Validate

The Validate page is where you review detections, correct mistakes, and export a validated dataset.

Choose a validation sample

The first time you open Validate, PAMalytics presents the validation sampling wizard. Use it to define which clips will be included in the review set before opening the validation cards.

The wizard records:

• the selected grouping structure, such as species, site, or species × confidence band;
• the target number of clips;
• any confidence-band structure;
• the sampling strategy.

Available sampling strategies include:

• Representative sampling: randomly samples across the selected groups or strata
• Likely classifier mistakes: prioritises lower-confidence detections
• Strongest detections: prioritises higher-confidence detections
• Equal allocation: distributes the sample evenly across confidence bands
• Custom selection: creates a user-defined stratified review set

Before applying the sample, review the preview showing the number of selected and available clips, the groups or strata represented, and any groups below the requested target.

Once applied, the selected review set is passed to the validation interface. The sampling design is retained separately from the validation decisions.

What Validate is for

Validation is designed for fast review:

• Track what proportion of clips have been validated by site × species (or any other grouping)
• Monitor classifier accuracy as you review
• Sort and filter by clip probability (min–max segment probability per file)
• Use high-resolution spectrograms for quick visual checks
• Use audio playback with automated time expansion for ultrasonic calls
• Mark detections as uncertain where a confident decision is not possible
• Review pending changes before saving

Validation summary and filters

At the top of the page you’ll see a summary of validation progress, including:

• How many detections have been ingested
• The proportion reviewed
• The percentage changed
• The percentage the classifier got correct

Below the headline figures, PAMalytics breaks these metrics down by species, site (and other metadata), so you can balance your validation effort in line with your sampling strategy.

Use the expandable Advanced filters panel to focus your review effort. For example, you can:

• Order clips from lowest to highest classifier confidence (useful for targeting potential false positives)
• Filter to a specific site or species
• Set a fixed frequency range for spectrogram visualisation
• Pre-select a preferred time expansion factor for audio
• Reduce the number of cards displayed per page when working with large review sets or on a slower computer

Review spectrogram cards

Below the summary and filters you’ll review detections using spectrogram “cards”, similar to the Detection Dashboard but with additional validation controls.

Each card includes two status indicators in the top-right:

• Review status (default: Not reviewed)
• Classifier performance (default: Not assessed)

Mark a card as reviewed

If all detections on the card are correct, select Mark card as reviewed. The indicators update (e.g. Reviewed and Classifier: all correct) and the summary table refreshes immediately.

Correct detections

If you disagree with any detection:

• Open the dropdown beneath the spectrogram
• Locate the detection you want to change
• Update it to Absent, Uncertain, or to the correct species

After making edits, select Mark card as reviewed. The classifier performance indicator will update (for example Classifier: mixed or Classifier: all incorrect).

Use Uncertain when the detection cannot be confidently confirmed or rejected. Marking a detection as uncertain does not by itself count as a classifier error unless the species label is also changed.

Any changes you make are recorded in the dynamic table beneath the spectrogram cards, so you can keep track of edits before exporting.

Export your validated results

When you’ve completed your review, download the validated CSV using the Download CSV option at the bottom of the page.

Your export will contain nine additional columns relating to the validation effort:

• species_name_original: the species label provided by the classifier
• presence_label_original: the presence/absence label provided by the classifier
• validation_state: whether the reviewer agreed with the classifier prediction (e.g. correct/incorrect/not reviewed)
• validation_label: the reviewer’s final presence/absence decision
• validation_species: the reviewer’s final species label decision
• validated_by: the reviewer name
• validated_at: the date/time of the review
• validation_notes: any notes recorded during validation
• uncertain_flag: whether the detection was marked as uncertain