autoresearch-with-docs

Automated iterative benchmarking against a local documentation corpus is achieved by running corpus-guided experiments that evaluate performance impact and log outcomes for continuous improvement. The workflow reads configuration files, dispatches focused changes, and keeps or reverts them based on results.

A perf-driven AutoResearch loop is an automated workflow that runs iterative experiments grounded in a local documentation corpus. It dispatches focused changes, evaluates their performance impact, and keeps or reverts modifications based on standardized results logged for auditability.

To start automated experimentation, supply a target argument to initiate the workflow, which prepares state, switches to a dedicated autoresearch branch, and runs the iterative evaluation loop. The workflow enforces contracts by reading configuration files like autoresearch.toml.

Yes, automated AutoResearch requires specific configuration files to enforce contracts, reading program.md, autoresearch.toml, and related state files. These files define experiment limits, guardrails, and state rescue parameters to ensure reproducible baselines and standardized result logging.

Benchmarking results are logged for auditability by emitting standardized outputs to a results.tsv file within a branch-scoped workflow. The automated loop creates dedicated autoresearch branches, evaluates performance impact, and records outcomes to track whether changes are kept or reverted.

Yes, you can rescue experiment state if an automated benchmarking workflow is interrupted because the loop reads related state files and enforces configurable experiment limits. This state rescue capability ensures continuous improvement workflows can recover and resume without losing prior progress.