Run several sessions through trial builder and QC

[arjunsridhar12345]

After meeting on 06/22 for a sync up, we are going to some test some sessions using the trial builder and produce csvs with trial table for them for further validation. Creating this issue to track that work. The sessions that will be processed are for this test subject 845023.

[arjunsridhar12345]

Adding to this, there is a test session that was collected, that was acquired with the following setup via a message in a teams thread:

• Start a session with no mouse. It doesn't matter to me what mouse ID to use
• Wait for the go cue, and then touch the R lickspout with a finger
• Repeat for a total of 5 go cues touching right, 5 go cues touching left, and 5 ignore go cues
• End the session.

This session was confirmed to be collected today 06/22. Will run the trials builder on this as well. This will allow for hopefully some ground truth validation as there should be 15 trials and such.

[arjunsridhar12345]

updated the issue to include QC as well. will post outputs in this issue

[arjunsridhar12345]

Here are the trials table for the sessions that have been uploaded as of now for 845023:

behavior_845023_2026-06-22_20-31-28_trials_table.csv
behavior_845023_2026-06-18_18-58-39_trials_table.csv
behavior_845023_2026-06-17_19-41-02_trials_table.csv
behavior_845023_2026-06-16_19-45-03_trials_table.csv
behavior_845023_2026-06-15_19-43-35_trials_table.csv
behavior_845023_2026-06-12_20-13-58_trials_table.csv
behavior_845023_2026-06-11_19-39-13_trials_table.csv
behavior_845023_2026-06-10_19-26-44_trials_table.csv

Some quick notes - For the 06/22 which is a ground truth, there is no right response even though there should have been I believe. The raw data confirms this as well from looking at the Response.json Software Event. Probably need to check with Micah. Side bias seems to be nan throughout - seems like its nan in the raw data as well.

From the discussion here: #15, I believe the lick spout positions are still outstanding, in addition to general validation.

Here are the logs also, there seems to be a mismatch for a couple of them. Could be some bug in the logic when making the table:

Processing behavior_845023_2026-06-17_19-41-02
Processing behavior_845023_2026-06-15_19-43-35
Processing behavior_845023_2026-06-22_20-31-28
Per-trial streams are misaligned with TrialOutcome; rows are paired by index so the table may be incorrect: QuiescentPeriod (start_time) has 15 events but there are 14 trials; Response has 15 events but there are 14 trials
Processing behavior_845023_2026-06-18_18-58-39
Processing behavior_845023_2026-06-12_20-13-58
Processing behavior_845023_2026-06-16_19-45-03
Processing behavior_845023_2026-06-10_19-26-44
Per-trial streams are misaligned with TrialOutcome; rows are paired by index so the table may be incorrect: QuiescentPeriod (start_time) has 785 events but there are 784 trials
Processing behavior_845023_2026-06-11_19-39-13

[arjunsridhar12345]

met with @micahwoodard today, 06/24 to discuss the above. Posting notes - Micah, please edit and correct anything that is not accurate!:

• Side bias being nan throughout for session on 06/22 (the ground truth) seems ok. There may not be enough trials
• I think Micah will try to look at the animal response and the mismatch in number of rows for the software streams

[alexpiet]

Change log

Starting a change log to document minor and intentional differences in NWB files here: AllenNeuralDynamics/dynamic-foraging-task#1754

Trials table index name

Arjun, on the existing trials table, when I do this:

df = nwb.trials.to_dataframe()

The index of the dataframe is named id. We should replicate this when you write the table to the NWB file

Side bias

• side_bias is NaN for all trials in behavior_845023_2026-06-10_19-26-44_trials_table.csv
• side_bias is NaN for all trials in behavior_845023_2026-06-22_20-31-28_trials_table.csv (this is the test session, probably just too few trials)
• side_bias should be strictly bounded between -1 and 1, yet it exceeds that for every session. Here is min, max for each session. @micahwoodard this could be acquisition

845023_2026-06-10_19-26-44: nan nan
845023_2026-06-11_19-39-13: -3.9907189567195833 1.2744730559506616
845023_2026-06-12_20-13-58: -1.2713705840242473 3.8203644355726207
845023_2026-06-15_19-43-35: 0.00114741577143 2.6238343831947817
845023_2026-06-16_19-45-03: 0.000449547948209 2.737163266092617
845023_2026-06-17_19-41-02: -0.3771616918618208 1.542499134453883
845023_2026-06-18_18-58-39: -0.0001982520406844 3.816731162620141
845023_2026-06-22_20-31-28: nan nan

lickspout position

• still issue with the values having different ranges (values like 10 instead of like 1)
• All of the new sessions have the same lickspout position for all trials, but the anti-bias algorithm states the lickspout was moved.

[micahwoodard]

• side_bias should be strictly bounded between -1 and 1, yet it exceeds that for every session. Here is min, max for each session. @micahwoodard this could be acquisition

@alexpiet definitely an acquisition bug! Will create an issue

[alexpiet]

Start time values are incorrect relative to go cue

col='start_time'
df_new[col+'_from_goCue'] = df_new[col] - df_new['goCue_start_time']
df_old[col+'_from_goCue'] = df_old[col] - df_old['goCue_start_time']
print('new: {}'.format(df_new['start_time_from_goCue'].mean()))
print('old: {}'.format(df_old['start_time_from_goCue'].mean()))

new: -0.002761265311665757
old: -3.90354462069106

[micahwoodard]

Start time values are incorrect relative to go cue

col='start_time'
df_new[col+'_from_goCue'] = df_new[col] - df_new['goCue_start_time']
df_old[col+'_from_goCue'] = df_old[col] - df_old['goCue_start_time']
print('new: {}'.format(df_new['start_time_from_goCue'].mean()))
print('old: {}'.format(df_old['start_time_from_goCue'].mean()))

new: -0.002761265311665757
old: -3.90354462069106

I bet this has to do with the curriculum setting some timings to 0. I will update the version on the rig today to update all of the corrected curriculum fields

[alexpiet]

Ok, lets check again with tomorrow's data. A lot of timing looks odd

new session:

old session:

[alexpiet]

That run at the end with the no responses looks like the baiting is causing the stop time to be triggered early?