Functional connectomics spanning multiple areas of mouse visual cortex (DANDI:000402)

Claims

From the paper Functional connectomics spanning multiple areas of mouse visual cortex

Reproducing the MICrONS Consortium (2025) from dandiset 000402 13

Notebook reproduce_microns_2025.ipynb

Reproduced10

ReproducedResults, 2P calcium imaging, paragraph 1; Methods, 2P functional imaging
The calcium imaging data comprise 14 scans of one mouse, collected between P75 and P81, spanning a volume of about 1,200 x 1,100 x 500 um.
The dataset contains 14 individual scans, collected between postnatal day 75 (P75) and P81, spanning a volume of approximately 1,200 × 1,100 × 500 µm 3
Results, 2P calcium imaging, paragraph 1; Methods, 2P functional imaging
The 14 scans of Extended Data Table 1 are all in the dataset, recorded on 6 days (4 to 9 March 2018), which are P75 to P80 against the birth date of 19 December 2017. The dataset holds 19 scans, the 19 completed scans of the Methods. Each scan's fields tile 1,200 x 1,100 um (1,200 x 1,000 um in scan 9-6) at depths of 80 to 570 um; the files give no common frame for the scans, so the extent of the whole volume is not measured.
How the test differed from the paper’s method
- Which 14 of the 19 files are the paper's scans was taken from Extended Data Table 1 (Methods, 2P functional imaging, "Of 19 completed scans over 6 days of imaging, 14 are described here (Extended Data Table 1)"). The text the claims file was read from holds only the table's caption; its cells were read from the table's image in PMC11981939 (41586_2025_8790_Tab1_ESM.jpg). They are 4-7, 5-6, 5-7, 6-2, 6-4, 6-6, 6-7, 7-3, 7-4, 7-5, 8-5, 9-3, 9-4 and 9-6.
ReproducedMethods, 2P functional imaging, paragraph 1; Results, 2P calcium imaging, paragraph 2
Eleven scans were imaged at 6.3 Hz with four depths (eight fields), two at 8.6 Hz with three depths (six fields) and one at 9.6 Hz with two depths (four fields); in the 11 scans the four planes span roughly 300 to 400 um.
For 11 scans, imaging was performed at 6.3 Hz, collecting eight 620 × 1,100 μm fields per frame ... For 2 scans, imaging was performed at 8.6 Hz, collecting six 620 × 1,100 μm fields per frame ... For 1 scan, imaging was performed at 9.6 Hz, collecting four 620 × 1,000 μm fields per frame
Methods, 2P functional imaging, paragraph 1; Results, 2P calcium imaging, paragraph 2
Of the 14 scans, 11 are at 6.30 Hz with 4 depths and 8 fields, 2 at 8.61 Hz with 3 depths and 6 fields, and 1 at 9.62 Hz with 2 depths and 4 fields, two fields per depth in every scan. The depth spans of the 11 scans are 280 to 420 um, all within 270 to 440 um (reading a); their median is 420 um (reading b, given as measured). In scan 8-5 two of the four depths are 10 um apart (210 and 220 um). All 19 files fall in one of the three groups (16 at 6.3 Hz). The table of fields by depth shows exactly two fields at every depth in all 19 files.
ReproducedResults, 2P calcium imaging, paragraph 2; Methods, 2P functional imaging, paragraph 3
In total 125,413 masks were segmented across the 14 scans, of which 115,372 were classified as somatic by a trained classifier (8.1% excluded).
In total, 125,413 masks were generated across 14 scans, of which 115,372 were automatically classified as somatic masks by a trained classifier
Results, 2P calcium imaging, paragraph 2; Methods, 2P functional imaging, paragraph 3
The dataset holds every mask with a soma or artifact label (reading a). Over the 14 scans there are 125,413 masks, 115,372 somatic, so 8.01% are excluded; both counts equal the paper's. Over all 19 files the counts are 168,971 and 155,153.
ReproducedMethods, Stimulus composition; Results, Behavioural tracking and visual stimulation, paragraph 1
Each scan's stimulus lasted about 84 min and was made of 6 oracle clips of 10 s repeated 10 times (10 min), 144 unique natural clips of 10 s (24 min), 90 natural clips of 10 s shown twice (30 min), and 20 Trippy and 20 Monet2 seeds of 15 s each shown twice (10 min each).
Each scan stimulus was approximately 84 min in duration and comprised
Methods, Stimulus composition; Results, Behavioural tracking and visual stimulation, paragraph 1
All 14 scans have 384 clip trials (6 clips shown 10 times, 90 shown twice, 144 once), and 20 Monet2 and 20 Trippy seeds shown twice each; clips last 10 s and parametric trials 15 s. The total is 83.76 min from the trial times (reading a, given as measured) and 84.0 min nominal (reading b). Each category (cinematic, Sports-1M, rendered) has 2 oracle, 30 repeated and 48 unique clips in every scan, as the Methods give.
ReproducedResults, Behavioural tracking and visual stimulation, paragraph 2
The six oracle clips were repeated in the same order ten times per scan.
6 natural film stimuli clips totalling 1 min (oracle natural videos) were repeated in the same order 10 times per scan
Results, Behavioural tracking and visual stimulation, paragraph 2
In every scan the 10 repeats each hold the 6 oracle clips consecutively and in one order (one distinct order per scan), with at most 0.633 s between clips, and no trial of any other stimulus starts inside a repeat. The 6 oracle clips are the same in all scans, but their order differs between scans (14 distinct orders), which the claim does not address.
ReproducedMethods, Global directional parametric stimulus; Fig. 3g legendverifier disagreed at first
Each 15 s Monet2 block holds 16 equal periods of motion, one for each of 16 directions between 0 and 360 degrees, at 42.8 degrees per second.
Each 15-s block consisted of 16 equal periods of motion along one of 16 unique directions of motion between 0–360° with a velocity of 42.8° s −1 at the nearest point on the monitor
Methods, Global directional parametric stimulus; Fig. 3g legend
Every Monet2 trial has num_directions 16 and duration 15 s in the table. From the video, the period fitted to the direction changes is 929.6 to 945.4 ms (median 937.1) in all 560 trials, within tolerance. With the direction of each period from the global optical flow, all 560 trials have 16 distinct directions, both at least 10 degrees apart and after snapping to the 22.5 degree grid. The 8,960 measured directions lie within 7.9 degrees of the grid (99% within 5.0 degrees; 36 more than 6.25 degrees off). Both showings of every seed give the same 16 directions. The phase correlation estimate of the first round gave 15 directions in 4 trials (10 degree rule) or 6 (grid), with deviations up to 11.0 degrees, in 8 trials (both showings of seeds 7637 in scan 5-6, 5145 and 2618 in scan 6-7 and 6154 in scan 9-6); in all 8 the optical flow hits all 16 grid directions, and the period the phase correlation put at 124.9 degrees in seed 7637 measures 111.3. The spacing is the 22.5 degree grid. The speed is not in the dataset and is not tested.
How the test differed from the paper’s method
- The table gives no direction or onset per period, so both were measured from the stimulus video in each file; a method of this attempt, not the paper's. Boundaries come from phase correlation between successive frames, and each period's direction from the global optical flow over the period (gradient method of Lucas and Kanade, frames smoothed by a 1 pixel Gaussian). The boundary detection, the 10 degree tolerance for distinct directions, the smoothing and the 0.1 s left out at each end of a period are standard choices, stated in the helper.
