EthoTrace

Open a session folder containing the behavioral video (MP4) and the photometry file exported by the acquisition software (Fluorescence.csv: a configuration line, a header row, then one row per sample with a millisecond timestamp and paired 410-nm and 470-nm values). Both streams are loaded onto one timeline. Behavior is scored on a frame grid whose rate you set (30 frames/s by default), independently of the video's native frame rate. The analyses place photometry samples by their own timestamps rather than by frame index; the trace drawn under the video while annotating is a navigation aid spread evenly across the video's length. Annotations and analysis settings (analysis_settings.json) are written back into the same folder as plain files, so each session carries its own analysis configuration alongside its raw data.

The video plays above the ΔF/F trace and the behavior track, all sharing one zoom window, so signal and behavior are navigated together. Five categories sit on keys 1–5 (pup retrieval, nest-based care, pup approaching, pup investigation, and out of nest), and a bout is marked by pressing I at its first frame and O at its last.

• Bouts occupy one exclusive track, so no frame can carry two labels. They can be relabeled, resized, or deleted, with undo.
• Annotations save automatically as a human-readable JSON file listing each bout's inclusive frame range and category.
• Any bout can be exported as a video clip for inspection, or for training an automated classifier elsewhere.
• The ΔF/F trace can be hidden so that scoring stays blind to the signal.

The 470-nm signal is regressed on the 410-nm isosbestic channel by ordinary least squares across the whole session. The baseline is the fitted control offset by the 10th percentile of the regression residuals, and the resulting ΔF/F is smoothed with a Savitzky–Golay filter (11-sample window, cubic). No other detrending, z-scoring, or per-trial baseline subtraction is applied.

The Annotate tab shows a fast display ΔF/F (baseline = fitted control) for navigation; every reported number uses the analysis ΔF/F defined above. Keeping the two explicitly separate avoids treating a trace read off the annotation screen as the analyzed signal. (If a session has only a precomputed ΔF/F and no raw Fluorescence.csv, the analysis runs on that trace and the report warns that the baseline convention and timing differ.)

A bout becomes an analysis object as soon as it is marked. Every retrieval is drawn as its own trial, from 5 s before to 10 s after onset, with neighboring bouts shaded by behavior. When behaviors occur as a stereotyped sequence, a signal change near onset may belong to the scored behavior or to what immediately preceded it. Drawing each trial in its own behavioral context makes that ambiguity visible on the trial itself, which an average across trials cannot.

Repeated bouts of each of the four pup-directed behaviors (approach, investigation, retrieval, and nest-based care) are then aligned to their own onsets and shown as a heat map (longest bout first, one shared color scale) with the mean ± SEM beneath. Each bout keeps its full observed duration. Where an aligned window extends beyond a bout's available data, the cells are left empty rather than padded, held, or extrapolated: they appear hatched, and the mean is drawn dashed wherever fewer than half the bouts contribute, so declining data coverage is visible in the figure.

Every bout enters the statistics as an individual observation, and the test for each comparison is selected from assumption checks computed on the values being compared:

• Before vs. after onset: each bout contributes its mean ΔF/F over a window before onset and a window after onset (bouts that do not cover both windows are excluded). A Shapiro–Wilk test on the paired differences selects a paired t-test or a Wilcoxon signed-rank test.
• Between behaviors: each bout is reduced to the mean of its analysis ΔF/F from onset to offset and shown as an individual point on a box plot. Shapiro–Wilk tests on each group and a median-centered Levene's test select Student's t-test, Welch's t-test, or the Mann–Whitney U test.
• When a deciding assumption test returns P between 0.04 and 0.06, the choice is flagged as marginal and the P value of the test not selected is reported alongside it.
• Selection can instead be forced to parametric or nonparametric for a session, and the forced choice is recorded.

Tests are two-sided at α = 0.05, each with an effect size and a 95% confidence interval of the mean difference, and a Holm correction can be applied across the comparisons on a panel (on by default in the current build). The Stats .txt button saves a plain-text report listing every assumption check, every test, and every excluded bout, and figures export as PDF, PNG, or SVG.