalbumin, n = 76), Npy -Cre/Ai9 (neuropeptide Y, n = 2), Vipr2 -Cre/Ai9 (vasoactive intestinal peptide receptor 2, n = 7), Scl17a8 -Cre/Ai9 (VGLUT3, vesicular glutamate transporter 3, n = 6), Gnb4 -Cre/Ai9 ( n = 1), and Slc17a8 -iCre/Ai9 ( n = 2) mice",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "partially_reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "Total 266 mice, one genotype each. Nine of ten lines match. Wild type is 26 against 27, and one mouse is Scnn1a-Tg3-Cre/wt; Ai9/wt (Cre label SCNN1), a line the paper does not list; the paper's code counts SCNN1 with the wild type (ttype-assignment.ipynb, not run here). The two Slc17a8-iCre mice are spelled Vglut3-Cre in the genotype and SLC17a8-iCre in the Cre field.",
     "deviations": [],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "stimulus-protocol",
     "statement": "Each neuron was stimulated with 600 ms current pulses from -200 pA upwards in 20 pA steps up to 1,380 pA (sometimes stopped earlier), with 1.3 or 1.4 s between pulses, and for most neurons the series was repeated several times.",
     "source": "Methods, Patch-seq recording procedures, paragraph 4",
     "quote": "Each neuron was injected with 600-ms-long current pulses starting from \u2212200 pA and up to 1,380 pA with 20-pA increment steps (in some cases stimulation was stopped before reaching 1,380 pA). There were 1.3- or 1.4-s intervals between successive current pulses, depending on the used setup. For most neurons, the stimulation was then repeated multiple times from the beginning.",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "partially_reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "Every step lasts 600 ms (samples 2,500 to 17,500 at 25 kHz) and every series rises in 20 pA steps. 1,183 cells (89.1%) start at -200 pA, 90 at -100 pA and the rest between -320 and -80 pA; 112 cells go above 1,380 pA (87 to 1,480, one to 1,780). 89.0% conform on the steps, between the criterion's 80% and 95%. The departures are mostly one shift of the whole series, as 87 of the 90 cells that start at -100 pA end at 1,480 pA. The intervals between pulses cannot be tested, as no series has a start time, and no file holds a second series (0 cells), so the repeated stimulation is not in the dandiset.",
     "deviations": [
      "The interval part of the criterion could not be applied; conformance is judged on the steps alone."
     ],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "soma-depth-count",
     "statement": "Soma depth was measured for 1,284 of the 1,329 neurons; 45 neurons were recorded with too little biocytin, and the depth could be measured for only 2 of them.",
     "source": "Methods, Cortical thickness normalization and layer assignment",
     "quote": "We obtained soma depth information for 1,284 neurons out of 1,329 (45 neurons were mistakenly recorded using a solution with insufficient biocytin concentration, and we could measure soma depths for only 2 of those; for 2 other neurons the measurements could not be carried out because the slices were lost).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "1,283 cells have a soma depth, in the field cortical_thickness_4x (4x image units), which the data check shows to be a depth (its ratio to cortical_thickness_um lies between 0.031 and 0.977 and gives the files' inferred layer for 1,264 of 1,283 cells) and the link shows to be the table's 4x soma depth. The field soma_depth_um is present in 43 files and empty in all. The cell missing from the dandiset has a depth, so 1,283 is the count the criterion accepts (the table has 1,284).",
     "deviations": [
      "The depth is read from a field whose name says cortical thickness; the identification rests on the values and on the link to the paper's table."
     ],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed4-example-cell",
     "statement": "The features of the example cell of Extended Data Fig. 4 are an input resistance of 274.80 MOhm, a membrane time constant of 21.95 ms, ISI adaptation index 1.26, ISI average adaptation index 1.15, AP amplitude adaptation index 0.91 and its average 0.99, AP threshold -40.18 mV, AP amplitude 81.17 mV, AP width 0.80 ms, AHP -12.60 mV, ADP 0 mV, UDR 1.62, latency 69.28 ms, rheobase 20.44 pA, 32 APs in the highest firing trace, ISI CV 0.27, ISI Fano factor 0.0014, AP CV 0.17, AP Fano factor 1.32, sag ratio 1.17, sag time 0.26, sag area 31.16 mV ms and rebound 17.84 mV.",
     "source": "Extended Data Figure 4a-f legend",
     "quote": "Hyperpolarizing currents were used to compute the input resistance (274.80 MOhm) and membrane time constant tau (21.95 ms). ... It was used to compute AP threshold (\u221240.18 mV), AP amplitude (81.17 mV), AP width (0.80 ms), AHP (\u221212.60 mV), ADP (0 mV), UDR (1.62) and latency of the first spike (69.28 ms). d , Regression line gives the rheobase estimate (20.44 pA).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "All 23 values within tolerance, each equal to the paper's own feature table. The AP threshold is -43.18 mV as in that table, against -40.18 mV in the legend, inside the 10% tolerance.",
     "deviations": [],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ephys-feature-count",
     "statement": "All 17 features used for statistics could be computed for 1,320 cells, of which 1,219 had a t-type assignment.",
     "source": "Figure 3a legend; Methods, t-SNE visualization of the morpho-electric phenotypes",
     "quote": "n = 1,320 cells used to construct the embedding, 1,219 cells with t-type labels shown.",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "Counting each recording once, 1,317 cells have all 17 features and 1,216 of them a t-type and no exclusion; with all cells, as the paper, exactly 1,320 and 1,219. Without the paper's exclusions, which the files do not mark, the second count is 1,221 (1,224 with all cells).",
     "deviations": [
      "The three cells whose recording duplicates another cell's are left out (each recording counted once); the counts with all cells are given.",
      "The five excluded cells are taken from the paper's table (link-metadata-table)."
