Runtime
Console errors, failed requests, final URLs, viewport outcomes, and screenshot provenance.
For engineering and delivery teams
AuditDeck records what rendered, where it rendered, and how each finding was produced. Interpretation never replaces the underlying capture.
Versioned contract. Run, route, navigation, screenshot, element, copy, and evidence identifiers preserve the chain from capture to recommendation.
Evidence model
The engine keeps complete machine-readable findings separate from curated presentation output, with deterministic omission reasons and explicit evidence references.
Console errors, failed requests, final URLs, viewport outcomes, and screenshot provenance.
Titles, headings, meta descriptions, visible links, buttons, and page copy.
Rendered-page axe rules plus bounded deterministic checks without certification claims.
Passive route graph, CTA classification, conversion endpoints, loops, and dead ends.
Availability / September 25, 2026
AuditDeck is a private DRKNYT pilot. The GitHub repository is private and no public AuditDeck npm package has been published. The command below is for operators who already have access; it is not a public installation instruction.
Bounded passive audits produce screenshots, structured findings, journey data and HTML/PDF reports. Human review is needed before sharing conclusions.
Public audit requests, automatic hosted runs, customer accounts, private report delivery and payments. The planner and account/checkout pages are demonstrations, not an order or service commitment.
Artifact shape
Current local engine output includes structured findings, journey data, report HTML/PDF, technical appendices, screenshots, and an AI-friendly review packet.
auditdeck --site https://example.com \ --depth 3 \ --rate-limit-ms 1000 \ --max-pages 50 \ --output ./output
Requires private operator access, Node.js and a compatible Playwright browser. This site does not execute the command.
For AI assistants
Read the capability description and plain-text guide for current availability and limits. These describe AuditDeck; they do not let an assistant run it. No public API, Model Context Protocol (MCP) server or installable plugin is available yet.
Planned integrations will distinguish reading an authorized report from creating an audit job. JSON-LD describes the product; it does not grant access, authenticate a user or start work. Technical severity, business priority and evidence of a verified fix are separate concepts.
Source boundary
Source Review is designed around exact authorization, immutable commit identity, allowlisted text files, secret redaction, provenance, and deletion policy. Customer repository connections are not available in this preview.
| Evidence | Claim level | Required label |
|---|---|---|
| Captured public state | Observed | Deployed evidence |
| Authorized source match | Observed | Source evidence |
| Likely code relationship | Inferred | Correlation |
| Reproduced causal explanation | Confirmed with scope | Root cause |