Building Helmsman as a platform
How one research engine grew into a shared data layer, an operator workspace, and a place to test competing ideas.
Building an independent trading research platform: data, experiments, the operator workspace, and the decisions that shaped it.
How one research engine grew into a shared data layer, an operator workspace, and a place to test competing ideas.
One synthetic corrected value traced through the real observation writer and schema: revisions as new rows, an as-known-at query, frozen run inputs, and freshness judged against schedule and consumer.
The contract between model output and deterministic code in Helmsman, traced with synthetic replies: what a model may emit, what the validator rejects, and what happens when the handoff fails.
A synthetic identity record traced through the real score store: which change needs new model work, which only reruns deterministic math, and why replaying stored scores does not recover the past.
Stable identity, evidence gates, and visible deployment states in the Helmsman operator workspace.
One vague chart claim turned into three versioned definitions and replayed on synthetic data through Structure Lab's real detector and registry. The replay shows what a new version does to earlier annotations and evidence.
Polymarket research, unreliable causal graphs, and the value of preserving a rejection with its evidence.
One synthetic headline traced through a partial observation, an upgrade, duplicate deliveries, an edit and a retraction, using the ingestion writer's real storage contract.
A cutover trace from Helmsman's move to Kubernetes CronJobs: one declared owner per job, a revision-pinned switch, a rollback that could not be a revert, and alerts that had to move too.
The design of an explicit promotion boundary: frozen definitions, concurrent requests, durable evidence gates, and the limits of a deployed research workspace.