Design preview

Process Review

Review a complete photographic encounter at a time, understand how species suggestions were produced, and correct decisions without losing the surrounding context.

Start the guide See the page map

Encounter 18

Varied Thrush 2 bursts · 5 photos

Burst 1

Burst 2

About this guide

This is a preview of the planned encounter-preserving Process Review redesign. It deliberately documents the intended experience, including editable scopes and expanded species evidence. Current Vireo releases may not have every control shown here; this status will be updated as the redesign becomes available.

Overview

One place to make the final review decisions

Process Review turns the output of a completed Process run into an editable review workspace. It keeps related photos together so that you can compare poses, verify species, correct grouping, and choose what to keep with the evidence still visible.

Decide

Accept or change automated keep, review, and reject recommendations.

Identify

Confirm one or more species while preserving the predictions that informed you.

Organize

Fix burst and encounter boundaries when the automatic grouping is not right.

The central rule: filters choose encounters, not loose photos

When one photo matches a filter, Vireo can show its complete encounter. This protects the sequence you need for comparison and prevents a narrower view from silently changing the grouping.

The mental model

Photos live inside bursts, and bursts live inside encounters

Understanding these three levels makes every Process Review action predictable.

EncounterA continuous sighting or photographic situation

BurstsShort, rapid sequences inside the encounter

PhotosThe individual files you inspect and triage

Vireo preserves this hierarchy while you filter and edit.

Photo

One image file. A photo can contain several visible subjects and therefore several species observations.

Burst

Photos taken close together, usually with little change in subject or composition. Burst-level review is useful for choosing the strongest frames.

Encounter

One or more related bursts that belong to the same sighting. It is the primary unit shown, filtered, paged, and counted.

Subject

A distinct detected animal or region in a photo. Species evidence belongs to a subject before it is summarized for the photo, burst, or encounter.

Orientation

Page map

The page keeps high-frequency review tools close to the photographs and moves detailed evidence or tuning into panels that can be opened when needed.

Encounter cardspecies · progress · summary

Next encounter

34 encounters loadedLoading more as you scroll…

EvidenceSubjectsModelsAlternativesHistory

A Coverage and summary B Filters and view controls C Encounter list D Optional evidence panel

Page header

States which Process result is being reviewed, how much of the workspace it covers, and whether changes still need sidecar synchronization.

Summary strip

Keeps Process-result inventory, triage, and species-review totals canonical while showing matching photos, complete encounters, and context photos separately for the current view.

Review toolbar

Searches and narrows the encounter list without modifying the saved Process result.

Encounter list

Loads complete encounters progressively in bounded pages as you approach the end of the list. Loading more never changes grouping.

Advanced controls

Adjust grouping and scoring, inspect why an encounter was formed, and control optional overlays.

Evidence panel

Explains species suggestions by subject, model, and label list while leaving the encounter visible.

Working with large reviews

Narrow the list without breaking encounters apart

A review view is a lens over one saved Process result. Changing the view decides which complete encounters appear; it does not rerun grouping and does not create a second, read-only result.

Start from the latest Process result

Every view is a lens over the same canonical result. Begin with All processed photos, and optionally narrow to one collection.

Add conditions

Narrow by triage, species-review state, encounter match rule, or a search across species, scientific name, filename, model name, label-list name, or photo ID.

Choose how encounters qualify

Use Any photo for inclusive discovery or Every photo for strict containment.

Review complete encounters

Once an encounter qualifies, all of its photos remain together and every edit uses the same stable identities.

Encounter A3 of 4 photos match

✓✓—✓

Any photoShow the complete encounterAt least one photo matches

Every photoDo not show the encounterOne photo does not match

The matching rule changes which encounters qualify, never which photos are removed from a qualifying encounter.

Setting	Use it when	An encounter appears when	What is displayed
Any photo	You want to find every encounter touched by a condition.	One or more photos match.	The complete encounter, including nonmatching photos.
Every photo	You want encounters wholly contained in a condition.	Every photo matches.	The complete encounter.

