A stale branch makes your test harness lie about the product

Building a test environment that compiles the app from the working tree means the harness measures YOUR BRANCH, not the codebase — and its output reads exactly like a statement about the product. Fourteen commits behind main I concluded a UI feature was unbuilt and a decision was blocking it; both had shipped, and the harness faithfully reproduced their absence because it built the old code. Three habits that would have caught it: check free identifiers (proposal numbers, migration numbers) with git ls-tree origin/main rather than ls on the branch, because a number free locally may be claimed upstream; before reporting any negative finding from a locally-built stack, confirm the same absence on origin/main; and check which container actually owns the port you curled — a neighbouring stack answered one of my probes and returned a plausible 200 that had nothing to do with my build.