ReproducedFig. 3f; Results, Behavioural tracking and visual stimulation, paragraph 2; Methods, Oracle score
Neurons differ in how reliably they respond to the repeated oracle clips; units with high oracle scores give repeat-to-repeat responses that are consistent, and units with low scores do not.
Raster of deconvolved calcium activity for three neurons to repeated stimulus trials (oracle trials; ten repeats of six sequential clips, with each repeat normalized independently). Rasters for high (top), medium (middle) and low (bottom) oracle scores
Fig. 3f; Results, Behavioural tracking and visual stimulation, paragraph 2; Methods, Oracle score
The score is computed for all 115,372 somatic masks of the 14 scans under both readings. The units at the 34th, 79th and 99th percentiles score 0.208, 0.396 and 0.702 under reading (a) and 0.151, 0.275 and 0.465 under reading (b), increasing, with the 99th above 0. In the variant, the mean pairwise correlation between single repeats of the 100 units nearest each percentile is 0.084, 0.203 and 0.524 (a) and 0.096, 0.205 and 0.394 (b). Median scores by scan range from 0.208 to 0.348. The criterion is met by any score whose 99th percentile is above 0, so it tests little. The 10,041 artifact masks score 0.235, 0.402 and 0.650 at the three percentiles, as high as the somatic masks, and the mean trace of a field scores 0.499 (median of 104 fields, range 0.161 to 0.780). With each mask's fit on its field's mean trace removed, the somatic scores are 0.200, 0.380 and 0.665 and the artifact scores 0.229, 0.380 and 0.613, barely changed, so a signal shared across the field does not drive them. What makes the artifact masks score as high is not shown (a local neuropil signal or real neurites could account for it), and the score on fluorescence cannot separate the reliability of a neuron from that of what else its mask holds.
How the test differed from the paper’s method
- Fluorescence traces replace the deconvolved traces of the paper, which are not in the dataset.
- Frames are taken from each field's time stamps from the onset of each repeat; the per-unit delay within a frame that the paper's code adds is not in the dataset and is not applied. All fields of a scan share one array of time stamps in the files.
ReproducedFig. 3g; Results, Behavioural tracking and visual stimulation, paragraph 2
Monet2 drives direction-tuned responses; neurons can be found preferring each of the 16 directions, and their trial-averaged responses peak at their preferred direction.
Trial-averaged raster (central 500 ms of trial-average raster for each direction, out of 937 ms) of deconvolved calcium activity for 80 neurons in 40 Monet2 trials (16 randomly ordered directions) grouped by preferred direction (5 neurons per direction
Fig. 3g; Results, Behavioural tracking and visual stimulation, paragraph 2
Every direction is preferred by at least 5,205 somatic masks under reading (a) and 5,084 under reading (b), and by at least 4,698 with the windows shifted by +0.47 s, where the split-half reliability of the tuning peaks in all 104 fields. The count tests little with this many units; the argmax of the difference between the two halves of the trials, which holds no shared tuning, puts 6,153 to 8,217 units on each direction. The split-half variant gives the evidence of tuning. Over all somatic masks the response at the first-half preferred direction exceeds the mean of the other 15 in the second half for 62.9% (0.5 without tuning), and for 63.7% of 80 units chosen on the first half alone. The 80 units shown are chosen on all 40 trials, so their 91.2% is raised by the selection. The test does not separate neural tuning from a stimulus-locked signal that masks share. The artifact masks pass it as often (65.9% of 10,041), and the most common first-half preferred direction of a field holds a median 11.3% of its somatic masks (8.6% to 17.7% over 104 fields), against 6.25% if spread evenly.
How the test differed from the paper’s method
- Fluorescence replaces the deconvolved traces, which are not in the dataset.
- The direction of each period comes from the video (see monet2-directions), snapped to the 22.5 degree grid.
- The 80 units shown are ranked by the preferred response minus the mean over directions, divided by the standard deviation over directions, in place of the claims file's ratio, which is not defined for fluorescence with its arbitrary offset. This choice only picks the units shown.
ReproducedResults, Automated reconstruction, paragraph 2data in CAVE table nucleus_detection_v0, datastack minnie65_phase3_v1 (MICrONS Explorer); the paper names no materialization version
Nuclei were automatically segmented in subvolume 65, 144,120 in all.
Nuclei were also automatically segmented ( n = 144,120) within subvolume 65
Results, Automated reconstruction, paragraph 2
The released nucleus_detection_v0 table holds 144,120 detections, all at sections 14,834 to 27,871, within subvolume 65.
ReproducedMethods, Cell classificationdata in CAVE tables nucleus_neuron_svm and the cell class tables of datastack minnie65_phase3_v1 (MICrONS Explorer); the paper names no materialization version
An SVM on nucleus features detects neurons with 96.9% precision and 99.6% recall, predicting 82,247 neurons in subvolume 65, of which 69,957 have well-segmented somata; a multilayer perceptron separates excitatory from inhibitory neurons with 97% accuracy (96% precision and 94% recall for inhibitory calls).
This model predicted 82,247 neurons detected within the larger subvolume ... We estimate from this test that the classifier had an overall accuracy of 97% with an estimated 96% precision and 94% recall for inhibitory calls
Methods, Cell classification
Of the 144,120 nuclei of subvolume 65, 144,117 are in the SVM table and 89,056 are called neurons, within the tolerance of the paper's 82,247 (8.3% higher). The excitatory and inhibitory table holds 69,957 neurons, the paper's count of well-segmented neurons, all among the SVM neurons. Against the manual labels of the 100 um column (2,029 cells labelled neuron or non-neuron, matched by position at 0 um), the SVM has precision 95.5% and recall 99.6%. The multilayer perceptron's accuracy was measured by the paper on 350 further cells that are not released, so it is not tested; on the column cells, its training and validation data, the calls agree with the labels for 98.2%. The paper's 82,247 equals 69,957 plus 12,290, while the released SVM table calls 89,056 neurons, so the count passes on the tolerance only.
How the test differed from the paper’s method
- The tables are the 2021 releases on BossDB, not the CAVE tables the claims file names; the paper gives no version.
- The SVM table holds 27,698 rows that are not nuclei of nucleus_detection_v0 (27,692 called non-neurons); the count is restricted to subvolume 65's nuclei, as the paper's is.

