morphiq-scan（形态扫描）

Pipeline Position

Step 1 of 4 — entry point.

• Input: A domain URL from the user.
• Output: Scan Report (JSON) → consumed by morphiq-rank.
• Data contract: See PIPELINE.md §1 for the Scan Report schema.

Purpose

Morphiq Scan audits a website's readiness for AI visibility. It answers two questions: "Can AI systems parse and understand this site?" (Technical Score) and "Where are the gaps preventing AI citations?" (issue identification). The output feeds morphiq-rank for prioritization.

Workflow

Step 1: Discover Pages

1. Fetch {domain}/robots.txt — extract sitemap URLs
2. Fetch {domain}/sitemap.xml if not found in robots.txt
3. Classify discovered pages by type using URL pattern matching
4. Select up to 10 marketing-relevant pages, prioritized: home → pricing → features → product → solutions → about → blog → other → documentation
5. Exclude non-marketing pages from scoring (contact, login, signup, legal, demo, careers, changelog)

For page type classification and URL patterns, read references/page-type-rules.md.

Step 2: Audit Policy Files (Category 5 — Domain Level)

1. robots.txt — Validate existence, format, AI crawler access (GPTBot, Google-Extended, Anthropic-AI, PerplexityBot)
2. llms.txt — Validate existence and quality (≥500 chars = good, <500 = thin)
3. llms-full.txt — Check existence
4. sitemap.xml — Validate XML structure

Score on 10-point scale. Generate issues for findings.

For detection rules and scoring, read references/policy-files.md.

Step 3: Score Each Page — Per-Page Technical Score (0–100)

For each selected page, compute across four dimensions:

For sub-check methodology, read references/agentic-readiness.md.

Step 4: Score Content Quality (Category 2 — Per Page, 20 pts)

Evaluate citation-readiness: title clarity (3), TL;DR placement (4), E-E-A-T signals (6), statistics & citations (5), real examples (2).

For criteria, read references/content-quality.md.

Step 5: Score Chunking & Retrieval (Category 3 — Per Page, 15 pts)

Evaluate LLM retrieval optimization: heading hierarchy (3), section scope (3), paragraph self-containment (2.25), answer-first openings (2.25), vocabulary/lists/FAQ/summary (4.5).

For criteria, read references/chunking-retrieval.md.

Step 6: Simulate Query Fanout (Category 4 — Domain Level, 10 pts)

1. Identify core topics from page content
2. Generate simulated sub-queries using per-model rules (GPT-5.4 two-phase, Claude bundled, Gemini systematic)
3. Check site coverage per sub-query
4. Apply citation weights (citation-producing 1.5x, silent 0.5x, site: 2x)
5. Score: (weighted answered / weighted total) × 10

For simulation rules and coverage scoring, read references/query-fanout.md.

Step 7: Compute Aggregate Score (0–100)

overall = agentic_readiness(45) + content_quality(20) + chunking_retrieval(15) + query_fanout(10) + policy_files(10)

Categories 1–3 averaged across pages then scaled. Categories 4–5 are domain-level.

For the full rubric, read references/scoring-rubric.md.

Step 8: Generate Issues

Create issues with: id, category, severity, summary, detail, affected_urls, remediation_hint. Use issue ID patterns from the pipeline contract.

Step 9: Produce Scan Report

Assemble Scan Report JSON (PIPELINE.md §1): domain metadata, per-page scores, issues, schema detection, policy file status, query fanout analysis.

SaaS Detection

Before scoring, detect SaaS by matching 2+ of 3 content sources against indicator terms (platform, saas, cloud, api, etc.). Changes expected schemas for product/pricing/features/solutions pages.

For detection logic, read references/page-type-rules.md.

Reference Files