Confidence Scores Explained

What a confidence score actually represents

A confidence score reflects how strongly the visible evidence supports a specific finding. It is not the same thing as truth, and it should never be read as a guarantee. A score is a review signal inside the report, not a final decision about what happened.

In practice, confidence tells you how stable a deduction is under the information available in the photo. Clear evidence usually pushes the score up. Occlusion, blur, weak lighting, missing context, or multiple plausible explanations pull it down. The same score can feel more or less important depending on the consequence of the finding.

Confidence is attached to a claim

Read the score together with the exact wording of the deduction. A report may be highly confident that an object is visible, but less confident about how that object came to be there. It may be confident that a package corner is crushed, but less confident about whether the damage happened before delivery, during transit, or after arrival.

This distinction keeps the review honest. The score should weight the claim that is actually being made, not a broader story you might be tempted to infer from it.

How to think about high, medium, and low confidence

What changes a score

Confidence rises when the image is sharp, the subject is visible, and the surrounding context makes the finding easier to interpret. It falls when the scene is cropped, cluttered, poorly lit, compressed, or missing critical details. Scores can also fall when several explanations fit the same visual pattern.

How to use scores well

The safest habit is to review the confidence score together with the annotated overlay and the written deduction. If all three line up cleanly, the finding is easier to use as a working note. If they feel mismatched, take more photos, check the original frame, or review manually.

High confidence can help you move faster through ordinary findings. Lower confidence often tells you exactly where more evidence would help. In a web-first workflow, use the larger screen to inspect the overlay region and decide whether the score feels consistent with the visible support.

A practical review checklist

Why low confidence can still be useful

A low-confidence finding is not useless. It can still point you toward an anomaly, a missing angle, or a question worth checking. In many real scenes, the best next step comes from noticing where the evidence is thin.

That makes confidence useful as a review tool. It shows not only what the system sees, but where uncertainty remains. A low-confidence note can become a capture prompt: get a wider angle, photograph the detail straight on, improve lighting, or collect a supporting record outside the photo.

Common confidence mistakes

The first mistake is treating a percentage-like signal as a legal or factual conclusion. Probe does not determine liability, validate claims, or authenticate events with certainty. The second mistake is ignoring low confidence because it feels less authoritative. Low confidence can be the most useful part of the report when it tells you what to verify next.

The third mistake is comparing scores across unrelated scenes as if they are universal grades. A medium score on a dark, cluttered scene may still be a careful read. A high score on a simple visible object may not support any broader conclusion about cause or intent.

How Probe uses confidence in case reports

Probe presents confidence so the web platform can support review instead of blind acceptance. The score sits beside deductions and overlays because the report is meant to be checked. When you reopen a case later, the score helps you remember which findings were visually strong and which ones required caution.

This matters for shared work. If another person reviews the report, confidence gives them a starting point for discussion, not an instruction to agree. The most useful reports make strong findings clear and uncertain findings easy to spot.

When to capture more evidence

Capture more evidence when a low or medium score is attached to the main reason you created the report. A second image from a wider angle, a close-up of the marked area, or a cleaner shot under better lighting can change the review from guessing to checking. If new photos do not resolve the uncertainty, keep that limitation visible in your notes.