{
 "dandiset": "001097",
 "page": "dandisets/001097/",
 "generated": "2026-10-19T16:23:39+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": 7,
  "partially_reproduced": 2,
  "not_reproduced": 3,
  "blocked": 12
 },
 "outcome_counts_with_outside_data": {
  "reproduced": 0,
  "partially_reproduced": 0,
  "not_reproduced": 0,
  "blocked": 0
 },
 "outcome_counts_elsewhere": {
  "reproduced": 0,
  "partially_reproduced": 0,
  "not_reproduced": 0,
  "blocked": 0
 },
 "attempts": [
  {
   "id": "001097-liu-2024-01",
   "dandiset": "001097",
   "dandiset_version": "0.260624.0330",
   "paper": "10.1038/s41586-024-07916-w",
   "paper_title": "Encoding of female mating dynamics by a hypothalamic line attractor",
   "date": "2026-10-14",
   "kind": "agent",
   "accountable": "CatalystNeuro",
   "agent": {
    "harness": "Claude Code (GitHub Actions)",
    "model": "claude-opus-5-5",
    "protocol": "pilot-3"
   },
   "tier": "agent_verified",
   "entry": "001097-liu-2024",
   "entry_title": "A hypothalamic line attractor for female mating (Liu, Nair et al. 2024)",
   "description": "Tests the findings of Liu, Nair et al. (2024) on dandiset 001097, which holds miniscope imaging of VMHvl alpha-cells in female mice during mating, with annotations of male and female behaviour. It tests the behavioural dynamics of mating, single-cell tuning, GLM fits and persistence, and, with the authors' deposited rSLDS weights for two mice, how the receptive-day integration dimension and its cells behave across the oestrus cycle. Claims that need new rSLDS fits are blocked, as the model cannot be built or fit within the registry's environment and time limit.",
   "notebooks": [
    {
     "path": "reproduce_liu_2024.ipynb",
     "title": "Reproducing Liu, Nair et al. (2024) from dandiset 001097",
     "executed": "notebooks/001097-liu-2024/reproduce_liu_2024.html"
    }
   ],
   "outcomes": [
    {
     "claim": "fig1b-accepting-escalates",
     "statement": "During mating the probability of accepting behaviours (lordosis, wiggle) increases gradually over the interaction, while resistance behaviours (dart, top up, kick, turn) first increase and then slowly decrease.",
     "source": "Figure 1b, Extended Data Figure 1a, Results \"Dynamics of female behaviours in mating\"",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "reproduce_liu_2024.ipynb",
     "notes": "Over the first 600 s after male entry the mean probability of accepting behaviour rises (Spearman 0.74, p = 3e-6, 69 annotated trials; 0.59, p = 6e-4, 58 receptive trials), from 0.11 in the first third of the 20-s bins to 0.19 in the last. Resistance peaks in the third bin (0.14) and falls to 0.07 in the last third. Later bins hold fewer trials (69 down to 11); with the same 63 trials in each of the first 15 bins the rise remains (Spearman 0.63, p = 0.012, 0.080 to 0.151). Within single mice the accepting correlation is positive in 12 of 23, and within single trials in 33 of 58 (median 0.01, sign test p = 0.36), so the rise is a property of the pooled trials.",
     "deviations": [
      "69 distinct annotated trials from 23 mice, where the paper has 74 from 28; the paper's trials are not identified.",
      "Time is counted from male entry and the curves are cut at 600 s; the paper's figure does not state either."
     ],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig1c-responsive-dominates",
     "statement": "Females spend 83.9% of their mating behaviour time in male-responsive behaviours (accepting, resistance, staying) and 16.1% in self-initiated ones (approach, sniff, check genital area).",
     "source": "Figure 1c, Results",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "reproduce_liu_2024.ipynb",
     "notes": "Male-responsive behaviours take 84.1% of female mating time on average over the 69 annotated trials (83.5% pooled; 82.9% over receptive trials), against 83.9%. Per mouse the receptive-trial average runs from 65% to 94%.",
     "deviations": [
      "69 distinct annotated trials from 23 mice, where the paper has 74 from 28."
