{ "dataset": { "archive": "crcns", "id": "ia-1", "label": "CRCNS:ia-1" }, "page": "datasets/crcns/ia-1/", "generated": "2026-11-05T06:30:02+00:00", "caution": "A failure to reproduce is not a finding that the paper is wrong. It may be a failure of the attempt.", "outcome_counts": { "reproduced": 11, "partially_reproduced": 5, "not_reproduced": 7, "blocked": 3 }, "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 }, "outcome_counts_links": { "reproduced": 0, "partially_reproduced": 0, "not_reproduced": 0, "blocked": 0 }, "attempts": [ { "id": "crcns-ia-1-eyherabide-2008-01", "dataset": { "archive": "crcns", "id": "ia-1", "version": "unversioned", "doi": "10.6080/K0BG2KWB" }, "key": "crcns:ia-1", "label": "CRCNS:ia-1", "dataset_version": "unversioned", "paper": "10.3389/neuro.10.003.2008", "paper_title": null, "date": "2026-11-04", "kind": "agent", "accountable": "CatalystNeuro", "agent": { "harness": "Claude Code", "model": "claude-opus-5-5", "protocol": "pilot-13" }, "runner": { "platform": "GitHub Actions", "url": "https://github.com/catalystneuro/encore/actions/runs/41885596452" }, "tier": "runs", "deprecated": "", "deprecated_on": null, "entry": "crcns-ia-1-eyherabide-2008", "entry_title": "Burst coding in locust auditory receptors (Eyherabide et al. 2008)", "description": "Tests the findings of Eyherabide, Rokem, Herz and Samengo (2008) on CRCNS data set ia-1, which holds the spike times of locust auditory receptor cells under amplitude-modulated noise (Data1) and the stimuli. It identifies bursts from the spike-train correlation function, counts them per stimulus protocol, computes the information carried by bursts of each size and by the direct method, and the burst-triggered averages of the stimulus. The cells' thresholds, which the paper's model needs, are not in the data set.", "notebooks": [ { "path": "reproduce_eyherabide_2008.ipynb", "title": "Reproducing Eyherabide et al. (2008) from CRCNS ia-1", "executed": "notebooks/crcns-ia-1-eyherabide-2008/reproduce_eyherabide_2008.html" } ], "outcomes": [ { "claim": "data-sessions-and-trials", "statement": "Thirty-seven receptor cells were recorded from 23 animals, giving 132 sessions, of which 86 met the firing-rate and stationarity conditions; each stimulus was repeated 98 to 503 times, 172 on average.", "source": "Materials and Methods, Electrophysiology and stimulus design, paragraphs 1, 4 and 5", "quote": "Thirty-seven receptor cells were recorded, from 23 animals. Each cell was tested with two or more stimuli, resulting in 132 data sets in total ... There were 86 out of 132 data sets that fulfilled these two conditions ... N repetitions of each stimulus were presented, with N ranging between 98 and 503 (average 172)", "statistics": [], "kind": "finding", "inferred": "", "outcome": "partially_reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "The 116 sessions of the seven protocols come from 37 cells (rows of singlecells.txt) recorded on 23 dates; the 122 Gaussian AM sessions of any cutoff are within 10% of the paper's 132 (all 150 Data1 sessions are not). Under the criterion's readings the SD < 35 Hz condition keeps 115 of the 116 (113 with a 100 Hz target, reading b; 116 with whole-trial rates), against 86; over the 115 trials range from 69 (outside 98 +/- 9.8) to 503 and average 160.3, and the mean trial-to-trial SD of the rate is 6.48 Hz (within 6 +/- 0.6). Selection (c), which also leaves out the 9 sessions under 98 trials and the off-line sorted sessions, keeps 89 sessions with 98 to 503 trials (mean 168.7) and a mean SD of 6.00 Hz, within tolerance, but they come from 29 cells on 18 dates, outside 37 +/- 3.7 and 23 +/- 2.3 (the 37 cells and 23 dates hold over all seven-protocol sessions); it is not a reading of the criterion, which defines the selection by the rate conditions, and its fewest trials are 98 by construction, so it does not decide this outcome. The trial blocks were checked: late spikes (up to 2.76 s) occur only in off-line sorted sessions, in the pause, no block restarts its times or holds twice the usual spike count, and the data set's firing_rates file gives the same trial counts in 136 of 138 files.", "deviations": [ "Data1 does not record the animal; the 23 recording dates stand in for the 23 animals, assuming one animal per date.", "Sessions recorded are read as the 122 Gaussian AM sessions; the paper's 132 is not otherwise identifiable." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "table1-sessions-per-protocol", "statement": "The 86 kept sessions divide among the seven stimulus protocols as 43 (6 dB, 200 Hz), 6 (3 dB, 200 Hz), 7 (12 dB, 200 Hz), 7 (6 dB, 25 Hz), 8 (6 dB, 100 Hz), 7 (6 dB, 400 Hz) and 8 (6 dB, 800 Hz).", "source": "Table 1, row Recorded sessions", "quote": "", "statistics": [], "kind": "finding", "inferred": "The numbers are read from Table 1; no sentence states them. The legend calls the row recorded sessions; the counts sum to 86, so they are read as the kept sessions.", "outcome": "not_reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Kept sessions per protocol are 60, 9, 8, 10, 11, 8 and 9 (reading a, 200 to 1000 ms) against 43, 6, 7, 7, 8, 7 and 8; none within 10% under any reading of the criterion. Selection (c), which also leaves out sessions under 98 trials and off-line sorted sessions, gives 47, 6, 7, 7, 8, 7 and 7, six of seven within tolerance, but was chosen in part from these counts. Leaving out only the off-line sessions gives 3 of 7, only the sessions under 98 trials 1 of 7, and the six 450 ms pause sessions 2 of 7. The data set holds 55 experiments pairing the reference with each other protocol (8 to 11 per protocol).", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "firing-rate-uniform", "statement": "The kept sessions had a mean firing rate of 113 Hz (SD 16 Hz) that did not differ significantly across stimulus conditions (one-way ANOVA, p = 0.58).", "source": "Materials and Methods, Electrophysiology and stimulus design, paragraph 4", "quote": "The resulting firing rates had a mean of 113 Hz (SD = 16 Hz), and they did not show any significant variation in the different stimulus conditions, as assessed by a one-way ANOVA ( p = 0.58)", "statistics": [ { "test": "one-way ANOVA across stimulus conditions", "p": 0.58 } ], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Over the 89 sessions of selection (c) the mean rate is 107.1 Hz (SD 14.5 Hz); one-way ANOVA across