DANDI:000055 · Encore Science

Claims 24

From the paper Behavioral and Neural Variability of Naturalistic Arm Movements · claims file · checked 2026-10-23: ready

subject-days
The dataset's subjects and recording days are the paper's subjects S01 to S12 and the recording days the paper lists for each.
How it will be judged
The days per subject in Table 2 are S01 days 3, 4, 5 and 7; S02 and S03 days 3 to 6; S05 days 3, 4 and 7; S04 and S06 to S12 days 3 to 7 (55 subject-days). The link holds (reproduced) when the dataset's subjects 01 to 12 have exactly these days in order. When it does not, the subjects are matched by their day sets where these are unique, and the mapping and any mismatch are reported; subjects whose day sets are identical cannot be told apart this way, so the event counts per subject (claim fig2-event-counts) are reported beside them.
cohort
Twelve subjects (8 male, 4 female, 29.4 +- 7.9 years) had on average 87.0 +- 12.9 surface electrodes; five also had 23.2 +- 12.1 depth electrodes; five were implanted in the right hemisphere and seven in the left.
We analyzed opportunistic clinical recordings from 12 subjects (8 males, 4 females) during their clinical epilepsy monitoring. Subjects were (mean ± SD) 29.4 ± 7.9 years old ... resulting in 87.0 ± 12.9 total surface electrodes per subject (mean ± SD). In addition, five subjects had 23.2 ± 12.1 intracortical depth electrodes (mean ± SD). Electrodes were implanted primarily within one hemisphere for each subject (five right hemisphere, seven left hemisphere)
Methods, Subject information, paragraph 1
How it will be judged
Tolerances on the means are the SDs the paper states, as the protocol allows; the counts are exact. Surface electrodes are grid and strip electrodes, depth electrodes the rest (Extended Data Figure 1-1 legend). Values the dataset does not hold are reported as not testable.
fig2-event-counts
Each subject contributed between 151 and 947 movement events (median 640 across subjects).
The total number of events for each subject was between 151 and 947 (median, 640 across subjects).
Figure 2 legend
How it will be judged
Which events are counted is not fixed by the text. Readings are (a) all contralateral wrist movement events the dataset holds for a subject; (b) those left after dropping events marked false positive and ECoG segments with power below 0 dB or with 115-125 Hz power more than 3 SD above all segments (Methods, ECoG-event synchronization; the Figure 2 code reads its events from the spectral power files, which have these removed). Reproduced when all three values are within tolerance under one reading.
fig2-features-differ-across-subjects
All ten behavioural and environmental features differ significantly across subjects.
All 10 metadata features significantly differed across subjects ( p < 0.001 for every feature, one-way Kruskal–Wallis test).
Results, Behavior during naturalistic movements, paragraph 3; Figure 2
How it will be judged
For each of the ten features, a one-way Kruskal-Wallis test of the event values grouped by subject, on the events of claim fig2-event-counts (reading b, with reading a as a variant), reach angle folded as in the opening comment and day of recording as a number. Reproduced when p < 0.001 for all ten features; not reproduced otherwise, with every p reported.
fig2-reach-shape
Reach duration and reach magnitude vary little across subjects, with most reaches lasting under 2 s and covering under 200 pixels, and reach angle distributions tend to be bimodal with peaks near +90 and -90 degrees, with few side-to-side reaches.
Reach duration and reach magnitude show minimal intersubject variability, with most reaches lasting <2 s and covering <200 pixels ... For reach angle, the distributions tend to be bimodal, with peaks at ±90
Results, Behavior during naturalistic movements, paragraph 3; Figure 2
How it will be judged
The paper gives no measure for these words, so they are read as follows. "Most" is more than half of the events, judged in every subject and in the events pooled over subjects; reach duration in seconds and reach magnitude in pixels. Reach angle (folded to -90 to 90 degrees) is judged on a Gaussian kernel density with Scott's bandwidth, the seaborn kdeplot default the Figure 2 code uses, evaluated on -90 to 90; a subject is bimodal with peaks at +-90 when the density's two highest local maxima lie within 15 degrees of -90 and of +90 and the density at 0 degrees is below both. "Minimal intersubject variability" is read as the duration and magnitude parts holding in at least 10 of 12 subjects; the per-subject medians are reported. Reproduced when the duration and magnitude parts hold pooled and in at least 10 of 12 subjects, and the angle part holds pooled and in more than half of the subjects ("tend to"). Partially reproduced when one of the two parts holds.
fig2-bimanual-skew
Bimanual ratio and bimanual overlap are skewed toward unimanual movement of the contralateral arm, less so for subjects 02, 04 and 05, while the bimanual class is mostly bimanual.
