# Arwp Search Maturity > Reverse-engineer observable, evidence-bearing patterns from currently visible Search/AI reference pages, compare them with a target site, and route repeated gaps into evidence-backed experiments without pretending correlations are ranking factors. Use when the goal is competitive/reference analysis, site maturity benchmarking, fast-riser analysis or finding repeatable characteristics of discoverable pages. - Skill: `dkharlanau/arwp-search-maturity` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add dkharlanau/arwp-search-maturity` - Raw SKILL.md: https://api.skillmd.com/api/skills/dkharlanau/arwp-search-maturity/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - License: Apache-2.0 - Author: dkharlanau (https://skillmd.com/u/dkharlanau) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/dkharlanau/arwp-search-maturity --- # ARWP Search Maturity Use this specialist when the Growth Loop needs a **reference cohort**, not another generic checklist. Use `docs/ANTI-PATTERNS.md` to review fabricated maturity, weak evidence and measurement errors; preserve unknown states and legitimate countercases. Record findings with `templates/growth/anti-pattern-review.md`. ## Goal Turn volatile Search/AI observations into an explainable comparison: `intent → timestamped visible references → observable evidence vector → cohort pattern → target gap → experiment → verify → measure` The output is a hypothesis generator and prioritization input. It is not a list of secret ranking factors. ## Sources of truth Read: - `docs/SEARCH-MATURITY-BENCHMARK.md`; - `docs/INTERNAL-DISCOVERY-EVIDENCE.md` when reviewing `internalTopicalGraph`; - `docs/INTENT-OWNERSHIP.md` when owner query/grounding evidence may imply a new or competing canonical page; - `schema/search-maturity-corpus.schema.json`; - the relevant reviewed corpus under `benchmarks/search-maturity/`; - current primary platform guidance; - the target site's actual content, entity, Search Surface and owner evidence where available. ## Workflow 1. **Define the intent family first.** Do not benchmark unrelated high-authority sites merely because they rank for something. 2. **Resolve intent ownership before creating URLs.** When owner Search/AI evidence exists, map the intent family to a reviewed disposition and zero/one/multiple canonical owners with `arwp-intent-ownership`. Strengthen an existing owner first; query or fan-out variation is not a page-creation instruction. Deliberately decline query mismatches instead of turning them into owner gaps. 3. **Capture the visibility observation exactly.** Preserve provider/surface, query, locale/device when known and observation time. Store a numeric rank only when explicit `rankEvidence` independently supports that exact rank. 4. **Separate ownership.** Independent references can inform the external cohort. ARWP-owned/project-reference sites remain implementation/dogfood evidence. 5. **Review only observable dimensions.** Use the 16 Search Maturity dimensions. Omit anything not actually checked; omission is `unknown`, never failure. 6. **Classify evidence.** Keep `documented-platform`, `observed-correlation`, `experiment` and `unknown` distinct. 7. **Compile the cohort vector.** Prefer repeated patterns and state distributions over a universal score. 8. **Compare the target.** A target gap is interesting only when the reference pattern is sufficiently repeated and the target evidence is real. 9. **Perform applicability review.** Ask whether closing the gap improves the actual user/evidence artifact. Reject cosmetic copying and cargo cult. 10. **Route accepted work through existing ARWP machinery.** Growth hypothesis → Adaptive Upgrade → safe transformation/manual edit → verification receipt. 11. **Probe owner production surfaces after source changes.** Capture the bounded HTTP/robots/noindex/canonical observation in a derived ledger. A successful probe means evidence was collected, not that production is ready. 12. **Gate served intent owners separately.** Run the deterministic ownership gate over the observed ledger. A served family must have exactly one currently `indexable` owner; preserve the proof artifact on failure. Reviewed `decline` families are intentionally excluded. 13. **Verify source→production parity separately.** Technical owner eligibility does not identify the deployed commit. Use deployment/change receipts when that distinction matters. 14. **Measure external outcomes separately.** Search visibility, AI retrieval/citation, referral traffic and business outcomes are different signals. 15. **Preserve negative results.** A change with no effect is useful evidence and should influence future priors. ### Intent ownership evidence When query, grounding or referral observations are being used to decide content architecture: - declare the bounded intent family before reviewing phrases; - explicitly decide whether the site should serve or decline that family; - keep Google, Bing, referral and manual observations provider-scoped; - treat exactly one indexable reviewed owner as `owned`; - treat zero owners on a served family as a review gap, not an automatic new-page request; - treat reviewed query/navigational mismatches as `declined`, with no canonical owner and an explicit reason; - treat multiple owners as `fragmented`, not automatic proof of cannibalization; - treat off-owner observations as a routing/relevance review signal, not proof of ranking harm; - probe the public canonical inventory before claiming current technical eligibility; - use `gate` on the observed ledger when CI/release policy requires