the seven protocols F = 0.49, p = 0.814. Over the 115 of selection (a) 110.2 Hz (SD 16.7), F = 0.49, p = 0.818; with whole-trial rates and over all seven-protocol sessions p is 0.471 to 0.809. No population is significant.", "deviations": [ "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it. The outcome is the same under every population shown, so it does not rest on this assumption." ], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "threshold-population", "statement": "The mean threshold of the cells at their best frequency was 58 dB (SD 14 dB).", "source": "Materials and Methods, Electrophysiology and stimulus design, paragraph 2", "quote": "The mean threshold across the population was 58 dB (SD 14 dB)", "statistics": [], "kind": "finding", "inferred": "", "outcome": "blocked", "blocked_by": "data_missing", "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Data1 holds spike times, cell.dat (stimulus names, played intensities, cutoffs, same-cell links) and the stimulus files; no threshold, best-frequency or tuning measurement. The data descriptor says the best frequency was determined from f-I curves but does not deposit them for Data1.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "table1-bursting-sessions", "statement": "The number of sessions with bursts of n > 1 depends on the stimulus, from none for 3 dB and for 800 Hz to all 7 for 12 dB and 7 of 8 for 100 Hz; 40 of 43 reference sessions, 6 of 7 at 25 Hz and 3 of 7 at 400 Hz.", "source": "Table 1, row Sessions with n > 1; Results, Stimulus characteristics modulate burst probability, paragraph 8", "quote": "The fraction of bursting sessions, the percentage of isolated spikes (1-bursts), and the maximum n -value depend strongly on the standard deviation and cutoff frequency of the stimulus", "statistics": [], "kind": "finding", "inferred": "", "outcome": "not_reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Under the main burst reading the sessions of selection (c) with bursts of n > 1 are 40/47, 1/6, 7/7, 6/7, 7/8, 3/7 and 0/7 against 40/43, 0/6, 7/7, 6/7, 7/8, 3/7 and 0/8: six of seven counts equal the paper's, and one 3 dB session bursts where the paper has none. The plateau-excluded first maximum (peak c) gives 37/47 and the same other counts. Over the 115 sessions of selection (a) at most 2 of 7 are within tolerance under any combination (52/60, 2/9, 8/8, 9/10, 10/11, 3/8, 0/9 under the main reading), the first round's result. Peak reading a gives 5 of 47 bursting reference sessions, and N the number of trials none.", "deviations": [ "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it. For this claim selection (c) was chosen in part from the expected counts, so it is a variant and the outcome follows the selections the criterion supports, (a) and (b), under which at most 2 of 7 counts are within tolerance under any burst reading.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": false, "disagreed": true, "data": "dataset", "where": null, "proposed": null }, { "claim": "table1-isolated-spikes", "statement": "The percentage of isolated spikes among all bursts of the bursting sessions depends on the stimulus (74, 100, 62, 60, 55, 96 and 100% for protocols 1 to 7), and in every protocol isolated spikes are the most probable burst.", "source": "Table 1, rows % Isolated spikes and Most probable n; Results, Stimulus characteristics modulate burst probability, paragraph 8", "quote": "Notice, however, that in all cases, isolated spikes are more frequent than any other burst of n > 1.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "not_reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Under the main reading over selection (c) the pooled percentages of isolated spikes are 80.7, 71.7, 66.1, 78.0, 71.3, 95.0 and 100%; 4 of 7 within tolerance (protocols 1, 3, 6, 7), at most 5 under any combination and 4 over selection (a). n = 1 is the most probable n in every protocol under the main reading. The criterion requires every value and gives no partial level; 4 of 7 is below the five of seven at which the other Table 1 criteria call a result partial, so the first round's partially_reproduced is changed to not_reproduced, as the verifier gave it.", "deviations": [ "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it. The outcome is the same under every population shown, so it does not rest on this assumption.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": false, "disagreed": true, "data": "dataset", "where": null, "proposed": null }, { "claim": "table1-burst-length", "statement": "The highest n and the average n depend on the stimulus; the longest bursts (up to 15 spikes, average 2.10 with SD 3.67) occur at 25 Hz, and no bursts occur at 3 dB or at 800 Hz.", "source": "Table 1, rows Highest n and Average n", "quote": "", "statistics": [], "kind": "finding", "inferred": "The values are read from Table 1; the sentence of paragraph 8 states the dependence on the stimulus but not the values.", "outcome": "not_reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Under the main reading over selection (c) highest n 7, 9, 8, 14, 7, 5, 1, average n 1.23, 1.39, 1.49, 1.90, 1.39, 1.05, 1.00, SD of n 0.51 (p1), 0.82 (p3), 2.22 (p4), 0.71 (p5), 0.23 (p6); 8 of 19 within tolerance, at most 9 under any combination of readings and either selection (7 over selection (a) under the main reading). The longest bursts are at 25 Hz and 800 Hz has none, as in the paper; 3 dB has bursts of up to 9 spikes in its one bursting session.", "deviations": [ "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it. The outcome is the same under every population shown, so it does not rest on this assumption.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig5-burst-probability-vs-stimulus", "statement": "The probability of generating isolated spikes is lowest, and high-n bursts most frequent, for large amplitude fluctuations (12 dB) and cutoff frequencies around 100 Hz.", "source": "Figure 5; Results, Stimulus characteristics modulate burst probability, paragraph 10", "quote": "The probability of generating isolated spikes is minimal for large amplitude fluctuations and cutoff frequencies around 100 Hz.