nx_dwline_corpus_gate.nx

nx_dwline_corpus_gate.nx -- DOES STATEMENT LINE INFO HOLD ACROSS THE REAL CORPUS, OR ONLY ON A TOY? WHY THIS EXISTS. nx_dwline_stmt_gate proved statement granularity on runtime/nx_probe_dbgtiny.nx: ELEVEN LINES, TWO FUNCTIONS, four .loc rows. That is a sample size of ONE, and a feature that works on one hand-written probe is not a feature that works. The estate compiles thousands of real files with deep nesting, matches, loops, multi-line expressions and 20-argument calls -- none of which the probe contains. ★ A PROOF ON A FIXTURE YOU WROTE TO BE EASY IS A PROOF ABOUT YOUR FIXTURE. WHAT THIS MEASURES, over every .nx in runtime/ up to a DECLARED cap: - how many files the compiler can build at all with -g (the honest denominator) - how many emit ANY .loc, and how many .loc rows in total - the same numbers from the OLD blessed function-granularity compiler, as the control - the RATIO of statement rows to function rows -- which is the actual claim under test - files where the new compiler emits FEWER rows than the old one (a regression, individually named) NO SILENT CAPS: the file count scanned, the cap, and any skipped/failed files are all REPORTED. A gate that quietly samples 20 files and prints a percentage is the thing this gate exists to replace. license_tier: ORIGINAL expect_exit: 0