Partially reproduced3

Partially reproducedMethods, Stimulus composition; Results, Behavioural tracking and visual stimulation, paragraph 2
The oracle clips, the 90 twice-repeated clips and 10 of the 20 seeds of each parametric stimulus were the same in every scan; the 144 unique clips and the other 10 seeds differed between scans.
The stimulus composition included a mixture of unique stimuli for each scan, some that were repeated across every scan, and some that were repeated within each scan
Methods, Stimulus composition; Results, Behavioural tracking and visual stimulation, paragraph 2
The 96 clips shown 2 or 10 times in a scan occur in all 14 scans and the 2,016 clips shown once each occur in one scan only. 10 Monet2 and 10 Trippy seeds occur in all 14 scans, and the other 140 Monet2 seeds each occur in one scan. One Trippy seed (6831, condition hash Xe6/vzamVqYgecKk7y51) is shown twice in each of scans 6-4 and 7-5; the other 138 non-conserved Trippy seeds each occur in one scan. The criterion's requirement that every remaining seed occur in one scan fails for this seed.
Partially reproducedMethods, Stimulus composition, last paragraphverifier disagreed at first
Each scan was preceded by 0.15 to 5.5 min with the monitor on and followed by 8.3 to 21.2 min with the monitor off, to record spontaneous activity.
Each scan was also preceded by 0.15–5.5 min with the monitor on, and followed by 8.3–21.2 min with the monitor off, in order to collect spontaneous neural activity
Methods, Stimulus composition, last paragraph
Over the 14 scans the time from the first frame to the first stimulus is 0.145 to 4.841 min and from the last stimulus to the last frame 8.93 to 21.19 min, so every scan lies within the widened ranges, the criterion's test. The bounds of the range over the 14 scans do not all match the expected values; the longest period before the stimulus, 4.841 min, is outside 5.5 +- 0.55 min, while 0.145, 8.93 and 21.19 min are within tolerance of 0.15, 8.3 and 21.2. Over all 19 files the four bounds are 0.145, 5.539, 8.255 and 21.193 min, all within tolerance, so the paper's range was probably taken over all 19 recordings. The dataset does not say whether the monitor was on or off in these periods.
Partially reproducedResults, Behavioural tracking and visual stimulation, paragraph 1; Methods, Eye and face camera; Methods, Treadmill
Treadmill velocity, and the pupil position and diameter from video of the left eye, were recorded throughout each scan.
Treadmill rotation (single axis) and video of the left eye were captured throughout the scan, yielding locomotion velocity, eye movements and pupil diameter data
Results, Behavioural tracking and visual stimulation, paragraph 1; Methods, Eye and face camera; Methods, Treadmill
All 14 scans hold treadmill velocity (57.0 to 100.2 Hz) and pupil position and major and minor radii (19.96 to 20.01 Hz). The treadmill covers the imaging period in all 14 scans; the eye tracking covers it in 13, and in scan 4-7 it ends 968 s before the last imaging frame (85% coverage). File 4-7 holds one eye position series and two pupil radius series, all ending at 5,392.9 s, so no other series covers the rest of the scan. A pupil is fitted in 84.9% to 99.9% of eye frames. By the same 1 s rule, the eye tracking misses the imaging period in all 5 files outside the paper's 14. It ends 3,264 to 4,582 s early in 4-10, 5-3 and 8-7, and 3.9 and 6.2 s early in 4-9 and 8-9 (in 4-9 it also starts 1.77 s late).

EM sections and synapse counts of the MICrONS volume (MICrONS Consortium 2025) 2

Notebook em_volume_counts.ipynb

Partially reproduced2

Partially reproducedResults, The EM volume, paragraphs 1 and 2verifier disagreed at firstquestion to the readerdata in BossDB project microns-minnie (https://bossdb.org/project/microns-minnie) and MICrONS Explorer (https://www.microns-explorer.org/cortical-mm3); no version is given
The block was cut into 27,972 sections of 40 nm, 26,652 were imaged by five autoTEMs, and an 800 um region of sections 7,931 to 27,904 was processed, split into two subvolumes of about 35% and 65% of the sections.
The tissue sample was trimmed and sectioned into 27,972 serial sections (nominal thickness 40 nm) ... A total of 26,652 sections were imaged by 5 customized automated TEMs
Results, The EM volume, paragraphs 1 and 2
The imagery is public on BossDB in sharded precomputed format, and em_volume_counts.ipynb reads it. The metadata give bounds in whole chunks of 64 sections, 7,904 to 14,847 for subvolume 35 and 14,816 to 27,903 for subvolume 65, 20,000 sections with a share of 34.56% before subvolume 65. Sampling the chunk layers at each end on a grid of 2,288 chunks at 64 nm per pixel, the imagery holds data from section 7,931 to 14,847 in subvolume 35 and from 14,825 to 27,882 in subvolume 65; a sparser grid and 16 chunks at 16 nm give the same last section. The 8 nm scale was not checked. The first section, 7,931, matches the paper. The region with data is 19,952 sections against the expected 19,974 (tolerance 0), so the count fails. The share of subvolume 35 is 34.47% to 34.67% under the four definitions printed (34.55% with subvolume 35 up to the first section of subvolume 65, given as measured), within 35 +- 3.5%, so the share passes. The metadata give no count of sections cut or imaged. The paper gives section indices, not a count, and 27,904 is the metadata's exclusive end of subvolume 65 (14,816 + 13,088), so the paper's upper bound may be exclusive or a bounding box; this is asked of a reader in questions/000402.yaml. In the first two verification rounds the outcome was blocked (too_large); round 3 agreed with the change to partially_reproduced.
How the test differed from the paper’s method
- The first and last sections with data are read from the imagery at 64 nm per pixel on a grid of every 8th chunk in x and y (2,288 chunks a layer), a choice of this attempt; the release gives no list of processed sections.
Partially reproducedResults, Automated reconstruction, paragraph 3; Extended Data Fig. 1verifier disagreed at firstquestion to the readerdata in CAVE synapse tables of minnie65 and minnie35 (MICrONS Explorer, https://www.microns-explorer.org/cortical-mm3); the paper names no materialization version and does not say whether the 8,611 manual annotations are released
Automated detection found 524 million synaptic clefts (186 million in subvolume 35, 337 million in subvolume 65), with 96% precision and 89% recall against 8,611 manually annotated synapses in 70 subvolumes, and 98% partner assignment accuracy on 191 held-out synapses.
We automatically detected and associated a total of 524 million synaptic clefts across both subvolumes ... We manually identified synapses in 70 small subvolumes ( n = 8,611 synapses) distributed across the dataset, giving the automated detection an estimated precision of 96% and recall of 89%
Results, Automated reconstruction, paragraph 3; Extended Data Fig. 1
em_volume_counts.ipynb streams each table once and resumes a broken stream at the byte reached. The export of subvolume 65 at materialization 1181 (synapse_graph/cave_exports/v1181/synapses_pni_2_v1_filtered_view.csv.gz, 20.0 GB, no header line) holds 337,312,429 synapses (expected 337 +- 33.7 million), and the table of subvolume 35 (minnie35/synapse_graph/assigned.csv.gz, 13.8 GB, 2021, one header line) holds 186,268,895 (expected 186 +- 18.6 million); together 523.6 million against the paper's 524 million. Both counts pass. No manual synapse annotations were found on the BossDB prefixes searched, so precision, recall and F1 are not tested, and measured gives the two counts only. The 98% partner assignment accuracy on 191 held-out synapses in the statement has no expected quantity and is not tested either. In the first two verification rounds the outcome was blocked (too_large); round 3 agreed with the change to partially_reproduced.
How the test differed from the paper’s method
- The paper names no materialization version. Subvolume 65 is counted on the export of version 1181, the version the paper's code reads; subvolume 35 on its only public table, assigned.csv.gz (2021).