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Under the main reading over selection (c) P(n = 1) is 0.816, 0.749, 0.837, 0.980 and 1.000 at 25, 100, 200, 400 and 800 Hz, lowest at 100 Hz, and 0.953, 0.837 and 0.668 at 3, 6 and 12 dB, lowest at 12 dB; the same with bursting sessions only, over selection (a) and with the plateau-excluded peak. P(n = 2) and P(n = 3) are highest at 100 Hz and 12 dB over all kept sessions. With minimum reading b the lowest P(n = 1) moves to 25 Hz.", "deviations": [ "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig3-correlation-shape", "statement": "For a cell tested with all stimuli, larger stimulus SD gives correlation functions with sharper peaks, and faster stimuli lead from multi-modal (25 Hz) to single-peaked (100, 200 Hz) to shallower and broader correlation functions (400, 800 Hz).", "source": "Figure 3; Results, Stimulus characteristics modulate burst probability, paragraph 3", "quote": "Increasing the standard deviation of the amplitude distribution (from Figure 3 A to D to G) results in correlation functions that exhibit progressively sharper peaks ... Decreasing the typical time scale of the stimulus fluctuations (going right from Figure 3 B to F) leads from multi-modal (Figure 3 B) to single-peaked (Figures 3 C,D) to increasingly shallower and broader correlation functions (Figures 3 E,F).", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "The cell 03-04-23-ad to -aj of singlecells.txt was recorded with all seven protocols (six reference sessions). The first peak of C_s/rbar^2 above its baseline is 0.50 at 3 dB, 1.14 to 2.35 in the six 6 dB sessions and 5.06 at 12 dB (reading a, sharper with SD); its width does not order (reading b). C_s has 4 to 14 peaks at 25 Hz, one at 200 Hz, and one at 100 Hz at prominence 0.2 (two at 0.05 and 0.1); 3 dB, 400 and 800 Hz have no separable first peak. The model panels of Figure 3 are not tested (no thresholds).", "deviations": [ "The cell is identified from singlecells.txt as the only one with all seven protocols; the paper does not name it.", "Peaks are counted on C_s/rbar^2 smoothed over 1 ms as local maxima with a prominence above a threshold the paper does not give; 0.05, 0.1 and 0.2 are shown and the outcome rests on 0.2 for the single peak at 100 Hz.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "method-correlation-error", "statement": "The temporal average of the estimated error of the correlation function was on average 3.4% (SD 1.5%) of the span of C_s.", "source": "Materials and Methods, Burst identification, paragraph 6", "quote": "The population mean of the temporal average of this estimated error was 3.4% (SD 1.5%) of the total span of C s (\u03c4) (that is, the difference between the maximum and the minimum)", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "With sigma the SD of the product terms and N their number (reading a), over the whole lag range (reading b) and without tau = 0, the error averages 2.63% (SD 1.39%) of the span of C_s over the 89 sessions of selection (c) and 2.69% over the 115 of selection (a); the third reading of sigma and N gives 2.61% and 2.67%. With tau = 0 in the span all readings give 0.06 to 0.16% or, with N the number of trials, 6.3 to 7.9%; over 0 to 1.25 / cutoff without tau = 0 the mean is unstable (30 to 4500%).", "deviations": [ "The span of C_s leaves out tau = 0, where C_s holds the self term (two orders of magnitude above the rest); the paper does not say. Assumed because the claims file's own detail (1) excludes the zero-lag self term from the peak.", "A third reading of sigma and N (sigma the SD over trials of each trial's C_s, N the number of trials), from paragraph 6 (\"C_s can be interpreted as an average\"), is shown beside the claims file's two.", "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "model-matches-classification", "statement": "A threshold-linear Poisson model with refractoriness and no free parameters classifies sessions as bursting or not in agreement with the real data in 86% of sessions, and where both are bursting their limiting ISIs differ by less than 1 ms in 81% of cases.", "source": "Results, Stimulus characteristics modulate burst probability, paragraph 11; Figure 3 middle and lower subpanels", "quote": "When the analysis is extended to the whole population of cells, this agreement is observed in 86% of all sessions. Moreover, in those sessions where both real and simulated data are classified as bursting, the limiting ISI calculated with real and simulated data differ by <1 msec in 81% of the cases.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "blocked", "blocked_by": "data_missing", "notebook": null, "notes": "The model thresholds the stimulus at the measured threshold s_TH before filtering (Eq. 2); the criterion makes the claim untestable as the paper ran it without s_TH, and Data1 holds none.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig6-bursts-less-correlated", "statement": "Representing the spike train as n-bursts reduces correlations between time bins; the mean squared Pearson coefficient over t from 200 to 990 ms and tau from 0 to 10 ms is 2.89 times larger (SD 1.49) for spikes than for bursts, on average over the bursting sessions (2.94 for the example cell).", "source": "Results, Quantitative description of the information transmitted by bursts, paragraph 2; Figure 6", "quote": "For the cell shown in Figure 6 , the mean value of c s 2 ( t , \u03c4 ) averaged over all t \u2208 [200, 990 msec] and \u03c4 \u2208 [0, 10 msec] is 2.94 times larger than the corresponding mean of c b 2 ( t , \u03c4 ) . The population average of this ratio on all bursting sessions is 2.89 (SD 1.49).", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Over the 64 bursting sessions of selection (c) the mean square Pearson coefficient is larger for spikes than for bursts in every session. Without tau = 0 the mean ratio is 1.99 (SD 0.92) with 0.1 ms bins and 2.17 (SD 1.86) with 0.4 ms bins, within 2.89 +/- 1.49; with tau = 0, 1.29 and 1.21, outside. Over the 84 of selection (a), 1.99 and 2.13. All bursting sessions have bursts of n > 1.", "deviations": [ "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig7ab-information-per-burst-increases", "statement": "The information carried by each n-burst increases with n; every bursting session has a significantly positive slope of I_n(1) against n, with an average slope of 1.5 bits (SD 0.7 bits) over 59 bursting cells.", "source": "Results, Quantitative description of the information transmitted by bursts, paragraph 5; Figure 7A and 7B", "quote": "All sessions had significantly positive slopes. Figure 7 B shows