     ],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig2-family-features",
     "statement": "Pvalb neurons are characterized by high firing rates, while Sst neurons have high hyperpolarization sag and rebound.",
     "source": "Results, Distinct phenotypes of major families, paragraph 2; Figure 2 (right biplots)",
     "quote": "These groups had distinct electrophysiological properties: for example, as expected, Pvalb neurons were characterized by high firing rates while Sst neurons had high values of the hyperpolarization sag and rebound",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "Reading (b), five groups, holds in full; Pvalb median max APs 71 (others at most 27), Sst median sag ratio 1.154 and rebound 5.70 mV, the highest. Under reading (a), seven families, ET has the highest median sag ratio (1.252), so two of three hold. Same with all cells and with per-mouse medians. 1,216 cells from 264 mice. Mann-Whitney p (information only) below 1e-37 for the three comparisons.",
     "deviations": [
      "Each recording counted once; the result with all cells is shown and is the same."
     ],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig3d-families-ephys",
     "statement": "In a k-nearest-neighbour classification of cells into seven transcriptomic families from electrophysiological features, the confusion matrix is mostly diagonal, with only the ET and IT subclasses strongly overlapping.",
     "source": "Results, Distinct phenotypes of major families, paragraph 3; Figure 3a, d",
     "quote": "We quantified this separation using a confusion matrix for k -nearest neighbours (kNN) classification of cells into families: it was mostly diagonal, with only the ET and IT subclasses strongly overlapping (Fig. 3d ).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "Both handlings of the excluded cells give ET to IT 49.0% (IT to ET 1.2%) and every other off-diagonal entry below 25%, the largest Lamp5 to Sst 14.4%. Diagonal from 49.0% (ET) and 65.6% (Lamp5) to 97.6% (Pvalb). Mean diagonal 0.808 against 0.157 with permuted labels. The 24 ET cells classified as IT come from 16 mice. All cells give ET to IT 49.0% and largest other entry 14.3%.",
     "deviations": [
      "Each recording counted once; the result with all cells is shown and is the same."
     ],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed6-it-et-room-features",
     "statement": "At room temperature, the features that differ most between layer 5 IT and ET neurons by a two-sided Wilcoxon-Mann-Whitney test, after omitting the upstroke-to-downstroke ratio, sag time and sag area, are the sag ratio, input resistance, AP width and rebound.",
     "source": "Extended Data Figure 6a-e legend; the code, physiological-temperature.ipynb, cells 22 and 23",
     "quote": "The first four panels correspond to features showing the largest IT/ET differences at room temperature, according to the two-sided Wilcoxon-Mann\u2013Whitney test statistic (and omitting several features that are very correlated with the shown ones ... upstroke-to-downstroke ratio, sag time, and sag area).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "Layer as the claims file's opening comment defines it, the files' normalized depth cut at the Nissl boundaries (the targeted layer without a depth). 111 L5 IT cells (61 mice) against 46 L5 ET cells (26 mice). After leaving out UDR, sag time and sag area, the four smallest p are sag ratio (7.9e-14), input resistance (5.9e-12), AP width (6.4e-11) and rebound (1.8e-05); the fifth is the AHP (8.7e-05). With the files' inferred_layer, or the same rule on the paper's \u00b5m depths (45 L5 ET cells each), the same four come first, rebound 2.9e-05 against AHP 7.7e-05. Same with all cells.",
     "deviations": [
      "The normalized depth is the ratio of the files' 4x image measurements; the paper's code uses \u00b5m depths, which are not in the dandiset. The \u00b5m reading from the paper's table is shown and gives the same.",
      "Cells with an undefined feature are dropped for that feature, the reading the claims file names; here no L5 IT or ET cell lacks a feature among the top ranks."
     ],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed6-physiological-temperature",
     "statement": "Follow-up experiments at 34 degrees C confirmed the electrophysiological overlap between IT and ET neurons and the continuous variation of rebound across neighbouring layer 5 Sst t-types.",
     "source": "Results, Distinct phenotypes of major families, paragraph 3, and Continuous phenotypic variation, paragraph 2; Methods, Experiments at physiological temperature; Extended Data Figure 6",
     "quote": "We confirmed the electrophysiological overlap between IT and ET neurons in follow-up experiments at 34 \u00b0C (Extended Data Fig. 6 ).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "blocked",
     "blocked_by": "methods_underspecified",
     "notebook": null,
     "notes": "The claims file records this claim for coverage and its criterion leaves nothing to test; the 34 degree recordings are in DANDI:000035, and the paper gives no measure of the IT/ET overlap at 34 degrees (Extended Data Fig. 6f, g show it in two-dimensional plots only). The data are public, so the claim is not blocked for data_missing, which the protocol rules out for a claim with data elsewhere; what is missing is a measure to test.",
     "deviations": [],
     "disagreed": true,
     "data": "elsewhere",
     "where": "DANDI dandiset 000035 (physiological temperature recordings); features and metadata in mini-atlas data/m1_patchseq_phys_temp_* at commit 3bf98a524b8e48aa8647e1b9ae8f2351d9c712c9",
     "proposed": null
    },
    {
     "claim": "fig1a-ttype-coverage",
     "statement": "After quality control 1,227 neurons had a t-type (817 inhibitory, 410 excitatory), covering 77 of the 90 neuronal t-types of the reference taxonomy.",
     "source": "Results, Patch-seq of mouse primary motor cortex, paragraph 3; Figure 1a",
     "quote": "leaving 1,227 neurons for further analysis (817 inhibitory, 410 excitatory; 369 and 269 with morphological reconstructions, respectively). The resulting data set covered 77 out of the 90 neuronal t-types (Fig. 1a )",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "partially_reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "The files give 1,226 cells with a t-type after the exclusions, 817 inhibitory and 409 excitatory, in 76 t-types. The cell the dandiset lacks (20190211_sample_10) is excitatory, so 1,226 and 409 are accepted by the criterion. It is the only L6 CT Gpr139 cell, so the t-type count is 76, not 77; the paper's table, with that cell, gives 1,227 cells and 77 t-types. Three of the four counts match. Whether the allowance for the missing cell extends to the t-type it alone covers is not in the criterion; the question is in questions/000008.yaml.",
     "deviations": [
      "The five exclusions come from the paper's table (link-metadata-table); the files alone give 1,231 cells."
