1. Start with a physical clue
The first selection receives the greatest weight because an observed sign is more specific than a room or season. A candidate that does not plausibly produce the selected evidence is capped even when several context fields match.
2. Add independent context
Location, scale, distribution, nearby material and timing contribute separate weights. They are deliberately not collapsed into a vague “similarity” score. Each field remains visible in the result so a user can challenge the ranking.
| Dimension | Role | Maximum contribution |
|---|---|---|
| Evidence type | Can the candidate produce the sign? | 24 points |
| Location | Does the environment fit normal activity? | 18 points |
| Scale | Is the reported size compatible? | 18 points |
| Pattern | Does the distribution fit a route, nest or exit? | 14 points |
| Context | Are food, moisture, wood or fabric relevant? | 14 points |
| Timing | Does activity timing support the profile? | 12 points |
3. Preserve contradictions
A high clue fit does not erase conflicting observations. Result cards show both aligned answers and profile-specific evidence that would point away from the candidate. This prevents the interface from presenting every selected field as confirmation.
4. Choose the next observation by information value
The next check is designed to split plausible candidates, not merely collect another similarity. Scale beside a ruler, a companion egg case, a specific damage pattern or the origin of falling material may change the ranking more than another general photograph.
5. Keep claims within the evidence
PestDecoder does not diagnose bites, allergic reactions, infections, animal disease, pesticide exposure or structural safety. It also does not certify that evidence was produced at a particular time. Those boundaries are part of the product, not fine print.
It is the percentage of available rule weight aligned with the selected answers. It is not an empirical probability that the candidate is present.
Review and correction process
Profiles require characteristic evidence, contradictions, a next observation, a safe first response, escalation criteria and cited sources before they can be indexed. Corrections are documented in version control, and high-consequence safety language is checked against current public-health or extension guidance.