the distribution of slopes throughout the population. The average slope across the 59 bursting cells was 1.5 bits (SD 0.7 bits).", "statistics": [], "kind": "finding", "inferred": "", "outcome": "partially_reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Over the 64 bursting sessions of selection (c), within 59 +/- 5.9, the mean slope of I_n(1) on n is 2.22 bits (SD 0.87) with 0.1 ms bins, just outside 1.5 +/- 0.7, and 1.91 (SD 0.75) with 0.4 ms bins in bits per event (reading i), 1.78 and 1.53 as Eq. 10 prints it (reading ii), within. All 64 slopes are positive, 63 significant by one standard error and 62 by the t-test, so not every session. Over the 84 of selection (a): 2.14, 1.85, 1.71 and 1.48, 83 and 82 significant. A baseline with each trial's burst onsets shifted at random gives a mean slope of 2.59 bits (0.1 ms bins) and 2.23 (0.4 ms), above the data's; the data's slope exceeds its shifted copy in 3 of 64 sessions (9 with 0.4 ms bins). A shifted train has lost its stimulus locking, so its plug-in bias is at least that of the data; the comparison shows that the estimator's bias for rare events is of the order of the rise of I_n(1) with n, not that the whole rise is bias. The criterion fixes the formula, so the outcome rests on it.", "deviations": [ "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination.", "A baseline not in the paper: I_n(1) and its slope recomputed with each trial's burst onsets shifted by a random offset, wrapped within 200 to 1000 ms, ten shifts per session." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig7cd-occurrence-decreases", "statement": "High-n bursts are rare; in every session the frequency of occurrence decreases significantly with n, with an average slope of -2.6 (SD 0.7).", "source": "Results, Quantitative description of the information transmitted by bursts, paragraph 6; Figure 7C and 7D", "quote": "This result was also observed in all other recorded cells ... the frequency of occurrence always decreased significantly with n . The population data in Figure 7 D had an average slope of \u22122.6, with SD of 0.7.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "partially_reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Over the 64 bursting sessions of selection (c) the mean slope of log2 occurrence on n is -2.82 (SD 1.12), within -2.6 +/- 0.7 (natural log -1.96, log10 -0.85). All 64 slopes are negative; 63 are significant by one standard error and 61 by the t-test, so not every session, but above 90%. Over selection (a): -2.71, 83 and 81 of 84.", "deviations": [ "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig7ef-jitter-independent-of-n", "statement": "The jitter of the first spike of a burst is roughly independent of n in 80% of the bursting sessions, and the mean slope of jitter against n is -0.03 ms (SD 0.2 ms).", "source": "Results, Quantitative description of the information transmitted by bursts, paragraph 6; Figure 7E and 7F", "quote": "At the population level, in 80% of the bursting sessions the amount of jitter was roughly independent from n (the best linear fit had a slope that was not significantly different from 0) ... The mean slope was \u22120.03 msec (SD 0.2 msec).", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Over the bursting sessions of selection (c), events read as onsets within the first-peak domain around an event centre (reading b) give a mean slope of 0.005 ms (SD 0.052) and 86.9% of the 61 sessions with a test not significant by the two-sided t-test, within 80 +/- 8. Events between histogram troughs (reading a) give -0.001 ms and 88.9% (just outside); with significance read as one standard error 32.8% and 44.4%. Over selection (a), 87.7% and 91.6%. Median jitter of 1-bursts 0.389 ms (b) and 0.058 ms (a). Three sessions have fewer than three n values with a jitter and no test.", "deviations": [ "How events are formed is open (Rokem et al., 2006 is cited for the definition). Reading b is implemented with event centres taken as the 0.1 ms onset-histogram bins from the highest down and a window of the limiting ISI centred on each; reading a fragments events (median jitter 0.058 ms, below the paper's 0.2 ms error bar).", "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig7gh-information-rate", "statement": "The information rate of n-bursts decreases with n (average slope -11 bits/s, SD 16 bits/s), yet bursts with n > 1 together carry on average 47% of the total information over the bursting cells; in the example cell the total is 220 bits/s.", "source": "Results, Quantitative description of the information transmitted by bursts, paragraph 7; Figure 7G and 7H", "quote": "I \u2032 n strongly decreases with n (Figure 7 G). Similar results were obtained in all other recorded sessions (Figure 7 H), with an average slope of \u221211 bits/s (SD 16 bits/s) ... The population average of this fraction among all bursting cells was 47%.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "not_reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "The information rate falls with n in all 64 bursting sessions of selection (c); mean slope -46.5 bits/s (SD 39.0) with 0.1 ms bins and -38.1 (SD 33.1) with 0.4 ms bins, outside -11 +/- 16; bursts with n > 1 carry 35.5% and 37.7%, outside 47 +/- 4.7. Total rates average 241.4 and 199.8 bits/s. With each trial's burst onsets shifted at random the same formula gives 106.5 and 37.5 bits/s, slopes of -16.1 and -3.9 bits/s and fractions of 46.8% and 63.2%, so much of these rates is estimator bias. Over selection (a): -44.7 and -36.6 bits/s, 36.3% and 38.6%.", "deviations": [ "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination.", "A baseline not in the paper: I_n(1) and its slope recomputed with each trial's burst onsets shifted by a random offset, wrapped within 200 to 1000 ms, ten shifts per session." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "direct-method-comparison", "statement": "Information rates from the burst code are highly correlated with those of the direct method (R = 0.95), averaging 191 against 222 bits/s, the direct method being higher in all sessions but one by 31 bits/s on average, so the burst code keeps about 85% of the information.", "source": "Results, Quantitative description of the information transmitted by bursts, paragraph 10; Discussion, Comparison with other neural codes, paragraph 3", "quote": "The information measures obtained by our method and by the direct method were highly correlated ( R = 0.95, using all sessions). The population average obtained with the direct method is 222 \u00b1 69 bits/s. With our method, instead, this average was 191 \u00b1 72 bits/s. In all cases but one, the information obtained with the direct method was higher than the one obtained with our method, the average difference being 31 \u00b1 16 bits/s.