     ],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig1b-depth-layers",
     "statement": "Within-type distributions of soma depth agree with the layer-specific names of the excitatory t-types.",
     "source": "Results, Patch-seq of mouse primary motor cortex, paragraph 3; Figure 1b",
     "quote": "Within-type distributions of soma depths (Fig. 1b ) agreed well with previous data 4 and with the layer-specific nomenclature of excitatory t-types, confirming the validity of our t-type assignment.",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "partially_reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "18 of 19 excitatory t-types with at least three depths (95%) have a median in a layer their name allows. L5 IT_4 (four cells, four mice, depths 0.753 to 0.835) has its median at 0.757, in L6. The paper's \u00b5m depths give the same (median 0.767).",
     "deviations": [
      "The normalized depth is the ratio of the files' 4x image measurements of soma depth and cortical thickness; the paper used \u00b5m measurements, which are not in the dandiset (median absolute difference 0.005, 14 cells differ by more than 0.05). The \u00b5m reading from the paper's table is shown."
     ],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "l45-it1-boundary",
     "statement": "The L4/5 IT_1 t-type is located on the boundary between L2/3 and L5, and probably corresponds to the quasi-L4 neurons described before in motor cortex.",
     "source": "Results, Patch-seq of mouse primary motor cortex, paragraph 4; Figure 1b",
     "quote": "We were also able to identify a t-type, L4/5 IT_1, that was located on the boundary between L2/3 and L5 and probably corresponds to the quasi-L4 neurons described previously in motor cortex 31 .",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "22 L4/5 IT_1 cells from 14 mice; median depth 0.279, inside 0.235 to 0.40 (reading a); interquartile range 0.259 to 0.306, which contains 0.29 (reading b). With the paper's \u00b5m depths, 0.277 and 0.252 to 0.303. The correspondence to quasi-L4 neurons is not tested.",
     "deviations": [
      "Normalized depth from the files' 4x fields, as in fig1b-depth-layers."
     ],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig4c-distance-correlation",
     "statement": "Across 200 pairs of t-types from the same family (50 t-types with at least 5 cells), the transcriptomic and electrophysiological distances between t-types are correlated with r = 0.60.",
     "source": "Results, Continuous phenotypic variation, paragraph 3; Figure 4c and legend",
     "quote": "Pooling the pairs across all families, we found that these two distance measures were correlated, with r = 0.60 (Fig. 4c , n = 200 pairs; Extended Data Fig. 7 ).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_transcriptomics.ipynb",
     "notes": "Under reading (a), the code's, 50 t-types and 200 within-family pairs, r = 0.597 (p = 9.9e-21). Reading (b) gives 52 t-types, 222 pairs and r = 0.563, so the paper's counts are reading (a)'s. All cells give 0.597 and 0.562.",
     "deviations": [],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed7-within-families",
     "statement": "The correlation between transcriptomic and electrophysiological distances is also seen within several individual families and for many individual features.",
     "source": "Results, Continuous phenotypic variation, paragraph 3; Extended Data Figure 7c-n",
     "quote": "The correlation was also observed within multiple individual families and for many individual electrophysiological features (Extended Data Fig. 7 ).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_transcriptomics.ipynb",
     "notes": "All six families have at least three t-types and a positive correlation (Pvalb 0.747, Sst 0.494, Vip 0.403, Lamp5 0.781, IT 0.328, CT 0.967), so reading (b) holds. 16 of 17 features correlate positively in at least half of the families. Same with all cells.",
     "deviations": [],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig4d-it-depth",
     "statement": "Among IT t-types (eight with at least five cells), transcriptomic distances are strongly correlated with differences in average soma depth, r = 0.70.",
     "source": "Results, Continuous phenotypic variation, paragraph 4; Figure 4d legend",
     "quote": "Transcriptomic distances between t-types were strongly correlated with the average soma depth differences ( r = 0.70; Fig. 4d ).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_transcriptomics.ipynb",
     "notes": "Eight IT t-types with at least five cells with a depth (249 cells); r = 0.690 (p = 4.9e-05) with the files' 4x depths; 0.698 with the paper's \u00b5m depths.",
     "deviations": [
      "Normalized depth from the files' 4x fields; the paper's \u00b5m depths are shown as a variant."