For bimanual ratio and overlap features, the distributions are skewed toward unimanual movements of the contralateral limb, as expected from Extended Data Fig. 2-1 c , with less skew for subjects 02, 04, and 05. In contrast, the bimanual class categorical feature is primarily skewed toward bimanual movements
Results, Behavior during naturalistic movements, paragraph 4; Figure 2
How it will be judged
The paper gives no measure, so it is read as follows. Bimanual ratio (missing values 0.5, code) skewed toward unimanual when its median is below 0.5; bimanual overlap (seconds) skewed toward unimanual when its sample skewness (Fisher-Pearson) is positive; bimanual class mostly bimanual when more than half the events are bimanual. The paper's "less skew" for S02, S04 and S05 refers to both ratio and overlap, so it holds when each of the three is among the five subjects with the highest median bimanual ratio and among the five with the lowest overlap skewness; the two features are reported separately. Reproduced when the ratio and overlap parts hold pooled and in at least 9 of 12 subjects, the class part holds pooled and in more than half the subjects, and the S02, S04, S05 part holds; partially reproduced when some but not all of these hold.
extfig2-1c-contralateral-larger
Across all 12 subjects, the median displacement of the contralateral wrist after movement onset is substantially greater than that of the ipsilateral wrist, confirming that event detection mainly picked contralateral movements.
Because we selected movement initiation events of the wrist contralateral to the hemisphere with implanted electrodes, the median contralateral wrist displacements (orange lines) across all 12 subjects are substantially greater than ipsilateral wrist displacements (gray lines).
Extended Data Figure 2-1c legend; Results, Behavior during naturalistic movements, paragraph 2
How it will be judged
Per subject, the median over events of each wrist's displacement in pixels from its rest position, with events aligned at movement onset (Extended Data Figure 2-1b and c legends). The current code holds no script for this panel. The rest position is open; readings are (a) the mean position over -1.5 to -1 s, the behavioural baseline of the code (config.baseline_vals_movement; compute_wrist_radial_distance, used by archived figure code, takes the absolute change of the radial distance from the image origin over that baseline), and (b) the position at movement onset (the Methods' reach definition). Displacement is the Euclidean distance from the rest position. The window is 0 to 1.5 s, the end of the time axis of the archived plotting code (plot_behavior_traces). "Substantially greater" has no measure; the part holds in a subject when the peak of the contralateral median exceeds the peak of the ipsilateral median, and the ratio is reported. Event set as in fig2-event-counts reading b. Reproduced when this holds in all 12 subjects under both readings; partially when it holds in 10 or 11. Blocked as methods_underspecified when the readings give different outcomes.
extfig2-2-feature-correlations
Over all events, reach magnitude correlates with reach duration (r = 0.26) and onset speed (r = 0.56), reach duration with bimanual overlap (r = 0.48), and the bimanual features with one another (overlap and ratio 0.51, class and ratio 0.50, overlap and class 0.61).
Reach magnitude positively correlates with reach duration ( r = 0.26) and onset speed ( r = 0.56) ... Reach duration is also positively correlated with bimanual overlap ( r = 0.48) ... overlap vs ratio, r = 0.51; class vs ratio, r = 0.50; and overlap vs class, r = 0.61
Results, Behavior during naturalistic movements, paragraph 5; Extended Data Figure 2-2
How it will be judged
Pearson correlation over the events of all 12 subjects pooled ("group-level"), on untransformed features in the units of Figure 2 (duration and overlap in seconds, magnitude in pixels, speed in pixels per second, missing bimanual ratios 0.5), as plot_feature_scattermatrix in the code computes it (pandas DataFrame.corr). Event set as in fig2-event-counts reading b, reading a shown as a variant. Reproduced when all six are within tolerance; partially when four or five are.
fig2-days-and-waking
Reach counts were fairly even across recording days except for subjects 04, 05 and 09, each with one day holding most events, and events occurred mostly during waking hours.
The number of reaches detected across days of recording was fairly consistent, with the exceptions of subjects 04, 05, and 09, who each had 1 d representing most of the total events. As expected, detected movement events often occurred mostly during waking hours.