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "partially_reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "With the burst code in 0.4 ms bins over the 89 kept sessions of selection (c), R = 0.967, direct (8 bins) 198.4 bits/s, burst code 178.1, mean difference 20.3 (SD 21.2), all within tolerance; 10.2% lost, outside 14 +/- 1.4; the direct method is higher in 74 of 89 (83%), not all but one. Over the bursting sessions R = 0.951, 220.7 against 199.8, difference 21.0, higher in 52 of 64. Over selection (a) R = 0.965, 204.3 against 183.6, 10.1% lost, higher in 94 of 115. Extrapolated direct rates differ by under 1 bit/s. With 0.1 ms bins the burst code exceeds the direct method on average (220.1 against 198.4).", "deviations": [ "NSB entropies computed by this entry's implementation (ia1.nsb_entropy), not the authors' code, which is not public.", "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig8-nbta-differ", "statement": "In every bursting cell the burst-triggered averages for n = 1 to 4 differ significantly (in the example cell their central peaks grow higher and wider with n); 50 of the 58 bursting sessions show significantly different central peaks in the time domain, the other eight having too few bursts.", "source": "Results, Qualitative description of the information transmitted by bursts, paragraphs 3 and 6; Figure 8D", "quote": "All recorded and simulated bursting cells exhibited significantly different n -BTAs, for n ranging between 1 and 4 ... out of the 58 sessions where bursting cells were found, 50 exhibit n BTAs whose central peaks were significantly different from one another", "statistics": [], "kind": "finding", "inferred": "", "outcome": "partially_reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Of the 64 bursting sessions of selection (c) (just outside 58 +/- 5.8), 53 have at least two bursts of each n = 1 to 4 and all 53 pass the ANOVA, with all frequency bins and with bins up to the cutoff. With the central maximum searched before the onset, 38 have marked points in every central peak (reading a), outside 50 +/- 5, and 5 have most of every central peak marked (reading b); over the whole window 42 and 7. With at least 20 bursts of each n 33 of 39, with at least 50, 25 of 28. Over selection (a), 54 (60 over the whole window) of 71. The ANOVA part holds; the time-domain count does not.", "deviations": [ "The central maximum of an nBTA, and so tau_n, is searched from -25 to 0 ms, before the burst onset, as the Results call tau_n the delay between stimulus upstroke and burst generation (Qualitative description, paragraph 3); the first round searched -25 to +15 ms, which is shown as a variant.", "Sessions need at least two bursts of each n = 1 to 4, the least that gives a variance; the paper says only that eight sessions had too few bursts (Results, Qualitative description, paragraph 6). Assumed; the results with at least 20 and 50 bursts of each n are shown.", "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig8-latency-decreases", "statement": "The latency between the stimulus upstroke and burst onset decreases with n, so that high or wide deflections produce prompt responses with high-n bursts.", "source": "Results, Qualitative description of the information transmitted by bursts, paragraph 3; Figure 8D; Discussion, Implications for the neural code, paragraph 4", "quote": "Moreover, the mean delay between stimulus upstroke and burst generation decreases systematically with n . This implies that stimulus deflections that are either high or wide tend to produce prompt responses, with high- n bursts.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "With the central maximum searched before the onset, mean tau_n is 5.80, 4.23, 3.32 and 2.67 ms for n = 1 to 4 over the 53 sessions of selection (c) (reading b). No tau_n is negative, by construction, as the search ends at 0 ms; in 13 of the 53 sessions 17 values sit on that edge (1, 5 and 11 for n = 2, 3, 4), and the mean tau_4 includes them. Without the values on the edge the means are 5.80, 4.32, 3.67 and 3.37 ms; over the 40 sessions with none on the edge 5.77, 4.33, 3.77 and 3.40 ms; of the 48 sessions with tau_1 to tau_3 off the edge, 32 have tau_1 > tau_2 > tau_3. 51 of 53 have tau_4 < tau_1 (Wilcoxon p = 4.5e-9) but only 20 of 53 decrease at every step (reading a, not a majority). Searched over the whole window, as in the first round, 5.80, 4.22, 3.14 and -0.50 ms, with a negative tau_n in 33 sessions. With at least 20 or 50 bursts of each n the means fall from 6.01 to 2.12 and from 5.88 to 2.05 ms; over selection (a) from 5.75 to 2.46 ms.", "deviations": [ "The central maximum of an nBTA, and so tau_n, is searched from -25 to 0 ms, before the burst onset, as the Results call tau_n the delay between stimulus upstroke and burst generation (Qualitative description, paragraph 3); the first round searched -25 to +15 ms, which is shown as a variant.", "Sessions need at least two bursts of each n = 1 to 4, the least that gives a variance; the paper says only that eight sessions had too few bursts (Results, Qualitative description, paragraph 6). Assumed; the results with at least 20 and 50 bursts of each n are shown.", "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig8e-sigma-minimum", "statement": "The standard deviation of the stimuli preceding an n-burst has a minimum at a position nearly the same for all n (about 7 ms before burst onset in the example cell, with an SD of 0.33 ms across n); over the bursting cells the SD of the minimum's position across n = 1 to 4 averages 0.4 ms (SD 0.7 ms), roughly 0.05 times the inverse cutoff frequency.", "source": "Results, Qualitative description of the information transmitted by bursts, paragraph 4; Figure 8E", "quote": "The constancy of the location of the minima also holds at the population level ... Its average among all bursting cells is 0.4 msec (SD 0.7 msec) considering 1 \u2264 n \u2264 4.