01 / PROBLEM
The comparison was easy. Knowing when to trust it was not.
A scheduled system can collect prices and calculate differences. The harder question begins when a value is missing, a page changes shape or two sources disagree.
A blank field may mean unchanged, unavailable or broken. Treating those states as equivalent turns faster automation into faster uncertainty.
Missing evidence is a state to investigate, not a value to assume.
02 / EVIDENCE
Every candidate change needed its own evidence packet.
The review object was designed to answer five questions before asking a person to decide.
When was this evidence collected?
Which expected values are present?
Are like-for-like values being compared?
What caused this item to be flagged?
What does the reviewer need next?
03 / WORKFLOW
A six-stage path from collection to judgment.
Capture source evidence on schedule.
Check freshness and required fields.
Align formats and comparison units.
Calculate candidate changes.
Explain anomalies and uncertainty.
Ask for judgment before action.
04 / BOUNDARY
The system prepares the decision. A person owns it.
AUTOMATION HANDLES
- Scheduled collection
- Schema validation
- Normalized comparison
- Evidence packet assembly
- Candidate change flags
HUMAN DECIDES
- Whether the source is trustworthy
- Whether the anomaly is real
- Whether the impact matters
- Whether action is safe
- When the rule itself should change
05 / OUTCOME
The value was not removing people. It was focusing their attention.
LESS MANUAL COMPARISON
Routine evidence arrives already organized.
CLEARER UNCERTAINTY
Missing and conflicting states stay visible.
SAFER ACTION BOUNDARY
No uncertain candidate becomes an automatic decision.
BETTER AUDIT TRAIL
Evidence and review context remain connected.
06 / REFLECTION
A dependable system needs a designed way to say, “not enough evidence yet.”
The useful part of this workflow is not the comparison engine alone. It is the explicit stop condition, the review object and the ownership boundary around the final action.
This case describes operating logic and decision boundaries. It does not display client data, private systems or fabricated performance metrics.