Not tested in a notebook 21

4 data missing from the dataset, 1 metadata insufficient, 16 other

Blocked21

BlockedResults, 2P calcium imaging, paragraph 4; Methods, 2P structural stack, paragraph 2· Data missing from the dataset
Consolidating masks imaged in more than one scan by their 3D position in the structural stack gives an estimated 75,909 unique functionally imaged neurons.
we estimate the functional imaging volume contains 75,909 unique functionally imaged neurons consolidated from 115,372 segmented somatic masks, with many neurons imaged in 2 or more scans
Results, 2P calcium imaging, paragraph 4; Methods, 2P structural stack, paragraph 2
The method groups masks by their 3D centroids in the structural stack. The dataset holds neither the stack nor the registration of the fields into it nor any stack coordinate of a mask; each file gives its field origins on its own axes only. The EM root ids of the dataset cover half the fields and cannot stand in for the grouping. The BossDB open data bucket serves a structural stack of this mouse (functional_data/two_photon_structural_stacks/two_photon_stack_17797_9_19_v1.tiff), and the paper lists the 2P structural images among its public resources (Fig. 1 legend). The claims file limits this claim to the dandiset, so the stack was not used, and a reader is asked whether the claim should be data elsewhere (questions/000402.yaml).
BlockedMethods, Imaging site selection, paragraph 1; Methods, 2P structural stack, paragraph 3; Fig. 3a,d· Metadata insufficient
The imaging volume is roughly 50% VISp and 50% higher visual areas (VISlm, VISrl and VISal), with each functional unit assigned an area from its position in the stack.
This resulted in an imaging volume that was roughly 50% VISp and 50% higher visual area (HVA)
Methods, Imaging site selection, paragraph 1; Methods, 2P structural stack, paragraph 3; Fig. 3a,d
No mask or field carries a visual area. The location of every imaging plane is the same list, "VISp,VISrl,VISlm,VISal", and the area masks and sign map the Methods use are not in the dataset.
BlockedMethods, 2P structural stack, paragraph 2· Data missing from the dataset
All scans drifted less than 10 um in z over the recording, and most less than 5 um.
All scans had less than 10-μm drift over the 1.5-h recording, and for most of them drift was limited to <5 μm
Methods, 2P structural stack, paragraph 2
Drift was measured by registering sections of each scan into the structural stack, which is not in the dataset. The raw frames are there, but without the stack there is nothing to register them to in z. The BossDB open data bucket serves a structural stack of this mouse (functional_data/two_photon_structural_stacks/two_photon_stack_17797_9_19_v1.tiff), and the paper lists the 2P structural images among its public resources (Fig. 1 legend). The claims file limits this claim to the dandiset, so the stack was not used, and a reader is asked whether the claim should be data elsewhere (questions/000402.yaml).
BlockedResults, 2P calcium imaging, paragraph 3; Methods, 2P structural stack, paragraph 1· Data missing from the dataset
A two-channel structural stack (green GCaMP and red vasculature) of 1,412 x 1,322 x 670 um at 0.5 pixels per um was collected after Texas Red injection.
a 2-channel (green, red) 1,412 × 1,322 × 670 μm 3 (anteroposterior × mediolateral × radial depth) structural stack was collected at 0.5 pixels per μm
Results, 2P calcium imaging, paragraph 3; Methods, 2P structural stack, paragraph 1
The dataset holds no structural stack, in any file or as a separate asset (19 assets, one per functional scan). The BossDB open data bucket serves a structural stack of this mouse (functional_data/two_photon_structural_stacks/two_photon_stack_17797_9_19_v1.tiff), and the paper lists the 2P structural images among its public resources (Fig. 1 legend). The claims file limits this claim to the dandiset, so the stack was not used, and a reader is asked whether the claim should be data elsewhere (questions/000402.yaml).
BlockedMethods, Stimulus alignment· Data missing from the dataset
The binary flip code of the photodiode signal, used to align stimulus and behaviour clocks, was recovered for 89% of trials.
The encoded binary signal was reconstructed for 89% of trials
Methods, Stimulus alignment
The files hold the aligned stimulus times only, with no photodiode signal and no per-trial flag of whether the flip code was read.
BlockedResults, Functional–structural co-registration, paragraph 2· Otheruses data from outside the dataset
Experts manually matched 19,181 functional ROIs from 14 scans to 15,439 individual EM neurons.
we used the results of the transform to guide a group of experts to manually match 19,181 functional ROIs from 14 scans to 15,439 individual EM neurons
Results, Functional–structural co-registration, paragraph 2
The match table is the CAVE table coregistration_manual_v4. CAVE serves its tables only to users with an account token (the materialization API redirects to an authorization page), and this run has none. The BossDB bucket holds only a 2021 match release (func_unit_em_match_release.csv, 14 kB), not this table. As the variant the claims file names, the notebook counts the EM identifiers the dataset holds. Only 52 of the 104 fields of the 14 scans have EM columns (one of the two fields at each depth), holding 8,863 ROIs with a root id (8,326 distinct values, among them 0 in 2 ROIs; 395 non-zero root ids occur in more than one scan), so they are not comparable with the paper's 19,181 and 15,439.
BlockedResults, Functional–structural co-registration, paragraph 2; Methods, Evaluating manual matches; Extended Data Fig. 4f–h· Otheruses data from outside the dataset
Units independently matched to the same EM neuron in two scans have higher signal correlation than each matched unit with the nearest unmatched unit in the other scan, more clearly for units with oracle score above 0.2.
ROIs with at least moderate visual responses that are independently matched to the same neuron across multiple scans have higher signal correlations than adjacent neurons
Results, Functional–structural co-registration, paragraph 2; Methods, Evaluating manual matches; Extended Data Fig. 4f–h
Depends on manual-coreg-19181, whose CAVE table needs an account token. As a variant that decides nothing, the notebook runs the criterion's test under reading (b) (nearest somatic mask in the same field, by the 2D centroid of the image masks) on the matches the dataset holds, masks of different scans with the same pt_root_id; reading (a) needs stack coordinates, which the dataset lacks. Over 538 pairs of scans (395 root ids) the matched signal correlation on fluorescence (median 0.585) exceeds both controls (medians 0.239 and 0.233) in 71.9% of pairs, and the median differences, 0.273 and 0.273 over all pairs, are 0.303 and 0.353 when both masks score above 0.2 (77.7% of 399 pairs above both controls). Two-sided Wilcoxon p is below 1e-43 in every case. In 50 pairs the EM cell body points of the two masks lie more than 10 um apart; without them the shares are 75.2% (488 pairs) and 80.4% (368). On the dataset's own matches the test would pass, but these matches cover one field per depth and are not the paper's table.