     ],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig1d-male-drives-mating",
     "statement": "Males spend 11.3 times more time in self-initiated mating behaviours (sniff, mount, intromission) than females do in theirs (approach, sniff, check genital area).",
     "source": "Figure 1d, Results",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "reproduce_liu_2024.ipynb",
     "notes": "The mean per-trial ratio of male to female self-initiated time is 11.75 over the 69 annotated trials, within the tolerance of 11.3. Only this one of the six readings matches; the ratio of summed times is 7.5 (7.1 on receptive trials), the median per-trial ratio 9.2 (8.2), and the mean ratio over receptive trials 9.3. Every reading is well above 1.",
     "deviations": [
      "69 distinct annotated trials from 23 mice, where the paper has 74 from 28.",
      "The paper does not say how the ratio is averaged; six readings were set out before the analysis and the match rests on one of them."
     ],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig1e-interbout-interval",
     "statement": "Male copulation (mounting and intromission) comes in bouts separated by interbout intervals with a mean of 13.7 s (1,685 bouts and 1,611 intervals in 74 trials).",
     "source": "Figure 1e, Results",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "reproduce_liu_2024.ipynb",
     "notes": "Merging mount and intromission annotations less than 0.1 s apart, the mean interbout interval is 14.0 s over the 69 annotated trials (1,540 bouts, 1,471 intervals), 14.9 s over receptive trials and 14.6 s over all 88 distinct sessions, against 13.7 s. Merging gaps up to 1 s gives 15.5 to 16.3 s and not merging gives 9.8 to 10.8 s.",
     "deviations": [
      "The rule for joining annotations into bouts is not stated; three were set out before the analysis.",
      "69 distinct annotated trials from 23 mice, where the paper has 74 from 28."
     ],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig1g-social-in-bouts-and-intervals",
     "statement": "Females show social behaviours (accepting, resistance and appetitive) for 62% of the duration of each male copulation bout and continue them for 23% of each interbout interval.",
     "source": "Figure 1g, Results",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "reproduce_liu_2024.ipynb",
     "notes": "Over the 69 annotated trials females show social behaviour for 63.3% of each copulation bout and 21.8% of each interbout interval on average, against 62% and 23%. This is the only one of the four readings that matches; over receptive trials the interval value is 17.4%, outside the tolerance, and pooled over time the values are 74% and 11%. Leaving out m930's oestrogen days 3 and 4, the two closest to its receptive session, gives 21.7% to 21.9% for the interval value, still within tolerance.",
     "deviations": [
      "69 distinct annotated trials from 23 mice, where the paper has 74 from 28.",
      "Social behaviour follows the Methods' grouping, which leaves out staying."
     ],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig2c-few-tuned-cells",
     "statement": "Only a small percentage of VMHvl alpha-cells (approximately 2 to 13%) are tuned to any one mating behaviour by choice probability; most show mixed selectivity.",
     "source": "Figure 2c, Extended Data Figure 2a,b, Methods \"Choice probability\"",
     "outcome": "not_reproduced",
     "blocked_by": null,
     "notebook": "reproduce_liu_2024.ipynb",
     "notes": "In the 15 receptive sessions (1,335 cells), 6.9% (approach) to 37.4% (dart) of cells are tuned to a behaviour, median 22.0% over the 11 behaviours with enough bins, against about 2 to 13%. Eight of the 11 exceed 15%. The shuffle test alone passes 21% to 75% of cells. Comparing pairs of behaviours, as the Methods describe, tunes 15.4% to 57.6% over 20 pairs (median 39.3%), with none within 1 to 15%.",
     "deviations": [
      "Each behaviour is compared with every other 1-s bin of the interaction. The Methods compare two behaviours without saying which pairs Figure 2c uses, and only this comparison was set out before the analysis. Pairs of behaviours are shown as a variant, added after the first round of verification.",
      "A bin belongs to a behaviour when the behaviour covers at least half of it; the shuffle permutes bin labels. Neither is stated in the paper."