Results, Behavior during naturalistic movements, paragraph 3; Figure 2
How it will be judged
"Most of the total events" is read as more than half. Reproduced when the subject's busiest day holds more than half of its events for S04, S05 and S09 and for none of the other nine subjects. Waking hours are not defined; they are read as 08:00 to 24:00, the two daytime bins of the paper's time-of-day feature, and the part holds when more than half of each subject's events fall in them. Partially reproduced when the day part holds for the three named subjects but also for one or two others, or when only one of the two parts holds. Event set as in fig2-event-counts reading b.
fig3-group-lfb-decrease-hfb-increase
At the group level, low-frequency power decreases and high-frequency power increases at movement initiation in several cortical regions, as in controlled movement experiments.
Generally, we find the expected pattern of low-frequency (∼4–30 Hz) spectral power decrease and high-frequency (∼50–120 Hz) power increases during movement initiation across multiple cortices ( Fig.
Figure 3; Results, Intracortical spectral power during naturalistic movements, paragraph 1
How it will be judged
Group spectrogram per region as the Methods and the code (plot_specs_group) make it. Per subject, the median over events of region power (spectral method of the opening comment); the mean over the 12 subjects; the group median over -1.5 to -1 s subtracted again (code). Masking as in the code (tfr_boot_sig_mask). Each subject's baseline time bins are shuffled within frequency, the median over subjects is taken, repeated to about 2000 values per frequency, two-sided percentile p, Benjamini-Hochberg over all frequency and time bins of the region at 0.05; non-significant values set to 0. "At movement initiation" is read two ways. (a) The mean of the masked values over 0 to 0.5 s, the window of the paper's banded analyses. (b) The time bin nearest 0 s. Reproduced when, under both readings, the masked mean over the 4-30 Hz bins is below 0 and over the 50-120 Hz bins above 0 in each of precentral, postcentral and inferior parietal. Partially reproduced when this holds in one or two of them. Blocked as methods_underspecified when the readings disagree.
fig3-sensorimotor-largest
Spectral power changes are largest in precentral, postcentral and inferior parietal regions, and the three temporal regions show minimal movement-related changes.
Spectral power deviations are largest in magnitude in precentral, postcentral, and inferior parietal regions, which are located in sensorimotor areas of the brain ... As expected, all three temporal cortical regions contain minimal movement-related spectral power fluctuations.
Results, Intracortical spectral power during naturalistic movements, paragraph 2; Figure 3
How it will be judged
On the masked group spectrograms of claim fig3-group-lfb-decrease-hfb-increase. The paper does not say how magnitude is measured. Readings are (a) the mean absolute masked value over all frequency bins and 0 to 0.5 s; (b) the same over the 4-30 Hz bins and, separately, over the 50-120 Hz bins, a region ranking high when it does in both; (c) as (a) over -0.5 to 0.5 s. Reproduced when, under every reading, precentral, postcentral and inferior parietal are the three largest of the eight regions and each temporal region is below all three. Partially reproduced when two of the three named regions are in the top three and the temporal part holds. Blocked as methods_underspecified when the readings give different outcomes.
fig3-lfb-more-widespread
Low-frequency power decreases are more spatially widespread than high-frequency increases and are also present in supramarginal and superior temporal regions.
In addition, low-frequency power decreases appear more spatially widespread than high-frequency power increases and are also present in supramarginal and superior temporal regions.
Results, Intracortical spectral power during naturalistic movements, paragraph 2; Figure 3
How it will be judged
On the masked group spectrograms, over 0 to 0.5 s. Readings for "widespread" are (a) the number of regions with any significant negative bin in 4-30 Hz against the number with any significant positive bin in 50-120 Hz; (b) the count of significant negative 4-30 Hz bins against significant positive 50-120 Hz bins, summed over the eight regions and divided by the number of frequency bins in each range. Reproduced when, under both readings, the low-frequency count is larger, and supramarginal and superior temporal each have significant negative 4-30 Hz bins. Blocked as methods_underspecified when the readings disagree.
fig4-precentral-subject-variability
In the precentral region every subject shows the low/high-frequency pattern, but amplitudes and frequency bands vary; subjects 03, 06, 07, 08 and 11 have increases up to 120 Hz, subjects 09 and 12 mainly at 60-80 Hz, and subjects 04 and 08 have narrower low-frequency decreases.
