decision_agent 0.1.2 → 0.1.4

data/lib/decision_agent/ab_testing/ab_test_manager.rb

@@ -0,0 +1,317 @@
+require "monitor"
+
+module DecisionAgent
+  module ABTesting
+    # Manages A/B tests and provides high-level orchestration
+    class ABTestManager
+      include MonitorMixin
+
+      attr_reader :storage_adapter, :version_manager
+
+      # @param storage_adapter [Storage::Adapter] Storage adapter for persistence
+      # @param version_manager [Versioning::VersionManager] Version manager for rule versions
+      def initialize(storage_adapter: nil, version_manager: nil)
+        super()
+        @storage_adapter = storage_adapter || default_storage_adapter
+        @version_manager = version_manager || Versioning::VersionManager.new
+        @active_tests_cache = {}
+      end
+
+      # Create a new A/B test
+      # @param name [String] Name of the test
+      # @param champion_version_id [String, Integer] ID of the champion version
+      # @param challenger_version_id [String, Integer] ID of the challenger version
+      # @param traffic_split [Hash] Traffic distribution
+      # @param start_date [Time, nil] When to start the test
+      # @param end_date [Time, nil] When to end the test
+      # @return [ABTest] The created test
+      def create_test(name:, champion_version_id:, challenger_version_id:, traffic_split: { champion: 90, challenger: 10 }, start_date: nil,
+                      end_date: nil)
+        synchronize do
+          # Validate that both versions exist
+          validate_version_exists!(champion_version_id, "champion")
+          validate_version_exists!(challenger_version_id, "challenger")
+
+          test = ABTest.new(
+            name: name,
+            champion_version_id: champion_version_id,
+            challenger_version_id: challenger_version_id,
+            traffic_split: traffic_split,
+            start_date: start_date || Time.now.utc,
+            end_date: end_date,
+            status: start_date && start_date > Time.now.utc ? "scheduled" : "running"
+          )
+
+          saved_test = @storage_adapter.save_test(test)
+          invalidate_cache!
+          saved_test
+        end
+      end
+
+      # Get an A/B test by ID
+      # @param test_id [String, Integer] The test ID
+      # @return [ABTest, nil] The test or nil if not found
+      def get_test(test_id)
+        @storage_adapter.get_test(test_id)
+      end
+
+      # Get all active A/B tests
+      # @return [Array<ABTest>] Array of active tests
+      def active_tests
+        synchronize do
+          return @active_tests_cache[:tests] if cache_valid?
+
+          tests = @storage_adapter.list_tests(status: "running")
+          @active_tests_cache = { tests: tests, timestamp: Time.now.utc }
+          tests
+        end
+      end
+
+      # Assign a variant for a request
+      # @param test_id [String, Integer] The A/B test ID
+      # @param user_id [String, nil] Optional user identifier for consistent assignment
+      # @return [Hash] Assignment details { test_id:, variant:, version_id: }
+      def assign_variant(test_id:, user_id: nil)
+        test = get_test(test_id)
+        raise TestNotFoundError, "Test not found: #{test_id}" unless test
+
+        variant = test.assign_variant(user_id: user_id)
+        version_id = test.version_for_variant(variant)
+
+        assignment = ABTestAssignment.new(
+          ab_test_id: test_id,
+          user_id: user_id,
+          variant: variant,
+          version_id: version_id
+        )
+
+        saved_assignment = @storage_adapter.save_assignment(assignment)
+
+        {
+          test_id: test_id,
+          variant: variant,
+          version_id: version_id,
+          assignment_id: saved_assignment.id
+        }
+      end
+
+      # Record the decision result for an assignment
+      # @param assignment_id [String, Integer] The assignment ID
+      # @param decision [String] The decision result
+      # @param confidence [Float] The confidence score
+      def record_decision(assignment_id:, decision:, confidence:)
+        @storage_adapter.update_assignment(assignment_id, decision_result: decision, confidence: confidence)
+      end
+
+      # Get results comparison for an A/B test
+      # @param test_id [String, Integer] The test ID
+      # @return [Hash] Comparison statistics
+      def get_results(test_id)
+        test = get_test(test_id)
+        raise TestNotFoundError, "Test not found: #{test_id}" unless test
+
+        assignments = @storage_adapter.get_assignments(test_id)
+
+        champion_assignments = assignments.select { |a| a.variant == :champion }
+        challenger_assignments = assignments.select { |a| a.variant == :challenger }
+
+        {
+          test: test.to_h,
+          champion: calculate_variant_stats(champion_assignments, "Champion"),
+          challenger: calculate_variant_stats(challenger_assignments, "Challenger"),
+          comparison: compare_variants(champion_assignments, challenger_assignments),
+          total_assignments: assignments.size,
+          timestamp: Time.now.utc
+        }
+      end
+
+      # Start a scheduled test
+      # @param test_id [String, Integer] The test ID
+      def start_test(test_id)
+        synchronize do
+          test = get_test(test_id)
+          raise TestNotFoundError, "Test not found: #{test_id}" unless test
+
+          test.start!