     ],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig2e-glm-behaviour-explains-little",
     "statement": "A linear model of each neuron's activity from behaviour explains little of its variance, with mean cross-validated R2 of 0.08 from female behaviours, 0.14 from male behaviours and 0.15 from both.",
     "source": "Figure 2e, Extended Data Figure 2c,d, Methods \"GLM\"",
     "outcome": "not_reproduced",
     "blocked_by": null,
     "notebook": "reproduce_liu_2024.ipynb",
     "notes": "The outcome is the one both verification rounds agreed with. After the second round the author proposed partially_reproduced, from a penalty selection added then; that change has not been verified (protocol pilot-4, \"Hand over for verification\"). With tenfold contiguous cross-validation and the ridge penalty chosen per cell by fivefold cross-validation within each training set, the mean cvR2 over 1,335 cells is -0.041 from female behaviours, -0.005 from male behaviours and -0.004 from both (mouse means -0.026, 0.014 and 0.016), against 0.08, 0.14 and 0.15. The ordering holds, both against male by a margin of 0.0003 over cells and 0.0014 over mouse means, and all three are below 0.25, so the partial rule is met by a thin margin. With the penalty chosen on the last fifth of each training set, the design used in the first two rounds, the means are -0.077, -0.013 and -0.057 (mouse means -0.050, 0.008 and -0.017) and both scores below male, so the ordering failure that gave not_reproduced came from that selection. A model with only an intercept scores -0.086 under the same folds, so the negative values come from the fold design; with the cross-validated penalty the gain over it is 0.045, 0.081 and 0.082 (mouse means 0.054, 0.095 and 0.096), below the paper's values. With the last-fifth penalty, R2 within single folds is dominated by the same offsets (medians -0.61 to -0.71, intercept -0.70). With frames assigned to folds at random, a variant not named in the criterion, the means are 0.229, 0.207 and 0.322, with neighbouring frames in both training and test sets; the value from both behaviours is then above 0.25. The test is weak, since no design here separates behaviour from the slow drift of the traces.",
     "deviations": [
      "Lags in 1-s steps over 0 to 10 s, not every frame, to keep the run short.",
      "Contiguous folds, the ridge penalty chosen per cell by fivefold contiguous cross-validation within each training set, and R2 on pooled held-out predictions; the Methods state none of these. The cross-validated penalty selection was added after the second round of verification; the outcome would change with it, and that change is proposed, not verified.",
      "Fit over the interaction (male entry to end of annotation) on z-scored, unsmoothed traces."
     ],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": {
      "outcome": "partially_reproduced",
      "reason": "With the ridge penalty chosen by fivefold cross-validation within each training set, a selection added after the second round, the ordering of the criterion holds (both above male by 0.0003 over cells) and all three values are below 0.25, which meets the partial rule."
     }
    },
    {
     "claim": "fig2f-glm-coupling",
     "statement": "Adding the activity of the other simultaneously imaged neurons as regressors raises the mean cross-validated R2 to 0.46.",
     "source": "Figure 2f, Methods \"GLM\"",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "reproduce_liu_2024.ipynb",
     "notes": "Adding the instantaneous activity of the other cells gives a mean cvR2 of 0.477 over cells (0.458 over mouse means) with the cross-validated penalty, and 0.450 (0.434) with the last-fifth penalty, against 0.46. With the last-fifth penalty 73 of 1,335 cells remain below zero and the gain over the intercept-only baseline is 0.536 (0.563 with the cross-validated penalty). With random folds it is 0.814. It shares the fold design questioned for fig2e-glm-behaviour-explains-little.",
     "deviations": [
      "As for fig2e-glm-behaviour-explains-little. Coupling uses the other cells at the same time point; the Methods do not say whether coupling filters had lags."
