Live · The research engine behind Automate Capture
The execution layer of public research.
Papers say what could work. Research Radar shows what does. An always-on engine reads the world's public research, builds the claims it finds, tests them, and writes the result to a permanent record. What you see below is the live output.
Feed
Newest first. Every reproduction, combination, and release across the pipeline.
No activity yet. The pipeline runs daily — check back tomorrow.
What the engine does
Outcomes, not architecture. Every artifact in the feed above is the end of a long chain that runs without a human in the loop.
→
Reads
Continuously ingests the world's public technical research as it's published. The corpus is curated by relevance, not by author.
→
Builds
Promising claims become real, runnable code. The engine writes the package, the tests, and the documentation in one pass.
→
Validates
Every build faces an independent quality gate. The score is honest — passing, failing, and partial reproductions are all surfaced.
→
Combines
Survivors don't sit in isolation. The engine finds non-obvious connections between unrelated reproductions and tries to merge them into capabilities no single paper proposed.
→
Attests
Every passing reproduction writes a permanent record on chain — paper, build hash, score, tests. Citable forever, by anyone, with no gatekeeper.
The corpus compounds. Every week the engine reads more, builds more, and learns more about what's connectable to what. The value isn't any single reproduction — it's the accumulated map. Read the thesis →