+          @storage_adapter.update_test(test_id, status: "running", start_date: test.start_date)
+          invalidate_cache!
+        end
+      end
+
+      # Complete a running test
+      # @param test_id [String, Integer] The test ID
+      def complete_test(test_id)
+        synchronize do
+          test = get_test(test_id)
+          raise TestNotFoundError, "Test not found: #{test_id}" unless test
+
+          test.complete!
+          @storage_adapter.update_test(test_id, status: "completed", end_date: test.end_date)
+          invalidate_cache!
+        end
+      end
+
+      # Cancel a test
+      # @param test_id [String, Integer] The test ID
+      def cancel_test(test_id)
+        synchronize do
+          test = get_test(test_id)
+          raise TestNotFoundError, "Test not found: #{test_id}" unless test
+
+          test.cancel!
+          @storage_adapter.update_test(test_id, status: "cancelled")
+          invalidate_cache!
+        end
+      end
+
+      # List all tests with optional filtering
+      # @param status [String, nil] Filter by status
+      # @param limit [Integer, nil] Limit results
+      # @return [Array<ABTest>] Array of tests
+      def list_tests(status: nil, limit: nil)
+        @storage_adapter.list_tests(status: status, limit: limit)
+      end
+
+      private
+
+      def default_storage_adapter
+        # Use in-memory adapter by default
+        require_relative "storage/memory_adapter"
+        Storage::MemoryAdapter.new
+      end
+
+      def validate_version_exists!(version_id, label)
+        version = @version_manager.get_version(version_id: version_id)
+        return if version
+
+        raise VersionNotFoundError, "#{label.capitalize} version not found: #{version_id}"
+      end
+
+      def cache_valid?
+        return false unless @active_tests_cache[:timestamp]
+
+        # Cache is valid for 60 seconds
+        Time.now.utc - @active_tests_cache[:timestamp] < 60
+      end
+
+      def invalidate_cache!
+        @active_tests_cache = {}
+      end
+
+      def calculate_variant_stats(assignments, label)
+        with_decisions = assignments.select(&:decision_result)
+
+        if with_decisions.empty?
+          return {
+            label: label,
+            total_assignments: assignments.size,
+            decisions_recorded: 0,
+            avg_confidence: nil,
+            decision_distribution: {}
+          }
+        end
+
+        confidences = with_decisions.map(&:confidence)
+        decision_counts = with_decisions.group_by(&:decision_result).transform_values(&:size)
+
+        {
+          label: label,
+          total_assignments: assignments.size,
+          decisions_recorded: with_decisions.size,
+          avg_confidence: (confidences.sum / confidences.size.to_f).round(4),
+          min_confidence: confidences.min&.round(4),
+          max_confidence: confidences.max&.round(4),
+          decision_distribution: decision_counts
+        }
+      end
+
+      def compare_variants(champion_assignments, challenger_assignments)
+        champion_with_decisions = champion_assignments.select(&:decision_result)
+        challenger_with_decisions = challenger_assignments.select(&:decision_result)
+
+        return { statistical_significance: "insufficient_data" } if champion_with_decisions.empty? || challenger_with_decisions.empty?