     ],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig2i-persistent-cells",
     "statement": "About 30% of cells have an autocorrelation half-width (ACHW) longer than 25 s, and the mean ACHW is 20 s, longer than the mean interbout interval.",
     "source": "Figure 2i, Results, Methods \"Calculation of ACHW\"",
     "outcome": "not_reproduced",
     "blocked_by": null,
     "notebook": "reproduce_liu_2024.ipynb",
     "notes": "Over the whole recording, smoothed as in the authors' code, 20.0% of 1,335 cells have an ACHW above 25 s and the mean is 15.5 s, against 30% and 20 s. The first-crossing rule gives the same, and the other readings give 10 to 17% and 9.9 to 13.1 s. Per mouse the fraction runs from 0 to 0.48, and four mice reach 0.28 or more; the lower panel of Figure 2i has n = 4 mice, which are not identified. Ending the window at the last annotation, a variant that drops the unannotated imaging at the end of 7 sessions, gives 15.5 to 19.7% and 12.7 to 15.0 s. The mean ACHW is still longer than the mean interbout interval (14.0 s), as the paper says.",
     "deviations": [
      "The ACHW follows the authors' code (the lag closest to 0.5, up to 50 s), where the Methods say the lag where the function reaches 0.5; both are shown."
     ],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed3b-pencil-cup",
     "statement": "The ACHW of the same cells is shorter when the male is behind a perforated pencil cup (14.3 s) than during free interaction (19.6 s), 470 neurons from 5 mice.",
     "source": "Extended Data Figure 3a-c, Results",
     "outcome": "blocked",
     "blocked_by": "metadata_insufficient",
     "notebook": null,
     "notes": "No annotation marks the pencil-cup period. In 17 sessions imaging continues for up to 7 minutes after the last annotation, which may be that period, but the files do not say so. An exploratory look, not judged, at the 7 receptive sessions with at least 3 minutes of it (544 cells, where the paper has 470 from 5 mice) finds a shorter mean ACHW in the tail than over the interaction (4.0 to 11.5 s against 7.3 to 17.9 s), but against a stretch of the interaction of the same length the tail is longer in 5 of 7 sessions, so the difference comes from the window length.",
     "deviations": [],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig3a-rslds-fit",
     "statement": "An rSLDS fit to each receptive mating trial explains 64.08% (SD 6.8) of the variance of the neural trajectories, n = 15 mice.",
     "source": "Figure 3a, Results",
     "outcome": "blocked",
     "blocked_by": "other",
     "notebook": null,
     "notes": "Needs rSLDS fits. The ssm package is a Cython extension installed from GitHub that does not build in the registry's locked environment (it needs numpy at build time), and with the authors' settings one 10,000-frame session takes about 7.5 minutes to fit (10 of 150 iterations took 30 s in a separate environment), so the fits the blocked claims need would take hours against the 30-minute limit. A single linear dynamical system was tried as a stand-in and fitted rotations with two equal time constants in 14 of 15 receptive sessions, so it cannot separate an integration dimension and was not used. The Methods also leave the forward simulation interval open, and the authors' code computes this variance explained differently (a full free-running simulation). Neither timing nor stand-in is run in the notebook, so they cannot be checked from its outputs; the build failure alone blocks the fits here.",
     "deviations": [],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig3c-integration-dimension",
     "statement": "The rSLDS reveals one integration dimension with a long time constant (110.7 +- 13.6 s) compared with the next (24.5 +- 5.1 s), forming an approximate line attractor; n = 15 mice, P = 6.5e-5.",
     "source": "Figure 3b,c,e, Results",
     "outcome": "blocked",
     "blocked_by": "other",
     "notebook": null,
     "notes": "Needs rSLDS fits to the 15 receptive sessions, which are in the dandiset; see fig3a-rslds-fit for why they cannot be run here. The Methods also find the integration dimension by partial least squares onto a ramp (Extended Data Figure 3d,e). That gives a direction but no time constants, so it cannot test this claim, and using it for the ramp and persistence claims would assume the ramp they test; it is not used.",