Photo decisions

Keep, review, and reject are recommendations you control

Automated triage gives each photo a starting state. Your manual decision always takes precedence and remains distinct from the automated recommendation.

Keep

A strong frame you intend to retain or carry forward.

Review

An uncertain frame that deserves another look before the review is complete.

Reject

A frame you do not intend to keep. This marks the photo; it does not delete the file.

Where a decision can apply

One photo: change only the selected frame.
Selected photos: apply the same decision to a deliberate selection.
One burst: update all photos in that rapid sequence.
One encounter: update all photos in the sighting.

Vireo names the target in the command and confirmation text. A control never silently widens from photo to burst or from burst to encounter.

Identification

Suggestions are evidence; confirmations are your conclusion

Vireo keeps three ideas separate: what a model predicted, how those predictions were summarized, and what you confirmed. Confirming a species does not erase a model disagreement or rewrite its confidence.

Human confirmationYour accepted identification, including multiple species when needed

Review-unit suggestionA summary for the photo, burst, or encounter—not a new prediction

Prediction sourceThe exact model and label-list pairing that produced the alternatives

Subject evidencePredictions attached to a particular animal or image region

Species decisions

What happens when species predictions disagree

Disagreement is a review signal, not an error to hide. Vireo distinguishes compatible differences—such as genus versus species—from true alternatives that cannot all describe the same subject.

Model A · Birds v4Song Sparrow82%

Model B · Regional birdsLincoln's Sparrow74%

Model A · Genus listMelospiza91%

Compatible: Melospiza Independent models disagree at species level Needs confirmation

Human decision Song Sparrow All original evidence retained

A broader taxon may be compatible with both alternatives; it does not cast an extra independent vote.

Recommended review sequence

1
Open Species evidence.
Start with the subjects that have a disagreement badge or low agreement.
2
Check whether the predictions describe the same subject.
Two correct labels on two different birds are not a disagreement.
3
Compare independent sources.
Model–label-list pairs are shown separately, while related lists from one model are consolidated before cross-model agreement is calculated.
4
Consider taxonomic compatibility.
A genus-level label can support a species-level label without pretending to be equally specific.
5
Confirm, add another species, or leave unresolved.
Uncertainty remains visible until you have enough evidence to decide.

Compatible

“Sparrow” and “Song Sparrow” can both be true at different levels. Vireo presents the more specific supported result and retains the broader evidence.

Competing

“Song Sparrow” and “Lincoln's Sparrow” are exclusive alternatives for the same subject and require adjudication.

Independent subjects

“Song Sparrow” on subject 1 and “Dark-eyed Junco” on subject 2 can both be confirmed for the same photo.

Insufficient evidence

If no alternative is reliable, leave the unit unconfirmed or record the broadest defensible taxon.

Complex photographs

One image can contain several subjects and several species

Species review is not limited to one label per file. Evidence is attached to subjects first, allowing a mixed flock or predator-and-prey image to be represented honestly.

1 2 3 One photograph

Subject 1Snow Gooseconfirmed

Subject 2Snow Gooseconfirmed

Subject 3Ross's Gooseconfirmed

The photo summary contains two species, while each subject retains its own evidence and confirmation.

Confirm each visible subject separately when detections can be distinguished.
Add a species manually if it is visible but was not detected.
Mark duplicate subject detections so they do not inflate support.
Use an image-level observation when subjects overlap and cannot be separated reliably.
At burst and encounter level, show the union of confirmed species with photo-level provenance.

Prediction provenance

How Vireo combines multiple models and label lists

Every prediction source has an identity: a model version paired with the exact label list it used. This matters because two lists run through the same model are related views of one system, while two independently trained models provide stronger evidence of agreement.