BlockedResults, Functional–structural co-registration, paragraph 3; Extended Data Fig. 5· Otherdata in CAVE tables coregistration_auto_phase3_fwd and coregistration_manual_v4, datastack minnie65_phase3_v1 (MICrONS Explorer, https://www.microns-explorer.org/cortical-mm3#f-coreg); the paper names no materialization version, its code uses 1181
The fiducial-based automatch assigns 84,198 functional ROIs to 37,364 EM neurons with 83% precision against manual matches; dropping the bottom 30% of separation scores gives 90% precision with 59,934 ROIs and 31,042 neurons.
This method (referred to as the fiducial-based automatch table) resulted in 84,198 functional ROIs matched to 37,364 EM neurons. Considering all matches, this method achieved 83% precision relative to manual matchers, but filtering out matches in the bottom 30% of separation scores yields 90% precision, while still including 59,934 functional ROIs and 31,042 EM neurons
Results, Functional–structural co-registration, paragraph 3; Extended Data Fig. 5
The CAVE tables coregistration_auto_phase3_fwd and coregistration_manual_v4 need an account token, which this run has none of; no public copy of them was found on the BossDB bucket.
BlockedResults, Functional–structural co-registration, paragraph 4; Extended Data Fig. 5· Otherdata in CAVE tables apl_functional_coreg_vess_fwd and coregistration_manual_v4, datastack minnie65_phase3_v1 (MICrONS Explorer, https://www.microns-explorer.org/cortical-mm3#f-coreg); the paper names no materialization version, its code uses 1181
The vessel-based automatch, without fiducials, assigns 75,856 functional ROIs to 34,712 EM neurons with 84% precision; dropping the bottom 30% of separation scores gives 90% precision with 53,248 ROIs and 28,233 neurons.
This table (referred to as the vessel-based automatch table) contains 75,856 functional ROIs matched to 34,712 EM neurons. Remarkably, this fiducial-free method performed as well as the fiducial-based method, achieving 84% precision with manual matches
Results, Functional–structural co-registration, paragraph 4; Extended Data Fig. 5
As for fiducial-automatch; the CAVE table apl_functional_coreg_vess_fwd needs an account token.
BlockedResults, Functional–structural co-registration, paragraph 5; Methods, Generating the fiducial-vessel agreement automatch table· Otherdata in CAVE tables coregistration_auto_phase3_fwd, apl_functional_coreg_vess_fwd and coregistration_manual_v4, datastack minnie65_phase3_v1 (MICrONS Explorer); the paper names no materialization version, its code uses 1181
Keeping only the matches on which both automated methods agree gives 89% agreement with manual matches, with 60,091 ROIs and 29,620 EM neurons.
this hybrid automated table achieves 89% agreement with no additional filtering, yielding 60,091 functional ROIs and 29,620 EM neurons
Results, Functional–structural co-registration, paragraph 5; Methods, Generating the fiducial-vessel agreement automatch table
As for fiducial-automatch.
BlockedResults, Functional–structural co-registration, paragraph 1; Methods, Co-registration Transform· Otherdata in the fiducial points released with the MICrONS resource (MICrONS Explorer, https://www.microns-explorer.org/cortical-mm3) and the 2P structural stack; the paper gives no version or table name for the fiducials
A co-registration transform built from 2,934 expert-matched fiducials (1,994 somata and 942 blood vessels) has an average residual of 3.8 um; the staged transform lowers the residual from about 10 um to 5.6, 4.6, 3.9, 3.5, 3.1 and 2.9 um.
In the first step, we generated a co-registration transform using a set of 2,934 expert-matched fiducials ... The average residual was 3.8 μm
Results, Functional–structural co-registration, paragraph 1; Methods, Co-registration Transform
The paper gives no table name or version for the fiducials, and no public copy was found on the BossDB bucket; the CAVE tables need an account token.
BlockedResults, Proofreading, paragraphs 1 to 3· Otherdata in CAVE table proofreading_status_and_strategy and the ChunkedGraph edit history of datastack minnie65_public (MICrONS Explorer); the code reads version 1181
The released segmentation holds 1,046,656 proofreading edits as of 16 September 2024; 1,433 neurons have proofread axons, 85 excitatory neurons have fully proofread axons and dendrites, and 1,188 excitatory neurons have proofread dendrites.
The released segmentation now contains all 1,046,656 edits of the proofreading that had occurred as of 16 September 2024
Results, Proofreading, paragraphs 1 to 3
The CAVE table proofreading_status_and_strategy and the ChunkedGraph edit history need an account token. The BossDB bucket holds only a 2021 proofreading status release, neither of the claims file's readings of the version. The BossDB bucket serves without a token the synapse table exports of versions 1078, 1181 and 1300 (about 20 GB each) and a connection table with nucleus ids at version 1300, but not proofreading_status_and_strategy, which chooses the neurons; its proofreading release (proofreading_status_public_release.csv, 2021, 600 rows) holds root ids of an early version and no axon strategy.
BlockedResults, Proofreading, paragraph 3; Results, Integrated analysis, paragraph 5· Otherdata in CAVE synapse table and proofreading_status_and_strategy, with ChunkedGraph root history, datastack minnie65_public (MICrONS Explorer); no version is given
Comparing proofread with automated dendrites, 99% of inputs assigned to a postsynaptic soma in the automated segmentation were correct.
From the proofread dendrites, we determined that 99% of inputs were correct when assigned to a postsynaptic soma in the automated segmentation
Results, Proofreading, paragraph 3; Results, Integrated analysis, paragraph 5
Needs the CAVE synapse and proofreading tables with the root history, which need an account token. The BossDB bucket serves without a token the synapse table exports of versions 1078, 1181 and 1300 (about 20 GB each) and a connection table with nucleus ids at version 1300, but not proofreading_status_and_strategy, which chooses the neurons; its proofreading release (proofreading_status_public_release.csv, 2021, 600 rows) holds root ids of an early version and no axon strategy.
BlockedResults, Proofreading, paragraph 3· Otherdata in CAVE tables proofreading_status_and_strategy and synapse tables, with skeletons from the MICrONS release, datastack minnie65_public; no version is given
The longest proofread excitatory axon measures 18.9 mm with 2,483 synaptic outputs; inhibitory axons range from 1.1 to 32.3 mm in length with a mean of 2,754 outputs (range 99 to 14,019).
the longest excitatory axon measuring 18.9 mm with 2,483 synaptic outputs and inhibitory axons ranging from in length from 1.1 to 32.3 mm with a mean of 2,754 synaptic outputs (range 99–14,019)
Results, Proofreading, paragraph 3
Needs proofreading_status_and_strategy from CAVE (account token) to choose the axons; skeletons on BossDB are of version 661 and alone do not say which axons are proofread. The BossDB bucket serves without a token the synapse table exports of versions 1078, 1181 and 1300 (about 20 GB each) and a connection table with nucleus ids at version 1300, but not proofreading_status_and_strategy, which chooses the neurons; its proofreading release (proofreading_status_public_release.csv, 2021, 600 rows) holds root ids of an early version and no axon strategy.