     ],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig4a-vip-tau",
     "statement": "Within the Vip subclass the membrane time constant varies continuously across the transcriptomic landscape, largest near the Sncg subclass and decreasing towards Vip Gpc3.",
     "source": "Results, Continuous phenotypic variation, paragraph 1; Figure 4a",
     "quote": "For example, electrophysiological properties of the t-types within the Vip subclass varied continuously across the transcriptomic landscape; the membrane time constant, for instance, had its largest values close to the Sncg subclass and gradually decreased towards Vip Gpc3 (Fig. 4a ).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "blocked",
     "blocked_by": "methods_underspecified",
     "notebook": "scala_2021_transcriptomics.ipynb",
     "notes": "The two readings disagree, and the criterion then blocks the claim. Reading (a), 153 Vip cells from 51 mice, Spearman rho between tau and position from the Sncg to the Vip Gpc3 centroid = -0.582, one-sided p = 1.6e-15. Reading (b), nine Vip t-types, rho between median tau and transcriptomic distance to Sncg = -0.383, one-sided p = 0.154 (-0.367, p = 0.166 with the mean of the Sncg t-type means). The direction agrees in both; the t-type reading has nine points.",
     "deviations": [
      "The Sncg subclass mean of reading (b) is taken over the Sncg reference cells; the mean of the Sncg t-type means is shown and gives the same."
     ],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig4b-sst-rebound",
     "statement": "In layer 5 Sst neurons (Sst Chodl excluded), morpho-electric properties vary continuously across the transcriptomic landscape, neighbouring t-types having similar rebound values.",
     "source": "Results, Continuous phenotypic variation, paragraph 2; Figure 4b; Extended Data Figure 6i",
     "quote": "Here we also found that morpho-electric properties varied continuously across the transcriptomic landscape, with neighbouring t-types consistently showing similar morphologies and similar rebound values (Fig. 4b ).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_transcriptomics.ipynb",
     "notes": "Layer as the claims file defines it, from the files' normalized depth. 11 L5 Sst t-types with at least five cells (143 cells, 63 mice), 55 pairs; r = 0.339, p = 0.0113 with all 17 features. Rebound alone r = 0.275, p = 0.042. With the files' inferred_layer or the paper's \u00b5m depths (64 mice), r = 0.347, p = 0.0094. The morphological part is not tested.",
     "deviations": [
      "The normalized depth that gives the layer is the ratio of the files' 4x image measurements; the paper's \u00b5m depths are shown as a variant."
     ],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig4e-pvalb-axon-shape",
     "statement": "In layer 5 Pvalb neurons the shape of the axonal arbor changes continuously across the transcriptomic landscape, from shrub-like to horizontally and vertically elongated basket cells, with intermediate forms in between.",
     "source": "Results, Continuous phenotypic variation, paragraph 5; Figure 4e",
     "quote": "The shape of the axonal arbor showed continuous changes across the transcriptomic landscape (Fig. 4e ): small shrub-like basket cells, horizontally elongated basket cells, and vertically elongated classical basket cells were located in different corners of the t-SNE embedding, with intermediate morphologies in between.",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "blocked",
     "blocked_by": "methods_underspecified",
     "notebook": null,
     "notes": "Shown only as a colour overlay; the paper gives no measure and the criterion names none. The reconstructions are at the Brain Image Library, whose server did not answer from the runner (see morph-reconstructions), and the paper's morphometric table is public in its code repository, so the data are elsewhere and the claim is not blocked for data_missing; the lack of a measure blocks it.",
     "deviations": [],
     "disagreed": true,
     "data": "elsewhere",
     "where": "Brain Image Library, https://download.brainimagelibrary.org/3a/88/3a88a7687ab66069/ (SWC reconstructions); morphometric features in mini-atlas data/m1_patchseq_morph_features.csv at commit 3bf98a524b8e48aa8647e1b9ae8f2351d9c712c9",
     "proposed": null
    },
    {
     "claim": "fig3d-morph-combined",
     "statement": "Between inhibitory families, separability is weaker with morphological features than with electrophysiological ones, and it is strongest with the two combined.",
     "source": "Results, Distinct phenotypes of major families, paragraph 4; Figure 3b-d",
     "quote": "The separability between inhibitory families was weaker than with electrophysiological features (Fig. 3d ). The between-family separability was strongest when we had combined electrophysiological and morphological features into a joint representation (Fig. 3c, d ), showing that these sets of properties are not redundant.",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "Inhibitory mean diagonal 0.737 with morphology against 0.851 with electrophysiology; seven-family mean highest for the combined representation, 0.872 (16 PCs) or 0.874 (17 PCs), against 0.808 and 0.753 (627 cells with both). Both handlings of the excluded cells, both block sizes and all cells give the same orderings; with all cells, as the paper, 628 cells have both, the criterion's count, with inhibitory means 0.737 against 0.847 and the combined seven-family mean 0.869 or 0.872. The morphology covers 635 cells (367 inhibitory, 268 excitatory); the paper's 636 includes the cell missing from the dandiset.",
     "deviations": [
      "Each recording counted once; the result with all cells is the same."