     "deviations": [],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed3s-integrator-model",
     "statement": "A single-state linear dynamical system driven by male sniff, mount and intromission predicts activity in the integration dimension with cvR2 of 0.88 +- 0.02 and an intrinsic time constant above 200 s.",
     "source": "Extended Data Figure 3r-u, Results",
     "outcome": "blocked",
     "blocked_by": "other",
     "notebook": null,
     "notes": "Needs the integration dimension from rSLDS fits; see fig3a-rslds-fit.",
     "deviations": [],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed4b-single-cell-ramps",
     "statement": "Of neurons contributing to the integration dimension, 56% correlate with an ideal ramp with r2 above 0.5, but they peak at different times.",
     "source": "Extended Data Figure 4a-d, Results",
     "outcome": "blocked",
     "blocked_by": "other",
     "notebook": null,
     "notes": "Needs the integration dimension from rSLDS fits; see fig3a-rslds-fit. The Methods do not define the ideal ramp either. An exploratory look, not judged, in the two natural-cycle mice with deposited weights, with a ramp chosen here (0 before male entry, rising linearly to the end of the annotation), finds r2 above 0.5 in 0 of 44 weighted cells in m541 and 1 of 14 in mouse 543, against 56%; these are not the paper's 15 receptive sessions.",
     "deviations": [],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig3k-perturbation-relaxation",
     "statement": "Transient optogenetic inhibition of VMHvl pushes population activity off the line attractor, after which it relaxes back to the attractor, as the flow field fit to unperturbed data predicts; n = 3 mice.",
     "source": "Figure 3h-k, Extended Data Figure 5c-j, Results",
     "outcome": "blocked",
     "blocked_by": "data_missing",
     "notebook": null,
     "notes": "The sessions with 10-s optogenetic inhibition of VMHvl are not in the dandiset; no label marks such an inhibition. The eight male-inhibition sessions have one or two inhibition epochs of 106 to 351 s, the male manipulation of Figure 4c,d. The authors' code repository carries one such session (sub-M1188) as example data, outside the archive.",
     "deviations": [],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig4b-persistence-during-intervals",
     "statement": "Activity in the integration dimension during copulation interbout intervals is as high as in the preceding bout, so a decoder cannot tell bouts from intervals (0.52 against a shuffle of 0.49, n = 4 mice, P = 0.2).",
     "source": "Figure 4a,b, Results",
     "outcome": "blocked",
     "blocked_by": "other",
     "notebook": null,
     "notes": "Needs the integration dimension from rSLDS fits; see fig3a-rslds-fit. The four mice of the figure are not identified.",
     "deviations": [],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig4d-persistence-male-inhibition",
     "statement": "When male mating is stopped for minutes by optogenetic activation of the male's VMH, activity in the female's integration dimension stays elevated; n = 4 mice.",
     "source": "Figure 4c,d, Results",
     "outcome": "blocked",
     "blocked_by": "other",
     "notebook": null,
     "notes": "The male-inhibition sessions are in the dandiset (8 mice, where the figure has 4), but the claim needs the integration dimension from rSLDS fits; see fig3a-rslds-fit.",
     "deviations": [],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig4h-peak-before-ejaculation",
     "statement": "Activity in the integration dimension peaks just before ejaculation and drops immediately after; n = 4 mice.",
     "source": "Figure 4g,h, Results",
     "outcome": "blocked",
     "blocked_by": "other",
     "notebook": null,
     "notes": "Twelve sessions annotate an ejaculation, but the claim needs the integration dimension from rSLDS fits; see fig3a-rslds-fit.",
     "deviations": [],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig5c-attractor-receptive-days",
     "statement": "Over the natural oestrus cycle, rSLDS models fit on receptive days have a line attractor and those fit on unreceptive days do not (line-attractor score 1.9 +- 0.2 against 0.05 +- 0.02, n = 4 mice, P = 0.02).",
     "source": "Figure 5b-f, Results",
     "outcome": "blocked",
     "blocked_by": "other",
     "notebook": null,
