Keep source changes visible and bound evaluation calls consistently

Why: Updated or missing sources must remain visible as limitations, and comparison calls need stable prompt grouping.

Evidence: SELF card/server suites and live-tag compilation pass.

Impact: Adds neutral source limitations and refines test-only provider accounting.

Refs: fast-slow-u2-u3 Tasks 6 and 7

internal/cardsvc/task_visibility.go

@@ -74,6 +74,7 @@ func (s *Service) VisibleTaskCard(owner string, c core.ActivityCard) core.Activi
 		task.Inputs = nil
 		task.Findings = nil
 		task.Conclusion = "相关来源当前不可用，结果暂不展示"
+		task.Limitations = []string{"相关来源当前不可用，请核实访问权限与文件状态"}
 		task.Remaining = "请核实材料访问权限与文件是否仍可读取"
 	} else {
 		task.Findings = append([]core.TaskFinding(nil), task.Findings...)
@@ -86,6 +87,7 @@ func (s *Service) VisibleTaskCard(owner string, c core.ActivityCard) core.Activi
 			}
 		}
 		task.Conclusion = "依据文件已更新，当前结论需要重新核验"
+		task.Limitations = []string{"依据文件已更新，历史发现不能直接作为当前结论"}
 		task.Remaining = "展开历史记录可查看处理时的合法引用；当前文件与当时版本可能不同"
 	}
 	task.OutcomeCompleteness = "partial"

internal/server/work_checkpoint_eval_test.go

@@ -232,7 +232,19 @@ func evalProviderStream(route map[string]any, dir string) agentllm.StreamFunc {
 	unknown := false
 	records := []map[string]any{}
 	return func(ctx context.Context, m string, msgs []agentllm.Message, decls []agentllm.ToolDecl, opts agentllm.StreamOptions) <-chan agentllm.Event {
-		runKey := fmt.Sprintf("%p", ctx)
+		fingerprint := []agentllm.Message{}
+		for _, msg := range msgs {
+			if msg.Role == agentllm.RoleSystem && len(fingerprint) == 0 {
+				fingerprint = append(fingerprint, msg)
+			}
+			if msg.Role == agentllm.RoleUser {
+				fingerprint = append(fingerprint, msg)
+				break
+			}
+		}
+		fp, _ := json.Marshal(fingerprint)
+		hash := sha256.Sum256(fp)
+		runKey := hex.EncodeToString(hash[:])
 		mu.Lock()
 		a := groups[runKey]
 		if a.Started.IsZero() {