qualspec 0.1.1 → 0.2.0

data/exe/qualspec

@@ -66,10 +66,10 @@ parser = OptionParser.new do |opts|
     puts opts
     puts
     puts 'Environment variables:'
-    puts '  QUALSPEC_API_URL   API endpoint (default: http://localhost:11434/v1)'
-    puts '  QUALSPEC_API_KEY   API key for authentication'
-    puts '  QUALSPEC_MODEL     Default model for candidates'
-    puts '  QUALSPEC_JUDGE_MODEL  Model to use as judge'
+    puts '  QUALSPEC_API_URL   API endpoint (default: https://openrouter.ai/api/v1)'
+    puts '  QUALSPEC_API_KEY   API key for authentication (falls back to OPEN_ROUTER_API_KEY)'
+    puts '  QUALSPEC_MODEL     Default model for candidates (default: openrouter/auto)'
+    puts '  QUALSPEC_JUDGE_MODEL  Model to use as judge (default: same as QUALSPEC_MODEL)'
     puts
     puts 'Example:'
     puts '  qualspec eval/model_comparison.rb'

data/lib/qualspec/client.rb

@@ -53,8 +53,8 @@ module Qualspec
       return if @config.api_key_configured?
 
       raise Qualspec::Error, <<~MSG.strip
-        QUALSPEC_API_KEY is required but not set.
-        Set it via environment variable or Qualspec.configure { |c| c.api_key = '...' }
+        No API key set. Set QUALSPEC_API_KEY (or OPEN_ROUTER_API_KEY) as an
+        environment variable, or use Qualspec.configure { |c| c.api_key = '...' }
       MSG
     end
 
@@ -70,6 +70,10 @@ module Qualspec
       # Set temperature if provided
       payload[:temperature] = temperature if temperature
 
+      # Ask OpenRouter to include usage accounting (cost + token details).
+      # Only when metadata is requested, so cost-less calls stay lean.
+      payload[:usage] = { include: true } if with_metadata
+
       start_time = Process.clock_gettime(Process::CLOCK_MONOTONIC)
 
       response = @conn.post('chat/completions', payload)
@@ -108,12 +112,15 @@ module Qualspec
     end
 
     def extract_cost(response, data)
-      # OpenRouter includes cost in response or headers
-      header_cost = response.headers['x-openrouter-cost']
-      return header_cost.to_f if header_cost
+      # OpenRouter returns cost under usage.cost when usage accounting is
+      # requested (usage: { include: true }). Fall back to other shapes for
+      # other OpenAI-compatible providers.
+      usage = data['usage'] || {}
+      cost = usage['cost'] || usage['total_cost'] || data['cost']
+      return cost.to_f if cost
 
-      # Check response body (some providers include it)
-      data.dig('usage', 'total_cost') || data['cost']
+      header_cost = response.headers['x-openrouter-cost']
+      header_cost&.to_f
     end
 
     def extract_tokens(data)

data/lib/qualspec/configuration.rb

@@ -2,15 +2,20 @@
 
 module Qualspec
   class Configuration
-    attr_accessor :api_url, :api_key, :default_model, :judge_model, :cache_enabled, :cache_dir, :judge_system_prompt,
+    attr_accessor :api_url, :default_model, :judge_model, :cache_enabled, :cache_dir, :judge_system_prompt,
                   :request_timeout
+    attr_writer :api_key
 
     DEFAULT_API_URL = 'https://openrouter.ai/api/v1'
-    DEFAULT_MODEL = 'google/gemini-3-flash-preview'
+    # Universal fallback. `openrouter/auto` routes to a sensible model for any
+    # request, so qualspec works even with no model configured anywhere.
+    DEFAULT_MODEL = 'openrouter/auto'
 
     def initialize
       @api_url = ENV.fetch('QUALSPEC_API_URL', DEFAULT_API_URL)
-      @api_key = ENV['QUALSPEC_API_KEY']
+      # Default nil: set explicitly via Qualspec.configure { |c| c.api_key = ... }.
+      # When unset, #api_key falls back to env vars (see reader below).
+      @api_key = nil
       @default_model = ENV.fetch('QUALSPEC_MODEL', DEFAULT_MODEL)
       @judge_model = ENV.fetch('QUALSPEC_JUDGE_MODEL') { @default_model }
       @cache_enabled = false
@@ -19,14 +24,22 @@ module Qualspec
       @request_timeout = 120
     end
 