+
+        champion_confidences = champion_with_decisions.map(&:confidence)
+        challenger_confidences = challenger_with_decisions.map(&:confidence)
+
+        champion_avg = champion_confidences.sum / champion_confidences.size.to_f
+        challenger_avg = challenger_confidences.sum / challenger_confidences.size.to_f
+
+        improvement = ((challenger_avg - champion_avg) / champion_avg * 100).round(2)
+
+        # Calculate statistical significance using Welch's t-test approximation
+        sig_result = calculate_statistical_significance(champion_confidences, challenger_confidences)
+
+        {
+          champion_avg_confidence: champion_avg.round(4),
+          challenger_avg_confidence: challenger_avg.round(4),
+          improvement_percentage: improvement,
+          winner: determine_winner(champion_avg, challenger_avg, sig_result[:significant]),
+          statistical_significance: sig_result[:significant] ? "significant" : "not_significant",
+          confidence_level: sig_result[:confidence_level],
+          recommendation: generate_recommendation(improvement, sig_result[:significant])
+        }
+      end
+
+      def calculate_statistical_significance(sample1, sample2)
+        n1 = sample1.size
+        n2 = sample2.size
+
+        return { significant: false, confidence_level: 0 } if n1 < 30 || n2 < 30
+
+        mean1 = sample1.sum / n1.to_f
+        mean2 = sample2.sum / n2.to_f
+
+        var1 = sample1.map { |x| (x - mean1)**2 }.sum / (n1 - 1).to_f
+        var2 = sample2.map { |x| (x - mean2)**2 }.sum / (n2 - 1).to_f
+
+        # Welch's t-statistic
+        t_stat = (mean1 - mean2) / Math.sqrt((var1 / n1) + (var2 / n2))
+
+        # Simplified p-value approximation (for demonstration)
+        # In production, use a proper statistical library
+        t_stat_abs = t_stat.abs
+
+        confidence_level = if t_stat_abs > 2.576
+                             0.99 # 99% confidence
+                           elsif t_stat_abs > 1.96
+                             0.95 # 95% confidence
+                           elsif t_stat_abs > 1.645
+                             0.90 # 90% confidence
+                           else
+                             0.0
+                           end
+
+        {
+          significant: confidence_level >= 0.95,
+          confidence_level: confidence_level,
+          t_statistic: t_stat.round(4)
+        }
+      end
+
+      def determine_winner(champion_avg, challenger_avg, significant)
+        return "inconclusive" unless significant
+
+        challenger_avg > champion_avg ? "challenger" : "champion"
+      end
+
+      def generate_recommendation(improvement, significant)
+        if !significant
+          "Continue testing - not enough data for statistical significance"
+        elsif improvement > 5
+          "Strong evidence to promote challenger"
+        elsif improvement.positive?
