braintrust 0.4.1 → 0.4.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 1047e872fe3a0acf40117bd9a0fcbf4e5024edbe3c1eb3d3bd1e32f60e5ffb18
-  data.tar.gz: e4c412abf3cf748ff157248feb821724ba4ffb76b67aecd8d1735179572afa93
+  metadata.gz: 1afc7372056311aec7ccaaae11c5d7a9204cdfc826a34db1bd67126769d253e3
+  data.tar.gz: 6492ea96b8f616c05241983baaeb935cdc3d206ced063781c44981b6d0817e64
 SHA512:
-  metadata.gz: 41f183ac2bb20f64bca5c576f0decabdea9b7a4bcb46cbcf0c66292eb34f7e75a3b91516cd5520b4a3e207669eff0fa212afd1f75c1d6b172a44ea27bbc91e4b
-  data.tar.gz: 8a98b9ea710b69ac0b4e22cbdbed2b55bfab0d25db5fda2afc872931ce27a4eb9183433db03be6cedcb295dfca2f1fa43583620d848c55e9fa8b19629ecbfb8e
+  metadata.gz: e6471fc6f3b8beba508dfef0deebfa5f10b1044aaf44817cfb341a71396655ec034b9eca70aaf2e4efb0afa9007a5c284564b3d3471ba1dd18edfd336ea51cb8
+  data.tar.gz: 333ad63d354f902c5e4c5d25d2d67e5ed4f4843fcbfeb7c33a6f273de9dcd7b1c77202417816dd9b834881a3d3a833c68418e81c9a1365ec2e503da3ada027ac

data/lib/braintrust/dataset.rb

@@ -166,14 +166,14 @@ module Braintrust
       record
     end
 
-    # Build origin JSON for tracing/linking
+    # Build origin pointer for tracing/linking
     # @param raw [Hash] Raw record from API
     # @param dataset_id [String] Dataset ID (fallback if not in record)
-    # @return [String, nil] JSON-serialized origin, or nil if record lacks required fields
+    # @return [Hash, nil] Origin pointer, or nil if record lacks required fields
     def build_origin(raw, dataset_id)
       return nil unless raw["id"] && raw["_xact_id"]
 
-      Internal::Origin.to_json(
+      Internal::Origin.build(
         object_type: "dataset",
         object_id: raw["dataset_id"] || dataset_id,
         id: raw["id"],

data/lib/braintrust/eval/runner.rb

@@ -95,7 +95,7 @@ module Braintrust
           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
+          set_json_attr(eval_span, "braintrust.origin", kase.origin) if kase.origin
 
           # Run task
           begin
@@ -275,9 +275,7 @@ module Braintrust
       def report_progress(eval_span, kase, **fields)
         return unless eval_context.on_progress
         progress = {"id" => eval_span.context.hex_span_id}.merge(fields.transform_keys(&:to_s))
-        if kase.origin
-          progress["origin"] = kase.origin.is_a?(String) ? JSON.parse(kase.origin) : kase.origin
-        end
+        progress["origin"] = kase.origin if kase.origin
         eval_context.on_progress.call(progress)
       rescue => e
         Braintrust.logger.warn("on_progress callback error: #{e.message}")

data/lib/braintrust/eval.rb

@@ -336,7 +336,7 @@ module Braintrust
         # Use pinned version if available, otherwise compute from max(_xact_id)
         version = dataset_obj.version
         version ||= cases
-          .filter_map { |c| c[:origin] && JSON.parse(c[:origin])["_xact_id"] }
+          .filter_map { |c| c[:origin]&.dig("_xact_id") }
           .max
 
         {cases: cases, dataset_id: dataset_obj.id, dataset_version: version}

data/lib/braintrust/internal/origin.rb

@@ -1,27 +1,27 @@
 # frozen_string_literal: true
 
-require "json"
-
 module Braintrust
   module Internal
-    # Origin provides serialization for source object pointers in Braintrust.
-    # Used internally to link spans back to their source records (e.g., dataset rows).
+    # Origin builds source object pointers, which link spans back to the record
+    # they came from (e.g., a dataset row). Pointers stay Hashes internally and
+    # are serialized once at the boundary that needs them, so that inbound
+    # pointers from the wire and ones we build here have the same shape.
     module Origin
-      # Serialize an origin pointer to JSON
+      # Build an origin pointer
       # @param object_type [String] Type of source object (e.g., "dataset", "playground_logs")
       # @param object_id [String] ID of the source object
       # @param id [String] ID of the specific record within the source
       # @param xact_id [String] Transaction ID
       # @param created [String, nil] Creation timestamp
-      # @return [String] JSON-serialized origin
-      def self.to_json(object_type:, object_id:, id:, xact_id:, created:)
-        JSON.dump({
-          object_type: object_type,
-          object_id: object_id,
-          id: id,
-          _xact_id: xact_id,
-          created: created
-        })
+      # @return [Hash] Origin pointer with string keys, matching the wire format
+      def self.build(object_type:, object_id:, id:, xact_id:, created:)
+        {
+          "object_type" => object_type,
+          "object_id" => object_id,
+          "id" => id,
+          "_xact_id" => xact_id,
+          "created" => created
+        }
       end
     end
   end

data/lib/braintrust/server/services/eval_service.rb

@@ -2,12 +2,19 @@
 
 require "json"
 
+require_relative "../../eval/case"
+
 module Braintrust
   module Server
     module Services
       # Framework-agnostic service for running evaluations and streaming SSE results.
       # Must be long-lived (not per-request) to preserve the @state_cache across requests.
       class Eval
+        # Row fields this SDK understands, derived from the Case contract so that
+        # adding a field there carries it through here. Anything else on an
+        # inbound row is a field this version has no code for, and is ignored.
+        CASE_FIELDS = Braintrust::Eval::Case.members.map(&:to_s).freeze
+
         def initialize(evaluators)
           @evaluators = evaluators
           @state_mutex = Mutex.new
@@ -177,9 +184,10 @@ module Braintrust
         # Returns [cases, dataset] where exactly one is non-nil.
         def resolve_data_source(data)
           if data.key?("data")
-            cases = data["data"].map do |d|
-              {input: d["input"], expected: d["expected"]}
-            end
+            # Rows arrive inline from the Playground carrying tags, metadata and
+            # an origin pointer back to the row they came from. Carry them all:
+            # the Playground matches streamed results to its grid by origin.
+            cases = data["data"].map { |row| row.slice(*CASE_FIELDS).transform_keys(&:to_sym) }
             [cases, nil]
           elsif data.key?("dataset_id")
             [nil, Braintrust::Dataset::ID.new(id: data["dataset_id"])]

data/lib/braintrust/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Braintrust
-  VERSION = "0.4.1"
+  VERSION = "0.4.2"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: braintrust
 version: !ruby/object:Gem::Version
-  version: 0.4.1
+  version: 0.4.2
 platform: ruby
 authors:
 - Braintrust