BlockedResults, Proofreading, paragraph 4; Extended Data Fig. 3· Otherdata in CAVE tables nucleus_detection_v0 and nucleus_ref_neuron_svm, datastack minnie65_phase3_v1 (MICrONS Explorer); the before counts need the first materialization
The automated segmentation had 7,050 objects merging 17,753 neurons; after manual and automated splitting the total of individually segmented neurons is 84,035, and NEURD made more than 164,000 edits.
there were 7,050 segmented objects consisting of a total of 17,753 neurons that were merged together (based on nucleus segmentation) ... bringing the total number of individually segmented neurons to 84,035
Results, Proofreading, paragraph 4; Extended Data Fig. 3
Needs nucleus_ref_neuron_svm and root ids at a recent version from CAVE (account token); the 2021 nucleus table on BossDB holds root ids of an early version.
BlockedFig. 6f; Results, Integrated analysis, paragraph 5· Otherdata in CAVE synapse table, nucleus_detection_v0 and proofreading_status_and_strategy, datastack minnie65_public (MICrONS Explorer); no version is given
Between 60% and 95% of the output synapses of proofread axons map to postsynaptic targets with a known soma, depending on cell type and location.
between 60 and 95% of those outputs can be accurately mapped onto their postsynaptic targets with a known soma location, depending on the cell type and its spatial location in the volume
Fig. 6f; Results, Integrated analysis, paragraph 5
Needs proofreading_status_and_strategy from CAVE (account token) to choose the neurons, besides the 20 GB synapse table. The BossDB bucket serves without a token the synapse table exports of versions 1078, 1181 and 1300 (about 20 GB each) and a connection table with nucleus ids at version 1300, but not proofreading_status_and_strategy, which chooses the neurons; its proofreading release (proofreading_status_public_release.csv, 2021, 600 rows) holds root ids of an early version and no axon strategy.
BlockedFig. 6g,h; Results, Integrated analysis, paragraphs 5 and 6· Otherdata in CAVE tables synapses_pni_2, nucleus_detection_v0 and coregistration_manual_v4, datastack minnie65_phase3_v1, materialization 1181 (as in the code, Fig 6/panel_k/get_data.ipynb); nucleus id 294657
For the example proofread L2/3 VISp pyramidal cell, 74.5% (1,053 of 1,412) of its output synapses are onto objects with a single nucleus (275 onto inhibitory, 662 onto excitatory and 116 onto unclassified cells), 359 onto orphan fragments, and 365 onto functionally matched cells.
For this particular proofread neuron, 74.5% (1,053 out of 1,412 synapses) are onto objects with a single nucleus (as determined from automated detection), with 275 synapses onto cells classified as inhibitory, 662 synapses onto cells classified as excitatory, and 116 synapses onto cells whose soma did not pass classification quality control
Fig. 6g,h; Results, Integrated analysis, paragraphs 5 and 6
Needs the root id of nucleus 294657 at version 1181 and coregistration_manual_v4, both from CAVE (account token), besides the 20 GB synapse table. The BossDB connection table of version 1300 (connections_with_nuclei) gives outputs by presynaptic nucleus, but at version 1300, not 1181, and the class and match tables the counts need are only in CAVE.
BlockedFig. 6k; Results, Integrated analysis, paragraph 6· Otheruses data from outside the dataset
The oracle responses of the example cell's functionally co-registered postsynaptic targets can be retrieved from the functional data and set beside the presynaptic cell's.
By filtering the synaptic targets with functionally matched neurons (Fig. 6k ), one can further identify which targets have been matched to the functional experiments (365 out of 1,412) and use DataJoint to query the functional data or read NWB files deposited in the DANDI data archive
Fig. 6k; Results, Integrated analysis, paragraph 6
Depends on manual-coreg-19181 (coregistration_manual_v4, CAVE, account token) and on the targets of nucleus 294657 from the CAVE synapse table. The BossDB bucket holds a connection table with nucleus ids at version 1300 (connections_with_nuclei), which could give the targets of nucleus 294657 at that version, but the match of targets to units still needs coregistration_manual_v4.
BlockedDiscussion, Importance of functional connectomics, paragraph 3· Otherdata in CAVE tables proofreading_status_and_strategy, synapse table and nucleus_detection_v0, datastack minnie65_public; no version is given
From 94 proofread excitatory axons one can query 69,962 output synapses, onto 20,112 distinct neuron somata.
from just 94 proofread excitatory axons, one can query 69,962 output synapses, which map to 20,112 distinct neuron soma in the volume
Discussion, Importance of functional connectomics, paragraph 3
Needs proofreading_status_and_strategy from CAVE (account token) to choose the 94 axons. The BossDB bucket serves without a token the synapse table exports of versions 1078, 1181 and 1300 (about 20 GB each) and a connection table with nucleus ids at version 1300, but not proofreading_status_and_strategy, which chooses the neurons; its proofreading release (proofreading_status_public_release.csv, 2021, 600 rows) holds root ids of an early version and no axon strategy.
BlockedFig. 7a–c; Results, Cell types, paragraph 2· Otherdata in CAVE datastack minnie65_public, materialization 1181, tables proofreading_status_and_strategy and aibs_metamodel_mtypes_v661_v2 and the synapse table (code, Fig 7/Fig7abc/connectivity.ipynb, with the edge list it saved)
The outputs of proofread neurons, more than 900,000 synaptic connections, show differences in interlaminar communication between excitatory and inhibitory neurons.
Examining the output of proofread neurons, which includes more than 900,000 synaptic connections between neurons, reveals key differences in the interlaminar communication between excitatory and inhibitory neurons
Fig. 7a–c; Results, Cell types, paragraph 2
Needs proofreading_status_and_strategy and aibs_metamodel_mtypes_v661_v2 at version 1181 from CAVE (account token), besides the 20 GB synapse table. The BossDB bucket serves without a token the synapse table exports of versions 1078, 1181 and 1300 (about 20 GB each) and a connection table with nucleus ids at version 1300, but not proofreading_status_and_strategy, which chooses the neurons; its proofreading release (proofreading_status_public_release.csv, 2021, 600 rows) holds root ids of an early version and no axon strategy.
BlockedFig. 7d; Results, Cell types, paragraph 2· Otherdata in the csv files of the paper's repository (Fig 7/Fig7-VISp-L3a-order1-E-conn.csv and Fig7-VISp-L3a-Inh-conn-mastersheet.csv, github.com/AllenInstitute/MicronsFunctionalConnectomics at commit be24e51) and CAVE datastack minnie65_public
Seven layer 3 pyramidal cells can be followed through their first-order excitatory and inhibitory targets to the second-order excitatory targets of those inhibitory cells, grouped by inhibitory targeting type (DTC, PTC, ITC, STC).
we can follow a collection of layer 3 pyramidal neurons and compare their first-order (direct) connectivity onto excitatory cell types and inhibitory neurons as well as the second-order (two-hop) connectivity of those inhibitory neurons that are targeted by the layer 3 cells
Fig. 7d; Results, Cell types, paragraph 2
The csv files are public in the paper's repository, but the recount the criterion asks for needs CAVE at version 1181 (account token).