     "notes": "Needs rSLDS fits on each day; see fig3a-rslds-fit. The files also do not say which day of a natural-cycle series was receptive, and the female's behaviour is not annotated on those days; the receptive day can be found only by matching a day to the mouse's separately deposited receptive session. The matching rule (same count of male annotations, offsets within 0.5 s) finds it for 4 of the 7 mice (543, m541, m542, m551); m438 day 1 has the same frame count and interaction length as m438's receptive session but different male annotations (60 against 76), and m610 and m764 have no receptive session.",
     "deviations": [],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed6e-projection-unreceptive-days",
     "statement": "Projected onto the integration dimension found on the receptive day, activity on unreceptive days does not ramp and stays low during male mounting (normalised value 0.09 on unreceptive against 0.69 on receptive days, n = 4 mice).",
     "source": "Figure 5g,h, Extended Data Figure 6c-f, Results",
     "outcome": "partially_reproduced",
     "blocked_by": null,
     "notebook": "reproduce_liu_2024.ipynb",
     "notes": "With the authors' weights, the normalised integration value during mounting in m541 is 0.61 on the receptive day against 0.23 and 0.19 on the unreceptive days (paper 0.69 against 0.09; the unreceptive values here are about twice the paper's). In mouse 543 it is 0.61 against 0.64 and 0.63, not higher. The ramp part of the claim holds in both mice; the projection rises by 2.5 and 2.1 receptive-day SDs on the receptive day and by -0.4 to 0.5 on unreceptive days. The outcome depends on which factor is taken as the integration factor. On the other factor the receptive day is higher in both mice (m541 factor 1, 0.62 against 0.48 and 0.49; 543 factor 2, 0.66 against 0.54 and 0.51), which would meet the criterion in both, though on 543's factor 2 the unreceptive days also ramp (1.07 and 1.03 SDs).",
     "deviations": [
      "Two mice, where the paper has four; the weights of only two are deposited, and the file labelled m542 holds mouse 543's cells.",
      "The authors' weights are applied to the stored traces of each day; the paper's projection procedure and normalisation are not described, and min-max over the three days is used. The cell order was checked after the first round of verification; 171 of 176 cells (m541) and 71 of 72 (543) best match the session cell of the same index. The per-mouse files' own traces were processed differently (same-index median correlation 0.86 and 0.80), but the projections from them and from the stored traces correlate at 0.94 to 0.96.",
      "The integration factor is not labelled in the files; it is chosen by the longer ACHW of its receptive-day projection. Both factors reach the code's 50-s cap in m541, so the tie was broken with lags up to 300 s, a step not set out in the criterion. The outcome changes with the factor chosen.",
      "The receptive day is the one matching the mouse's deposited receptive session."
     ],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed7-achw-across-cycle",
     "statement": "Cells weighted on the integration dimension have shorter ACHW on unreceptive than on receptive days (16.1 +- 0.8 s against 25.2 +- 1.5 s, P < 0.001), while cells that do not contribute to it do not change.",
     "source": "Extended Data Figure 7a-h, Results",
     "outcome": "partially_reproduced",
     "blocked_by": null,
     "notebook": "reproduce_liu_2024.ipynb",
     "notes": "In both mice weighted cells have longer ACHWs on the receptive day, 28.4 against 17.2 s in m541 (44 cells, KS p = 2e-5) and 26.7 against 9.1 s in 543 (14 cells, KS p = 2e-4), against 25.2 and 16.1 s in the paper. Other cells do not differ in m541 (KS p = 0.23) but do in 543 (12.8 against 10.8 s, KS p = 0.005). The top 25% of weights give the same pattern; with the top 10% the 8 weighted cells of 543 no longer differ (p = 0.13). The change is not specific to the integration factor; in m541 the 44 cells above one SD on the other factor, 23 of them shared, lengthen as much (30.8 against 17.5 s, KS p = 7e-6). In 543 the 20 cells of the other factor, 7 shared, do not differ at p 0.05 (21.0 against 12.9 s, p = 0.07).",
     "deviations": [
      "Two mice where the paper has four, with the factor and receptive day chosen as for ed6e-projection-unreceptive-days."
