Add per-section verbatim fidelity for markdown files

[ptr727]

Closes the propagation gap #305 names for the docs class: a universal rule block could silently rot — or fall behind a newly added rule — downstream while its heading still passed the presence check, because AGENTS.md is intent as a whole and the audit only grepped for headings. The new history-rewrite rule (#394) was the live test case: nothing flagged that Financial-Modeling's paraphrased "Git and Commit Rules" was missing it.

What

A carried markdown file can now be intent overall while individual ## sections are verbatim — byte-identical to the hub canonical, EOL-normalized, reusing the existing region-verbatim engine (stale-vs-modified by git history).

• audit.py — extract_section (cuts a ## heading body at the next sibling H2, keeps nested ###), verbatim_sections, per-section check_verbatim wiring, and a --selftest case.
• files.schema.json / validate.py — a section object may carry fidelity (intent default, or verbatim, markdown-only).
• files.json — mark AGENTS.md's three universal rule sections verbatim: Repository Boundaries and Write Safety, Git and Commit Rules, Verification Discipline. Verified repo-agnostic (no SHAs, no ptr727/<repo> refs, no placeholders). Branching Model and the rest stay intent (Branching Model cites this repo's own historical SHAs — the trap that makes it not universal).
• fidelity-model.md — documents the section granularity.

Verification

validate.py green; audit.py --selftest passes (incl. the new section case); a live audit of Financial-Modeling flags all three sections as DRIFT — its AGENTS.md carries older paraphrases. DRIFT is advisory (0 defect/letter), so it surfaces the re-vendor need without redding the gate — the idempotent artifact recording current state.

Follow-up (next PR)

Extend fidelity_honesty.py + reports/divergences.md to show the fleet-wide section burn-down (which repos diverge on which sections), so the propagation state is one regenerable artifact.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR extends the spec/audit fidelity model to support per-section verbatim fidelity for Markdown documents, allowing a file to remain intent overall while specific ## sections are enforced as byte-identical (EOL-normalized) to the hub canonical. This closes a propagation gap where downstream paraphrases could silently miss newly added universal rules while still passing a heading-presence check.

Changes:

• Add per-section fidelity support (intent | verbatim) for sections[] entries in spec/files.json, validated by schema and spec/validate.py.
• Teach spec/audit.py to exclude verbatim sections from heading-presence checks and instead hash-compare extracted ## section bodies via the existing verbatim engine (with a new --selftest case).
• Mark select AGENTS.md universal rule sections as verbatim in spec/files.json and document the new granularity in spec/fidelity-model.md.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
spec/validate.py	Validates per-section fidelity values and enforces that section-level verbatim is Markdown-only.
spec/files.schema.json	Extends the schema to allow a fidelity field on section objects.
spec/files.json	Marks selected AGENTS.md sections as verbatim while keeping the file overall intent.
spec/fidelity-model.md	Documents that verbatim can apply to Markdown section regions in addition to files and workflow job regions.
spec/audit.py	Implements section extraction and per-section verbatim checks, plus self-test coverage for the new section region behavior.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

spec/audit.py:171

• extract_section() only recognizes H2 headings that start with the literal "## " (single space) and matches the target via exact string equality on the stripped line. Markdown allows arbitrary whitespace (multiple spaces/tabs) after the hashes, so a valid heading like "## Git and Commit Rules" would fail to match and be reported as "region absent downstream" rather than a stale/modified verbatim diff.

        elif not fenced and stripped.lower().startswith("## "):
            if capturing:
                break  # a sibling H2 ends the section
            if stripped.lower() == target:
                capturing = True

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (1)

spec/audit.py:158

• extract_section() docstring says the heading locate is "case- and whitespace-folded", but the implementation only case-folds and strips surrounding whitespace (it does not fold internal whitespace). Tighten the wording so the doc matches the actual matching behavior, which avoids misleading future readers about what variations are tolerated.

    EOL-normalized to `\\n`. The match that locates the heading is by its parsed text (the text after the `## `
    marker, case- and whitespace-folded), so a re-cased or re-spaced heading is still found rather than read as
    a missing section. The heading line's exact bytes are then part of the hashed region, so that re-casing or