dot-iu-cutter v0.5 Constitution Source Snapshot Capture E1 BLOCKED — GPT Ruling

Date: 2026-05-18 Reviewer / decision authority: GPT Reviewed package: knowledge/dev/laws/dieu44-trien-khai/v0.5-constitution-source-snapshot-capture-execution/

Reviewed files:

files:
  - dot-iu-cutter-v0.5-constitution-source-snapshot-capture-execution-log-2026-05-18.md
  - dot-iu-cutter-v0.5-constitution-source-snapshot-capture-verification-result-2026-05-18.md
  - dot-iu-cutter-v0.5-constitution-source-snapshot-capture-report-2026-05-18.md

Verdict

E1_capture_execution: BLOCKED_CORRECTLY
agent_behavior: PASS_CORRECT
artifact_written: false
production_mutation: none
DML_executed: false
rollback_needed: false
blocker: SC3_parser_profile_reproduction_divergence

The Agent correctly refused to freeze a snapshot artifact whose normalized content hash was not produced by the ratified canonical parser/checksum path. This is fail-closed behavior and preserves source-version integrity.

---

Accepted evidence

ratified_canonical:
  checksum: 17660443e0f23e994e1807cf8e22920951a9e70c598956dbd0e752f4f5cae80c
  length: 17522

this_environment:
  checksum: 072983ac6cf44ee0722d0fff68731135a00f224db459bc43e51d1c0c31b3bdea
  length: 17657
  delta_length: +135
  deterministic_3_of_3: true

marker_counts:
  enacted: 19
  controlled_draft: 1
  draft: 1
  obsolete: 1

Marker counts remain stable, so the blocker is not marker-structure drift. The issue is parser-profile reproduction divergence: different executors can apply the prose spec differently and produce different byte-level normalized content.

---

Interpretation

The system has exposed a missing layer:

missing_layer: canonical_parser_reference_implementation

A parser profile described only in prose is not sufficient to produce production source-version identity. For source_document_version, the identity hash must be generated by a byte-exact, executable, versioned parser implementation or by a formally pinned canonical executor.

This is not a production incident:

production_impact:
  registry_rows_added: 0
  snapshot_artifacts_written: 0
  source_versions_seeded: 0
  dryrun_cut_verify: 0

But it is a blocker for B5 because E2 seed must bind to a snapshot artifact whose checksum is reproducible.

---

Ruling

write_noncanonical_072983_snapshot: REJECTED
rerun_E1_by_prose_spec_only: REJECTED
seed_from_noncanonical_snapshot: REJECTED

GPT selects the durable path:

selected_path: R2_RATIFY_REFERENCE_IMPLEMENTATION_FIRST

Reason:

reason:
  - rerunning E1 on a specific executor may unblock once, but leaves future agents vulnerable to the same divergence
  - source_version identity must be reproducible across agents/executors
  - operations-first goal requires one-command operation, not executor-specific hidden behavior

A limited R1 canonical executor rerun may be used only as a comparison source during reference implementation authoring, not as the final long-term solution.

---

Required next phase

Open a narrow technical-authoring phase, still operations-first in purpose:

next_phase: v0_5_constitution_nuxt_parser_reference_implementation_authoring
nature: code_or_script_authoring_and_test_design__no_capture_no_seed_no_dml

Purpose:

purpose:
  - convert parser profile nuxt-incomex-portal-constitution-v1 from prose into a byte-exact reference implementation
  - define deterministic inputs/outputs
  - include fixture/test vectors for known cases
  - reproduce the ratified canonical checksum if the live source content is still equivalent
  - if live content has drifted, classify drift and produce a new candidate only after the reference implementation is fixed

The implementation must output at least:

required_outputs:
  - normalized_content
  - normalized_content_checksum
  - normalized_content_length
  - marker_counts
  - extraction_span_diagnostics
  - parser_version

It must also record the exact algorithm steps:

algorithm_must_pin:
  - fetch/HTML decoding rules
  - authoritative span selection
  - script/style/hydration removal
  - HTML entity decoding
  - Unicode normalization
  - line ending normalization
  - whitespace normalization
  - CHANGELOG inclusion boundary
  - BEGIN/END sentinel content-region semantics
  - emoji marker preservation by codepoint

---

Still forbidden

still_forbidden:
  - snapshot artifact write
  - source_document INSERT
  - source_document_version INSERT
  - DML execution
  - production DB mutation
  - checksum persistence into registry
  - dry-run
  - CUT
  - VERIFY
  - schema change
  - GRANT/REVOKE
  - Directus mutation
  - deploy/restart
  - git commit unless later implementation phase explicitly authorizes repo work
  - self-advance to E1 capture or E2 seed

Final status

status: E1_BLOCKED_BY_PARSER_IMPLEMENTATION_DIVERGENCE__REFERENCE_IMPLEMENTATION_NEXT
next_action: author_and_test_canonical_parser_reference_implementation