+    # Explicitly configured key wins; otherwise fall back to env vars.
+    # Prefer QUALSPEC_API_KEY, then OPEN_ROUTER_API_KEY (default backend is
+    # OpenRouter). The env vars are a convenience fallback, not a requirement —
+    # pass api_key in Qualspec.configure to avoid relying on them.
+    def api_key
+      @api_key || ENV['QUALSPEC_API_KEY'] || ENV['OPEN_ROUTER_API_KEY']
+    end
+
     def api_headers
       headers = { 'Content-Type' => 'application/json' }
-      headers['Authorization'] = "Bearer #{@api_key}" unless @api_key.to_s.empty?
+      headers['Authorization'] = "Bearer #{api_key}" unless api_key.to_s.empty?
       headers
     end
 
     def api_key_configured?
-      !@api_key.to_s.empty?
+      !api_key.to_s.empty?
     end
   end
 end

data/lib/qualspec/judge.rb

@@ -141,7 +141,7 @@ module Qualspec
       parts << '## Responses:'
 
       responses.each do |candidate, response|
-        parts << "\n### #{candidate}:\n#{response}"
+        parts << "\n### #{candidate}:\n```\n#{response}\n```"
       end
 
       parts << "\nScore each candidate (#{candidate_names}) from 0-10."

data/lib/qualspec/model_registry.rb

@@ -0,0 +1,62 @@
+# frozen_string_literal: true
+
+require 'yaml'
+
+module Qualspec
+  # Loads a curated list of named models from a YAML config file and resolves
+  # names to their full provider slugs. Unknown/blank names fall back to the
+  # configured default (ultimately Configuration::DEFAULT_MODEL, openrouter/auto),
+  # so model lookups always return something usable.
+  #
+  # @example config/models.yml
+  #   default: openrouter/auto
+  #   models:
+  #     glm: z-ai/glm-5.2
+  #
+  # @example
+  #   Qualspec.model(:glm)      # => "z-ai/glm-5.2"
+  #   Qualspec.model(:nope)     # => "openrouter/auto"
+  #   Qualspec.model            # => "openrouter/auto"
+  class ModelRegistry
+    DEFAULT_CONFIG_PATH = 'config/models.yml'
+
+    def initialize(path: nil, default: nil)
+      @models = {}
+      @default = default
+      load_file(path || ENV['QUALSPEC_MODELS_FILE'] || DEFAULT_CONFIG_PATH)
+    end
+
+    # Resolve a model name to its slug, falling back to the default.
+    #
+    # @param name [Symbol, String, nil] the configured name (or nil for default)
+    # @return [String] a model slug
+    def resolve(name = nil)
+      return default if name.nil? || name.to_s.empty?
+
+      @models.fetch(name.to_s, default)
+    end
+
+    # @return [Hash{String=>String}] all configured name => slug pairs
+    def all
+      @models.dup
+    end
+
+    # @return [String] the universal fallback model
+    def default
+      @default || Configuration::DEFAULT_MODEL
+    end
+
+    private
+
+    def load_file(path)
+      return unless path && File.exist?(path)
+
+      data = YAML.safe_load_file(path) || {}
+      @default ||= data['default']
+      (data['models'] || {}).each { |name, slug| @models[name.to_s] = slug }
+    rescue StandardError
+      # A malformed config file should never break a run; defaults still apply.
+      nil
+    end
+  end
+end

data/lib/qualspec/recorder.rb

@@ -13,16 +13,32 @@ module Qualspec
       def setup(cassette_dir: '.qualspec_cassettes')
         require_vcr!
 
+        recorder = self
         VCR.configure do |config|
           config.cassette_library_dir = cassette_dir
           config.hook_into :faraday
           config.default_cassette_options = {
             record: :new_episodes,
-            match_requests_on: %i[method uri body]
+            match_requests_on: %i[method uri body_without_model]
           }
-          # Filter out API keys
-          config.filter_sensitive_data('<API_KEY>') { Qualspec.configuration.api_key }
+          # Filter out API keys — guard against adding duplicate filters
+          unless @api_key_filter_registered
+            config.filter_sensitive_data('<API_KEY>') { Qualspec.configuration.api_key }
+            @api_key_filter_registered = true
+          end
         end
+
+        # Register custom matcher once — ignores the `model` field so cassettes
+        # recorded with one model work in CI where a different model is configured.
+        unless @matcher_registered
+          VCR.configure do |config|
+            config.register_request_matcher(:body_without_model) do |r1, r2|
+              recorder.send(:normalize_body_for_match, r1.body) == recorder.send(:normalize_body_for_match, r2.body)
+            end
+          end
+          @matcher_registered = true
+        end
+
         @configured = true
       end
 
