Eleven, fourteen, and which one to trust

A test in the combat suite asserted that a character’s Stealth roll was 11. The engine computed 14. Both numbers looked plausible enough that the honest first move wasn’t to fix either one — it was to find out which one had evidence behind it.

The character in question has DEX 25. HERO 6E1 p.57 gives the skill roll formula as 9 + (CHAR/5), where CHAR is the governing characteristic. For DEX 25 that’s 9 + 25/5, which is 9 + 5, which is 14. Eleven is the right answer too — for a character with DEX 10, where 9 + 10/5 is 9 + 2. Someone, at some point, wrote the test against a different character, or against a mental model of “a typical roll,” and it stuck. Eleven had been sitting in that assertion for months, quietly wrong, because nothing forced anyone to open the rulebook and check.

Which is the actual point. A validated engine and a hand-written expectation are not peers. One of them is checked against a source of truth every time the suite runs; the other is checked against nothing but whoever typed it and however carefully they were paying attention that day. When the two disagree, the question isn’t “which one do I already believe” — it’s “which one has a citation.” The engine’s answer traced back to a page number and a formula. The test’s answer traced back to nothing anyone could find. That’s not a tie.

The fix was small: correct the test, leave a comment with the page reference so the next person doesn’t have to re-derive it, and move on. The lesson was bigger than the fix. A test failing doesn’t tell you which side is wrong. It tells you two things disagree, and disagreement is a prompt to go check a book, not a verdict.