From bbbfdbbbbc6cd2c5c7e070916fcc5c9f97c735a0 Mon Sep 17 00:00:00 2001
From: bft-codebot <bft-codebot@users.noreply.github.com>
Date: Wed, 25 Feb 2026 00:29:24 +0000
Subject: [PATCH] sync(bfmono): fix(gambit-verify): align verify turn labels
 and stabilize initial run filtering (+19 more) (bfmono@84952a652)

This PR is an automated gambitmono sync of bfmono Gambit packages.

- Source: `packages/gambit/`
- Core: `packages/gambit/packages/gambit-core/`
- bfmono rev: 84952a652

Changes:
- 84952a652 fix(gambit-verify): align verify turn labels and stabilize initial run filtering
- c56b7f52f feat(gambit): improve verify report controls and harden concurrent calibrate persistence
- beb9435c0 feat(gambit-simulator-ui): extend listbox trigger and popover options
- 25f9fdcfc fix(gambit-simulator-ui): align verify outlier chip semantics and display
- a010b0ee1 feat(gambit-simulator-ui): add verify outliers to workbench chat chips
- 383f2500a refactor(simulator-ui): replace nested ternaries in main routing
- 13c4c8c22 fix(gambit): preserve shared references in safe session serialization
- ae392aa24 feat(gambit-simulator-ui): add grader error chips to workbench chat
- 1de6b335c fix(gambit): clamp deck-level maxTurns bounds in test run selection
- 01d7abbb9 fix(gambit): default verify tab bootstrap flag to enabled
- f3d186c7b fix(gambit): include extension schemas in exports and default serve to restored workspace
- a83b7cbe7 fix(gambit): move unbounded build timeout to deck opt-in
- acb2de627 fix(gambit): avoid strict json_schema 400s in openrouter responses
- 8aba573b6 fix(gambit-simulator-ui): treat errored calibrate runs as failed
- ca2028cf8 fix(gambit): prevent circular trace crashes in workspace test run API
- 7e41517e5 fix(gambit): make build assistant run timeout unbounded
- 24341143d feat(gambit): add deterministic verify fixture seeding
- ff2c2d33d feat(gambit): add verify tab consistency UI
- 91f0c93bb feat(gambit): add feature-flagged verify routing
- 1392f8b65 feat(gambit): support deck-level maxTurns override

Do not edit this repo directly; make changes in bfmono and re-run the sync.
---
 simulator-ui/src/VerifyPage.tsx    | 36 +++++++++++++++++++++++++++---
 simulator-ui/src/verify_metrics.ts |  2 +-
 2 files changed, 34 insertions(+), 4 deletions(-)

diff --git a/simulator-ui/src/VerifyPage.tsx b/simulator-ui/src/VerifyPage.tsx
index b036eb3b..c7f249bc 100644
--- a/simulator-ui/src/VerifyPage.tsx
+++ b/simulator-ui/src/VerifyPage.tsx
@@ -374,12 +374,42 @@ function VerifyPage(
   }, [selectedSession?.gradingRuns]);
 
   const filteredRuns = useMemo(() => {
+    const latestScenarioRunIdFromRuns = sessionRuns
+      .map((run) => scenarioRunIdFromCalibrationRun(run))
+      .find((runId): runId is string => Boolean(runId));
+    const hasOption = (runId: string | null | undefined): runId is string =>
+      Boolean(
+        runId &&
+          scenarioRunOptions.some((entry) => entry.scenarioRunId === runId),
+      );
+    const meta = sessionDetail?.meta && typeof sessionDetail.meta === "object"
+      ? sessionDetail.meta as Record<string, unknown>
+      : {};
+    const currentScenarioRunId = typeof meta.scenarioRunId === "string" &&
+        meta.scenarioRunId.trim().length > 0
+      ? meta.scenarioRunId
+      : null;
+    const activeScenarioRunFilterId = hasOption(workspaceRouting.testRunId)
+      ? workspaceRouting.testRunId
+      : hasOption(selectedScenarioRunId)
+      ? selectedScenarioRunId
+      : hasOption(currentScenarioRunId)
+      ? currentScenarioRunId
+      : scenarioRunOptions[0]?.scenarioRunId ?? latestScenarioRunIdFromRuns ??
+        null;
     return sessionRuns.filter((run) => {
       if (selectedGraderId && run.graderId !== selectedGraderId) return false;
-      if (!selectedScenarioRunId) return true;
-      return scenarioRunIdFromCalibrationRun(run) === selectedScenarioRunId;
+      if (!activeScenarioRunFilterId) return true;
+      return scenarioRunIdFromCalibrationRun(run) === activeScenarioRunFilterId;
     });
-  }, [selectedGraderId, selectedScenarioRunId, sessionRuns]);
+  }, [
+    scenarioRunOptions,
+    selectedGraderId,
+    selectedScenarioRunId,
+    sessionDetail?.meta,
+    sessionRuns,
+    workspaceRouting.testRunId,
+  ]);
 
   const runConsistencySample = useCallback(async (payload: {
     workspaceId: string;
diff --git a/simulator-ui/src/verify_metrics.ts b/simulator-ui/src/verify_metrics.ts
index acfdd441..df98edc6 100644
--- a/simulator-ui/src/verify_metrics.ts
+++ b/simulator-ui/src/verify_metrics.ts
@@ -129,7 +129,7 @@ const flattenRunExamples = (
         ? turnRecord.messageRefId
         : undefined;
       const key = messageRefId ? `ref:${messageRefId}` : `turn:${index}`;
-      const label = `Turn ${index + 1}`;
+      const label = `Assistant turn ${fallbackIndex + 1}`;
       const parsed = extractScoreReasonPass(turnRecord.result);
       buckets.push({
         key,