@@ -40,6 +56,20 @@ module Qualspec
         VCR.use_cassette(name, record: :none, &block)
       end
 
+      # Replay a cassette if it already exists (no API key required), otherwise
+      # record a fresh one. Ideal for examples that ship a committed cassette so
+      # they run for free, but still record on first run.
+      def use_cassette(name, &block)
+        setup unless configured?
+        mode = cassette_exists?(name) ? :none : :new_episodes
+        VCR.use_cassette(name, record: mode, &block)
+      end
+
+      def cassette_exists?(name)
+        require_vcr!
+        File.exist?(File.join(VCR.configuration.cassette_library_dir, "#{name}.yml"))
+      end
+
       private
 
       def require_vcr!
@@ -50,6 +80,14 @@ module Qualspec
           Add to your Gemfile: gem 'vcr'
         MSG
       end
+
+      def normalize_body_for_match(body)
+        parsed = JSON.parse(body)
+        parsed.delete('model')
+        JSON.generate(parsed)
+      rescue JSON::ParserError
+        body
+      end
     end
   end
 end

data/lib/qualspec/suite/candidate.rb

@@ -5,14 +5,16 @@ module Qualspec
     class Candidate
       attr_reader :name, :model, :system_prompt, :options
 
-      def initialize(name, model:, system_prompt: nil, **options)
+      def initialize(name, model: nil, system_prompt: nil, **options)
         @name = name.to_s
-        @model = model
+        # Fall back to the configured default model (ultimately openrouter/auto)
+        # so a candidate works even when no model is specified.
+        @model = model || Qualspec.configuration.default_model
         @system_prompt = system_prompt
         @options = options
       end
 
-      def generate_response(prompt:, system_prompt: nil, temperature: nil)
+      def generate_response(prompt:, system_prompt: nil, temperature: nil, with_metadata: false)
         messages = []
 
         sys = system_prompt || @system_prompt
@@ -23,7 +25,8 @@ module Qualspec
           model: @model,
           messages: messages,
           json_mode: false, # We want natural responses, not JSON
-          temperature: normalize_temperature(temperature)
+          temperature: normalize_temperature(temperature),
+          with_metadata: with_metadata
         )
       end
 

data/lib/qualspec/suite/dsl.rb

@@ -11,16 +11,31 @@ module Qualspec
         @scenarios_list = []
         @variants_config = nil
         @temperature_list = [nil] # nil means use model default
+        @track_cost = false
 
         instance_eval(&block) if block_given? # rubocop:disable Style/EvalWithLocation -- DSL pattern requires eval
       end
 
+      # DSL: capture per-call cost + token metadata so cost/value analysis works.
+      # Off by default — evaluations that don't look at cost skip the overhead.
+      #
+      # @example
+      #   track_cost
+      def track_cost(value = true) # rubocop:disable Style/OptionalBooleanParameter -- reads as a DSL toggle
+        @track_cost = value
+      end
+      alias capture_metadata track_cost
+
+      def track_cost?
+        @track_cost
+      end
+
       # DSL: define candidates
       def candidates(&block)
         instance_eval(&block) # rubocop:disable Style/EvalWithLocation -- DSL pattern requires eval
       end
 
-      def candidate(name, model:, system_prompt: nil, **options)
+      def candidate(name, model: nil, system_prompt: nil, **options)
         @candidates_list << Candidate.new(name, model: model, system_prompt: system_prompt, **options)
       end
 

data/lib/qualspec/suite/html_reporter.rb

@@ -481,10 +481,12 @@ module Qualspec
 
         scenario_blocks = scenarios.map do |scenario|
           response_cards = responses.map do |candidate, candidate_responses|
-            response = candidate_responses[scenario]
-            next unless response
+            variant_map = candidate_responses[scenario]
+            next unless variant_map
 
-            response_text = response.to_s.strip
+            contents = variant_map.flat_map { |_v, tm| tm.values.map { |d| d[:content] } }.compact
+            response_text = contents.join("\n\n---\n\n").strip
+            next if response_text.empty?
 