     ],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig5a-ttypes-between-families",
     "statement": "Most t-types are classified into their own family from electrophysiology, but some lie between two families; many Sst Pvalb Calb2 neurons are classified as Pvalb, Lamp5 Egln3_1 neurons have Vip- and Sst-like firing, and Vip Mybpc1 neurons often have Sst-like firing.",
     "source": "Results, Variability in individual t-types, paragraph 1; Figure 5a and legend",
     "quote": "Most t-types could be unambiguously placed into the correct family (Fig. 5a ), but some t-types were in between two families. For example, many Sst Pvalb Calb2 neurons were classified as belonging to the Pvalb subclass on the basis of electrophysiology. ... Vip Mybpc1 neurons often had Sst -like firing.",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "28 of 35 t-types with at least 10 cells have at most 25% classified into another family (i). Sst Pvalb Calb2 as Pvalb 38.5% (ii); Lamp5 Egln3_1 as Vip or Sst 90% of 10 cells (iii); Vip Mybpc1_3 as Sst 37.5% (iv, reading a), Vip Mybpc1_2 23.5%, pooled 25.4% (reading b). The code's handling (36 t-types, 29 at most 25% wrong) and all cells give the same.",
     "deviations": [],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig5b-ttype-variability",
     "statement": "Many t-types have a normalized total variance of electrophysiological features well above that of k-means clusters; some, such as Pvalb Vipr2_2 (chandelier cells), appear homogeneous, while Vip Mybpc1_2 is highly variable.",
     "source": "Results, Variability in individual t-types, paragraphs 2 and 3; Figure 5b",
     "quote": "We found that many t-types had total variance substantially above the variances of the k -means clusters (Fig. 5b ) ... some of them, such as Pvalb Vipr2_2 (chandelier cells), appeared morpho-electrically homogeneous. By contrast, Vip Mybpc1_2 was marked as having high electrophysiological variability",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "With n_init=10 and the Methods' membership, the k-means band is 0.173 to 0.468; 15 of 35 t-types (43%) are above it, many under both readings; Pvalb Vipr2_2 (0.449) is not above it and Vip Mybpc1_2 (0.692) is. With all cells, 17 of 35. The result depends on the k-means initialization; with n_init='auto' only 8 of 35 (23%) are above the band, and with the code's membership of the excluded cells n_init=10 gives 8 of 36 and 'auto' 16 of 36.",
     "deviations": [
      "n_init=10 is set explicitly. The criterion says scikit-learn defaults as the code; the code ran under Python 3.7.9 (notebook metadata), for which every scikit-learn release has the default n_init=10, while the installed scikit-learn defaults to 'auto'. 'auto' is shown as a variant.",
      "Each recording counted once; results with all cells are shown."
     ],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed8-leiden-entropy",
     "statement": "An alternative analysis using the entropy of Leiden cluster assignments within each t-type often highlights the same t-types as variable.",
     "source": "Results, Variability in individual t-types, paragraph 3; Extended Data Figure 8g-l",
     "quote": "an alternative analysis using entropies of Leiden clustering 35 often highlighted the same t-types as variable (Extended Data Fig. 8 ).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "not_reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "Leiden at resolution 2 gives 15 to 20 clusters per run. Across 35 t-types the Spearman correlation between normalized variance and median entropy is 0.266, one-sided p = 0.061, short of 0.05. With the code's membership of the excluded cells it is 0.313 (p = 0.031); with all cells, as the paper, 0.166 (p = 0.171) and 0.222 (p = 0.097). Four t-types are among the ten most variable by both measures (L5 ET_1, L5 IT_1, Sst Chodl, Vip Mybpc1_3). The paper gives no measure of agreement; the Spearman test is the reader's reading, and the result is near its threshold and moves with three cells.",
     "deviations": [
      "Each recording counted once; results with all cells are shown. leidenalg is a current release; the paper names no version."
     ],
     "disagreed": false,
     "data": "dataset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig5c-vip-mybpc1-rebound",
     "statement": "Vip Mybpc1_2 neurons vary in input resistance, membrane time constant and rebound, and this variability is not random; cells with low rebound lie close to the boundary with the low-rebound Vip Sncg type.",
     "source": "Results, Variability in individual t-types, paragraph 3; Figure 5c",
     "quote": "This variability was not random: overlaying the rebound values on the t-SNE embedding (Fig. 5c ) showed that cells with low rebound were located close to the boundary with the low-rebound Vip Sncg type.",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "blocked",
     "blocked_by": "methods_underspecified",
     "notebook": "scala_2021_transcriptomics.ipynb",
     "notes": "The two readings disagree. 34 Vip Mybpc1_2 cells (22 mice), rebound 0.3 to 19.8 mV, median 11.31 mV against 2.26 mV for Vip Sncg. Reading (a), rebound rises with the distance to the nearest Vip Sncg reference cell (rho = 0.356, one-sided p = 0.019). Reading (b), rebound does not fall with the correlation to the Vip Sncg mean (rho = 0.067, one-sided p = 0.648; 0.066 and 0.644 with the log taken after the mean). Scaling the Patch-seq counts to a million per cell first gives rho = 0.068, p = 0.648, and raw counts per million rho = 0.016, p = 0.536.",
     "deviations": [
      "The transform of reading (b) is not given by the claims file; the Methods' log2(x + 1) of length-normalized Patch-seq counts (the code's rnaseqTools.map_to_tsne with normalizeNew=True) against the mean of log2(UMI + 1) of the reference is the main reading; log2 of the mean UMI, and two library-size normalizations of the Patch-seq counts, are variants, and all give the same."
