Defects found in the published benchmark artifacts, recorded here rather than left for a reader to discover. An entry stays on this page after it is fixed, because the scored manuals in this repository are the artifacts as they were scored, not as they would be regenerated today.
Status: producer defect fixed. The affected scored artifacts are unchanged and remain as published.
What it is. The inbound-call section of a generated manual lists the programs that transfer control into the documented program, with the source line of each transfer site. In the affected manuals, some listed sites are wrong: the calling program named is real and does call the documented program, but one or more of the cited line numbers point at a transfer that targets a different program.
Worked example. COTRN01C (CICS Inquiry, scored 97.4, zero entity fabrications) lists its inbound callers as:
COTRN00C— CICS XCTL at L1103 and L1429 ofCOTRN00C(2 sites)
COTRN00C is a real caller and L1103 is a real transfer site into COTRN01C. L1429 is not; that XCTL targets a different program.
Scope. Eight of the 42 scored programs are affected, concentrated in two archetypes:
| Program | Archetype | Score |
|---|---|---|
COADM01C | CICS Menu | 99.2 |
COTRTUPC | CICS Inquiry | 99.2 |
COTRTLIC | CICS Menu | 99.0 |
COSGN00C | CICS Inquiry | 98.5 |
COUSR00C | CICS Menu | 98.5 |
COTRN01C | CICS Inquiry | 97.4 |
COMEN01C | CICS Menu | 97.1 |
COTRN00C | CICS Menu | 97.0 |
Five of the six CICS Menu programs and three of the eight CICS Inquiry programs carry it. No batch program is affected. The concentration is consistent with the defect's mechanism: menu programs are the ones that navigate by CICS XCTL, so they have the most inbound transfer sites to resolve.
Why it was not caught, and what that tells you. Every entity involved is real, so the fabrication gate does not fire: the gate checks that a named reference resolves to something that exists, not that the line attached to it is the right one. The independent cross-vendor grader also scored all eight of these manuals and did not flag the defect either. That is the honest boundary of both controls, and it is stated in METHODOLOGY.md §5.
What it does not affect. The entity-fabrication result is unchanged: all eight manuals contain zero fabricated entities, because no invented name is involved. The published scores are the scores these artifacts actually received and have not been adjusted.
Reported: internal review, 2026-08. Fixed in producer: 2026-08.