     ],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed9j-ovx-hormone-priming",
     "statement": "Attractor dynamics disappear after ovariectomy and return after hormone priming.",
     "source": "Extended Data Figure 9j-n, Results",
     "outcome": "blocked",
     "blocked_by": "other",
     "notebook": null,
     "notes": "The ovariectomy series of three mice is in the dandiset, but the claim needs rSLDS fits on each day; see fig3a-rslds-fit. The deposited weights of m541 cover 176 cells, its ovariectomy days 184, so they cannot be applied.",
     "deviations": [],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "fig5k-receptivity-correlation",
     "statement": "In oestrogen-injected females, the area under the curve of the integration dimension correlates with the percentage of time spent in accepting behaviours (r2 = 0.62, P < 0.0001, n = 50 trials), and not with resistance or appetitive behaviours.",
     "source": "Figure 5i-l, Extended Data Figure 10b, Results",
     "outcome": "blocked",
     "blocked_by": "other",
     "notebook": null,
     "notes": "The oestrogen-injected cohort is in the dandiset (42 trials where the paper has 50), but the claim needs an rSLDS fit per trial; see fig3a-rslds-fit.",
     "deviations": [],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    },
    {
     "claim": "ed10c-population-mean-uncorrelated",
     "statement": "In the same trials, the area under the curve of the population mean of all neurons does not correlate with the percentage of time spent in accepting behaviours.",
     "source": "Extended Data Figure 10c,d, Results",
     "outcome": "reproduced",
     "blocked_by": null,
     "notebook": "reproduce_liu_2024.ipynb",
     "notes": "Over 42 trials from 6 mice the area under the population mean does not correlate with the percentage of time in accepting behaviour under any of three readings (r2 0.015, 0.003 and 0.027; p 0.45, 0.73 and 0.30). Within mice the Spearman correlations run from -0.67 to 0.52. Without the two sessions that fail the clock check (m928 day 2, m940 day 7), r2 is 0.011, 0.005 and 0.025 over 40 trials.",
     "deviations": [
      "42 trials where the paper has 50.",
      "The window of the area under the curve is not stated; three readings were set out before the analysis."
     ],
     "disagreed": false,
     "data": "dandiset",
     "where": null,
     "proposed": null
    }
   ],
   "findings": [
    {
     "kind": "data_missing",
     "statement": "No rSLDS fits or latent factors are deposited apart from the per-cell weights of two mice, though the dandiset's description says it contains low-dimensional latent factors from dynamical models. The per-mouse files hold under analysis only a table of two weight columns, and under processing one trace series and the annotations; there are no latent time series or dynamics matrices. The paper's central line-attractor results can only be checked by refitting."
    },
    {
     "kind": "data_missing",
     "statement": "The sessions with optogenetic inhibition of VMHvl during imaging (Figure 3h-k, Extended Data Figure 5) are not in the dandiset."
    },
    {
     "kind": "metadata_error",
     "statement": "The per-mouse file sub-m542/sub-m542_behavior.nwb is labelled m542 but its neural data are those of mouse 543 (72 cells and 10,527 frames, matching 543's receptive day cell by cell with a median correlation of 0.80); m542's own sessions have 129 cells."
    },
    {
     "kind": "identifier_missing",
     "statement": "One mouse is labelled 543 in its natural-cycle files and m543 in its receptive session."
    },
    {
     "kind": "metadata_error",
     "statement": "The same recording is filed under two experiments five times. m764's natural-cycle day 1 has the same traces and annotations as its primed ovariectomy session; m776's primed session has the same traces as its male-inhibition session, with different annotations; and day 3 of the ovariectomy series of m438, m541 and m551 has the same traces as the mouse's primed session, with annotations that differ in m541 and m551 and an interaction of 290.7 against 274.6 s in m438."
