diff --git a/docs/wiki/Audit-Skill.md b/docs/wiki/Audit-Skill.md
index 000da8d..c6fe618 100644
--- a/docs/wiki/Audit-Skill.md
+++ b/docs/wiki/Audit-Skill.md
@@ -1,10 +1,10 @@
# Audit Skill
-The audit skill scores a site across five content layers — SEO, AEO, GEO, perf and
-security — plus a site-level cross-URL pass, an indexability-contradictions pass, and
-sitemap hygiene, returning findings that carry an observed value, an expected value, a
-fix and a `fixTier`. It never edits the site; it only diagnoses, then hands the list to a
-human, CI job, or `omnirank fix`.
+The audit skill scores a site across **five** content layers — SEO, AEO, GEO, perf and
+security — plus a site-level cross-URL pass, an indexability-contradictions pass, and one
+sitemap-hygiene check, returning findings that carry an observed value, an expected
+value, a fix and a `fixTier`. It never edits the site; it only diagnoses, then hands the
+list to a human, CI job, or `omnirank fix`.
Everything on this page is verified against `scripts/py/omnirank/audit.py`,
`scripts/py/omnirank/gates/`, `scripts/py/omnirank/report.py`, and `skills/audit/`.
@@ -19,7 +19,7 @@ phrases and what will *not* trigger it: [[Claude-Code-Setup#what-should-i-say-to
## CLI reference
-Verbatim `--help` output, unchanged since v0.3.0:
+Verbatim `--help` output from v0.4.0:
```
$ python3 -m omnirank.cli audit --help
@@ -64,42 +64,42 @@ Two behaviours worth being precise about:
- **Passing `--fail-on` with zero gate names** explicitly overrides the config to an empty
gate list, so the run always exits `0` regardless of what `audit.failOn` says.
-**Console output as of v0.2.1** groups findings by `id`, shows a count, the shared
-`expected`/`fix` text, and up to three example URLs — every group is shown, nothing is
-silently truncated. `--detail` keeps the historical ungrouped, 25-finding-capped view. The
-JSON report is unaffected by either flag: it always holds every finding individually.
+**Console output** groups findings by `id`, shows a count, the shared `expected`/`fix`
+text, and up to three example URLs — every group is shown, nothing is silently truncated.
+`--detail` keeps the historical ungrouped, 25-finding-capped view. The JSON report is
+unaffected by either flag: it always holds every finding individually.
## What are the five layers?
| Layer | Audience | What it wants |
|---|---|---|
-| SEO | Googlebot, Bingbot | Crawlable, canonical, correctly sized metadata, valid structured data, no self-contradictions between what the site submits and what it forbids |
+| SEO | Googlebot, Bingbot | Crawlable, canonical, correctly sized metadata, valid structured data, and no self-contradictions between what the site submits and what it forbids |
| AEO | AI Overviews, Copilot, voice assistants | A short, liftable, factual answer near the top of the page, sized for the page's script |
| GEO | ChatGPT, Claude, Perplexity, Gemini | Machine-ingestible ground truth (`llms.txt`, `facts.json`) plus explicit permission to cite |
| perf | Every crawler and user agent | Fast full response time, reasonable HTML weight, compression, and no excess render-blocking `` scripts — derived from one HTTP response, no browser involved |
-| security | Browsers rendering the page | No mixed content and an http→https redirect; response headers reported as inventory, never graded (v0.4.0) |
+| security | Browsers rendering the page | No mixed content and an http→https redirect (the two that break crawling/rendering); response headers reported as inventory, never graded — **new in v0.4.0**, see [[Security-Layer]] |
-Crawl hygiene, the site-level cross-URL pass, and the indexability-contradictions pass all
-carry `layer: "seo"` in the report — there is no separate `"hygiene"`, `"site"` or
-`"contradictions"` value. `report.py` defines `Layer = Literal["seo", "aeo", "geo",
-"offsite", "smm", "perf", "security"]`. `perf` has been a real, populated layer since
-0.2.0 — every audited page runs `perf.run(page)`. `security` is new in v0.4.0 and
-populated the same way (`security.run_page(page)` per page, plus one site-level
-`http://` redirect probe). `offsite` and `smm` remain reserved for roadmap skills and are
-unused by any gate today.
+Crawl hygiene, the site-level cross-URL pass, and the indexability-contradictions pass
+([[Contradictions]]) all carry `layer: "seo"` in the report — there is no separate
+`"hygiene"`, `"site"` or `"contradictions"` value. `report.py` defines `Layer =
+Literal["seo", "aeo", "geo", "offsite", "smm", "perf", "security"]`. `perf` has been a
+real, populated layer since v0.2.0. `security` is new in v0.4.0 and is populated the same
+way (`security.run_page(page)` per page, plus one site-level `http://` redirect probe).
+`offsite` and `smm` remain reserved for roadmap skills and are unused by any gate today.
As of 0.2.0, `audit_site()` keeps every fetched page's HTML alive as a `PageData` record
instead of discarding it after the per-page gates run — that is what makes the site-level
-pass below possible.
+and contradiction passes below possible.
## Every gate, by layer
Every finding carries a `gate` (what `--fail-on` matches against), an `id` (a stable,
dotted identifier), and a `fixTier` (`mechanical`/`templated`/`drafted`/`advisory`/
-`infrastructure`, described in [[Fix-Tiers-and-Applicability]]). For the complete,
-registry-generated table of all 67 ids see [[Finding-Reference]]. 42 distinct gate names
-exist in `audit.failOn`'s schema enum; only 21 of them are error-capable and can actually
-gate a build — see [[CI-Recipes#which-gates-can-actually-fail-a-build-with---fail-on]].
