Wave 46 tick2#10 — sitemap-companion-md regen + 17-gap fix¶

Date: 2026-05-12 (UTC)
Branch: feat/jpcite_2026_05_12_wave46_sitemap_regen
Memory keys honored: feedback_dual_cli_lane_atomic (lane = /tmp/jpcite-w46-sitemap.lane), feedback_destruction_free_organization (no rm/mv; sitemap is overwritten via regen), feedback_overwrite_stale_state (new STATE doc; no historical edits).

Reported delta¶

signal	before
`find site -type f -name '*.md' \\| wc -l`	10,282
`grep -c '<loc>' site/sitemap-companion-md.xml`	10,265
nominal gap	17

Root-cause analysis¶

The "17 gap" headline is two off-by-one mismatches stacked on top of an in-scope vs out-of-scope drift:

grep -c '<loc>' over sitemap-companion-md.xml matches the literal substring <loc> anywhere — including the XML comment header which contains the prose "Each <loc> points at …". That contributes +1 to the headline count. The XML-pair regex <loc>[^<]+</loc> yields 10,264 real URL pairs.
find site -type f -name '*.md' returns every .md file under site/, including 18 non-companion files: top-level page companions (about.html.md, pricing.html.md, etc.), press/*.md, legal/subprocessors.md, security/policy.md, and two repo-internal docs (README.md, assets/BRAND.md). The existing generator (scripts/generate_sitemap_companion_md.py) only ever covered the cases / laws / enforcement directories per its docstring intent.
Disk inventory in those three companion categories is 10,264 — exactly matching the sitemap's real URL count. Diff disk_md_urls - sitemap_urls was 0 before the fix.

So the literal "17 gap" was a measurement artefact, not a missing-URL problem inside the existing 3-category scope. But it surfaced a real sitemap coverage hole: 16 first-class public companion .md surfaces (about / pricing / facts / transparency / data-licensing / legal-fence / compare / index .html.md plus the 6 press/*.md files plus legal/subprocessors.md plus security/policy.md) were not exposed in any sitemap. README.md and assets/BRAND.md are intentionally repo-internal and stay out.

Fix¶

scripts/generate_sitemap_companion_md.py:

Added ROOT_INCLUDE_GLOBS (*.html.md, press/*.md, legal/*.md, security/*.md) + ROOT_EXCLUDE_NAMES (README.md, BRAND.md, index.md).
New helper _enumerate_root_page_urls() emits a ("root", url) entry per matched file.
New flag pair --include-root-pages / --no-include-root-pages (default ON) wires the helper into main().
Promoted --scan-md-only to default ON (with explicit --no-scan-md-only opt-out) so the default sitemap matches the on-disk .md inventory (~10,264 across the 3 companion categories) rather than the legacy HTML-derived ~9,178 figure (which under-counts cases by 1,086).

Post-regen verification¶

$ python3 scripts/generate_sitemap_companion_md.py
[sitemap-companion-md] wrote site/sitemap-companion-md.xml
  (10280 URLs, 1877495 bytes, lastmod=2026-05-12)
        cases: 2286 URLs
        laws:  6493 URLs
  enforcement: 1485 URLs
        root:    16 URLs

check	result
`xmllint --noout site/sitemap-companion-md.xml`	OK (rc=0)
`len(re.findall(r'<loc>[^<]+</loc>', xml))`	10,280 URL pairs
disk in-scope .md count (3 cat + 4 root globs)	10,280
`disk - sitemap` gap	0
`sitemap - disk` orphans	0
brand grep `税務会計AI \\| AutonoMath \\| zeimu-kaikei`	not introduced (none)

The 2 disk files intentionally excluded (README.md, assets/BRAND.md) are repo-internal and not part of the public companion-md surface.

Files touched¶

scripts/generate_sitemap_companion_md.py (~50 LOC added: 1 const block, 1 helper, 3 argparse args, 2 wire lines).
site/sitemap-companion-md.xml (full regen; lastmod=2026-05-12; +16 root entries, +1086 cases entries vs HTML-derived legacy, total 10,280 <url> blocks).
docs/research/wave46/STATE_w46_sitemap_regen.md (this file).

PR¶

To be filled by the push step (see git log / PR description).