             <<~CARD
               <div class="response-card">
@@ -660,13 +662,11 @@ module Qualspec
       end
 
       def get_candidate_model(candidate)
-        # Try to find the model from the suite
-        @results.evaluations.find { |e| e[:candidate] == candidate }&.dig(:model) || 'unknown'
+        @results.candidate_models[candidate] || 'unknown'
       end
 
-      def get_scenario_prompt(_scenario)
-        # This would need to be stored in results - for now return nil
-        nil
+      def get_scenario_prompt(scenario)
+        @results.prompts[scenario]
       end
     end
   end

data/lib/qualspec/suite/runner.rb

@@ -11,6 +11,12 @@ module Qualspec
         @definition = definition.is_a?(String) ? Suite.find(definition) : definition
         @results = Results.new(@definition.name)
         @judge = Qualspec.judge
+
+        @definition.candidates_list.each do |c|
+          @results.candidate_models[c.name] = c.model
+        end
+
+        @results.metadata_captured = @definition.track_cost?
       end
 
       def run(progress: true)
@@ -52,6 +58,8 @@ module Qualspec
         responses = {}
         errors = {}
 
+        @results.prompts[scenario.name] ||= scenario.compose_prompt(variant)
+
         # Phase 1: Collect all candidate responses
         @definition.candidates_list.each do |candidate|
           log_candidate_progress(candidate, scenario, 'generating') if progress
@@ -100,7 +108,8 @@ module Qualspec
         response = candidate.generate_response(
           prompt: final_prompt,
           system_prompt: final_system_prompt,
-          temperature: effective_temperature
+          temperature: effective_temperature,
+          with_metadata: @definition.track_cost?
         )
 
         duration_ms = ((Process.clock_gettime(Process::CLOCK_MONOTONIC) - start_time) * 1000).round
@@ -217,7 +226,9 @@ module Qualspec
 
     # Results container with multi-dimensional support
     class Results
-      attr_reader :suite_name, :evaluations, :responses, :started_at, :finished_at, :timing, :costs
+      attr_reader :suite_name, :evaluations, :responses, :started_at, :finished_at, :timing, :costs,
+                  :candidate_models, :prompts
+      attr_accessor :metadata_captured
 
       def initialize(suite_name)
         @suite_name = suite_name
@@ -225,8 +236,36 @@ module Qualspec
         @responses = {} # Nested: {candidate => {scenario => {variant => {temp => response}}}}
         @timing = {}
         @costs = {}
+        @candidate_models = {} # {candidate_name => model_string}
+        @prompts = {}          # {scenario_name => prompt_string}
         @started_at = Time.now
         @finished_at = nil
+        @metadata_captured = false # set true when the suite enables track_cost
+      end
+
+      # Whether per-call cost/token metadata was captured this run.
+      def costs_tracked?
+        @metadata_captured
+      end
+
+      # Total cost per candidate. Raises if cost tracking wasn't enabled.
+      def cost_by_candidate
+        ensure_cost_tracking!
+        @costs.dup
+      end
+
+      # Rank candidates by quality-per-dollar (avg score / total cost), best
+      # first. Candidates with zero recorded cost sort last. Raises a helpful
+      # error if cost tracking wasn't enabled for the run.
+      def value_ranking
+        ensure_cost_tracking!
+
+        ranked = scores_by_candidate.map do |candidate, stats|
+          cost = @costs[candidate].to_f
+          score_per_dollar = cost.positive? ? (stats[:avg_score] / cost).round : nil
+          [candidate, { avg_score: stats[:avg_score], cost: cost, score_per_dollar: score_per_dollar }]
+        end
+        ranked.sort_by { |_, v| -(v[:score_per_dollar] || 0) }.to_h
       end
 