Findings about the data

What attempts found about the dataset itself, for its owner and the archive.

Selection not documentedThe dandiset holds all 19 completed scans, while the paper describes 14 (Extended Data Table 1). Nothing in the files or the metadata marks which 14, so its counts (for example 125,413 masks) can be matched only after reading the table's image. The five other files are 4-9, 4-10, 5-3, 8-7 and 8-9.
Data missingThe files hold fluorescence traces only. The deconvolved traces that the paper's analyses use, and that the paper says can be read from the NWB files, are not in them; nor are the oracle scores.
Data missingThe structural stack, the registration of each field into it, and the stack coordinates of the masks are not in the dataset, so masks imaged in more than one scan cannot be grouped and the z drift cannot be measured.
Description missingNo mask or field has a visual area; every imaging plane's location is the same list of four areas. The area masks the Methods transform into the stack are not in the dataset.
Identifier missingThe EM columns (cave_ids, pt_root_id, pt_supervoxel_id and positions) exist only in one of the two fields at each depth (52 of the 104 fields of the paper's 14 scans) and in none of the fields of files 4-9, 4-10, 8-7 and 8-9.
Description missingThe Monet2 table gives num_directions but not the direction or onset of each period, which Fig. 3g needs. The directions can be measured from the stimulus video in the files, as this attempt did.
Metadata errorThe imaging_rate of every imaging plane is NaN (the rate is only in the field description), and the subject age is the range P75D/P81D in every file, though each scan was recorded on one day.
OtherIn scan 4-7, one of the paper's 14, the eye tracking ends 968 s before the last imaging frame; in files 4-10, 5-3 and 8-7 it ends 54 to 76 minutes early, and in 4-9 and 8-9 3.9 and 6.2 s early. Several files have missing (NaN) time stamps in the treadmill and eye series (up to 190).
OtherTrippy seed 6831 is shown in both scans 6-4 and 7-5, while the Methods say the 10 seeds that are not conserved are unique to each scan.
OtherAll fields of a scan share one array of frame time stamps, though the fields are imaged one after another within each frame; the per-field offset is not recorded.
OtherThe EM imagery of subvolume 65 on BossDB holds data up to section 27,882, while the paper gives the processed region as sections 7,931 to 27,904. The imagery of the two subvolumes overlaps from section 14,825 to 14,847, and the metadata give the bounds of each subvolume only in whole chunks of 64 sections (7,904 to 14,847 and 14,816 to 27,903).
File format errorpt_root_id and pt_supervoxel_id are stored as 64-bit floats, which hold integers exactly only up to 2**53; root ids near 8.65e17 are spaced by 128 in float64, so ids that differ in their last digits can collide. Of 538 pairs of masks in different scans that share a root id, 50 have EM cell body points more than 10 um apart.

Questions

Where the author or a verifier believed the claims file misread the paper, they asked, citing the paper, and a reader who had not seen the data answered from the paper alone, amending the claims file where it was wrong. The author then ran the claim as answered. Attempts before protocol pilot-11 also asked here for method details the claims file did not give.

openasked 2026-10-30about structural-stack, z-drift-under-10um, unique-neurons-75909
Should these claims be marked as data elsewhere? They are marked as dataset claims and blocked as data_missing, because the dandiset holds no structural stack. The BossDB open data bucket serves a structural stack of the mouse without an account (iarpa_microns/minnie/functional_data/two_photon_structural_stacks/two_photon_stack_17797_9_19_v1.tiff, 1.25 GB, with a technical document), and the paper lists the 2P structural images among the publicly available data resources.
Passage. Fig. 1 legend, "a, The nine data resources that are publicly available at https://www.microns-explorer.org/ . b, ... The primary in vivo data resource consists of 2P calcium images, 2P structural images, natural and parametric video stimuli used as visual input, and behavioural measurements"; Data availability, "EM imagery, segmentation and annotation data is available via https://www.micronsexplorer.org/cortical-mm3 and from https://bossdb.org/project/microns-minnie".
Why it matters. Under the protocol, data that are public elsewhere are marked as data elsewhere, never data_missing. structural-stack could then be tested on the stack's size and resolution. The other two also need the registration of each field into the stack, which this attempt did not find on the bucket, so they might stay blocked for another reason.
No reader has answered it yet.
openasked 2026-10-30about em-sections
Should the expected quantity be the section range 7,931 to 27,904 given in the paper, with the count of 19,974 dropped or marked as the reader's arithmetic? And should the criterion still say that the released imagery gives these ranges?
Passage. Results, The EM volume, "sections 7,931–27,904" and "800-µm"; the paper gives no count of processed sections.
Why it matters. The verifier of round 3 found that the tolerance-0 count of 19,974 is the reader's inclusive arithmetic, not a number in the paper. On BossDB, 27,904 is the metadata's exclusive end of subvolume 65 (14,816 + 13,088), and the imagery holds data up to section 27,882, so the paper's upper bound may be exclusive or a bounding box and not the last section with data. The outcome was tested on the claim as written.
No reader has answered it yet.
openasked 2026-10-30about synapse-detection
Should the 98% partner assignment accuracy on 191 held-out synapses get an expected quantity, or be dropped from the statement?
Passage. Results, Automated reconstruction, paragraph 3, and Extended Data Fig. 1, as quoted in the statement.
Why it matters. The statement includes it, but no expected quantity tests it, so it is not tested and is listed as untested in the attempt (verifier, round 3).
No reader has answered it yet.