+          "Moderate evidence to promote challenger"
+        elsif improvement > -5
+          "Results are similar - consider other factors"
+        else
+          "Keep champion - challenger performs worse"
+        end
+      end
+    end
+
+    # Custom errors
+    class TestNotFoundError < StandardError; end
+    class VersionNotFoundError < StandardError; end
+  end
+end

data/lib/decision_agent/ab_testing/ab_testing_agent.rb

@@ -0,0 +1,152 @@
+module DecisionAgent
+  module ABTesting
+    # Agent wrapper that adds A/B testing capabilities to the standard Agent
+    # Automatically handles variant assignment and decision tracking
+    class ABTestingAgent
+      attr_reader :ab_test_manager, :version_manager
+
+      # @param ab_test_manager [ABTestManager] The A/B test manager
+      # @param version_manager [Versioning::VersionManager] Version manager for rules
+      # @param evaluators [Array] Base evaluators (can be overridden by versioned rules)
+      # @param scoring_strategy [Scoring::Base] Scoring strategy
+      # @param audit_adapter [Audit::Adapter] Audit adapter
+      def initialize(
+        ab_test_manager:,
+        version_manager: nil,
+        evaluators: [],
+        scoring_strategy: nil,
+        audit_adapter: nil
+      )
+        @ab_test_manager = ab_test_manager
+        @version_manager = version_manager || ab_test_manager.version_manager
+        @base_evaluators = evaluators
+        @scoring_strategy = scoring_strategy
+        @audit_adapter = audit_adapter
+      end
+
+      # Make a decision with A/B testing support
+      # @param context [Hash, Context] The decision context
+      # @param feedback [Hash] Optional feedback
+      # @param ab_test_id [String, Integer, nil] Optional A/B test ID
+      # @param user_id [String, nil] Optional user ID for consistent assignment
+      # @return [Hash] Decision result with A/B test metadata
+      def decide(context:, feedback: {}, ab_test_id: nil, user_id: nil)
+        ctx = context.is_a?(Context) ? context : Context.new(context)
+
+        # If A/B test is specified, use variant assignment
+        if ab_test_id
+          decide_with_ab_test(ctx, feedback, ab_test_id, user_id)
+        else
+          # Standard decision without A/B testing
+          agent = build_agent(@base_evaluators)
+          decision = agent.decide(context: ctx, feedback: feedback)
+
+          {
+            decision: decision.decision,
+            confidence: decision.confidence,
+            explanations: decision.explanations,
+            evaluations: decision.evaluations,
+            ab_test: nil
+          }
+        end
+      end
+
+      # Get A/B test results
+      # @param test_id [String, Integer] The test ID
+      # @return [Hash] Test results and statistics
+      def get_test_results(test_id)
+        @ab_test_manager.get_results(test_id)
+      end
+
+      # List active A/B tests
+      # @return [Array<ABTest>] Active tests
+      def active_tests
+        @ab_test_manager.active_tests
+      end
+
+      private
+
+      def decide_with_ab_test(context, feedback, ab_test_id, user_id)
+        # Assign variant
+        assignment = @ab_test_manager.assign_variant(test_id: ab_test_id, user_id: user_id)
+
+        # Get the version for the assigned variant
+        version = @version_manager.get_version(version_id: assignment[:version_id])
+        raise VersionNotFoundError, "Version not found: #{assignment[:version_id]}" unless version
+
+        # Build evaluators from the versioned rule content
+        evaluators = build_evaluators_from_version(version)
+
+        # Create agent with version-specific evaluators
+        agent = build_agent(evaluators)
+
+        # Make decision
+        decision = agent.decide(context: context, feedback: feedback)
+
+        # Record the decision result
+        @ab_test_manager.record_decision(
+          assignment_id: assignment[:assignment_id],
+          decision: decision.decision,
+          confidence: decision.confidence
+        )
+
+        # Return decision with A/B test metadata
+        {
+          decision: decision.decision,
+          confidence: decision.confidence,
+          explanations: decision.explanations,
+          evaluations: decision.evaluations,
+          ab_test: {
+            test_id: ab_test_id,
+            variant: assignment[:variant],
+            version_id: assignment[:version_id],
+            assignment_id: assignment[:assignment_id]
+          }
+        }
+      end
+
+      def build_agent(evaluators)
+        Agent.new(
