Resources
Everything we publish, for a person reading and for an agent enumerating. Every list on this page has a JSON twin, because half of what reads it is not a browser.
Datasets
Each one names its licence and, on its own page, the limit that makes it the wrong dataset for some question. All 9.
Neurarch architecture corpus
36 neural network architectures kept as typed graphs rather than diagrams.
Neurarch structural check catalogue
The 41 structural checks Neurarch runs on a model graph, as data: id, severity, category, the condition that triggers it, why it costs something, and the fix.
Verifier grounding study (264 graphs)
Clean reference architectures plus systematically corrupted variants (broken attention head divisibility, linear width mismatches, severed connections), each built as a real PyTorch model and run on a GPU.
Arch-Bench arena results
Frontier language models scored on design-from-spec tasks by a deterministic verifier rather than a human or an LLM judge.
Arch-Bench task set
The task definitions behind the benchmark: design-from-spec and repair-and-extend instances for agents that build neural network architectures.
arch-design-sft: verified architecture-design SFT data
Supervised fine-tuning data for neural architecture design treated as structured graph editing.
Structure, verdict and trained outcome triples
The corpus that pairs what a design looks like with what it did.
Verified architecture-design reasoning traces (Claude)
Spec to reasoning to design triples where the design is re-graded by the same deterministic verifier the benchmark uses, and only passing traces are kept.
Verified architecture-design reasoning traces (Grok)
The same verified spec to reasoning to design triples as the Claude split, rejection-sampled from a different frontier model, so the two can be pooled for volume or held apart to see how much of the reasoning style is model-specific.
Call it
Verifier API
POST a graph to /api/v1/check and get the verdict the app shows a human: readiness, cost, deploy target. Deterministic, no model call.
Structural lint API
POST /api/v1/lint returns the structural findings alone, when the full pipeline verdict is more than you need.
Candidate ranking API
POST /api/v1/rank orders k unexecuted designs for an agent that proposed more candidates than it can afford to train.
MCP server
The checks as tools an agent can call over stdio, on a laptop, offline.
GitHub Action
The same checks on every pull request that touches a model.
Open source
neurarch-arch-bench
The benchmark and RL environment: tasks, the deterministic grader, an env server, and the training data on this page. MIT.
neurarch-mcp
MCP server on npm. MIT.
neurarch-lint
The CI action, built to be copied. MIT.