Catch a production distribution changing under your model
one move: turn change into a signal
Open the code pane immediately, then use the Spec tab, optional design question, tests, and artifact when you want them.
Finish each move before carrying the artifact forward.
Monitoring compares a trusted reference distribution with current traffic and separates drift from malformed input.
Deliverable: State the invariant in one sentence.Build a small histogram monitor that reports per-bin proportions and the maximum absolute gap.
Deliverable: Type the smallest deterministic implementation.Reference and current proportions sum to one, empty traffic is explicit, and a known shift crosses the alert threshold.
Deliverable: Record a normal case and a boundary case.Inject a missing value, a new category, and a gradual shift; distinguish quality failure from drift.
Deliverable: Name the failure signal and the guard you would add.A monitoring snapshot with alert thresholds, evidence, and a runbook link.
Deliverable: Save the contract, evidence, and handoff note.Reference and current proportions sum to one, empty traffic is explicit, and a known shift crosses the alert threshold.
Inject a missing value, a new category, and a gradual shift; distinguish quality failure from drift.
A monitoring snapshot with alert thresholds, evidence, and a runbook link.
Deriva-authored extension beyond the supplied PDF. The five moves are Deriva’s implementation contract for this project.