Technical reference.

Monroya computes two distinct scores. Conflating them is the most common source of confusion, so they're kept separate everywhere in the product.

The readiness score is a weighted 0–100 roll-up of seven signals, each scored independently against its own maximum:

• Schema — presence and validity of Organization, Product, Article, and FAQ JSON-LD.
• FAQ structure — question-shaped headings with direct answers beneath them.
• Answer-first formatting — a 40–60 word extractable answer near the top of the page rather than buried below narrative.
• Freshness — dated content and modified timestamps assistants can read.
• Crawlability — robots rules, AI-crawler allowances, render-blocking, and server-rendered content.
• E-E-A-T signals — named authorship, credentials, and verifiable sourcing.
• Live snippet — whether a provider, asked cold, can produce an accurate description of the brand.

Weights are not uniform: crawlability and answer-first formatting carry the most weight because they gate everything downstream, and the live snippet acts as a multiplier rather than a flat addition. The result is intentionally high-variance — a page with schema but no extractable answer scores materially lower than the average of its parts would suggest.

Stage tracking is prompt-level, not page-level. Each tracked prompt is classified into exactly one buyer stage at generation time, and the matrix is an aggregation over those classifications.

1. Classification. Prompts are generated from a stage-tagged template library — Discovery ("what is X", "how do I solve Y"), Evaluation ("best X for Y", "X vs Y"), Decision (pricing, reviews, switching risk, implementation effort). The stage comes from the template, not from a post-hoc classifier, so it's stable across scans.
2. Execution. Every prompt runs 3× per provider with buyer persona framing prepended.
3. Entity resolution. Each response is scanned for your brand and every tracked competitor using name aliases plus all registered domains, after URL normalization (lowercased, tracking params stripped, www canonicalized, short URLs resolved).
4. Mention classification. Each entity gets one of four outcomes per run: cited (linked source), named (mentioned without a link), category (category discussed, entity absent), absent.
5. Aggregation. The matrix cell for (entity × stage) is the share of runs in that stage where the entity was cited or named, across all providers. Cells backed by fewer than 5 runs are shown as raw counts ("2 of 3") rather than percentages, because a percentage on a sample of 2 is misleading.
6. Gap detection. A competitor cell materially above yours in the same stage generates an opportunity card weighted by stage: Decision-stage gaps rank above Discovery-stage gaps at the same magnitude.

Stability and variance guards apply here too — prompts whose dominant outcome repeats in fewer than 60% of runs are downweighted so the matrix doesn't move on noise. See methodology for the exact thresholds.

Stored per scan: raw provider responses, the extracted source ledger, mention classifications, score deltas, and the prompts you confirmed. Stored per account: site crawl metadata, brand aliases and domains, competitor configuration, and draft history.

Queries to providers are ordinary API calls and are not shared or labeled back to your account. We do not train models on your data. Account deletion purges everything within 30 days. Security detail lives on /security.

• CSV export — scans, source ledger, and the opportunity queue.
• Shareable briefing links — a signed, read-only URL for stakeholders with no account.
• Email delivery — weekly briefing and competitor-movement alerts to any address you add.
• Public API — not available yet. Contact sales if you need it scoped.

Current connections are listed on /integrations.