     ],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig5d-sst-pvalb-calb2-firing",
     "statement": "Sst Pvalb Calb2 cells vary widely in maximum firing rate, but high-firing cells are mostly grouped in one part of the transcriptomic landscape.",
     "source": "Results, Variability in individual t-types, paragraph 3; Figure 5d",
     "quote": "Similarly, Sst Pvalb Calb2 cells had high variability in terms of the maximum firing rate, but high-firing cells were mostly grouped in one part of the transcriptomic landscape (Fig. 5d ).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_transcriptomics.ipynb",
     "notes": "26 Sst Pvalb Calb2 cells (19 mice), max APs 15 to 87, none at the median (32.5), so 13 above it. Reading (a), max APs rise towards the Pvalb centroid (rho = 0.363, one-sided p = 0.034). Reading (b), the 13 high-firing cells (mean pairwise distance 7.05, against 8.49 for random subsets) under the criterion as the answer of 2026-10-21 fixes it. With 10,400,600 subsets of that size, more than 1,000,000, p is estimated from 100,000 random subsets, p = 0.0485 with a 95% Clopper-Pearson interval of 0.0472 to 0.0499, so (b) holds and both readings hold. What the answer changed, before it the criterion's 1,000 permutations gave p = 0.052 with this seed and 0.045 to 0.057 with five others, and the claim was blocked as methods_underspecified because the readings disagreed; the answer replaced them with 100,000 subsets and the bound, under which the claim is reproduced. The margin is small, and the seed still matters. The exact p-value over all 10,400,600 subsets, computed as a variant, is 0.0497, below 0.05, but the 100,000-subset estimate with five other seeds gives 0.0496 to 0.0514, and each interval reaches 0.05 or above, so (b) would not hold under the criterion with those seeds. The section's seed (55) was fixed before the answer. A follow-up question on this is in questions/000008.yaml.",
     "deviations": [
      "The seed of the 100,000 random subsets is a standard choice (55, the section's seed before the answer); it can decide reading (b) here, so five other seeds and the exact p-value are shown."
     ],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    },
    {
     "claim": "rnaseq-depth",
     "statement": "The 1,329 sequenced cells had median 1.3 million exonic and 0.7 million intronic reads (log10 mean 6.0 and 5.6) and 9,100 detected genes on average (s.d. 3,500).",
     "source": "Results, Patch-seq of mouse primary motor cortex, paragraph 1; Extended Data Figure 2",
     "quote": "The mRNA of these 1,329 cells was sequenced, yielding on average 1.3 million exonic and 0.7 million intronic reads (medians ... 6.0 \u00b1 0.6 and 5.6 \u00b1 0.8, respectively) and 9,100 \u00b1 3,500 (mean \u00b1 s.d.) detected genes per cell",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_sequencing.ipynb",
     "notes": "Over the 1,329 cells of the paper's count tables, median exonic count 1,281,492 and intronic 687,965 (log10 means 5.99 and 5.61), and 9,097 detected genes on average (s.d. 3,533). The read fields of the paper's table, which the dandiset's files repeat, give 1,190,890 and 688,140, also within 10%.",
     "deviations": [],
     "disagreed": false,
     "data": "elsewhere",
     "where": "GEO GSE163764 (also NeMO dat-kq04hua); count tables in mini-atlas data/ at commit 3bf98a524b8e48aa8647e1b9ae8f2351d9c712c9",
     "proposed": null
    },
    {
     "claim": "ttype-qc-counts",
     "statement": "In the first round of mapping, 78 cells had a highest correlation below 0.4, 14 mapped to non-neural types and 5 were ambiguous between large groups, leaving 1,232 cells; four with excitatory RNA but inhibitory phenotype and one outside MOp were excluded, leaving 1,227.",
     "source": "Methods, t-Type assignment, paragraphs 1 and 3",
     "quote": "cells with the highest correlation below 0.4 (78 cells); cells that would be assigned to non-neural t-types, presumably owing to RNA contamination 51 (14 cells; see also Extended Data Fig. 2j\u2013n ); cells with the highest correlation less than 0.02 above the maximal correlation in one of the other two large transcriptomic groups (5 cells). The remaining 1,232 cells passed quality control",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_sequencing.ipynb",
     "notes": "The code's first-round mapping to the Tasic et al. (2018) reference gives 78, 14, 0 (Meis2 or CR) and 5 failing cells, 1,232 passing; every cell's pass or fail agrees with the paper's table, and the 1,232 include the 5 excluded cells, leaving 1,227.",
     "deviations": [],
     "disagreed": false,
     "data": "elsewhere",
     "where": "GEO GSE163764 (also NeMO dat-kq04hua); count tables in mini-atlas data/ at commit 3bf98a524b8e48aa8647e1b9ae8f2351d9c712c9; reference Smart-seq2 data of Tasic et al. 2018",
     "proposed": null
    },
    {
     "claim": "ed2-smartseq-similarity",
     "statement": "Patch-seq expression profiles are most similar to the single-nucleus Smart-seq2 reference, and t-type assignments based on Smart-seq2 and 10x references agree well.",
     "source": "Results, Patch-seq of mouse primary motor cortex, paragraph 2; Extended Data Figure 2g-i",
     "quote": "We found that Patch-seq expression profiles were most similar to the single-nucleus Smart-seq2 data (Extended Data Fig. 2g, h ). At the same time, there was good agreement between t-type assignments based on Smart-seq2 and those based on 10x reference data (Extended Data Fig. 2i )",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_sequencing.ipynb",
     "notes": "Over 1,232 cells, the median maximal correlation is highest for the single-nucleus Smart-seq2 reference (0.703; then single-cell 10x v2 0.623 and single-cell Smart-seq2 0.621), and 75.7% of cells correlate best with it. The single-cell Smart-seq2 and 10x v2 references assign the same t-type for 78.6% of cells.",
     "deviations": [
      "The assignment is the direct one, without the bootstrap over genes that the paper uses for confidence."