For instance, the precentral region shows the same low/high-frequency power pattern for each subject ( Fig. 4 ), but the amplitudes and frequency bands of maximal power deviation differ widely across subjects. Subjects 03, 06, 07, 08, and 11 show increased power at high frequencies up to 120 Hz, while subjects 09 and 12 have increased power primarily between 60 and 80 Hz. For subjects 04 and 08, low-frequency power decreases occur across narrower frequency bands compared with the other subjects.
Results, Intracortical spectral power during naturalistic movements, paragraph 3; Figure 4
How it will be judged
Per subject, the unmasked median over events of precentral power (Figure 4 legend, no statistical masking), with the mean of that median over -1.5 to -1 s subtracted again, as plot_specs_single_region in the code does, averaged over 0 to 0.5 s (the window of the paper's banded analyses, Methods; Figure 4 itself shows whole spectrograms and fixes no window) to give a spectrum. The paper gives no measure, so the parts are read as follows. (1) Every subject has a negative mean over 8-32 Hz and a positive mean over 76-100 Hz. (2) For S03, S06, S07, S08 and S11 the 116 and 121 Hz bins are positive. (3) For S09 and S12 the largest value among the 56-121 Hz bins is at 61 to 76 Hz (bins within 60-80 Hz), and the mean over 101-121 Hz is below half of it. (4) The width of the decrease is the number of 1-46 Hz bins below half of the subject's most negative value; S04 and S08 are among the three subjects with the fewest. Reproduced when all four hold; partially when two or three hold. Part (1) failing in a subject is reported by subject.
fig5-band-power-differs-across-subjects
Precentral LFB and HFB power in the 0.5 s after movement onset differs significantly across subjects.
For both frequency bands, spectral power significantly differed across subjects ( p < 0.001 for both bands, one-way Kruskal–Wallis test).
Results, Intracortical spectral power during naturalistic movements, paragraph 4; Figure 5; Table 1
How it will be judged
Banded precentral power per event (opening comment; band mean as the main result, the code's median as a variant), all days pooled per subject. One-way Kruskal-Wallis test across the 12 subjects per band, as the Figure 5 notebook (pingouin kruskal). "Pooling over all electrodes" (Results) is the projection of all the subject's good electrodes onto the precentral region, as the code does. Reproduced when p < 0.001 in both bands.
table1-subject-band-power-ci
Each subject's precentral band power lies in the 95% bootstrap interval Table 1 gives; LFB power is below baseline and HFB power above it for most subjects.
How it will be judged
Table 1 intervals (dB), LFB then HFB. S01 -1.87 to -1.29, 0.03 to 0.23; S02 -1.12 to -0.31, 0.07 to 0.64; S03 -2.11 to -1.73, 1.39 to 1.67; S04 -0.17 to 0.01, 0.32 to 0.52; S05 -0.7 to -0.38, 0.2 to 0.48; S06 -1.63 to -1.11, 2.02 to 2.37; S07 -0.84 to -0.51, 1.14 to 1.34; S08 -0.3 to 0.04, 0.36 to 0.63; S09 -0.57 to -0.08, -0.11 to 0.23; S10 -0.87 to -0.35, 0.01 to 0.3; S11 -2.3 to -1.86, 0.09 to 0.34; S12 -0.66 to -0.44, 0.06 to 0.2. The table does not say which statistic the interval is for. Readings are (a) the mean and (b) the median over the subject's events of banded precentral power. Band power is taken as the band mean (Methods) or the band median (code), giving four readings in all. Reproduced when, under one reading, the point estimate lies inside the paper's interval for all 24 subject and band pairs. Partially reproduced when it does for at least 18. The attempt's own 5000-replicate bootstrap intervals are shown beside the paper's.
fig5-recording-day-effects
Precentral band power differs significantly across recording days in LFB for subjects 03, 05, 06, 07, 08 and 11, and in HFB for subjects 03, 04, 07, 08 and 10, despite baseline subtraction.
We also found a significant effect of recording day in LFBs for subjects 03 ( p < 0.001, Kruskal–Wallis test), 05 ( p = 0.013), 06 ( p < 0.001), 07 ( p = 0.041), 08 ( p = 0.013), and 11 ( p < 0.001), as well as in HFBs for subjects 03 ( p = 0.002), 04 ( p < 0.001), 07 ( p < 0.001), 08 ( p = 0.004), and 10 ( p = 0.006).