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "not_reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "The minimum of sigma_n lies near -6 ms for every n (medians -6.85, -6.3, -4.8, -6.6 ms) but over the 53 sessions of selection (c) the SD of its position across n = 1 to 4 averages 2.37 ms (SD 2.44) in window a and 2.84 ms in window b (2.05 and 2.46 ms with the population SD), outside 0.4 +/- 0.7; times the cutoff 0.35 to 0.48, against 0.05. With at least 20 bursts of each n 2.14 ms (39 sessions), with at least 50, 1.74 ms (28 sessions; 0.19 times the cutoff), and 1.46 ms over the 38 such sessions of selection (a).", "deviations": [ "Sessions need at least two bursts of each n = 1 to 4, the least that gives a variance; the paper says only that eight sessions had too few bursts (Results, Qualitative description, paragraph 6). Assumed; the results with at least 20 and 50 bursts of each n are shown.", "The SD across n is the sample SD (N - 1, MATLAB's default); the population SD is shown too.", "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "sigma-n-below-sta-sigma", "statement": "In 98% of bursting cells, for all n, the SD of the stimuli preceding an n-burst is smaller than the SD of the stimuli preceding all spikes, by a ratio of minima of 1.62 on average (SD 0.56).", "source": "Results, Qualitative description of the information transmitted by bursts, paragraph 5", "quote": "For 98% of the bursting cells and for all n -values, \u03c3 n (\u03c4) was smaller than the standard deviation \u03c3(\u03c4) of the stimuli preceding all spikes (prior to any classification). The population average of the ratio of the minimum value of \u03c3(\u03c4) to the n -average of the minimum values of \u03c3(\u03c4) was 1.62 (SD 0.56).", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "In 98.1% of the 53 sessions of selection (c) the minimum of sigma_n is below that of sigma for all n = 1 to 4 (reading a); the ratio of minima averages 1.596 (SD 0.280). At every tau (reading b) 1.9%. A size-matched control (sigma from random subsets of the spike-triggered segments of each n's size, 20 draws) has a minimum about 5% below that of all segments; min sigma_n is below the size-matched minimum for every n in 94.3% of sessions, with a ratio of 1.517, so the result is not an effect of sample size. With at least 20 or 50 bursts of each n, 97.4% and 1.562, 96.4% and 1.571. Ratios per n 1.27, 1.45, 1.62 and, for n = 4, 23.5, inflated by sessions with few 4-bursts (2.38 with at least 20).", "deviations": [ "Sessions need at least two bursts of each n = 1 to 4, the least that gives a variance; the paper says only that eight sessions had too few bursts (Results, Qualitative description, paragraph 6). Assumed; the results with at least 20 and 50 bursts of each n are shown.", "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination.", "A control not in the paper: sigma from random subsets of the spike-triggered segments of the size of each n's, 20 draws per n." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig9-nbta-not-concatenated-1bta", "statement": "The nBTA differs from n 1BTAs shifted by the burst's own ISIs and summed; for a large fraction of cells, real and simulated, the difference is significant, the fraction diminishing as n grows; results are similar with shifted STAs.", "source": "Results, Qualitative description of the information transmitted by bursts, paragraphs 7 and 8; Figure 9A and 9B", "quote": "A large fraction of the cells show a significant difference, for both real and simulated data. Hence, also at the population level, the n BTAs differ significantly from the convolved 1BTAs. As n increases, the number of sessions with significant differences diminishes.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Over the bursting sessions of selection (c) the ANOVA finds the nBTA different from the summed 1BTAs in 100% of sessions for n = 2, 3 and 4 paired (64, 63 and 53 sessions), and 100, 100 and 98.1% grouped, so the fraction does not grow; the share with over 70% of the interval significant falls from 93.8% to 84.1% and 52.8%. With the STA the ANOVA fractions are 100, 100 and 98.1% and the 70% share falls from 100% to 37.7%. The average of the shifted copies and selection (a) give the same picture. The ANOVA, which treats every segment and frequency bin as independent, rejects in nearly every session; the criterion specifies it. The model sessions are not tested.", "deviations": [ "The central maximum of an nBTA, and so tau_n, is searched from -25 to 0 ms, before the burst onset, as the Results call tau_n the delay between stimulus upstroke and burst generation (Qualitative description, paragraph 3); the first round searched -25 to +15 ms, which is shown as a variant.", "Eq. 15 sums 1BTAs of the stimulus in dB; the sum is taken after subtracting the session's mean stimulus, as the paper does for Figure 9C, since a sum of n absolute levels cannot compare with an nBTA. The average of the shifted copies is shown as a variant.", "The paired ANOVA (reading a) tests the mean difference between each burst's segment and its own f_n and its variation across frequency bins.", "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig9c-population-nbta", "statement": "In the seven cells driven at 100 Hz cutoff and 6 dB, the population-average nBTAs show that bursts with more spikes follow higher or wider stimulus deflections, with large spread across cells.", "source": "Results, Qualitative description of the information transmitted by bursts, paragraph 9; Figure 9C", "quote": "Figure 9 C demonstrates that also at the population level, high- n bursts are associated with either higher or wider stimulus deflections (or both).", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Selection (c) holds 7 bursting protocol-5 sessions with bursts of n = 1 to 3, from 7 distinct cells (rows of singlecells.txt; 03-04-23-aa and 03-04-23-ah are different cells), the paper's seven cells. Their population-average central peak is 3.13, 7.03 and 8.46 dB above the mean stimulus for n = 1, 2, 3, with half-height widths 5.00, 5.50 and 7.35 ms (12.55, 13.60, 17.80 ms between minima) and SD across sessions 0.54, 0.83 and 1.51 dB at the peak. Both searches for the maximum give these values; the 10 sessions of selection (a) give 3.22, 7.11 and 8.60 dB.", "deviations": [ "The central maximum of an nBTA, and so tau_n, is searched from -25 to 0 ms, before the burst onset, as the Results call tau_n the delay