dot-iu-cutter v0.5 — Snapshot MARK Test Transport Analysis

Phase: v0_5_constitution_snapshot_MARK_test_transport_fixture_CI_fix · Nature: analysis_only__no_commit__no_dryrun · Date: 2026-05-18 · doc 1 of 4

dryrun_py: UNCHANGED byte-exact f1f42e83… (QG1 honored — not modified)
test_file_current: 454d9fc8… ; prior_expected: 31143968…
verdict: mismatch = transport/whitespace artifact at KB-authoring time, NOT a content change
recommendation: OPT_2 — ratify current 454d9fc8 (side-by-side equivalent); 31143968 NOT byte-recoverable from KB

---

1. Authority

GPT ruling …/reviews/dot-iu-cutter-v0.5-branch-patch-apply-ci-blocked-gpt-ruling-2026-05-18.md: OPT_1 publish_byte_exact_test_blob: approved_if_available, OPT_2 ratify_current_test_hash_454d9fc8: conditionally_allowed_after_side_by_side_review. Patch SSOT = code-authoring doc 2 (verbatim test block; declared test_sha256: 31143968…).

2. Side-by-side review (repo file vs KB doc-2 verbatim test block)

method: Read repo tests/test_dryrun_snapshot_mark.py (219 lines) vs the
        ```python verbatim block in code-authoring doc 2 (full content).
result: LINE-FOR-LINE IDENTICAL — docstring, imports, ART/SHA/EM constants,
        _Args, TestGate(5), TestManifest(9), TestFailClosedSynthetic(4),
        TestNoDbImportIsolation(3). No semantic divergence; no logic edit;
        no added/removed test; identical assertions, identical synthetic
        fixtures, identical Vietnamese/em-dash/emoji literals.
file_hygiene: pure LF (CRLF=0), no trailing whitespace (per-line rstrip ==
        as-is), no tabs, single final newline, UTF-8, is_NFC=true.

3. Why 31143968 ≠ 454d9fc8 — bounded recovery experiment

Deterministic whitespace/newline/encoding variants of the faithfully transcribed KB block were hashed (no character guessing):

as-is (repo file)                : 454d9fc8…   (== current)
strip 1 final newline            : 3c268b36…
add a final newline              : 0c237699…
rstrip trailing ws per line      : 454d9fc8…   (proves NO trailing ws)
CRLF line endings                : 6da558f7…
strip all trailing newlines      : 3c268b36…
TARGET                           : 31143968…   (NO variant matches)

conclusion: 31143968 is NOT byte-recoverable from the KB doc-2 embedded
  block by any deterministic whitespace/newline/CRLF transform. Combined
  with: (a) repo file ≡ KB verbatim block line-for-line; (b) cutter_agent/
  dryrun.py from the SAME doc reproduced byte-exact f1f42e83 (transport is
  faithful for code); (c) 7/7 fixture-independent tests PASS (semantics
  correct) — the 31143968 hash corresponds to a pre-embed scratch copy
  whose exact whitespace/blank-line layout was normalized when the
  code-authoring agent embedded the block into the KB markdown doc, OR
  the declared test_sha256 was computed on that scratch copy. This is a
  KB-AUTHORING TRANSPORT/WHITESPACE ARTIFACT, not a semantic content change.
diffstat note: doc-2 `211` for the test (vs 219 lines) is git --stat bar
  scaling, NOT an exact line count — dryrun.py is likewise `559` vs 558.

4. Recommendation

OPT_1 (byte-exact 31143968 recovery): NOT POSSIBLE from KB via faithful
  transcription or any deterministic whitespace transform.
OPT_2 (RECOMMENDED): GPT/User ratify the current test artifact sha256
  454d9fc8… as canonical. Justification (GPT's "after side-by-side review"):
  it is line-for-line semantically identical to the patch-SSOT verbatim
  block; production module dryrun.py is byte-exact accepted; 7/7
  fixture-independent tests PASS; zero logic divergence. Update the KB
  hash-of-record from 31143968 → 454d9fc8 on ratification.
OPT_1' (durable fix, also recommended): future patch artifacts ship as a
  base64 / content-addressed blob in KB (markdown-fenced retype + KB store
  normalization is not byte-faithful for whitespace-significant files).
dryrun.py: NOT modified (no defect — byte-exact f1f42e83; QG1 honored).

doc 1 of 4. Self-advance PROHIBITED.