@ggui-ai/ui-gen 0.13.0 → 0.15.0

package/dist/{types-public-DIpjC-OA.d.ts → types-public-CkNWEov9.d.ts}

@@ -49,9 +49,72 @@ interface RuntimeProbeMeta {
     /** Populated for `infra-skipped` / `not-applicable` — why the probe didn't run. */
     readonly reason?: string;
 }
+/**
+ * Per-criterion execution status — three-valued so "the criterion
+ * silently produced no verdict" is never conflated with "the criterion
+ * ran clean", and neither is conflated with "not evaluated by design".
+ * Same doctrine and vocabulary family as `RuntimeProbeStatus` above.
+ *
+ *   - `ran`            — the LLM call completed and returned a verdict
+ *                        for this criterion; its findings (possibly
+ *                        none) are in `issues`.
+ *   - `skipped`        — the call errored or returned no tool call. The
+ *                        criterion carries ZERO evidence; the evaluator
+ *                        fail-opens it into `pass` to avoid blocking
+ *                        generation on flaky LLM calls, so consumers
+ *                        MUST consult this field before treating `pass`
+ *                        membership as a verdict. Counts AGAINST
+ *                        coverage.
+ *   - `not-applicable` — the evaluator was bypassed by design for this
+ *                        cell (e.g. the same-image low-risk bypass:
+ *                        axis checks clean, tier-1/2 never invoked).
+ *                        `reason` names the bypass. Excluded from
+ *                        coverage denominators — it is neither evidence
+ *                        nor a silent absence.
+ */
+type CriterionRunStatus = "ran" | "skipped" | "not-applicable";
+/** Coverage row for one evaluation criterion of one eval run. */
+interface CriterionCoverage {
+    readonly criterion: string;
+    readonly tier: 1 | 2;
+    readonly status: CriterionRunStatus;
+    /** Populated for `skipped` / `not-applicable` — why no verdict. */
+    readonly reason?: string;
+}
+/**
+ * The static criteria `runLLMEvaluation` runs, in stamp order — the
+ * ONE list both the evaluator's coverage stamp and the harness's
+ * bypass stamp derive from, so the two can never disagree on the set.
+ * Tier 1 = binary pass/fail gates; tier 2 = fail/warn/pass quality.
+ */
+declare const LLM_EVAL_STATIC_CRITERIA: ReadonlyArray<{
+    readonly criterion: string;
+    readonly tier: 1 | 2;
+}>;
+/**
+ * Coverage rows for a cell where the evaluator was BYPASSED BY DESIGN
+ * (never invoked) — one `not-applicable` row per static criterion, all
+ * carrying `reason`. Harness exits that skip tier-1/2 evaluation stamp
+ * this so the cell is neither a silent absence (field missing) nor
+ * evidence. Same shape contract as `runLLMEvaluation`'s stamp: always
+ * every criterion. Lives here (not in llm-evaluator) because the bypass
+ * exit runs before the evaluator module is lazily loaded.
+ */
+declare function notApplicableCoverage(reason: string): CriterionCoverage[];
 interface EvalResult {
     issues: EvalIssue[];
     pass: string[];
+    /**
+     * Per-criterion coverage for this eval run — one row per static
+     * criterion, ALWAYS all of them when present. Stamped by
+     * `runLLMEvaluation` (`ran` / `skipped`) and by the harness on
+     * evaluator-bypass exits (`not-applicable` + reason). The harness
+     * carries it unchanged through every later re-stamp of this result.
+     * Absent = coverage UNKNOWN (paths that predate the instrument, e.g.
+     * `parseEvalResponse` replays) — never assume `ran`, never assume
+     * bypass.
+     */
+    criteriaCoverage?: CriterionCoverage[];
     /**
      * Runtime-render probe execution meta. Stamped by the exit-decision
      * probe runner; absent on eval paths that never invoke the probe.
@@ -149,4 +212,4 @@ declare function getLLMCriteria(): EvalCriterion[];
  */
 declare function buildCodingCriteriaSummary(): string;
 
-export { type AxisCheck as A, CRITERIA as C, DEFAULT_QUALITY_CONFIG as D, type EvalIssue as E, type Priority as P, type QualityConfig as Q, type RuntimeProbeStatus as R, type AxisCheckInput as a, type EvalResult as b, type EvalTier as c, type EvalCategory as d, type EvalCriterion as e, type EvalOutcome as f, type QualityMode as g, buildCodingCriteriaSummary as h, getActionableIssues as i, getCriteriaByPriority as j, getCriterionById as k, getLLMCriteria as l, matches as m, isBlocked as n, priorityForIssue as p };
+export { type AxisCheck as A, CRITERIA as C, DEFAULT_QUALITY_CONFIG as D, type EvalIssue as E, LLM_EVAL_STATIC_CRITERIA as L, type Priority as P, type QualityConfig as Q, type RuntimeProbeStatus as R, type AxisCheckInput as a, type EvalResult as b, type EvalTier as c, type CriterionCoverage as d, type CriterionRunStatus as e, type EvalCategory as f, type EvalCriterion as g, type EvalOutcome as h, type QualityMode as i, buildCodingCriteriaSummary as j, getActionableIssues as k, getCriteriaByPriority as l, matches as m, getCriterionById as n, getLLMCriteria as o, isBlocked as p, notApplicableCoverage as q, priorityForIssue as r };