Results, Intracortical spectral power during naturalistic movements, paragraph 4; Figure 5; Table 2
How it will be judged
Per subject and band, a one-way Kruskal-Wallis test of banded precentral power across the subject's recording days, uncorrected, at p < 0.05 (Figure 5 legend; the notebook uses pingouin's uncorrected p). Band mean as the main result, band median (code) as a variant. The paper's pattern is the 11 significant pairs above and the 13 other subject and band pairs not significant. Reproduced when all 24 pairs agree; partially reproduced when at least 20 agree. Each p is reported, with the per-day 95% intervals of Table 2 for comparison.
extfig5-1-baseline-reduces-day-variability
Subtracting the -1.5 to -1 s baseline reduces the variability of band power across recording days, though clear variability remains.
Baseline subtraction reduces the variation across recording days, as expected, but clear variability remains.
Results, Intracortical spectral power during naturalistic movements, paragraph 4; Extended Data Figure 5-1 legend
How it will be judged
The legend gives the measure, the SD across recording days of the average LFB (8-32 Hz) and HFB (76-100 Hz) power, with and without baseline subtraction. The figure could not be read and no code for it is in the repository, so what is averaged is open. Readings are (a) precentral region power, per subject, the SD across days of the day means of event band power; (b) the same per good electrode, pooled over electrodes and subjects. Without baseline means the 10*log10 band power with no subtraction. Reproduced when, under both readings and in both bands, the SD with baseline is below the SD without it for more than half of the units (subjects or electrodes). The paper gives no measure for "clear variability remains"; the Results tie it to the recording-day effects that persist after baseline subtraction (paragraph 4), so that part is judged by claim fig5-recording-day-effects finding a significant day effect in at least one subject, and the SDs with baseline are reported. Blocked as methods_underspecified when the readings disagree.
discussion-day-changes-small
Day-to-day changes in baseline-subtracted spectral power are small (about 1-2 dB) compared with spectral power without baseline subtraction (about 10-50 dB).
It is also worth noting that these day-to-day changes in ECoG spectral power are small in magnitude (±1–2 dB) relative to spectral power without any baseline values subtracted
Discussion, paragraph 3
How it will be judged
On precentral banded power, per subject and band, the day means of the event values. The day-to-day change is the largest absolute difference between a day mean and the subject's overall mean. It holds when every such change is at most 2.2 dB (2 dB widened by 10%). The level without baseline is the mean 10*log10 band power without subtraction. It depends on the voltage unit, which the paper does not state; readings are microvolts and volts. That part holds when the levels lie in 9 to 55 dB (10-50 dB widened by 10% of each bound) under one unit reading. Reproduced when both parts hold; partially when only the first does.
fig6b-best-models-sensorimotor
Regression models with the highest withheld R2 are mainly at electrodes over frontoparietal sensorimotor areas, for both bands.
For both frequency bands, intracortical activity variability is best explained by the fit of our models to electrodes located in frontoparietal sensorimotor areas ( Fig. 6 b , Extended Data Fig. 6-1 ).
Figure 6b; Results, Modeling single-event spectral power with behavioral features, paragraph 2; Extended Data Figure 6-1
How it will be judged
Full-model withheld R2 per electrode and band (regression method of the opening comment). The paper shows this on brain maps and gives no measure. Each electrode is assigned to an area in one of two ways, (a) to the region of the eight with its largest projection weight, or (b) by projecting the electrode R2 values onto the eight regions with the projection weights, as for power, and averaging over subjects. Frontoparietal sensorimotor regions are precentral, postcentral and inferior parietal (the regions the Results call sensorimotor) with middle frontal and supramarginal. Under (a), the part holds when more than half of the electrodes with R2 > 0.05 are in these five regions and the median R2 of precentral, postcentral and inferior parietal electrodes exceeds that of temporal ones. Under (b), it holds when the three regions with the highest projected R2 are among the five frontoparietal ones. Reproduced when this holds in both bands under both readings. Blocked as methods_underspecified when the readings disagree.
fig6-r2-at-most-025
All regression models explain at most 25% of the variance of withheld single-event power.
However, all R 2 scores are at most 0.25, indicating that even the best models cannot explain >75% of the variance in the withheld data.
