Order Adjustment Bot

The problem

Order adjustments are the definition of work that shouldn’t need a person: the same handful of rules, applied over and over, to entries that differ only in their numbers. A human doing it is slow, and worse, a human doing it at the end of a shift is inconsistent. The hundredth adjustment gets less attention than the first.

What I built

A bot that applies the adjustment rules directly against the records: the same rules, in the same order, every time, without getting tired near the end of the batch.

How it works

Every case falls into one of three buckets:

1. Matches a rule cleanly: apply it and log what was done.
2. Matches nothing: leave it alone and flag it for a person.
3. Matches ambiguously: leave it alone and flag it, loudly.

Technical approach

The rules are applied in a fixed order, which is what makes the result reproducible: the same input produces the same output regardless of when it runs or what ran before it.

The design decision that mattered is that the bot does not guess. Automation that silently does something plausible with an edge case is worse than automation that stops. The whole value proposition is that you can stop checking its work, and you can only stop checking if it’s honest about what it couldn’t handle. Every applied rule is logged, so the run can be read back rather than trusted blindly.

Current state

In use. The recurring entry task is gone from the daily routine. What’s left is a short exception list, the cases that genuinely needed judgment, which is exactly the part a person should have been spending their attention on all along.