Name the independent review neutrally in the helper scripts, controls and contract comments

[azrollin]

The root helper scripts, the files under controls/ and pattern_field_contract.json carried comments that named the independent reviewer, used its case numbers and a fixture glob, named people or seats, and numbered review rounds. This is the second half of the wording change in #40. Those comments now say "the independent review", "an earlier version" or "a later review", describe each case in words, and no longer carry digit-numbered review labels ("round 5", "R5", "X1"). The change is in comments, docstrings and _comment strings only; no line of code, no key and no value changes.

The branch has two commits. The second one fixes the sites that the first one missed after review. The message of the first commit says more than it did; this description is the accurate one.

What changed

12 files, 67 lines in, 67 out:

• Root helpers: _compare_compiled.mjs (1), _source_parity.py (2), _vflag_disclosure_gate.mjs (7), compile_patterns.py (9), the differential parity script named after the reviewer (6), prep_recompile.sh (13), prep_recompile_v058.sh (6). Comments, plus the docstrings in the Python files.
• controls/: _metadata_one_key.mjs (2), evaluate_candidates.mjs (2), guard_selftest.sh (7), metadata_is_measured.mjs (8). Comments only.
• pattern_field_contract.json: 4 lines, all inside _comment strings (the arrays at lines 3 and 24 and the _comment string at line 100).

Case numbers (C01688, C01987, C02706, C02748), the fixture globs (gap_U1C_*, gap_UFEFF_*) and the reviewer's own case labels X1 and X2 (in controls/guard_selftest.sh line 99 and controls/metadata_is_measured.mjs line 111, one pair each) are dropped on purpose, as in #40. The numbers, hashes, dates and counts that were measured stay. In the second commit, the comments that attributed a finding to a named seat now say it happened on that date without naming anyone, and the comment in prep_recompile.sh that quoted an old error now describes the error in plain words.

Proof that no code changed

• git diff --stat origin/main: 12 files changed, 67 insertions, 67 deletions.
• For each file the whole file was checked against main after removing what is not code. .mjs: full line // comments removed, the rest byte identical. .sh: full line # comments removed, the rest byte identical. .py: the parsed syntax tree with docstrings removed is identical. pattern_field_contract.json: the parsed document with every _comment key removed is identical. All 12 pass on the branch head.
• Syntax: py_compile, bash -n, node --check and a JSON parse all pass on the changed files.
• node --test tests/*.test.mjs: 263 pass, 0 fail, the same as main.
• controls/guard_selftest.sh ends GUARD SELFTEST OK and controls/metadata_is_measured.mjs ends METADATA CONTRACT OK over 137 cases, both run on the first commit. The second commit touches only comments in both files, which the stripped comparison above covers. The README is untouched, so the gates that read it are unaffected.

Not changed, stated exactly

These still carry the reviewer's name, on purpose, because changing them changes run-time output, a file name or data:

• Three print lines in the differential parity script named after the reviewer (lines 45, 88 and 128), plus its fixture paths (lines 27 to 29) and an environment variable name. Those are code.
• The die message in prep_recompile_v058.sh (line 145) and the deploy checklist line in each of prep_recompile.sh (line 228) and prep_recompile_v058.sh (line 190).
• The temporary JavaScript runner file that the differential parity script writes on line 71 and passes to a subprocess on line 74 (its name carries the reviewer's name). It is a path in code, so it stays.
• File names: that parity script and its report, and the files under controls/candidates/ with the names inside them. Also RECOMPILE_CONTROLS_2026-09-20.md (recompile notes and history) and src/patterns.js.

Review rounds are written as ordinal words, not digits ("the fifth round"), and that wording stays on 14 comment, docstring or _comment lines: _source_parity.py 164 and 243, the differential parity script 22 and 23, controls/_metadata_one_key.mjs 40, controls/guard_selftest.sh 72 and 99, controls/metadata_is_measured.mjs 27, 53, 89 and 111, pattern_field_contract.json 3 and 24, and prep_recompile.sh 201. The claim in this description is about digits and names, not about ordinals. A comment is not run-time output, so no comment that names the reviewer or a seat is left on purpose; the retained mentions of the reviewer's name are the strings, paths and file names in the bullets above. The ordinal-word round references listed above are the only other comment wording kept on purpose. I searched the changed files for seat names, digit-numbered rounds and the reviewer's name. I did not certify that every pronoun that could refer to the reviewer is gone, because that is a judgment on prose and not something a search settles. The Worker and the scanner behave exactly as on main.