Excalibur: Building a 14-module marketing audit engine that runs on multiple LLMs and delivers 22-tab interactive HTML

What Excalibur is

Excalibur is a structured multi-LLM marketing audit framework. It assigns specific analytical modules to specific models, enforces an evidence-tagging protocol across all outputs, and produces a structured deliverable — currently a 22-tab interactive HTML report — that a client can navigate without the analyst being present.

How it evolved

V2 — Sologlass.ca (April 25, 2026). The first documented production deployment. A 14-module Business Growth Intelligence Report covering an ICBC-certified BC auto glass shop. Delivered as tabbed interactive HTML with six core sections. Evidence protocol: every finding tagged as VERIFIED / SITE CRAWL / AUTH SOURCE / NEEDS VALIDATION / CALCULATED. Data inputs: SEO PowerSuite Rank Tracker v8.50.18 [SEO PowerSuite]¹ geo-located from both business addresses, direct site crawl.

V3.1 — Distillery Events (May 2026). The primary LLM audit reported “zero JSON-LD anywhere.” A second independent LLM running the same audit caught what the primary crawler missed: the homepage carried a Restaurant schema block with a fabricated aggregateRating — 5 stars / 500 reviews [raw HTML crawl, May 2026] — against a real GBP rating of 3.8 stars on 32 reviews [Google Business Profile, May 2026].

Per Google’s review-snippet guidelines, fabricated review schema is a documented manual-action target ². This became Priority 0 in the V3.1 audit.

The methodology change locked in at V3.1: independent LLM verification is now a required step. WebFetch and site crawlers don’t reliably scan <script type="application/ld+json"> blocks — raw HTML fetch and grep for ld+json is now part of every schema audit.

The 14-module framework

• M0: Business context and competitive tier classification
• M1: Technical SEO (crawl, canonical, HTTP/HTTPS, dev artifacts)
• M2: On-page signals (title tags, H1s, meta, geo-targeting)
• M3: Schema and structured data (LocalBusiness, AutoRepair, FAQPage, Organization, etc.)
• M4: Page speed and Core Web Vitals [Google Search Central]³
• M5: Content depth and topical authority
• M6: Local Pack mechanics (GBP completeness, service area, category)
• M7: Review velocity and recency analysis
• M8: Competitive matrix (dedicated page structure, franchise vs independent)
• M9: Citation consistency and NAP audit
• M10: SWOT with evidence tagging
• M11: Keyword architecture by tier (core buyer, adjacent, comparison/review)
• M12: Revenue projection and conversion funnel modelling
• M13: Phased roadmap with dependency sequencing

Each finding is tagged to its evidence source. Claims without a source get a NEEDS VALIDATION tag — not a fabricated figure.

The deliverable format

Single-file interactive HTML. Tabbed navigation. No login required. No external dependencies beyond a browser.

The Sologlass V2 report: six tabs, 14 modules, site crawl + geo-located rank data, ICBC competitive matrix, 24-step roadmap [Excalibur V2 audit file, Apr 25, 2026].

The Distillery Events V3.1 report: 22 tabs, full multi-module coverage, schema discovery, brand SERP cannibalization analysis, $233K–$382K 12-month revenue projection [Excalibur V3.1 audit file, May 2026].

What the system can’t claim

Two live deployments are documented. Neither client has been tracked through full implementation of the audit roadmap. Revenue projections are modelled estimates — not realized outcomes.

Sources