Skip to content

Published benchmark

Manuals scored against the
products they emulate. Published.

Every help-center tool claims great docs. We score ours against the real market leaders each style imitates — with a structural scorer plus a 3-judge LLM panel — and publish the numbers. To our knowledge, no user-manual or help-center vendor publishes any documentation-quality score at all.

Mid-market

8.4/10

vs Odoo · Zoho · HubSpot

Enterprise

8.2/10

vs Dynamics · Salesforce · NetSuite

StyleModeled onBlendedStatus

Enterprise

User/admin split, learning paths, reference + concept tiers.

Dynamics 365 · Salesforce · NetSuite8.2

structural 9.0

validated

Mid-market

One section per app, config + usage colocated, decision-point framing.

Odoo · Zoho · HubSpot8.4

structural 9.0

validated

SaaS

Flat IA, "get started" per feature, friendly voice. Bounded only by screenshot density.

Monday · Notion · Intercom7.9

structural 9.0

at the gate

Corporate

Separate App-Help / Config / Operations books, O2C·P2P process chains. Blended validation ongoing (capture-level).

SAP · Oracle Fusion · Infor8.3★structural pass

★ Corporate: structural 8.3; blended pending capture-level work (see below). All scores are on live generated doc-sets.

Who else publishes a manuals-quality score?

AtloriaYes — published, reproducible harness, per-style
Zendesk (Knowledge Builder)No published quality score
Intercom (Fin)No published quality score
Tango / ScribeNo published quality score

Methodology

  • Per-style rubrics, calibrated to real products. Each of the four styles is scored against the public docs of the three market leaders it emulates — not one generic bar. An enterprise manual is graded like Dynamics; a SaaS one like Notion.
  • Structural scorer + LLM judge panel. A deterministic scorer measures IA depth, page-type mix, cross-links, callouts, word bands and screenshot density; a 3-judge LLM panel (median of 3, 25-page stratified sample) scores the qualitative dimensions — voice, role separation, learning curricula. The two blend into the published number.
  • Grounded in the real product. Every page is derived from the product's own screens and code — the manuals describe what exists, and go stale when the code changes. Screenshots are real captures, never generated.
  • Honest ceilings, published. SaaS sits at 7.9 and corporate's blended trails its 8.3 structural for one reason: screenshot density (real multi-shot capture is the remaining lever). We publish that rather than hide it.

See it on your own product.

Point Atloria at a repository and get manuals graded against this bar — plus a reference hub, API playground and grounded chat.

Try it free →