Package dev.dokimos.core.gate

Class GateConfig

java.lang.Object
dev.dokimos.core.gate.GateConfig
public final class GateConfig extends Object
Configuration for the server-free regression gate.

The defaults suit an LLM-judge gate: it flags only statistically significant aggregate and per-evaluator drops, so a noisy judge does not flake the build, while severityMargin still fails the gate on a single item that breaks hard. Pair the gate with a temperature-0 judge for stable scores.

Accessors follow the framework convention (no get prefix). Construct via defaults() or builder().

  • Method Details

    • defaults

      public static GateConfig defaults()
      Returns a configuration with all defaults.
      Returns:
      the default gate configuration

    • builder

      public static GateConfig.Builder builder()
      Creates a new builder.
      Returns:
      a new builder pre-populated with defaults

    • updateBaselineRequested

      public static boolean updateBaselineRequested()
      Whether an explicit baseline update was requested via DOKIMOS_UPDATE_BASELINE (env, the primary control because -D does not reach the test JVM under default Gradle or the IntelliJ runner) or the dokimos.updateBaseline system property.
      Returns:
      true if a baseline overwrite was requested out of band

    • alpha

      public double alpha()
      Returns:
      the significance level for the statistical tests

    • seed

      public long seed()
      Returns:
      the RNG seed for the permutation and bootstrap tests (pinned for reproducibility)

    • permutationIterations

      public int permutationIterations()
      Returns:
      the permutation-test iteration count

    • bootstrapIterations

      public int bootstrapIterations()
      Returns:
      the bootstrap iteration count

    • severityMargin

      public double severityMargin()
      Returns:
      the per-item worst-evaluator score-drop threshold that fails the gate (guard 2)

    • pairing

      public GateConfig.Pairing pairing()
      Returns:
      the item pairing strategy

    • failOnRegression

      public boolean failOnRegression()
      Returns:
      whether a detected regression fails the gate

    • failOnRemovedItems

      public boolean failOnRemovedItems()
      Returns:
      whether a removed item (present in baseline, absent in candidate) fails the gate

    • onRemovedEvaluator

      public GateConfig.RemovedEvaluatorPolicy onRemovedEvaluator()
      Returns:
      the policy for an evaluator present in the baseline but missing from the candidate

    • bootstrapPasses

      public boolean bootstrapPasses()
      Whether a missing-baseline bootstrap writes the file and passes (the default), versus the strict approval-test stance that writes it but fails once so the run is red until the new baseline is reviewed and committed. The CI no-baseline branch is unaffected either way: it never writes (the checkout is ephemeral) and always passes with a warning.
      Returns:
      true if a local missing-baseline bootstrap passes after writing

    • updateBaseline

      public boolean updateBaseline()
      Returns:
      whether the baseline should be overwritten from the candidate and the gate pass