     ],
     "disagreed": false,
     "data": "elsewhere",
     "where": "GEO GSE163764 (also NeMO dat-kq04hua); the seven MOp reference data sets of Yao et al. 2021 (doi:10.1038/s41586-021-03500-8)",
     "proposed": null
    },
    {
     "claim": "morph-reconstructions",
     "statement": "646 neurons had staining good enough for reconstruction, 369 inhibitory and 269 excitatory among cells with a t-type; 73 t-types have at least one reconstructed neuron.",
     "source": "Results, Patch-seq of mouse primary motor cortex, paragraphs 1 and 3",
     "quote": "Of these neurons, 646 had sufficient staining for their morphologies to be reconstructed.",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "blocked",
     "blocked_by": "data_unreadable",
     "notebook": null,
     "notes": "The SWC files are at the Brain Image Library (download.brainimagelibrary.org/3a/88/3a88a7687ab66069); the server did not answer from the runner (connection timed out, twice, on 2026-10-20, and again from the notebook on 2026-10-21), so the folders could not be listed. The paper's table and morphometric features (646 traced cells) are in its code repository but are not the data the criterion names.",
     "deviations": [],
     "disagreed": false,
     "data": "elsewhere",
     "where": "Brain Image Library, https://download.brainimagelibrary.org/3a/88/3a88a7687ab66069/",
     "proposed": null
    },
    {
     "claim": "fig2-rrr",
     "statement": "A sparse reduced-rank regression predicting the 17 electrophysiological features from gene expression, with 25 genes and a rank-5 latent space, reaches a cross-validated R2 of 0.38, with cross-validated correlations of 0.90, 0.74 and 0.67 between the first three pairs of projections.",
     "source": "Results, Distinct phenotypes of major families, paragraph 1; Figure 2; Methods, Reduced-rank regression",
     "quote": "The selected model used 25 genes with a 5-dimensional latent space and achieved a cross-validated R 2 of 0.38. ... The cross-validated correlations between the first three pairs of projections were 0.90, 0.74, and 0.67, respectively.",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_transcriptomics.ipynb",
     "notes": "Main reading (four log-transformed features), 1,216 cells from 264 mice. At lambda 0.5, 24 genes, cross-validated R2 0.370 and correlations 0.895, 0.725, 0.648; at lambda 0.45, 29 genes, 0.382, 0.898, 0.744, 0.669. All within tolerance at both lambdas; also under the code's three-feature transform and with all 1,219 cells. The folds are those of the paper's elastic_rrr_cv (seed 42, one permutation, ten contiguous test blocks of 121 cells, 6 cells never tested).",
     "deviations": [
      "The elastic net inside the paper's sparseRRR.py is fitted with scikit-learn (ElasticNet, MultiTaskElasticNet), which minimizes the same objective, because glmnet_python has no distribution for Python 3.13; the paper reports 25 genes at lambda 0.5, the substitute selects 24.",
      "Each recording counted once; the result with all cells is shown."
     ],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed5-rrr-ion-channels",
     "statement": "A reduced-rank regression restricted to ion channel genes does not perform much worse than the full model, with cross-validated R2 0.33 and correlations 0.86, 0.71 and 0.56.",
     "source": "Results, Distinct phenotypes of major families, paragraph 2; Extended Data Figure 5e, f",
     "quote": "A reduced-rank regression model restricted to using only ion channel genes (Extended Data Fig. 5 ) did not perform much worse than the full model (cross-validated R 2 = 0.33 and correlations 0.86, 0.71, and 0.56, respectively, with regularization set to select 25 genes).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_transcriptomics.ipynb",
     "notes": "307 ion channel genes detected in at least 10 cells, lambda 0.303, 26 genes selected; R2 0.341 and correlations 0.865, 0.706, 0.565, all within tolerance; the code's three-feature transform gives 0.335, 0.862, 0.707, 0.565.",
     "deviations": [
      "The scikit-learn substitute for glmnet, as in fig2-rrr; 26 genes are selected where the paper reports 25."
     ],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    },
    {
     "claim": "hold-time-yield",
     "statement": "The cDNA yield was not correlated with the recording hold time (Spearman correlation -0.01).",
     "source": "Methods, Patch-seq recording procedures, last paragraph",
     "quote": "The resulting cDNA yield was not correlated with the hold time (Spearman correlation \u22120.01).",
     "statistics": [],
     "kind": "finding",
     "inferred": "",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "scala_2021_ephys.ipynb",
     "notes": "Spearman correlation between hold time and log cDNA yield over the 1,319 dandiset cells with both values, -0.012 (p = 0.65); over all 1,320 cells of the table, also -0.012. Neither value is in the dandiset's files; the data check lists every DandiIcephysMetadata and subject field in all files and every entry under general of one file, and none is a hold time or a yield. The dandiset contributes only the cells.",
     "deviations": [],
     "disagreed": false,
     "data": "with_inputs",
     "where": null,
     "proposed": null
    }
   ],
   "findings": [
    {
     "kind": "other",
     "statement": "Three pairs of files hold identical voltage recordings although each pair is two different cells of one mouse, with different depths and read counts, two of the pairs from different slices, and in one pair different t-types (20190731_sample_5 and 20190731_sample_8; 20181116_sample_2 and 20181116_sample_4; 20190829_sample_2 and 20190829_sample_6). The paper's feature table has identical values for each pair, so the duplication predates the deposit and the paper analysed each recording twice. Which cell each recording belongs to is not recorded."
    },
    {
     "kind": "other",
     "statement": "Three files (20190128_sample_3, 20180215_sample_1, 20190222_sample_5) hold other traces than the ones the paper's published features were computed from; one of them has 41 sweeps. The features of 1,320 files reproduce the published table on every feature but the rheobase, which the paper fits with an unseeded random procedure."