Model version+Label-list version=Prediction source

Situation	How Vireo handles it	Why
One model, one exclusive list	Shows ranked alternatives for one subject.	The labels compete with one another.
One model, several related lists	Normalizes compatible taxa and consolidates them as one model family before agreement is measured.	Related outputs must not become duplicate votes.
Several independent models	Shows each source and calculates agreement from calibrated, independently weighted evidence.	Agreement is useful only when provenance remains visible.
Multi-label presence model	Allows several labels to be present simultaneously.	These outputs are not mutually exclusive alternatives.
Different taxonomies	Maps labels to stable taxon identifiers and reports incompatible or unmapped versions.	Matching text alone is not enough to establish biological equivalence.

The evidence panel always shows

Model and version

Which trained model produced the output.

Label list and version

Which possible labels the model was choosing among.

Output mode

Exclusive alternatives or independent multi-label presence.

Confidence and rank

The source's score and the candidate's position within that source.

Subject

Which animal or image region the prediction describes.

Compatibility notes

How broader, narrower, synonymous, or unmapped taxa were interpreted.

Focused comparison

Use Group Review to compare a burst without distractions

Group Review opens one burst in a larger comparison workspace. It is the fastest place to choose the strongest photos, inspect close calls, and optionally confirm species before moving on.

Group Review · Burst 28 photos

Keep

Selected photo

Reject

Species: Varied Thrush□ Confirm speciesApply and close

1
Compare at a useful size.
Select a frame to inspect focus, pose, occlusion, and other available quality evidence.
2
Move photos between decisions.
Arrange the burst into Keep, Review, and Reject without leaving the comparison.
3
Handle outliers structurally.
If selected frames do not belong in the burst, extract them to a new burst rather than merely changing their triage status.
4
Commit deliberately.
Apply only the checked changes. Closing without applying leaves the underlying decisions unchanged.

Grouping corrections

Splitting and detaching solve different problems

Structural commands use precise verbs because they change different boundaries. A split cuts an ordered sequence at a boundary; an extraction removes selected items and makes a new sibling group.

SplitCut at a boundary

ABCD

Use when everything after a chosen point belongs in a second burst or encounter.

ExtractRemove selected items

ABCD

Use when a proper subset of a burst—contiguous or noncontiguous—should become separate sibling groups. Vireo partitions the source into maximal selected and unselected chronological runs, and every run becomes its own burst so no group spans an intervening run. To move a complete burst out of its encounter, use Separate burst as encounter.

“Detach” is represented in the interface by a specific extraction command: Extract photos to new burst or Separate burst as encounter.

Burst

Split burst here

Cuts one burst between two adjacent photos. Photos before the boundary stay in the first burst; photos at and after it form the second.

Burst

Extract photos to new burst

Removes a selected proper subset from its current burst and places the photos into new sibling burst membership in the same encounter. When the selection is not contiguous, the source burst is partitioned into maximal selected and unselected chronological runs and each run becomes its own burst. Selecting every photo is invalid; use Separate burst as encounter for a complete burst.

Burst

Merge adjacent bursts

Combines neighboring bursts inside the same encounter while preserving chronological photo order.

Encounter

Split encounter here

Cuts an encounter between adjacent bursts. Bursts before the boundary stay together; the remainder becomes a new encounter.

Encounter

Separate burst as encounter

Removes one complete burst and turns it into its own encounter. Other bursts remain in their existing encounter.

Encounter

Merge encounters

Combines selected compatible encounters and then orders their bursts chronologically.

Persistence

View changes are temporary; review decisions are saved

Action	Saved?	Effect
Search, filter, sort, or change page	Saved as a view preference	Changes only what you see.
Change triage or confirm species	Saved immediately	Updates the canonical review result and marks metadata as pending synchronization when applicable.
Split, extract, or merge	Saved immediately	Creates a new structural revision while preserving stable photo identities.
Adjust grouping controls	Preview first	Shows the proposed result; applying it creates a new revision.
Synchronize sidecars	Written externally	Writes supported decisions to metadata that other applications can read.

Make a decisionVireo saves a revisionUndo remains availableSynchronize sidecars when ready

Undo and redo describe the exact target and result, such as “Moved 2 photos to a new burst.” If another session changes the same review unit, Vireo pauses the conflicting action, refreshes that unit, and asks you to try again rather than overwriting newer work.