       def record_response(candidate:, scenario:, response:, variant: 'default', temperature: nil, duration_ms: nil, cost: nil, variant_data: nil)
@@ -329,13 +368,15 @@ module Qualspec
       def scores_by_scenario
         @evaluations.group_by { |e| e[:scenario] }.transform_values do |evals|
           evals.group_by { |e| e[:candidate] }.transform_values do |candidate_evals|
-            eval_data = candidate_evals.first
+            total = candidate_evals.size
+            avg_score = (candidate_evals.sum { |e| e[:score] }.to_f / total).round(2)
+            first = candidate_evals.first
             {
-              score: eval_data[:score],
-              pass: eval_data[:pass],
-              reasoning: eval_data[:reasoning],
-              variant: eval_data[:variant],
-              temperature: eval_data[:temperature]
+              score: avg_score,
+              pass: candidate_evals.all? { |e| e[:pass] },
+              reasoning: first[:reasoning],
+              variant: first[:variant],
+              temperature: first[:temperature]
             }
           end
         end
@@ -374,6 +415,24 @@ module Qualspec
           responses: @responses
         }
       end
+
+      private
+
+      def ensure_cost_tracking!
+        return if @metadata_captured
+
+        raise Qualspec::Error, <<~MSG.strip
+          Cost data was not captured for this run, so cost/value analysis is unavailable.
+          Enable it with `track_cost` in the suite definition:
+
+            Qualspec.evaluation 'My Suite' do
+              track_cost
+              ...
+            end
+
+          (track_cost adds usage accounting to each request via with_metadata.)
+        MSG
+      end
     end
   end
 end

data/lib/qualspec/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Qualspec
-  VERSION = '0.1.1'
+  VERSION = '0.2.0'
 end

data/lib/qualspec.rb

@@ -7,6 +7,7 @@ module Qualspec
 end
 
 require_relative 'qualspec/configuration'
+require_relative 'qualspec/model_registry'
 require_relative 'qualspec/client'
 require_relative 'qualspec/evaluation'
 require_relative 'qualspec/prompt_variant'
@@ -37,6 +38,7 @@ module Qualspec
       @configuration = nil
       @client = nil
       @judge = nil
+      @models = nil
       Rubric.clear!
       Suite.clear!
       Suite::Behavior.clear!
@@ -50,6 +52,21 @@ module Qualspec
       @judge ||= Judge.new
     end
 
+    # Registry of named models loaded from config/models.yml (or
+    # QUALSPEC_MODELS_FILE). See ModelRegistry.
+    def models
+      @models ||= ModelRegistry.new
+    end
+
+    # Resolve a named model to its slug, falling back to the default
+    # (openrouter/auto). Returns the default when name is nil/unknown.
+    #
+    #   Qualspec.model(:glm)  # => "z-ai/glm-5.2"
+    #   Qualspec.model        # => "openrouter/auto"
+    def model(name = nil)
+      models.resolve(name)
+    end
+
     # Convenience method for defining rubrics
     def define_rubric(name, &block)
       Rubric.define(name, &block)

data/qualspec_structure.md

@@ -7,10 +7,12 @@ LLM-judged qualitative testing for Ruby. Evaluate AI agents, compare models, and
 
 ### Core Library Files (lib/qualspec/)
 - **builtin_rubrics.rb** - Built-in evaluation criteria
-- **client.rb** - API client for LLM interactions
+- **client.rb** - API client for LLM interactions (cost/token metadata optional)
 - **configuration.rb** - Configuration management
 - **evaluation.rb** - Core evaluation logic
 - **judge.rb** - LLM judge implementation
+- **model_registry.rb** - Named models from `config/models.yml` (`Qualspec.model`)
+- **prompt_variant.rb** - Variant value object (FactoryBot target)
 - **recorder.rb** - VCR integration for recording
 - **rspec.rb** - RSpec integration entry point
 - **rubric.rb** - Custom rubric definitions
@@ -23,10 +25,11 @@ LLM-judged qualitative testing for Ruby. Evaluate AI agents, compare models, and
 ### Configuration Environment Variables
 | Variable | Description | Default |
 |----------|-------------|---------|
-| QUALSPEC_API_KEY | API key (required) | - |
+| QUALSPEC_API_KEY | API key (falls back to OPEN_ROUTER_API_KEY) | - |
 | QUALSPEC_API_URL | API endpoint | https://openrouter.ai/api/v1 |
-| QUALSPEC_MODEL | Default model for candidates | google/gemini-3-flash-preview |
+| QUALSPEC_MODEL | Default model for candidates | openrouter/auto |
 | QUALSPEC_JUDGE_MODEL | Model used as judge | Same as QUALSPEC_MODEL |
+| QUALSPEC_MODELS_FILE | Named-models YAML | config/models.yml |
 
