qualspec 0.1.1 → 0.1.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: '069d64a5f846962da842ffe5336739761862d0b1a3ef4cf522bfc83932af504e'
-  data.tar.gz: 693d58eb17d4e495cc0e7f1ff0ed57fc925d81ee6b8060aef41f350bf1fdb2c9
+  metadata.gz: 004a0ca49d3bcb6890cadde4f994ddf6ce9585f06f4c81798644d4e02c74de5a
+  data.tar.gz: c3994806a042fc9693cdb7e6e9d0579a644edf811b4f529080b216506056aee0
 SHA512:
-  metadata.gz: b46c32e6c43fb4f52234db0f4d6038191e85f8c05f99d8082c0024ef39fe0d7d1fa93a213b9163e8eef1c06a88add8a66f8906fa219ce8c471117a9812778766
-  data.tar.gz: de83bb65616448ccca96a405a021b19f80cbe9673df8cf65c63af24d2bdaa96efece1ea485b2786cfa7b3507f7f0ea301bd30d193c3f4218d18b67f3d2cfd695
+  metadata.gz: dfad9e00f04bc7552f8641c1d12ce956d2e58f66d120a495274e4ca5e6af85a701866f0f8e3951c0a031e72e56f486ebcc041b52be61faadcd8f4822cfebee0d
+  data.tar.gz: d4192bc2492c27472169c30b024061ce969aa65d3855dc4c1b0b26d25db2f2fac46da1070c99418d61586cf475d8c759ee8db1a37673f17270ebcbc222f407ea

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/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 +11,10 @@ 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
       end
 
       def run(progress: true)
@@ -52,6 +56,8 @@ 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
@@ -217,7 +223,8 @@ 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
 
       def initialize(suite_name)
         @suite_name = suite_name
@@ -225,6 +232,8 @@ 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
       end
@@ -329,13 +338,15 @@ 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

data/lib/qualspec/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Qualspec
-  VERSION = '0.1.1'
+  VERSION = '0.1.2'
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: qualspec
 version: !ruby/object:Gem::Version
-  version: 0.1.1
+  version: 0.1.2
 platform: ruby
 authors:
 - Eric Stiens
 autorequire:
 bindir: exe
 cert_chain: []
-date: 2026-01-05 00:00:00.000000000 Z
+date: 2026-04-16 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: faraday
@@ -62,7 +62,6 @@ executables:
 extensions: []
 extra_rdoc_files: []
 files:
-- ".DS_Store"
 - ".qualspec_cassettes/comparison_test.yml"
 - ".qualspec_cassettes/quick_test.yml"
 - ".rspec"