[squid-protocol]

Part of #4711 (epic #4710, step 1).

What

• Catalogue: docs/language_status/cobolrt_rules.json. It has one entry per semantic rule of det/cobolrt: 143 rules in 12 families, covering all 28 runtime files. Each rule has:
  • a stable ID RT-<FAMILY>-<NAME>;
  • the IBM manual and section it implements (by section name; no page numbers invented). Where IBM is silent, ibm: null with a note, which is only allowed on a rule that is not MATCHED;
  • the options it depends on. Compile options are checked against the compile & runtime options as a first-class estate input: one resolver for oracle + det port, provenance, evidence fingerprint #4704 resolver's names, alongside source clauses, dialect switches and deployment facts;
  • its register entries;
  • a status against IBM (MATCHED / DIFFERS / REFUSED / ASSUMED);
  • a calibration status: all uncalibrated, because nothing is z/OS-measured yet (oracle policy: IBM is the reference, GnuCOBOL the instrument — audit register entries + z/OS calibration oracle #4702).
• Code sites: each rule's method carries // [rule ID] on the line above its declaration. This is a comment only.
• Two-way check: tests/tools/cobolrt_rules.py --check, run by tests/cobol_mainframe/test_cobolrt_rules.py. It fails when:
  • a marker has no entry;
  • an entry has no marker;
  • a marker stands before a method that isn't the rule's;
  • a method of a catalogued file is in no rule and in no plumbing list. A new runtime method can't land unclassified: this caught det: UNSTRING numeric operands and bad sign byte MOVE-to-X refused by name; NATIONAL already refused (Part of #5032) #5073's refuseUndefinedSign during this PR;
  • a named method doesn't exist;
  • a register ID that a family's comments cite is carried by none of its rules;
  • a runtime file is in no family;
  • an IBM, option, register or calibration field is bad.
• Docs: docs/language_status/cobolrt_rules.md is rendered from the catalogue. It is git-ignored and published to generated by generated-docs.yml, the same pattern as ci/docs: stop committing generated aggregate docs in PRs (oracle_assumptions.md et al.): render post-merge to the generated branch #5057. Each register entry's status is shown live from the register, not copied.
• det_port.py check: a carried cobolrt/*.java file is now compared as java_code, i.e. without comments, with whitespace outside literals collapsed. A string-, char- and text-block-aware lexer does this. Before this change, a comment in Cobol.java flagged every port that names Cobol as changed. The port's own code and generated/ are still compared byte for byte.

Families

family	files	rules
CBL	Cobol.java	43
NUM	Codec.java, Zoned.java	12
HFP	Hfp.java	12
EDIT	Editing.java	5
FUNC	Funcs.java	10
CICS	cics/DetCics.java	22
SQL	sql/DetSql.java	10
LE	le/*.java	13
DATA	Field, Storage, Lossless, Figurative, Hole	5
SORT	Sort.java	6
FILE	batch/DetFiles.java	3
SYS	standalone/*	2

Totals: 80 MATCHED, 22 DIFFERS, 10 REFUSED, 31 ASSUMED.

No behaviour change (evidence)

• det_port.py check against origin/main: 105/105 ports unchanged. With the old byte-for-byte runtime compare, the same output would have flagged 105/105 ports: every port names Cobol, and every runtime file now carries a marker.
• I compiled cobolrt at origin/main and at HEAD with javac -g:none (__PACKAGE__ replaced). The 47 classes are byte-identical. batch/ and cics/ were left out because they need the generated project's CobolFiles / CicsTask.
• All 28 runtime files are java_code-identical to origin/main, i.e. the code with comments removed.
• Gates:
  • pr_gates --fast: 6/6 pass;
  • pr_gates --ratchets: 7/7 pass, 0 skipped;
  • audit_check.py: all clear;
  • the new test (51) plus test_generated_docs.py: 63 pass, run with plain pytest from /tmp.
• CI note: det_port.py is a det-sweep input, so this PR's det-sweep runs in full once. Evidence records that use a runtime class go stale on generator, which is advisory and not in BLOCKING; the nightly evidence refresh picks them up. The evidence fingerprints were left alone on purpose, because changing them would re-stale every record.

Cross-repo

None.

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab

[squid-protocol]

Owner decision (2026-10-14): the edit to .github/workflows/generated-docs.yml is approved (path triggers for the rule catalogue sources + publishing cobolrt_rules.md to the generated branch, the #5057 pattern). Labelled for merge at c61850130696.

🤖 Generated with Claude Code

[squid-protocol]

No description provided.