    },
    {
     "kind": "metadata_error",
     "statement": "The field cortical_thickness_4x holds the soma depth and cortical_thickness_um the cortical thickness, both in 4x image units (the paper's table columns Soma depth (4x) and Cortical thickness (4x)). The field soma_depth_um is present in 43 files and empty in all. The paper's \u00b5m soma depths and cortical thicknesses, which its analyses use, are not in the dandiset."
    },
    {
     "kind": "metadata_error",
     "statement": "Voltage series are labelled volts with conversion 1 but hold millivolts; stimulus series are labelled amperes but hold picoamperes."
    },
    {
     "kind": "data_missing",
     "statement": "No series has a start time (starting_time is NaN in all 104,409 voltage series) and there are no timestamps, so the timing of sweeps, including the 1.3 or 1.4 s intervals the Methods give, is lost. Each file holds a single series of current steps, while the Methods say the series was repeated for most neurons."
    },
    {
     "kind": "metadata_insufficient",
     "statement": "The five cells with a t-type that the paper excludes (four with excitatory RNA but inhibitory firing, one outside MOp) are not marked in the files; they can be identified only from the paper's table. The t-types and families use an older name for the ET subclass (PT)."
    },
    {
     "kind": "data_missing",
     "statement": "The hold time and cDNA yield of each cell, used in the Methods, are not in the files; they are in the paper's metadata table."
    },
    {
     "kind": "other",
     "statement": "The step protocol departs from the Methods in part of the cells; 145 of 1,328 cells (10.9%) do not start at -200 pA (90 start at -100 pA), and 112 go above 1,380 pA, up to 1,780 pA."
    },
    {
     "kind": "other",
     "statement": "One file (20190625_sample_13) has voltages of about +-1,025 mV in some sweeps, values outside any physiological range."
    },
    {
     "kind": "metadata_insufficient",
     "statement": "One mouse has the genotype Scnn1a-Tg3-Cre/wt; Ai9/wt, which the paper does not list; the paper counts 27 wild-type mice where the files have 26."
    },
    {
     "kind": "paper_inconsistency",
     "statement": "The Extended Data Fig. 4 legend gives an AP threshold of -40.18 mV for the example cell; the paper's own feature table and the extraction from the file give -43.18 mV. The legend also gives the sag time in ms where the value is in seconds."
    }
   ],
   "verifications": [
    {
     "date": "2026-10-21",
     "kind": "agent",
     "by": "Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebooks)",
     "reviewed": "The handover of the first run, the three executed notebooks with the conclusions and summaries removed, the helper module scala.py and the claims file; the verifier also had the paper.",
     "agreed": 32,
     "of": 35,
     "disagreements": [
      {
       "claim": "cohort-cells-mice",
       "outcome": "reproduced",
       "reason": "1,328 distinct cells (reading b) and 266 subjects (cell 18). The archive summary gives the same counts."
      },
      {
       "claim": "ed6-physiological-temperature",
       "outcome": "blocked",
       "reason": "The 34 degree data are in DANDI:000035. The archive description of 000008 says so, and the paper gives no measure. The verifier gives blocked with data_missing."
      },
      {
       "claim": "fig4e-pvalb-axon-shape",
       "outcome": "blocked",
       "reason": "All 1,328 assets are NWB files, with no morphology. The paper gives no measure. The verifier gives blocked with data_missing."
      }
     ]
    },
    {
     "date": "2026-10-21",
     "kind": "agent",
     "by": "Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebooks)",
     "reviewed": "The full executed notebooks of the second run, with the conclusions and summaries, the README, the attempt record, the helper module scala.py and the claims file; the verifier also had the paper.",
     "agreed": 35,
     "of": 35,
     "disagreements": []
    }
   ]
  }
 ],
 "entries": [
  {
   "name": "000008-scala-2021",
   "type": "reproduction",
   "title": "Phenotypic variation of transcriptomic cell types in mouse motor cortex (Scala, Kobak et al. 2021)",
   "description": "Tests the findings of Scala, Kobak et al. (2021) on dandiset 000008, which holds the current-clamp recordings of 1,328 Patch-seq neurons of mouse primary motor cortex with each cell's t-type, family, layer and read counts. The paper's own feature extraction is run on every file; the cohort, the stimulus protocol, the features, the kNN family classification, the IT and ET differences, the t-type coverage and depths, and the variability of t-types are tested on the dandiset, and the claims that also need the paper's gene counts, morphology or transcriptomic reference data are tested with those inputs in separate notebooks.",
   "datasets": [
    "dandi:000008"
   ],
   "papers": [
    "10.1038/s41586-020-2907-3"
   ],
   "derived_from": [],
   "notebooks": [
    {
     "path": "scala_2021_ephys.ipynb",
     "title": "Recordings and metadata of dandiset 000008 (cohort, features, families, t-types)",
     "executed": "notebooks/000008-scala-2021/scala_2021_ephys.html"
    },
    {
     "path": "scala_2021_transcriptomics.ipynb",
     "title": "Dandiset 000008 with the paper's gene counts and 10x reference (RRR, within-family distances)",
     "executed": "notebooks/000008-scala-2021/scala_2021_transcriptomics.html"
    },
    {
     "path": "scala_2021_sequencing.ipynb",
     "title": "Sequencing claims on the paper's count tables and reference data",
     "executed": "notebooks/000008-scala-2021/scala_2021_sequencing.html"
    }
   ]
  }
 ]
}