+`infrastructure`, described in [[Fix-Tiers-and-Applicability]]). 42 gate names exist in
+`audit.failOn`'s schema enum (up from 28 pre-v0.4.0); 21 of them are error-capable and can
+actually trip `--fail-on`, 21 cannot — see [[CI-Recipes#which-gates-can-actually-fail-a-build-with---fail-on]]
+for the full breakdown. For the complete, registry-generated table of all 67 ids see
+[[Finding-Reference]].
### SEO gates
@@ -113,8 +113,9 @@ gate a build — see [[CI-Recipes#which-gates-can-actually-fail-a-build-with---f
| `hreflang` | If any `hreflang` alternates exist, one is `x-default` | warning | Add `` |
| `image-dims` | Every `` has both `width` and `height` | warning | Set explicit dimensions so the browser reserves space |
| `lastmod-inflation` | See [below](#sitemap-lastmod-inflation) | warning | Stamp `lastmod` from real publish/update timestamps |
-| `schema` | See [below](#structured-data-gates-schema-schema-fabrication) | error / warning | — |
-| `schema-fabrication` | See [below](#structured-data-gates-schema-schema-fabrication) | error | — |
+| `schema` | See [below](#structured-data-gates-schema-schema-fabrication-schema-required) | error / warning | — |
+| `schema-fabrication` | See [below](#structured-data-gates-schema-schema-fabrication-schema-required) | error | — |
+| `schema-required` | See [below](#structured-data-gates-schema-schema-fabrication-schema-required) | warning | — |
### AEO gates
@@ -123,16 +124,17 @@ gate a build — see [[CI-Recipes#which-gates-can-actually-fail-a-build-with---f
| `answer-block` | An element matches the configured selector (default `.answer-block`) | error if absent |
| `answer-block` | The block's text falls inside its script's word/character band | error if outside the range |
| `answer-block` | The block contains no `
`/``/`
` | error if list markup is present |
-| `faq` | At least 3 FAQ pairs, as `
`/`
`/`
` or `` elements | **warning as of v0.2.1** (downgraded from error — firing on pricing/about/404 pages was not defensible) |
+| `faq` | At least 3 FAQ pairs, as `
`/`
`/`
` or `` elements | warning if fewer (downgraded from error in v0.2.1) |
| `speakable` | Every `speakable.cssSelector` in any JSON-LD block resolves to a real element on the page | error if any selector matches nothing |
-**The band is script-aware as of 0.2.0, not a fixed 40–60 words.** `str.split()` returns a
-single token for an entire Chinese, Japanese, Thai or Khmer paragraph, so the old fixed
-band flagged every compliant CJK answer block as far too short. `bands.resolve_band()`
-now measures **characters** (80–200 by default) for the `cjk` script family, keyed off the
-page's ``, and keeps **word** counting for Latin, Brahmic (Bengali, Hindi,
-Tamil, ...), Arabic and Cyrillic scripts. Configure per script with `aeo.answerBlock` — see
-[[Configuration-Reference#aeoanswerblock]].
+**The band is script-aware, not a fixed 40–60 words.** `str.split()` returns a single
+token for an entire Chinese, Japanese, Thai or Khmer paragraph, so a fixed band would flag
+every compliant CJK answer block as far too short. `bands.resolve_band()` measures
+**characters** (80–200 by default) for the `cjk` script family, keyed off the page's
+``, and keeps **word** counting for Latin, Brahmic (Bengali, Hindi, Tamil,
+...), Arabic and Cyrillic scripts. Configure per script with `aeo.answerBlock` — see
+[[Configuration-Reference#aeoanswerblock]]. **A missing `` breaks this
+silently** — see `lang` under [On-page gates](#on-page-gates-new-in-v040) below.
### GEO gates
@@ -168,7 +170,7 @@ than the filename stem.
the wildcard (`*`) block blocks everything. A `Disallow: /some-path` does **not** count —
only a full-site block trips this gate.
-### Structured-data gates (`schema`, `schema-fabrication`)
+### Structured-data gates (`schema`, `schema-fabrication`, `schema-required`)
| Gate | Rule | Severity |
|---|---|---|
@@ -178,50 +180,56 @@ only a full-site block trips this gate.
| `schema` | Every top-level node has `@context` | warning |
| `schema-fabrication` | Every `AggregateRating` node has a non-zero `ratingCount` (or `reviewCount`) | error |
| `schema-fabrication` | Every `Review` node has an `author` | error |
+| `schema-required` | Google's documented rich-result properties are present for Article/NewsArticle/BlogPosting, Product, FAQPage, BreadcrumbList, Organization and LocalBusiness — **new in v0.4.0** | warning |
+
+**`schema-required` checks GOOGLE's requirement, not schema.org's.** schema.org marks no
+property required at all, so a node can be perfectly valid schema.org and still miss a
+Google rich result — every `seo.schema-required.missing-property` finding says so
+explicitly, and names both `RICH_RESULT_RULES` and the date the table was transcribed
+(`RICH_RESULT_RULES_AS_OF = "2026-08-04"` in `gates/jsonld.py`). It is `warning`, not
+`error`, because the table is a hand-transcribed snapshot with no freshness test yet, and
+a stale required-property table would produce fabricated errors at `error` severity. One
+finding per node, listing every missing property. A node carrying only `@id`/`@type`/
+`@context` is a reference to an entity declared elsewhere and is never checked — flagging
+it would report correct markup as broken.
"Top-level node" means each `
-+
++