FUNCTION-EVIDENCE.md

FUNCTION-EVIDENCE — does each ported function carry direct or classified evidence?

LIBCOB-PARITY proves a real Rust fn exists. This map classifies the evidence behind it: DIRECT = the fn name is referenced by a unit test, a Kani proof, a fuzz target, or an oracle-sweep mirror. TRANSITIVE = it is called (per CLANG-AST-PARITY) by a directly- covered function, so it is exercised through that public oracle/test path. LIFECYCLE = an init/exit/free/alloc/cache routine (RAII / deliberate no-op). UNEVIDENCED = none of these — a false-confidence risk. (Direct evidence is a token reference; court & transitive evidence are classified separately; byte parity remains the per-court oracle sweeps in lab/verify-sealed-courts.sh.)

Active ported: 1138 · direct 962 (incl. oracle-sweep 213) · transitive 150 · lifecycle 97 · evidenced (any) 1138 · unevidenced 0.

libcob file	active	direct	oracle	transitive	lifecycle	evidenced	unevidenced
numeric.c	98	69	20	26	6	98	0
move.c	56	39	23	15	2	56	0
strings.c	34	24	0	8	8	34	0
intrinsic.c	232	173	103	57	8	232	0
cconv.c	9	8	4	0	2	9	0
termio.c	18	15	3	2	1	18	0
screenio.c	105	105	0	0	6	105	0
call.c	67	58	1	6	5	67	0
fileio.c	180	156	36	21	18	180	0
mlio.c	43	38	5	3	2	43	0
reportio.c	39	27	0	8	7	39	0
common.c	253	249	17	1	31	253	0
cobgetopt.c	4	1	1	3	1	4	0

Ported but unevidenced (no direct / transitive / lifecycle coverage)

None — every active ported function carries at least one evidence reference.

How this is produced (reproducible)

gnucobol-rs-port-index evidence generate joins the committed parity-detailed.json with a scan of the Rust #[cfg(test)] / #[cfg(kani)] regions, the fuzz targets, the oracle-sweep mirrors (examples/*.rs + lab/oracle/*.sh), and the clang-callgraph.json (for transitive coverage). evidence check regenerates and diffs (anti-staleness). PORT-GOVERNANCE.3 closure rule (future Tier-2 gate): a 100%-closed file must have every active fn direct / transitive / lifecycle / court covered — no unclassified unevidenced active functions.