between stimulus upstroke and burst generation (Qualitative description, paragraph 3); the first round searched -25 to +15 ms, which is shown as a variant.", "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "7 protocol-5 sessions of selection (c), as the paper has seven cells; 10 over selection (a).", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "fig10-burst-n-vs-height", "statement": "The probability of a burst with n spikes after a stimulus maximum depends on its height, low excursions giving no response or an isolated spike and high ones doublets, triplets or 4-bursts; all bursting cells show this.", "source": "Results, Qualitative description of the information transmitted by bursts, paragraph 10; Figure 10", "quote": "Whereas fairly low excursions produce either no response (dotted line) or an isolated spike (black, solid line), large deflections are associated with doublets (dark grey), triplets (grey), or bursts with four spikes (light grey line). All cells showing a bursting behavior exhibited this phenomenon.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "not_reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "Over the 64 bursting sessions of selection (c), with tau_n from the maximum before the onset, the h at which P(n | h) peaks increases from n = 0 to the largest n (up to 4) in 7 sessions (10.9%) and the P-weighted mean h in 15 (23.4%), under both widths T (median 7.40 and 3.05 ms), against all bursting cells; tau_n over the whole window gives 7 and 16, selection (a) 8 and 19 to 21 of 84. Restricted to n = 0 to 3 (outside the criterion) 34 and 36 to 37 of 64 hold. The two T readings match different numbers of maxima in 58 of 64 sessions (median 8316 against 8173), but the peak flag changes in none and the mean flag in 2, one each way, which is why the counts are equal.", "deviations": [ "The central maximum of an nBTA, and so tau_n, is searched from -25 to 0 ms, before the burst onset, as the Results call tau_n the delay between stimulus upstroke and burst generation (Qualitative description, paragraph 3); the first round searched -25 to +15 ms, which is shown as a variant.", "A stimulus maximum whose closest burst was assigned to a nearer maximum counts as no response; the Methods do not say whether it is matched to another burst.", "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "firing-rate-no-effect-on-limit", "statement": "The firing rate has practically no effect on the limiting ISI; a 50 Hz increase in rate shifts it by less than 0.4 ms and raises the average n by less than 25%.", "source": "Results, Stimulus characteristics modulate burst probability, paragraph 9", "quote": "More specifically, a 50 Hz increase in firing rate shifts the limiting ISI by <0.4 msec, which is comparable to its estimated error bar. The average intra-burst spike count n , in turn, shows an increase of <25%.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "reproduced", "blocked_by": null, "notebook": "reproduce_eyherabide_2008.ipynb", "notes": "No cell was recorded at two target rates (headers hold 100 and 90 Hz), so reading a cannot be formed. Under reading b over selection (c) a 50 Hz higher rate goes with a limiting ISI shorter by 0.337 ms (p = 0.44, 64 sessions) and an average n higher by 15.7% (p = 0.132, 89 sessions); within protocol 1, 0.057 ms and 18.9%. Over selection (a), 0.189 ms and 13.6%. These are differences between cells.", "deviations": [ "The kept sessions are the 89 of the 116 seven-protocol sessions with a trial-to-trial rate SD below 35 Hz, at least 98 trials and spikes detected on-line (selection c), not the paper's 86, whose selection is not stated. Assumed because the Methods give 98 to 503 trials for the paper's sessions, and because with the off-line sorted sessions also left out the kept sessions per protocol (47, 6, 7, 7, 8, 7, 7) and the bursting sessions per protocol come close to Table 1; the off-line exclusion has no passage in the paper behind it. The 115 sessions of selection (a), the first round's population, are shown beside it.", "Burst identification takes the first peak as the largest value of C_s below 5 ms (peak reading b), the first local minimum after it that differs significantly from it (minimum reading a), and the error bar with sigma the SD of the product terms of Eq. 1 and N their number (reading a). Assumed because the first local maximum (reading a) falls on edge effects of the refractory plateau, where C_s has no spike pairs, which the Methods say the criteria reject (Burst identification, paragraph 5), and leaves 11 of 89 sessions bursting; started after the plateau (peak reading c, added after the first verification) it leaves 61, near the 64 of the main reading; with N the number of trials no session bursts; the main reading gives an error bar of 2.6% of the span of C_s, near the paper's 3.4% (paragraph 6). The notebook shows every combination." ], "assumed": true, "disagreed": false, "data": "dataset", "where": null, "proposed": null }, { "claim": "model-reproduces-coding-trends", "statement": "The threshold-linear model with refractoriness reproduces the coding trends of the data, at single-cell and population level, though it is less precise in time than the real cells.", "source": "Results, Quantitative description of the information transmitted by bursts, paragraph 8; Figure 7, right half; Figures 8C and 9D", "quote": "A comparison between the left and right panels of Figure 7 reveals that the model reproduces the general trends observed in the experimental data, both at the single-cell and population level.", "statistics": [], "kind": "finding", "inferred": "", "outcome": "blocked", "blocked_by": "data_missing", "notebook": null, "notes": "Depends on model-matches-classification, blocked because Data1 holds no thresholds s_TH.", "deviations": [], "assumed": false, "disagreed": false, "data": "dataset", "where": null, "proposed": null } ], "findings": [ { "kind": "selection_undocumented", "statement": "The paper keeps 86 of 132 sessions by a trial-to-trial rate SD below 35 Hz and a rate near 100 Hz. In Data1 the SD rule keeps 115 of the 116 sessions of the paper's seven protocols (and all 116 with whole-trial rates), and nothing marks which sessions the paper kept. Leaving out also the 9 sessions with fewer than 98 trials and the off-line sorted sessions keeps 89, with 47, 6, 7, 7, 8, 7 and 7 per protocol against Table 1's 