SDRAF.

Verified evidence over scraped noise.

Semantic Divergence Research & Adjudication Framework. When a research question hits a dead end, SDRAF fans out across multiple adversarial lanes, strips the SEO noise, and returns adjudicated evidence — running entirely on your own hardware. in build

AUTONOMOUS WORKFLOW | 5-STEP PIPELINE

Question

A research query, or a stuck point a tool couldn't resolve, enters the loop.

Divergence

Five adversarial research lanes run in parallel, each chasing a different angle.

Retrieval

A native Rust navigator pulls live sources — no third-party API, no data leaving your machine.

Adjudication

A local model filter strips SEO noise and extracts the math, code, and citations that matter.

Result

A concise, cited brief comes back — provenance attached, ready to act on.

      SAMPLE RUN
      sdraf research
    

[19:13:22] [INFO] question → divergence (5 lanes)
[INFO] retrieval (3/5) — pulling toolchain docs...
[ERR] reported blocker: `ValueError: Unsupported task: image_text_audio_to_text`
[INFO] adjudication — filtering noise, extracting citations...

> RESULT ADJUDICATED · confidence 0.88 · 3 sources cited
> FINDING: the official toolchain handles multimodal natively. Drop the --task flag.
> brief written · provenance attached