 ### Key Features
 1. **Model Comparison CLI** - Compare multiple models on the same prompts
@@ -36,6 +39,9 @@ LLM-judged qualitative testing for Ruby. Evaluate AI agents, compare models, and
 5. **Custom Rubrics** - Define your own evaluation criteria
 6. **VCR Recording** - Record and replay API calls for testing
 7. **HTML Reports** - Generate visual comparison reports
+8. **Named Model Registry** - Reference curated models by name (`Qualspec.model`)
+9. **Cost Tracking** - Opt-in per-call cost + quality-per-dollar `value_ranking`
+10. **Variant & Temperature Matrix** - Combinatorial prompt testing via FactoryBot
 
 ### Example: Model Comparison
 ```ruby

metadata

@@ -1,14 +1,13 @@
 --- !ruby/object:Gem::Specification
 name: qualspec
 version: !ruby/object:Gem::Version
-  version: 0.1.1
+  version: 0.2.0
 platform: ruby
 authors:
 - Eric Stiens
-autorequire:
 bindir: exe
 cert_chain: []
-date: 2026-01-05 00:00:00.000000000 Z
+date: 1980-01-02 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: faraday
@@ -62,7 +61,6 @@ executables:
 extensions: []
 extra_rdoc_files: []
 files:
-- ".DS_Store"
 - ".qualspec_cassettes/comparison_test.yml"
 - ".qualspec_cassettes/quick_test.yml"
 - ".rspec"
@@ -71,7 +69,9 @@ files:
 - CHANGELOG.md
 - README.md
 - Rakefile
+- config/models.yml
 - docs/.DS_Store
+- docs/alpha_readiness.md
 - docs/configuration.md
 - docs/evaluation-suites.md
 - docs/getting-started.md
@@ -80,7 +80,13 @@ files:
 - docs/rubrics.md
 - docs/to_implement/factory_bot_integration_design.md
 - docs/to_implement/variants_first_pass.md
+- examples/EXAMPLES.md
 - examples/README.md
+- examples/best_value.rb
+- examples/cassettes/best_value.yml
+- examples/cassettes/character_consistency.yml
+- examples/cassettes/customer_service_comparison.yml
+- examples/cassettes/date_awareness_gate.yml
 - examples/cassettes/qualspec_rspec_integration_basic_evaluation_evaluates_responses_with_inline_criteria.yml
 - examples/cassettes/qualspec_rspec_integration_basic_evaluation_provides_detailed_feedback_on_failure.yml
 - examples/cassettes/qualspec_rspec_integration_comparative_evaluation_compares_multiple_responses.yml
@@ -88,7 +94,10 @@ files:
 - examples/cassettes/qualspec_rspec_integration_vcr_integration_records_and_plays_back_api_calls_automatically.yml
 - examples/cassettes/qualspec_rspec_integration_with_context_uses_context_in_evaluation.yml
 - examples/cassettes/qualspec_rspec_integration_with_rubrics_evaluates_using_builtin_rubrics.yml
+- examples/character_consistency.rb
 - examples/comparison.rb
+- examples/customer_service_comparison.rb
+- examples/date_awareness_gate.rb
 - examples/model_comparison.rb
 - examples/persona_test.rb
 - examples/prompt_variants_factory.rb
@@ -105,6 +114,7 @@ files:
 - lib/qualspec/configuration.rb
 - lib/qualspec/evaluation.rb
 - lib/qualspec/judge.rb
+- lib/qualspec/model_registry.rb
 - lib/qualspec/prompt_variant.rb
 - lib/qualspec/recorder.rb
 - lib/qualspec/rspec.rb
@@ -131,7 +141,6 @@ metadata:
   homepage_uri: https://github.com/estiens/qualspec
   source_code_uri: https://github.com/estiens/qualspec
   changelog_uri: https://github.com/estiens/qualspec/blob/main/CHANGELOG.md
-post_install_message:
 rdoc_options: []
 require_paths:
 - lib
@@ -139,15 +148,14 @@ required_ruby_version: !ruby/object:Gem::Requirement
   requirements:
   - - ">="
     - !ruby/object:Gem::Version
-      version: 3.1.0
+      version: 3.3.0
 required_rubygems_version: !ruby/object:Gem::Requirement
   requirements:
   - - ">="
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 3.5.22
-signing_key:
+rubygems_version: 3.6.9
 specification_version: 4
 summary: RSpec DSL for qualitative LLM-judged testing
 test_files: []