every served family to be owned; - preserve probe artifacts even when the gate fails so the failure stays inspectable; - keep deployment commit parity independent from public HTTP eligibility; - preserve live owner query cohorts in private target evidence rather than public ARWP fixtures. Use `docs/INTENT-OWNERSHIP.md` for the full contract. ### Internal topical graph evidence When reviewing `internalTopicalGraph`: - prefer rendered/runtime or matching build-artifact anchors over repository-source guesses; - classify redirect-only aliases as URL transitions, not weak content nodes; - count shared-renderer links in source mode only when route → renderer → visible anchor behavior is deterministic and inspectable; - keep uncertain renderer/source relationships `unknown` instead of scoring them absent; - distinguish contextual/hub/action links from global navigation and footer reachability; - keep canonical owner identity, redirects, sitemap state and link relationships as separate observations. Use `docs/INTERNAL-DISCOVERY-EVIDENCE.md` for the full contract. ## Fast-riser handling A new page observed in retrieval can help study rapid discovery, but do not infer crawl/index/rank history from publication date. Record: - publication date if observable; - observation timestamp; - `retrievalAgeAtObservation`; - exact rank only with explicit `rankEvidence`. Never rename `age at observation` to `time to rank`. ## Dimension review The current dimensions cover: - retrieval clarity and topical focus; - answer-first structure; - evidence density and first-party evidence; - author/entity identity and site transparency; - semantic HTML and truthful structured data; - internal topical graph; - freshness and URL stability; - multimodal evidence and external corroboration; - real utility surfaces; - browser-agent accessibility. Do not force all dimensions onto every site type. ## Hard rules - Do not copy competitor prose, layouts, assets or branding. - Do not fabricate rank, traffic, citations, backlinks, authorship, dates or evidence. - Do not encode a cohort feature as a Google/Bing ranking factor unless the platform documents that claim. - Do not create one page per query variant. - Do not turn a reviewed query mismatch into a content gap merely because it generated impressions or retrieval. - Do not infer keyword cannibalization merely because an observed query/grounding phrase lands off the reviewed owner. - Do not treat successful `probe` execution as proof that served intent owners are ready; use the separate gate. - Do not treat a passing ownership gate as proof of actual Google indexing, ranking, AI citation or deployed commit identity. - Do not optimize the benchmark by selecting only supporting examples. - Do not collapse unknown into zero. - Do not hide misses or negative experiments. - Do not score redirect aliases as isolated weak pages or literal source imports as rendered links when the renderer contract is unverified. - Do not publish confidential algorithms, learned proprietary priors or candidate patent claims merely to document R&D. - Do not let a single composite score drive changes. ## CLI ```bash node bin/arwp-search-maturity.mjs check \ benchmarks/search-maturity/pilot-2026-09-07.json node bin/arwp-search-maturity.mjs cohort \ benchmarks/search-maturity/pilot-2026-09-07.json \ --intent=ai-search-optimization node bin/arwp-search-maturity.mjs diff \ benchmarks/search-maturity/pilot-2026-09-07.json \ target-search-maturity.json \ --intent=ai-search-optimization node bin/arwp-intent-ownership.mjs report \ intent-ownership.json node bin/arwp-intent-ownership.mjs probe \ intent-ownership.json \ --proof-output=surface-proof.json \ --ledger-output=intent-ownership.observed.json node bin/arwp-intent-ownership.mjs gate \ intent-ownership.observed.json ``` ## Done when - the reference cohort is timestamped and reviewable; - independent and owner-controlled evidence are separated; - unmeasured features remain unknown; - query/grounding variation maps to reviewed canonical ownership or explicit decline before any new-page decision; - repeated cohort patterns are explainable without a magic score; - target gaps retain evidence class and uncertainty; - accepted actions enter existing verification/measurement gates; - current served owner readiness is based on an observed production ledger, not source assumptions; - production probe, ownership gate and deployment parity remain separate evidence layers; - ranking/citation causality is not inferred from correlation; - negative outcomes remain available for future learning. ## Functional emulation and Evidence Relay When asked to reproduce maturity, find publication venues, connect DOI evidence or redesign growth analytics, read [Evidence Relay](../../docs/EVIDENCE-RELAY.md). Reproduce an observable useful function with the target site's own artifacts; do not imitate history or independent reputation. The venue register contains candidates and eligibility requirements, not completed submissions. Read DOI records for the exact supported claim; identifiers do not confer endorsement. For a reviewed artifact cohort, run `node skills/arwp-search-maturity/scripts/evidence-relay.mjs ` from the ARWP checkout. The script uses existing BraidGraph node references and evidence receipt references; it is not a second graph. It reports monitored yields, unknown coverage, immature assets, independent owner groups and citation support. Keep the synthetic example labeled; first-observed lags are not first-indexing times. Do not turn parallel stage observations into a visitor funnel or causal attribution.