43, 6, 7, 7, 8, 7 and 8, and bursting sessions per protocol equal to Table 1's in six of seven, which suggests the paper used only on-line sorted sessions; the paper does not say so.", "found": null, "status": null }, { "kind": "data_missing", "statement": "Data1 holds no threshold, best frequency or f-I measurement for its cells, though the paper's model (Eq. 2) uses the measured threshold s_TH and the descriptor says the best frequency was determined from f-I curves.", "found": null, "status": null }, { "kind": "identifier_missing", "statement": "Data1 does not record the animal; the paper's 23 animals can only be matched to the 23 recording dates of the seven-protocol sessions by assuming one animal per date. Two experiments (03-01-15-aa, 03-06-27-ae) are missing from singlecells.txt.", "found": null, "status": null }, { "kind": "metadata_insufficient", "statement": "28 sessions use the stimuli gaussModality_co200 and lmd_co200 and 6 use cutoffs of 50 and 1600 Hz; the descriptor does not describe these stimuli and the paper does not use them. 28 sessions sorted off-line have no header, so their target rate and pause are not recorded, and in 25 of them each trial block runs on into the pause (spikes up to 2.76 s), which the descriptor does not mention. The file Data1/firing_rates is not described; it gives each spike file's trial count and a rate with a +/- value of 57 to 189 Hz that is not the trial-to-trial SD of the rate.", "found": null, "status": null }, { "kind": "metadata_error", "statement": "The descriptor says the 1 s trials number 98 to 533; the files hold 69 to 503. The intensity in three trial headers differs from cell.dat by more than 1 dB (03-04-11-aa sig2, 03-04-16-ae sig2, 03-04-23-ae sig1, up to 6.6 dB), and six sessions of 03-01-30-ad, 03-02-25-ac and 03-02-25-aj had 450 ms pauses where the paper states 700 ms.", "found": null, "status": null }, { "kind": "methods_unavailable", "statement": "Questions for the paper's authors. (1) Burst identification; which maximum is the first peak of C_s (the first local maximum falls on edge effects of the refractory plateau and leaves 11 of 89 sessions bursting, 61 when the search starts after the plateau, the largest value below 5 ms 64), and what sigma and N are in Delta = sigma / sqrt(N) (the product terms and their number give 2.6% of the span when tau = 0 is left out of the span; N the number of trials gives no bursting session). (2) Which rule, besides the rate SD, removed the sessions not among the 86; whether the sessions with spikes detected off-line and those with fewer than 98 trials were left out. (3) How events were formed for the jitter (Rokem et al., 2006). (4) Whether the mean stimulus was subtracted before summing 1BTAs in Eq. 15. (5) What separates a peak of C_s for the multi-modal and single-peaked shapes of Figure 3. (6) Whether the maximum of the nBTA that gives tau_n was searched before burst onset, and how many bursts of each n a session needed. (7) Whether I_n(1) was corrected for the bias of the plug-in estimate, whose bias here is of the order of its rise with n.", "found": null, "status": null }, { "kind": "paper_inconsistency", "statement": "The paper counts bursting sessions three ways, 63 with n > 1 in Table 1, 59 bursting cells for Figure 7B and 58 bursting sessions for Figure 8, without saying why.", "found": null, "status": null } ], "reconciliation": null, "verifications": [ { "date": "2026-11-04", "kind": "agent", "by": "Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook)", "reviewed": "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 verifier also listed the archive and tested the helper functions on synthetic inputs.", "agreed": 25, "of": 26, "disagreements": [ { "claim": "table1-isolated-spikes", "outcome": "not_reproduced", "reason": "Cell 19: at most 4 of 7 within tolerance (baa: 79.5, 77.7, 67.7, 78.1, 73.5, 95.0, 100). n = 1 is most probable in every protocol only under some readings." } ] }, { "date": "2026-11-04", "kind": "agent", "by": "Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook)", "reviewed": "Round 2 (verifier-2), the full executed notebook of the first revision with its conclusions and summary, the README, the helper module, the attempt record and the claims file, with the paper under paper/.", "agreed": 25, "of": 26, "disagreements": [ { "claim": "table1-bursting-sessions", "outcome": "not_reproduced", "reason": "Under selection (c) 6 of 7 counts are within tolerance, but selection (c) drops off-line sorted sessions, which nothing in the paper supports, and cell 7 says the exclusion is taken because with it the kept and bursting sessions per protocol come close to Table 1, so the selection was tuned on this claim's own expected values. The record does not credit (c) for table1-sessions-per-protocol or data-sessions-and-trials, yet credits it here, which is inconsistent. Under every selection the criterion supports (a: 2 of 7; b: 2 of 7) the outcome is not reproduced; selection (c) should be shown as a variant." } ] }, { "date": "2026-11-04", "kind": "agent", "by": "Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook)", "reviewed": "Round 3 (verifier-3), the one proposed change of outcome, for table1-bursting-sessions, with the full executed notebook of the second revision, the attempt record and the claims file, with the paper under paper/.", "agreed": 1, "of": 1, "disagreements": [] } ] } ], "entries": [ { "name": "crcns-ia-1-eyherabide-2008", "type": "reproduction", "title": "Burst coding in locust auditory receptors (Eyherabide et al. 2008)", "description": "Tests the findings of Eyherabide, Rokem, Herz and Samengo (2008) on CRCNS data set ia-1, which holds the spike times of locust auditory receptor cells under amplitude-modulated noise (Data1) and the stimuli. It identifies bursts from the spike-train correlation function, counts them per stimulus protocol, computes the information carried by bursts of each size and by the direct method, and the burst-triggered averages of the stimulus. The cells' thresholds, which the paper's model needs, are not in the data set.", "datasets": [ "crcns:ia-1" ], "papers": [ "10.3389/neuro.10.003.2008" ], "derived_from": [], "notebooks": [ { "path": "reproduce_eyherabide_2008.ipynb", "title": "Reproducing Eyherabide et al. (2008) from CRCNS ia-1", "executed": "notebooks/crcns-ia-1-eyherabide-2008/reproduce_eyherabide_2008.html" } ] } ] }