Task recipes

Common workflows

01Review a new Process result

Open Process Review from the completed run.
Read the coverage summary and note any photos outside the result.
Start with Review or Conflicts to address uncertain units.
Open Group Review for close comparisons.
Confirm species and finish triage.
Synchronize sidecars when the pending count is ready.

02Review one species without losing context

Add a species condition in the toolbar.
Choose Any photo.
Vireo selects every encounter containing at least one matching photo.
Review the complete encounters, including nearby frames with different or uncertain suggestions.

03Resolve a disagreement between models

Set Species review to Conflicts.
Open the evidence panel for the flagged subject.
Compare independent models, related label lists, confidence, and taxonomy notes.
Inspect adjacent frames in the burst for a clearer view.
Confirm the supported species, add multiple species, or leave the unit unresolved.

04Correct an incorrectly grouped sequence

Decide whether the mistake is a boundary or an outlier.
For a boundary, use Split burst here or Split encounter here.
For outliers, use Extract photos to new burst or Separate burst as encounter.
Use Merge only for adjacent or explicitly selected compatible groups.
Check the resulting cards and undo if the new structure is not what you intended.

05Make a large review faster

Use a collection, triage, or species-review condition to reduce the qualifying encounters.
Keep Any photo selected when context matters most; use Every photo for strict containment.
Sort disagreements and unresolved review units first.
Keep the evidence panel closed until a decision needs explanation.
Let the encounter list load progressively in bounded pages rather than loading the complete workspace at once.

When something looks wrong

Troubleshooting

Why am I seeing photos that do not match my filter?

You selected Any photo. One or more photos qualified, so Vireo included the complete encounter for context. Change to Every photo if the whole encounter must satisfy the condition.

Why did an encounter disappear with Every photo selected?

At least one photo in that encounter did not satisfy all active conditions. The encounter is omitted rather than shown partially.

Why do two label lists not count as two model votes?

If both lists came from the same underlying model family, their errors may be correlated. Vireo consolidates those related outputs before measuring agreement across independent models.

Why does a photo show more than one confirmed species?

The photo contains multiple subjects or an image-level observation with multiple taxa. Species confirmation is intentionally not restricted to one label per file.

What does “Needs confirmation” mean?

There is usable automated evidence, but no human confirmation for the review unit—or a disagreement, ambiguity, or missing mapping still needs attention.

Does Reject delete my photo?

No. Reject is a review decision used by later workflows. Process Review does not delete the underlying file.

Why did a click take a long time?

Changing view conditions should update a bounded encounter query and feel immediate. Evidence and thumbnails load progressively. If an action waits several seconds without visible progress, that is a performance problem rather than expected behavior.

Why does my current version say a collection is view-only?

The released page currently protects its saved review cache by making Workspace and Collection scopes read-only. The redesign documented here replaces those alternate results with editable encounter-preserving views. Until that change ships, switch to Latest review before editing.

Reference

Glossary

Complete encounter: Every photo and burst belonging to an encounter at the selected structural revision.
Confirmation: A human identification accepted for a subject, photo, burst, or encounter.
Encounter matching: The rule that determines whether an entire encounter qualifies for the current review view.
Exclusive alternatives: Labels where only one is expected to be correct for a subject, such as the ranked species from a classifier.
Label list: The versioned set of labels a model was configured to predict.
Model family: Outputs treated as related because they share the same underlying model or training lineage.
Multi-label prediction: An output where several labels may be present at the same time.
Prediction source: A model version paired with a label-list version and output mode.
Process result: The canonical grouping, scoring, and model evidence produced by a completed Process run.
Review unit: The level at which a decision is being made: subject, photo, burst, or encounter.
Review view: A saved or temporary set of conditions that chooses which complete encounters appear.
Structural revision: A saved version of burst and encounter membership after a split, extraction, merge, or regrouping change.