Results, Modeling single-event spectral power with behavioral features, paragraph 2
How it will be judged
The largest split-averaged full-model withheld R2 over all good electrodes of all subjects in both bands (regression method of the opening comment). Reproduced when it is at most 0.275 (0.25 plus 10%). The value is also reported for the code's 80/20 split and median band power.
fig6c-angle-and-day-most-informative
Reach angle and day of recording are the most informative features for single-event power, especially for low-frequency power.
Among individual features, we find that reach angle and day of recording are the most informative ( Fig. 6 c ).
Figure 6c and legend; Results, Modeling single-event spectral power with behavioral features, paragraph 2
How it will be judged
Delta R2 per feature, electrode and band (opening comment), over electrodes with withheld R2 > 0.05 in that band, across all subjects. The summary across electrodes is the median, as the box plots of Figure 6c show (code). Reproduced when reach angle and recording day have the two largest median Delta R2 of the ten features in the LFB. The HFB ranking is reported, and the claim is partially reproduced when only one of the two is in the LFB top two. Shown as variants are the code's Delta R2 (shuffling test events too), the code's selection by LFB R2, the code's 80/20 split, and the code's inclusion of bad electrodes.
extfig6-2-angle-most-retained
Reach angle is the feature most often retained by forward selection in sensorimotor regions, followed by reach duration, reach magnitude, bimanual ratio and bimanual overlap.
Reach angle is the feature most often retained, followed by reach duration, reach magnitude, bimanual ratio, and bimanual overlap.
Extended Data Figure 6-2 legend; Results, Modeling single-event spectral power with behavioral features, paragraph 2
How it will be judged
Per electrode and band, whether each feature is kept by forward selection (code feat_select_prob; a day or time-of-day one-hot column counts as 1/number of columns of that feature). The legend projects the retention probability onto the eight regions and averages across subjects; the code of this step is not in the repository. It is read as the projection-weighted mean over the subject's good electrodes, then the mean over subjects, as for power. The angle part holds when reach angle has the highest retention probability among the ten features in precentral, postcentral and inferior parietal for both bands. The legend does not say over which regions and bands its order of the next four features holds. Readings are (a) the retention probability averaged over the three sensorimotor regions and both bands and (b) the same over all eight regions; the order part holds when the features ranked second to fifth are reach duration, reach magnitude, bimanual ratio and bimanual overlap in that order. Reproduced when the angle part holds and the order part holds under both readings; partially reproduced when only the angle part holds. Blocked as methods_underspecified when the readings give different outcomes for the order part.
fig6d-upward-reaches-enhance-pattern
Upward reaches are associated with lower low-frequency power and higher high-frequency power, that is, a larger version of the average movement response.
The coefficients for reach angle indicate that upward reaches are associated with decreased low-frequency power and increased high-frequency power compared with the average response.
Figure 6d and legend; Results, Modeling single-event spectral power with behavioral features, paragraph 2
How it will be judged
Reach angle coefficient (positive angle is upward, standardised feature), averaged over splits, over electrodes with withheld R2 > 0.05 in that band. Reproduced when the median coefficient is below 0 for LFB and above 0 for HFB; partially when it holds in one band. The code's selection by LFB R2 and its inclusion of bad electrodes are shown as variants.
fig6d-angle-and-day-largest-coefficients
Reach angle and day of recording have the largest coefficient magnitudes among the features, and the SD of the recording-day coefficients is large.
In addition, both reach angle and day of recording have the largest coefficient magnitudes among behavioral features in the regression models ( Fig. 6 c ).
Figure 6d and legend; Results, Modeling single-event spectral power with behavioral features, paragraph 2
How it will be judged
Per electrode with withheld R2 > 0.05 in the band, the split-averaged coefficient of each continuous or binary feature, and for day and time of day the SD of their one-hot coefficients (Figure 6 legend, code). The Results cite Fig. 6c for this sentence, but the coefficients are in Fig. 6d, whose box plots show the signed coefficients (code). The paper does not say how magnitude is summarised across electrodes. Readings are (a) the median of the absolute coefficients and (b) the absolute value of the median coefficient, the box-plot centre. Reproduced when, under both readings, reach angle and recording day have the two largest values in both bands; partially when in one band. Blocked as methods_underspecified when the readings give different outcomes. Variants as for fig6c-angle-and-day-most-informative.

Pipeline

Paper: done (3 of 3)
Claims: done
Claims check: done (ready)
Attempt: todo
Questions: none
Verified: none