Two code paths computing the same value will diverge

Symptom

"It's right on the review screen, but the sent / saved value is 0 or empty."

Cause

The review screen (GET) and the actual send (POST) each called the same builder function. Only one injected the exchange-rate parameter. GET displayed a price in the hundreds of thousands of won; POST got None → int(None or 0) → sent 0.

How to spot it

• Is the same domain value (price, category, key) built in more than one place?
• The difference in argument lists between the two call sites is the size of the divergence.

Fix

1. Collapse into a single-source helper
2. Compute derived values once and pass them through
3. Don't demote "unknown" to 0. int(x or 0) turns "I don't know" into "zero" and sends it quietly.
4. A contract test that pins displayed value == sent value. Testing the two sides separately will never catch the split.

Not only two code paths — the canonical system and a local copy diverge too

The sourcing map's krw disagreed with the marketplace's actual listed price. The marketplace is canonical; the local map is a copy. Compute margin from the copy and you get the wrong answer. The split isn't limited to two code paths; "canonical system vs local copy" is the same shape.

In the same map, 618 of 6,541 records had both usd and jpy empty. A source can be linked and still be useless: a record missing the price is equivalent to an unlinked record as far as the margin check is concerned. Treat a missing required input the same way you treat a missing link.

An upstream 0 turns a derived price negative

One channel had products whose sale-price field was 0 (a temporary-failure status, among others). Pulled as-is and fed into a derived price for another channel — cost − random — the result is a negative price. The derivation formula assumes the upstream value is sane. Put a lower-bound check right before the derivation. A 0 is often not "a value" but "no value" wearing different clothes.

---

---