`FINDINGS.md`

17 lines
# TK-10658 — DW Shopify sitemap health audit (read-only, $0)

**Verdict: mostly ALL-CLEAR** — the sitemap itself is healthy. Value is in cross-amplifying TK-10656 + one hand-off to gmc-agent. No NEW independent gated action (fixes route through existing tickets).

## Verified findings
1. **Product sitemap: HEALTHY.** 84 product sub-sitemaps (default locale) × 3 locales. Random sample of 250 product URLs across 12 sub-sitemaps → **250/250 resolve HTTP 200** (the 4 initial "code 0" were transient fetch timeouts; all 200 on retry — VERIFY-BEFORE-ACTING). No stale 404s, no redirects. Shopify keeps it current.
2. **Collections sitemap crawl-instructs the 19 empty pages (amplifies TK-10656).** The collections sitemap lists 577 URLs = all published collections. **19/19 of the TK-10656 empty-to-shopper collections ARE in the sitemap** (30/30 sampled collection URLs = 200). So Google is actively told to crawl all 19 dead/empty pages — INCLUDING the `versa-designed-surfaces-wallcoverings` private-label LEAK page. → Reinforces the TK-10656 fix: **unpublishing those collections auto-removes them from the sitemap** (Shopify regenerates it). No separate sitemap action needed.

## Secondary (hand-offs, not my lane to fix)
3. **~2% of sitemap products are `-sample-` pages (→ gmc-agent / TK-10635).** In a 17k-URL sample, 345 (~2%, est ~1,700 across ~84k published) are memo/`-sample-` products (e.g. `dwhf-70000-sample-…`) — thin/duplicate $4.25 sample pages Google indexes. Overlaps the known $4.25 GMC-leak class; handing to gmc-agent, not fixing here.
4. **en-gb sitemap persists (→ TK-10406).** en-ca + en-gb locale sitemaps both exist though Steve dropped UK from Google targeting (TK-10406). Minor — the GB market/sitemap is still generated. Flag to crash-recovery/TK-10406.
5. **Agentic-commerce discovery (informational).** New `sitemap_agentic_discovery.xml` → `/agents.md` (HTTP 200) — Shopify's AI-agent discovery surface. No action; noted for awareness.

## Honesty / limits
- Product-URL health is a 250-URL random sample (0 non-200 after retry) → high confidence the sitemap is clean, but not an exhaustive 84k crawl. The ~1,700 sample-product estimate is extrapolated from a 17k-URL sample (~2%).
- Self-gated (no full contrarian pass) because this is a low-risk all-clear with NO new independent gated action — the two actionable items route to their owning tickets (TK-10656 unpublish, gmc-agent samples). Applied contrarian discipline inline (retried the 4 code-0s; verified 19/19 directly).