+          evaluators: evaluators.empty? ? @base_evaluators : evaluators,
+          scoring_strategy: @scoring_strategy,
+          audit_adapter: @audit_adapter
+        )
+      end
+
+      def build_evaluators_from_version(version)
+        content = version[:content]
+
+        # If the version content contains evaluator configurations, build them
+        # Otherwise, use base evaluators
+        if content.is_a?(Hash) && content[:evaluators]
+          content[:evaluators].map do |eval_config|
+            build_evaluator_from_config(eval_config)
+          end
+        elsif content.is_a?(Hash) && content[:rules]
+          # Build a JsonRuleEvaluator from the full content (ruleset + rules)
+          [Evaluators::JsonRuleEvaluator.new(rules_json: content)]
+        else
+          # Fallback to base evaluators
+          @base_evaluators
+        end
+      end
+
+      def build_evaluator_from_config(config)
+        case config[:type]
+        when "json_rule"
+          Evaluators::JsonRuleEvaluator.new(
+            rules_json: config[:rules]
+          )
+        when "static"
+          Evaluators::StaticEvaluator.new(
+            decision: config[:decision],
+            weight: config[:weight] || 1.0,
+            reason: config[:reason]
+          )
+        else
+          raise "Unknown evaluator type: #{config[:type]}"
+        end
+      end
+    end
+  end
+end

data/lib/decision_agent/ab_testing/storage/activerecord_adapter.rb

@@ -0,0 +1,155 @@
+require "active_record"
+
+module DecisionAgent
+  module ABTesting
+    module Storage
+      # ActiveRecord storage adapter for A/B tests
+      # Requires Rails models: ABTestModel, ABTestAssignmentModel
+      class ActiveRecordAdapter < Adapter
+        # Check if ActiveRecord models are available
+        def self.available?
+          defined?(::ABTestModel) && defined?(::ABTestAssignmentModel)
+        end
+
+        def initialize
+          super
+          raise "ActiveRecord models not available. Run the generator to create them." unless self.class.available?
+        end
+
+        def save_test(test)
+          record = ::ABTestModel.create!(
+            name: test.name,
+            champion_version_id: test.champion_version_id,
+            challenger_version_id: test.challenger_version_id,
+            traffic_split: test.traffic_split,
+            start_date: test.start_date,
+            end_date: test.end_date,
+            status: test.status
+          )
+
+          to_ab_test(record)
+        end
+
+        def get_test(test_id)
+          record = ::ABTestModel.find_by(id: test_id)
+          record ? to_ab_test(record) : nil
+        end
+
+        def update_test(test_id, attributes)
+          record = ::ABTestModel.find(test_id)
+          record.update!(attributes)
+          to_ab_test(record)
+        end
+
+        def list_tests(status: nil, limit: nil)
+          query = ::ABTestModel.order(created_at: :desc)
+          query = query.where(status: status) if status
+          query = query.limit(limit) if limit
+
+          query.map { |record| to_ab_test(record) }
+        end
+
+        def save_assignment(assignment)
+          record = ::ABTestAssignmentModel.create!(
+            ab_test_id: assignment.ab_test_id,
+            user_id: assignment.user_id,
+            variant: assignment.variant.to_s,
+            version_id: assignment.version_id,
+            decision_result: assignment.decision_result,
+            confidence: assignment.confidence,
+            context: assignment.context,
+            timestamp: assignment.timestamp
+          )
+
+          to_assignment(record)
+        end
+
+        def update_assignment(assignment_id, attributes)
+          record = ::ABTestAssignmentModel.find(assignment_id)
+          record.update!(attributes)
+          to_assignment(record)
+        end
+
+        def get_assignments(test_id)
+          ::ABTestAssignmentModel
+            .where(ab_test_id: test_id)
+            .order(timestamp: :desc)
+            .map { |record| to_assignment(record) }
+        end
+
+        # rubocop:disable Naming/PredicateMethod
+        def delete_test(test_id)
+          record = ::ABTestModel.find(test_id)
+          ::ABTestAssignmentModel.where(ab_test_id: test_id).delete_all
+          record.destroy
+          true
+        end
+        # rubocop:enable Naming/PredicateMethod
+
+        # Get statistics from database
+        def get_test_statistics(test_id)
+          assignments = ::ABTestAssignmentModel.where(ab_test_id: test_id)
