Performance · Published 26 August 2026
Benchmarking a founder operating system without measuring clicks
A reproducible method for reconciliation accuracy, proposal quality, review time, recovery, and ownership.
Measure record correctness first
Use a fixed company fixture and compare imports, obligations, balances, decisions, and generated indexes against an expected canonical record. Count unexplained divergence, not only successful commands.
Fingerprint inputs and schema versions with the result.
Proposal quality needs a rubric
Score source citation, factual accuracy, stated uncertainty, risk classification, required approvals, and whether the proposed patch is minimal and reversible.
A fluent draft with missing provenance is a failed proposal.
Time the whole review loop
Measure sync, draft, human review, approve or reject, durable write, Git commit, and Run Log visibility separately. Include cold repositories, conflicts, and a failed vendor adapter.
Track the time to explain and recover a partial operation.
Ownership is testable
Restore into a clean directory, open the files without Lycato, reproduce a report, revoke an adapter, and export hosted records.
A workflow is not sovereign merely because it has an export button.
Production checklist
- Verify Founder ownership against the deployed environment, not a screenshot.
- Verify Proposal before effect against the deployed environment, not a screenshot.
- Verify Attributable Run Log against the deployed environment, not a screenshot.
- Verify Testable handback against the deployed environment, not a screenshot.