Notebooks

Every registered entry that uses this dataset. An entry can hold several notebooks, and can use more than one dataset.

Reproduction Two-photon functional data of the MICrONS mm3 volume (MICrONS Consortium 2025)

Tests the findings of the MICrONS Consortium (2025) on dandiset 000402, which holds the 19 two-photon scans of the MICrONS mouse, with the masks, fluorescence traces, stimulus tables, stimulus video and behaviour of each. It checks the scans, masks, stimulus design, spontaneous periods and behaviour against the paper, measures the Monet2 directions from the stimulus video, and computes oracle scores and direction tuning from the traces. Four EM claims are tested on the tables and imagery of the BossDB open data bucket. All files

Reproducing the MICrONS Consortium (2025) from dandiset 000402
reproduce_microns_2025.ipynbExecuted, with outputs · Source
EM sections and synapse counts of the MICrONS volume (MICrONS Consortium 2025)
em_volume_counts.ipynbExecuted, with outputs · Source

Verification

Round 1
2026-10-30
Agreed with 34 of 36 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: Round 1. The handover of the first run (the executed notebook with the conclusions and summary removed, the helper module and the claims file), with the paper under paper/; the attempt record was not available to the verifier.
- spontaneous-periods: the verifier’s outcome was “Partially reproduced”. Cell 20: every scan lies inside the widened ranges, which is the criterion's test. But the expected upper bound of the monitor-on period (5.5 +- 0.55 min) is not met; the 14 scans reach only 4.84 min. The 19 files match all four bounds (0.145, 5.54, 8.26, 21.19 min), so the paper's range was probably taken over all 19 recordings. Only the containment test is run; the four expected bounds and their tolerances are never compared.
- monet2-directions: the verifier’s outcome was “Partially reproduced”. Cell 25: the period is within tolerance in 560 of 560 trials (median 937.1 ms). 16 distinct directions in 556 of 560 trials by the 10 degree rule, or 554 of 560 after snapping to the 22.5 degree grid. The criterion requires every trial. The 10 degree merge rule is too tight for the measurement error (deviations from the grid up to 11 degrees); in a synthetic test, uniform noise of +-7 degrees gave a false 15 in 9.5% of trials. The two measures disagree on which trials have 15 directions, and the 16 directions of seed 7637 in scan 5-6 are never printed, so the outcome rests on an unvalidated proxy.
Round 2
2026-10-30
Agreed with 34 of 36 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: Round 2. The full executed notebook of the first revision (conclusions and summary included), the helper module, the claims file and the attempt record, with the paper under paper/.
- em-sections: the verifier’s outcome was “not given”. Does not fully agree with blocked (too_large). The record's metadata bounds (7,904; 14,816; 27,903) already decide the second expected quantity, the share of sections in subvolume 35: (14,816 - 7,904) / 20,000 = 34.6%, against 35 +- 3.5. Only the exact section count needs whole sections. The patch read that put the first section at 7,931 was run outside the notebook and has no printed output. The round-1 response says "No change of outcome", but the outcome changed from the verifier's round-1 blocked (other) to too_large.
- synapse-detection: the verifier’s outcome was “Blocked”. Does not fully agree with blocked (too_large). The record says a count of the same 20 GB file took 10 minutes in an earlier run and that the entry's run stalled; that is a failed run, not a file too large to process. The count for subvolume 65 should be retried, or the reason recorded as other. The subvolume 35 count and the precision and recall would stay untested either way. The earlier 10-minute count has no printed output.
Round 3
2026-10-30Agreed with 2 of 2 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: Round 3. The proposed changes of outcome for em-sections and synapse-detection, with the analysis behind them in em_volume_counts.ipynb (executed), the claims file and the attempt record, with the paper under paper/.

Method differences

How the tests differed from the paper’s method or data, and the claims each applies to.

Which 14 of the 19 files are the paper's scans was taken from Extended Data Table 1 (Methods, 2P functional imaging, "Of 19 completed scans over 6 days of imaging, 14 are described here (Extended Data Table 1)"). The text the claims file was read from holds only the table's caption; its cells were read from the table's image in PMC11981939 (41586_2025_8790_Tab1_ESM.jpg). They are 4-7, 5-6, 5-7, 6-2, 6-4, 6-6, 6-7, 7-3, 7-4, 7-5, 8-5, 9-3, 9-4 and 9-6.scans-14-p75-p81
The table gives no direction or onset per period, so both were measured from the stimulus video in each file; a method of this attempt, not the paper's. Boundaries come from phase correlation between successive frames, and each period's direction from the global optical flow over the period (gradient method of Lucas and Kanade, frames smoothed by a 1 pixel Gaussian). The boundary detection, the 10 degree tolerance for distinct directions, the smoothing and the 0.1 s left out at each end of a period are standard choices, stated in the helper.monet2-directions
Fluorescence traces replace the deconvolved traces of the paper, which are not in the dataset.fig3f-oracle-reliability
Frames are taken from each field's time stamps from the onset of each repeat; the per-unit delay within a frame that the paper's code adds is not in the dataset and is not applied. All fields of a scan share one array of time stamps in the files.fig3f-oracle-reliability
Fluorescence replaces the deconvolved traces, which are not in the dataset.fig3g-direction-tuning
The direction of each period comes from the video (see monet2-directions), snapped to the 22.5 degree grid.fig3g-direction-tuning
The 80 units shown are ranked by the preferred response minus the mean over directions, divided by the standard deviation over directions, in place of the claims file's ratio, which is not defined for fluorescence with its arbitrary offset. This choice only picks the units shown.fig3g-direction-tuning
The first and last sections with data are read from the imagery at 64 nm per pixel on a grid of every 8th chunk in x and y (2,288 chunks a layer), a choice of this attempt; the release gives no list of processed sections.em-sections
The paper names no materialization version. Subvolume 65 is counted on the export of version 1181, the version the paper's code reads; subvolume 35 on its only public table, assigned.csv.gz (2021).synapse-detection
The tables are the 2021 releases on BossDB, not the CAVE tables the claims file names; the paper gives no version.cell-classification
The SVM table holds 27,698 rows that are not nuclei of nucleus_detection_v0 (27,692 called non-neurons); the count is restricted to subvolume 65's nuclei, as the paper's is.cell-classification

Provenance

Attempt record: 000402-microns-2025-01
Paper: Functional connectomics spanning multiple areas of mouse visual cortex
Date: 2026-10-30
Made by: An agent, run by CatalystNeuro
Dataset: DANDI:000402
Version: 0.230307.2132 (doi:10.48324/dandi.000402/0.230307.2132)
Run on: GitHub Actions
Harness: Claude Code
Model: claude-opus-5-5
Protocol: pilot-12

Transcripts of the agent runs

Each run’s prompt, messages and tool calls, with long output cut and the text of the paper left out, and its final report.

← DANDI:000351 DANDI:000404 →