+
+          {
+            total_assignments: assignments.count,
+            champion_count: assignments.where(variant: "champion").count,
+            challenger_count: assignments.where(variant: "challenger").count,
+            with_decisions: assignments.where.not(decision_result: nil).count,
+            avg_confidence: assignments.where.not(confidence: nil).average(:confidence)&.to_f
+          }
+        end
+
+        private
+
+        def to_ab_test(record)
+          ABTest.new(
+            id: record.id,
+            name: record.name,
+            champion_version_id: record.champion_version_id,
+            challenger_version_id: record.challenger_version_id,
+            traffic_split: parse_traffic_split(record.traffic_split),
+            start_date: record.start_date,
+            end_date: record.end_date,
+            status: record.status
+          )
+        end
+
+        def to_assignment(record)
+          ABTestAssignment.new(
+            id: record.id,
+            ab_test_id: record.ab_test_id,
+            user_id: record.user_id,
+            variant: record.variant.to_sym,
+            version_id: record.version_id,
+            timestamp: record.timestamp,
+            decision_result: record.decision_result,
+            confidence: record.confidence,
+            context: parse_context(record.context)
+          )
+        end
+
+        def parse_traffic_split(value)
+          case value
+          when Hash
+            value.symbolize_keys
+          when String
+            JSON.parse(value).symbolize_keys
+          else
+            value
+          end
+        end
+
+        def parse_context(value)
+          case value
+          when Hash
+            value
+          when String
+            JSON.parse(value)
+          else
+            {}
+          end
+        end
+      end
+    end
+  end
+end

data/lib/decision_agent/ab_testing/storage/adapter.rb

@@ -0,0 +1,67 @@
+module DecisionAgent
+  module ABTesting
+    module Storage
+      # Base adapter interface for A/B test persistence
+      class Adapter
+        # Save an A/B test
+        # @param test [ABTest] The test to save
+        # @return [ABTest] The saved test with ID
+        def save_test(test)
+          raise NotImplementedError, "#{self.class} must implement #save_test"
+        end
+
+        # Get an A/B test by ID
+        # @param test_id [String, Integer] The test ID
+        # @return [ABTest, nil] The test or nil
+        def get_test(test_id)
+          raise NotImplementedError, "#{self.class} must implement #get_test"
+        end
+
+        # Update an A/B test
+        # @param test_id [String, Integer] The test ID
+        # @param attributes [Hash] Attributes to update
+        # @return [ABTest] The updated test
+        def update_test(test_id, attributes)
+          raise NotImplementedError, "#{self.class} must implement #update_test"
+        end
+
+        # List A/B tests
+        # @param status [String, nil] Filter by status
+        # @param limit [Integer, nil] Limit results
+        # @return [Array<ABTest>] Array of tests
+        def list_tests(status: nil, limit: nil)
+          raise NotImplementedError, "#{self.class} must implement #list_tests"
+        end
+
+        # Save an assignment
+        # @param assignment [ABTestAssignment] The assignment to save
+        # @return [ABTestAssignment] The saved assignment with ID
+        def save_assignment(assignment)
+          raise NotImplementedError, "#{self.class} must implement #save_assignment"
+        end
+
+        # Update an assignment
+        # @param assignment_id [String, Integer] The assignment ID
+        # @param attributes [Hash] Attributes to update
+        # @return [ABTestAssignment] The updated assignment
+        def update_assignment(assignment_id, attributes)
+          raise NotImplementedError, "#{self.class} must implement #update_assignment"
+        end
+
+        # Get assignments for a test
+        # @param test_id [String, Integer] The test ID
+        # @return [Array<ABTestAssignment>] Array of assignments
+        def get_assignments(test_id)
+          raise NotImplementedError, "#{self.class} must implement #get_assignments"
+        end
+
+        # Delete a test and its assignments
+        # @param test_id [String, Integer] The test ID
+        # @return [Boolean] True if deleted
+        def delete_test(test_id)
+          raise NotImplementedError, "#{self.class} must implement #delete_test"
+        end
+      end
+    end
+  end
+end