We shipped a pricing tool with the word "placeholder" printed on the screen. On purpose. In production. For a week. This week the real rates went in, I walked every number by hand, and the one defect I found is exactly why we build this way.
Our what-if pricing tool went live with a stand-in rate, and instead of hiding that, we labeled it. Right on the surface: this number is a placeholder. The math around it was real. The comparison logic was real. The one input we had not finished sourcing was named as unfinished.
Most software does the opposite. It renders every number with the same confident font, whether it came from a signed rate schedule or a developer's best guess at 2 AM. The user cannot tell the difference.
A tool that admits what it does not know yet is not a weaker tool. It is the only kind you can trust when it finally says it knows.
Then the real rates went in
This week we replaced the placeholder with actual negotiated card rates, blended properly by card brand, with floors and program minimums as guardrails that warn instead of block. Source documents never touch the codebase—only derived numbers ship, with a provenance note.
Before go-live, every number on the screen was walked against the math by hand. One summary label still showed an old figure while the math underneath had moved. Two numbers for one fact. The tests were green. The label was wrong anyway.
Green tests are not the gate
The fix was structural: the label now reads from the same single source the math reads from. But the catch did not come from the test suite. It came from a human walking the deployed screen with the actual numbers in hand.
Nothing merges until a person has eyeballed the live build—not a staging theory. Automated tests prove the code does what the code says. The walk proves the screen tells the truth.
What a merchant should take from this
If you evaluate software that shows money numbers, ask: how does it behave when it does not know something? Does it label estimates as estimates? Does it ever say "no data," or does every field always have a confident number?
We are building the slow, visible way: run it on our own company first, label what is unfinished, and let people using it catch what the tests miss. The placeholder dying this week is not the story. The story is that it lived honestly in the open until the real number earned its place.
Original: getvexis.ai