fix(conformance): stop reading header comments as receipt-key instantiations

[gHashTag]

Closes #877. Part of #851.

What was red, and why

On main, receipt-keys in trinet-portability.yml has failed on all 7 scheduled runs from 2026-09-28 to 2026-10-04. It reports two keyless instantiations of trinet_node_core in formal/tern_tc_layer_rtl_tb.v, at lines 7 and 18. Both lines are header comments that name the file fpga/portable/trinet_node_core.v. The real instantiation is at line 42, and it passes .RECEIPT_KEY(128'h0).

conformance/key_default_check.py matched module names in the raw source. This PR blanks comments and string literals before matching, keeping newlines so the reported line numbers are unchanged. The same change closes the reverse hole: a key written only in a comment inside #( … ) used to count as passed.

Measured

Every mutant ran on a copy of main's tree (57de4a1f5, 4,166 input files), in-process, on both checkers.

mutant	must	old checker	new checker
none	pass	rc 1, RULE 2 at :7 and :18	rc 0, 3 instantiations
tern_tc bench drops its key	fail	rc 1, 3 RULE lines	rc 1, one RULE 2 at :42
tern_tc key only in a comment	fail	rc 1, but only :7 and :18, nothing at :42	rc 1, one RULE 2 at :42
v2 bench drops its key (the workflow's own mutant)	fail	rc 1, 3 RULE lines	rc 1, one RULE 2
non-null default (the workflow's own mutant)	fail	rc 1, 3 RULE lines	rc 1, one RULE 1
non-null default only in a comment	pass	rc 1	rc 0
$display("trinet_node_core up") in the bench	pass	rc 1	rc 0

Old: 0 of 7 right. New: 7 of 7.

The new CI step covers the comment case. It finds the bench's instantiation line with grep rather than hard-coding 42. It checks that the mutation applied, then requires exactly one RULE 2, on that line.

The guard step mutates formal/tern_tc_layer_rtl_tb.v, so that bench is now in both paths: lists (e1b3d0a). Without it the Workflow path gate failed this PR with "BUILT BUT NOT WATCHED". The bench is the one file the new step names, and nothing more is widened here.

The new step's shell ran on Linux in this PR's own runs: receipt-keys passed, and step 7, "The guard must reject a key that is only in a comment", succeeded (job 111539931868, at 94bf141).

Not in this PR

• The rest of the trigger paths. The checker reads every formal/**/*_tb.v and fpga/**/*.v, but the workflow lists two testbenches and three RTL locations. That is how TRI-NET: three boards, ten families, and four numbers that were wrong #355 brought this bench to main without the gate running. Since 2026-09-04, 3 of 3 first-parent commits on main that changed checker inputs were outside the list. Widening to the full globs would also run the portability job, which takes about 2 minutes (2m00s to 2m02s on four runs, 2026-10-04). I first wrote "25-minute", which is the job's timeout-minutes, not its duration. ci: run nine workflows once per PR commit, on every path their push run watched #876 rewrites the same on: block, so this waits for it; see receipt-keys gate red on main since 2026-09-28: header comments read as keyless instantiations #877, step 2. Adding the one bench above merges cleanly with ci: run nine workflows once per PR commit, on every path their push run watched #876, ci: give the 102 push workflows a dispatch-safe concurrency group #870 and ci: run the nine document gates on push to main only #867 (git merge-tree).
• The path gate's parser keeps a trailing # ... as part of a paths: entry and reports it as dead, so the new entry's comment sits on its own line. That error is loud, not silent, and no workflow on main has such a line. Not fixed here.

🤖 Generated with Claude Code

[github-actions]

X Brain Health Check

Score: 100.0/100
Status: 🟢 HEALTHY
Threshold: 80/100

X Brain is above merge threshold

[github-actions]

🧪 Brain Stress Test Results

Score: not measured (no stress test prints a Score: line)
Status: 🟢 PASS
Threshold: 270/300

X Stress test passed

[tri-claw]

Summary by t27-bees

The PR fixes a false positive in conformance/key_default_check.py where header comments naming a module were mistaken for instantiations. It adds a code_only() filter that blanks comments and string literals while preserving newlines, so line numbers stay accurate. A new CI guard step verifies that a key written only in a comment is rejected and the error points at the real instantiation.

Worth a look

• conformance/key_default_check.py:38 — CODE_ONLY_RE uses ordered alternation (// then /* */ then strings). This works because each alternative starts with a distinct character, but a // inside a /* */ comment is handled correctly (non-greedy .*? stops at first */). No issue found in the diff.

• conformance/key_default_check.py:45–48 — code_only() replaces every non-newline character in a match with a space. This preserves line counts but changes column positions; the line-number calculation src.count("\n", 0, m.start()) + 1 (unchanged, line 83) still works because newlines are kept.

• conformance/key_default_check.py:75 — The instantiation regex \b + re.escape(name) + \s*(#\s*\()? matches name #( or name at word boundaries. After comment blanking, false matches in comments are gone. The pattern assumes the module name isn't a prefix of another identifier (e.g., trinet_node_core_extra); \b protects against that.

• .github/workflows/trinet-portability.yml:102 — sed mutation comments out all .RECEIPT_KEY(128'h0) occurrences. The preceding grep check (line 96) ensures exactly one trinet_node_core #( line exists, so multiple keys would imply multiple instantiations and the check would fail earlier. Acceptable for this bench.

• .github/workflows/trinet-portability.yml:108 — The check grep -q "RULE 2 $tb:$inst_line:" expects the checker's exact output format (RULE 2 file:line:). If the checker's message spacing changes, this guard will break loudly (which is correct behavior).

---

_{Read: 2 files, +50 -2, the whole diff. Head b40cbbf. Model: openai-compatible/nvidia/nemotron-3-ultra-550b-a55b. Commands: @t27-bees review | summary | help | pause | resume. Free for every repository.}