    },
    {
     "kind": "other",
     "statement": "The traces in the two per-mouse files, which carry the rSLDS weights, were processed differently from the session files of the same recordings (same-index median correlation 0.86 and 0.80, rising with smoothing of the session traces), and the files do not say how. The cell order matches the session files."
    },
    {
     "kind": "other",
     "statement": "Which natural-cycle days were receptive is not recorded, and the female's behaviour is not annotated on most of those days, so the paper's definition of a receptive day cannot be applied. The receptive session of each mouse is deposited separately, without a link to the day it repeats."
    },
    {
     "kind": "selection_undocumented",
     "statement": "The dandiset holds more animals or fewer trials than several analyses use, without saying which were used; 7 natural-cycle mice where Figure 5c has 4, 8 male-inhibition mice where Figure 4d has 4, 42 oestrogen-injected trials where Figure 5k has 50, and 69 annotated trials from 23 mice where Figure 1 has 74 from 28."
    },
    {
     "kind": "other",
     "statement": "In 17 sessions imaging continues for up to 7 minutes after the last annotation, which may be the pencil-cup period of the Methods, but nothing marks it. The labels other and eat, and the two names for the male-inhibition epoch (laser_on and mating_inhibition), are not described."
    },
    {
     "kind": "metadata_error",
     "statement": "Every session file describes a two-photon microscope with 920-nm excitation, where the paper used a one-photon miniscope; the raw traces are labelled lumens but are arbitrary units with negative values; the traces described as z-scored are not z-scored per file (per-cell means -1.6 to 2.4); experimenter, lab, institution and session start time are placeholders; image masks are 1 by 1 pixel; and the file names say ecephys for imaging data. The per-mouse files describe their session as a resident-intruder assay."
    },
    {
     "kind": "methods_unavailable",
     "statement": "The linked code holds demonstration notebooks, not the code for the paper's figures, so the GLM, choice-probability, decoder and projection procedures cannot be checked against it."
    }
   ],
   "verifications": [
    {
     "date": "2026-10-14",
     "kind": "agent",
     "by": "Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook)",
     "reviewed": "The first executed notebook with the conclusion cells and the summary removed, the helper module, the claims file and the paper's full text (PMC11499253). For the blocked claims the verifier checked the archive's asset list and metadata through the DANDI REST API. It did not run the notebook or stream any file.",
     "agreed": 24,
     "of": 24,
     "disagreements": []
    },
    {
     "date": "2026-10-14",
     "kind": "agent",
     "by": "Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook)",
     "reviewed": "The full re-executed notebook after the first round (79 cells) with its conclusions and summary, the README, the attempt record, the helper module, the claims file and the paper's full text (PMC11499253, re-downloaded from Europe PMC). It did not run the notebook or stream any file.",
     "agreed": 24,
     "of": 24,
     "disagreements": []
    }
   ]
  }
 ],
 "entries": [
  {
   "name": "001097-liu-2024",
   "type": "reproduction",
   "title": "A hypothalamic line attractor for female mating (Liu, Nair et al. 2024)",
   "description": "Tests the findings of Liu, Nair et al. (2024) on dandiset 001097, which holds miniscope imaging of VMHvl alpha-cells in female mice during mating, with annotations of male and female behaviour. It tests the behavioural dynamics of mating, single-cell tuning, GLM fits and persistence, and, with the authors' deposited rSLDS weights for two mice, how the receptive-day integration dimension and its cells behave across the oestrus cycle. Claims that need new rSLDS fits are blocked, as the model cannot be built or fit within the registry's environment and time limit.",
   "dandisets": [
    "001097"
   ],
   "papers": [
    "10.1038/s41586-024-07916-w"
   ],
   "derived_from": [],
   "notebooks": [
    {
     "path": "reproduce_liu_2024.ipynb",
     "title": "Reproducing Liu, Nair et al. (2024) from dandiset 001097",
     "executed": "notebooks/001097-liu-2024/reproduce_liu_2024.html"
    }
   ]
  }
 ]
}