friction: verify refreshed goal-autopilot markers against marker-free source

[ll7]

Archetype Metadata

archetype: workflow
evidence_tier: smoke
linked_policy:
  - docs/ai/open-issue-contract-preparation.md

Goal / Problem

Make preparation verification accept a legitimate refresh of an existing goal-autopilot-preparation:v1 marker while still detecting any change outside the marker.

_verify_batch currently removes the new marker and compares the remainder with body_sha256. When the audited source body already contained an older marker, that digest includes the old marker, so a correct marker replacement is reported as content drift outside marker. In the 2026-08-30 live pass this produced 104 false failures; revision-history comparison independently proved every edited body preserved its non-marker content.

A duplicate search for the marker name plus content drift outside marker, verifier, and refresh terms found no issue for this exact defect.

Inputs

• scripts/dev/prepare_open_issue_contracts.py: _MARKER_BLOCK_RE, _verify_batch, and plan digest fields.
• tests/dev/test_prepare_open_issue_contracts.py verify/idempotence fixtures.
• GitHub revision-history semantics are evidence only; tests must remain offline.

Scope

• Define one canonical digest for content outside the versioned marker, or retain enough audited source content to strip the marker on both sides.
• Verify first insertion, marker refresh, and an already-current idempotent body.
• Continue to fail on duplicate markers, malformed marker boundaries, or any byte change outside the documented newline boundary.
• Report a reason that distinguishes malformed markers from real non-marker drift.

Non-goals

• No weakening to a whole-body equality check that ignores user content.
• No network-dependent tests.
• No change to packet classification or labels.

Acceptance criteria

• Replacing one old well-formed marker with one new marker verifies successfully.
• First insertion and idempotent re-verification remain successful.
• A one-byte change before or after the marker fails.
• Duplicate or unterminated markers fail with a specific reason.
• The documented final-newline rule is deterministic and covered by tests.

Validation / Testing

uv run pytest tests/dev/test_prepare_open_issue_contracts.py -q
uv run ruff check scripts/dev/prepare_open_issue_contracts.py tests/dev/test_prepare_open_issue_contracts.py
uv run ruff format --check scripts/dev/prepare_open_issue_contracts.py tests/dev/test_prepare_open_issue_contracts.py
python -m compileall -q scripts/dev/prepare_open_issue_contracts.py tests/dev/test_prepare_open_issue_contracts.py
git diff --check

Dependency and terminal state

Blocked-by: #8080

Once the current-main gate clears, this is one local verifier/test pull request.

schema: goal_autopilot_preparation.v1
repository: ll7/robot_sf_ll7
issue: 8103
source_body_sha256: b9657856ca2d46809d4df14aee062383a5d6ffa0fc6fe9f1a040229815c8e001
source_comments_sha256: 
audit_schema: open_issue_contract_audit.v1
audit_digest: 3f3c51addec2c0ac8aad5ca07f84a63669ce53bf2603a0919267fd1bd198b7c6
audit_classification: blocked
next_action: resolve_named_blocker
authority: blocker_owner
execution_mode: blocker
preferred_worker: MaxRunner
expected_pr_runner_label: runner:max
implementation_admitted: False
state_ready_change_proposed: False
mutation_batch: open-issues-final-20260831

This packet is preparation evidence only. It never overrides live labels, exact claim state, branch state, typed dependencies, domain gates, compute authority, release authority, or scientific evidence rules.