update eval input/output/expected to conform to other SDKs

lib/braintrust/eval/runner.rb

@@ -91,8 +91,8 @@ def run_eval_case(kase, errors)
           # Set attributes known before task execution
           eval_span.set_attribute("braintrust.parent", eval_context.parent_span_attr) if eval_context.parent_span_attr
           set_json_attr(eval_span, "braintrust.span_attributes", build_span_attributes("eval"))
-          set_json_attr(eval_span, "braintrust.input_json", {input: kase.input})
-          set_json_attr(eval_span, "braintrust.expected", kase.expected) if kase.expected
+          set_json_attr(eval_span, "braintrust.input_json", kase.input)
+          set_json_attr(eval_span, "braintrust.expected_json", kase.expected) unless kase.expected.nil?
           set_json_attr(eval_span, "braintrust.metadata", kase.metadata) if kase.metadata
           eval_span.set_attribute("braintrust.tags", kase.tags) if kase.tags
           eval_span.set_attribute("braintrust.origin", kase.origin) if kase.origin
@@ -103,7 +103,6 @@ def run_eval_case(kase, errors)
           rescue => e
             # Error already recorded on task span, set eval span status
             eval_span.status = OpenTelemetry::Trace::Status.error(e.message)
-            set_json_attr(eval_span, "braintrust.output_json", {output: nil})
             errors << "Task failed for input '#{kase.input}': #{e.message}"
             report_progress(eval_span, kase, error: e.message)
             next
@@ -134,7 +133,7 @@ def run_eval_case(kase, errors)
           end
 
           # Set output after task completes
-          set_json_attr(eval_span, "braintrust.output_json", {output: kase.output})
+          set_json_attr(eval_span, "braintrust.output_json", kase.output)
 
           report_progress(eval_span, kase, data: kase.output)
         end
@@ -151,6 +150,7 @@ def run_task(kase)
           task_span.set_attribute("braintrust.parent", eval_context.parent_span_attr) if eval_context.parent_span_attr
           set_json_attr(task_span, "braintrust.span_attributes", build_span_attributes("task"))
           set_json_attr(task_span, "braintrust.input_json", kase.input)
+          set_json_attr(task_span, "braintrust.expected_json", kase.expected) unless kase.expected.nil?
 
           begin
             output = eval_context.task.call(

test/braintrust/eval/runner_test.rb

@@ -531,7 +531,7 @@ def test_runner_eval_span_has_case_metadata
     assert_equal "greeting", metadata["category"]
   end
 
-  def test_runner_eval_span_input_json_wrapped
+  def test_runner_eval_span_input_json_is_raw_value
     rig = setup_otel_test_rig
 
     context = Braintrust::Eval::Context.build(
@@ -550,7 +550,7 @@ def test_runner_eval_span_input_json_wrapped
     eval_span = rig.exporter.finished_spans.find { |s| s.name == "eval" }
     input_json = JSON.parse(eval_span.attributes["braintrust.input_json"])
 
-    assert_equal({"input" => "hello"}, input_json)
+    assert_equal "hello", input_json
   end
 
   def test_runner_eval_span_tags_as_array
@@ -576,7 +576,7 @@ def test_runner_eval_span_tags_as_array
     assert_equal ["fast", "regression"], tags
   end
 
-  def test_runner_eval_span_output_json_null_on_task_error
+  def test_runner_eval_span_omits_output_json_on_task_error
     rig = setup_otel_test_rig
 
     context = Braintrust::Eval::Context.build(
@@ -593,12 +593,11 @@ def test_runner_eval_span_output_json_null_on_task_error
     Braintrust::Eval::Runner.new(context).run
 
     eval_span = rig.exporter.finished_spans.find { |s| s.name == "eval" }
-    output_json = JSON.parse(eval_span.attributes["braintrust.output_json"])
 
-    assert_equal({"output" => nil}, output_json)
+    assert_nil eval_span.attributes["braintrust.output_json"]
   end
 
-  def test_runner_eval_span_output_json_wrapped
+  def test_runner_eval_span_output_json_is_raw_value
     rig = setup_otel_test_rig
 
     context = Braintrust::Eval::Context.build(
@@ -617,7 +616,7 @@ def test_runner_eval_span_output_json_wrapped
     eval_span = rig.exporter.finished_spans.find { |s| s.name == "eval" }
     output_json = JSON.parse(eval_span.attributes["braintrust.output_json"])
 
-    assert_equal({"output" => "HELLO"}, output_json)
+    assert_equal "HELLO", output_json
   end
 
   def test_runner_eval_spans_are_independent_roots
@@ -695,6 +694,7 @@ def test_runner_run_creates_task_spans
 
     assert_equal 1, task_spans.length
     assert_equal "experiment_id:exp-123", task_spans[0].attributes["braintrust.parent"]
+    assert_equal "HELLO", JSON.parse(task_spans[0].attributes["braintrust.expected_json"])
   end
 
   def test_runner_run_creates_per_scorer_spans

test/braintrust/eval_test.rb

@@ -358,13 +358,14 @@ def test_eval_run_with_tracing
       assert eval_span.attributes["braintrust.parent"]
       assert_match(/experiment_id:[0-9a-f-]{36}/, eval_span.attributes["braintrust.parent"])
       input_json = JSON.parse(eval_span.attributes["braintrust.input_json"])
-      assert_equal({"input" => "hello"}, input_json)
+      assert_equal "hello", input_json
       output_json = JSON.parse(eval_span.attributes["braintrust.output_json"])
-      assert_equal({"output" => "HELLO"}, output_json)
+      assert_equal "HELLO", output_json
 
       # Verify task span
       assert task_span.attributes["braintrust.span_attributes"]
       assert_includes task_span.attributes["braintrust.span_attributes"], "task"
+      assert_equal "HELLO", JSON.parse(task_span.attributes["braintrust.expected_json"])
 
       # Verify score span has scorer-specific attributes
       assert score_span.attributes["braintrust.scores"]