Skip to content

Initial implementation project

This project turns the feasibility study into a repeatable audit workflow. Its current deliverable is a development preview: portable skills, bounded browser/source collectors, evidence-linked judgments, deterministic scoring and per-site reports.

To use the audit, start with installation and your first audit. To review its outputs, use example reports. Model campaigns now have their own evaluation project.

What works

The agent selects scope and triggers collection, then reviews a frozen packet using quantitative, guideline and claims roles. Current checks cover public browser behavior and disclosures, with optional selected source files. A deterministic scorer preserves pass, fail and unknown counts; the report includes coverage and next auditing steps.

The check reference is the source of truth for implemented breadth and limits. Workflow and artifacts describes the current handoffs. Reporting defines human and machine outputs. Reusable measurement and interpretation principles live in the knowledge base.

Current assessment and next work

The project assessment compares the implementation with the intended audit experience. The next milestone is one reliable install-to-report audit, including collection recovery and portable delivery. Plugin manifests exist; clean installation and complete execution under another harness remain unvalidated. Qualitative evidence discovery, independent recommendation review, skill comparisons and Jev/Clef experiments are the other main gaps.

The roadmap states the next gates. The backlog keeps detailed ideas and feedback tied to pipeline components. The evidence record preserves executed experiments and validation results.

Historical implementation notes

These explain how the prototype developed; they are not current getting-started instructions.

Older reports and benchmark campaigns remain accessible through their own indexes. Historical source paths and immutable artifact bundles are retained for reproducibility.