Skip to content
finwave Docs

Confirmation Queue

What you’ll learn

Section titled “What you’ll learn”
  • What the confirmation queue is and why it exists
  • How identifications enter the queue
  • How items are prioritized for review
  • What actions are available for each evidence tier
  • What Disputed and Unverifiable mean

What is the confirmation queue?

Section titled “What is the confirmation queue?”

The confirmation queue is a prioritized work queue containing identifications that need human review before they become fully confirmed. It replaces the implicit “go find unconfirmed stuff” workflow with an explicit, filterable list ordered by review priority.

The queue contains identifications at evidence tiers 0 through 2 — attested records from contributor data, model suggestions, and model-corroborated matches that haven’t yet been confirmed by a human.

How identifications enter the queue

Section titled “How identifications enter the queue”

Identifications arrive in the queue through three paths:

  • Attested (Tier 0) — created during onboarding when a contributor’s historical records indicate an individual was present at an encounter. These have no bounding box.
  • ModelSuggested (Tier 1) — created when an ML model detects and identifies an animal but there is no corroborating evidence from the contributor’s records.
  • ModelCorroborated (Tier 2) — created when an ML model’s prediction agrees with the contributor’s attested records and the model confidence meets the population’s corroboration threshold.

Identifications that meet the auto-confirm threshold skip the queue entirely and are confirmed automatically.

Priority scoring

Section titled “Priority scoring”

The queue is ordered so that the easiest and most valuable items are reviewed first:

PriorityTierWhy
HighestModelCorroboratedLeast effort — has a bounding box and corroborating evidence. Just needs a human to verify.
MediumModelSuggestedHas a bounding box but no corroborating evidence. Needs more careful review.
LowerAttestedMost effort — no bounding box exists. The reviewer must find the animal in the encounter’s photos and draw a box from scratch.

Within each tier, more recent encounters and individuals flagged as active research subjects are prioritized higher.

Review actions

Section titled “Review actions”

The actions available depend on the identification’s evidence tier.

Attested records (no bounding box)

Section titled “Attested records (no bounding box)”

When reviewing an attested identification, the reviewer sees the encounter’s photos and the reported individual. Available actions:

  • Draw a bounding box and confirm — find the animal in one of the encounter photos, draw a box around the identifying feature, and confirm the identity. This promotes the record to Confirmed (Tier 3) and extracts the image crop.
  • Skip — leave the item in the queue for later review. It will be deprioritized.
  • Mark as Disputed — the reviewer believes the attested record is wrong and the animal was not actually present.
  • Mark as Unverifiable — the reviewer believes the record may be correct but the animal cannot be identified in any of the encounter’s photos (e.g., poor photo quality or the animal was not clearly captured).

ModelSuggested and ModelCorroborated records (has bounding box)

Section titled “ModelSuggested and ModelCorroborated records (has bounding box)”

When reviewing a model-assisted identification, the reviewer sees the staged crop alongside the full encounter photo. Available actions:

  • Confirm — the box and identity are correct. Promotes to Confirmed (Tier 3) and extracts the image crop.
  • Adjust box and confirm — the identity is correct but the box needs resizing. Creates a box-resized revision and promotes to Confirmed.
  • Change identity — the box is good but the model identified the wrong individual. Updates the identity, creates a revision, and promotes to Confirmed.
  • Reject — the detection is wrong (not a real animal feature, or completely misidentified). Deletes the record.
  • Skip — leave for later review.

Disputed and Unverifiable states

Section titled “Disputed and Unverifiable states”

Two states exist for identifications that cannot be confirmed:

Disputed

Section titled “Disputed”

A disputed identification means the reviewer believes the original record is wrong — the animal was not at this encounter. The record is removed from the animal’s sighting history but retained internally for audit purposes. The original metadata is preserved so the discrepancy can be investigated with the contributor.

Unverifiable

Section titled “Unverifiable”

An unverifiable identification means the reviewer believes the record may be correct, but the animal cannot be positively identified in any of the encounter’s photos. The record remains in the animal’s sighting history with a visual indicator that it is unverifiable. This is common for historical data where photo quality is poor or the animal was present but not captured clearly enough for annotation.

Section titled “Related”