completion-kit 0.28.45 → 0.28.46

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 7c6a9a433f596def84c665f6b6fbe2b5aa4ed1422ac923f20a2a2d246aad02c4
-  data.tar.gz: 3a0ad110587d51bad3b41f89439da3efa79c199887d58b76703b0d46d16d7f6d
+  metadata.gz: f620811d2032fa4aad51754dcee5aa92657cc193c890587845cb0c185060ef18
+  data.tar.gz: ea6426877ef769569438a48f3f2cc7848569e02c395db53e629898cda4c01545
 SHA512:
-  metadata.gz: 5f17ea2b059df5b6e2656c9a80a60d87ccd5a55b7d1d2a2599ca3b913227eca49d19a983015765b52accc0287a8e07e60dedb3e1d92d2744e81f5803a32f5b19
-  data.tar.gz: 673b44c3f1eb7edaf963acb85a0451067ed1f6fa9ee73fee156518c031e8fb2b74786aa119c88d39281cc36e4a5e56da3ab8354a3c0411fe1eca800ed96b275e
+  metadata.gz: c6f1a8afbf36d89f1dbd42919fc870732e6f5aeb14164303e62d68644489837f845fc50b19d5ea9735e55b5e1876e743f67ced221e9d3d720f08726d24b4f4f8
+  data.tar.gz: 4cf7b53081569d22ddb471e6b40418baeba32e3fcd11e284df14cd582810603f7e132165603e169cb19f3655fefc7b5f69a9fe437ee687be181d7dc4ced95447

data/app/assets/stylesheets/completion_kit/application.css

@@ -2311,6 +2311,15 @@ label.ck-checkbox[hidden] {
   flex: 1;
 }
 
+.ck-select-with-action > .field_with_errors {
+  flex: 1;
+  min-width: 0;
+}
+
+.ck-select-with-action > .field_with_errors > .ck-input {
+  width: 100%;
+}
+
 .ck-icon-btn:hover {
   background: rgba(255, 255, 255, 0.05);
   border-color: var(--ck-line-strong);
@@ -5175,11 +5184,13 @@ a.tag-mark {
 }
 
 
-.ck-input--error {
+.ck-input--error,
+.ck-input[aria-invalid="true"] {
   border-color: var(--ck-danger);
-  background: rgba(248, 113, 113, 0.05);
+  background-color: rgba(248, 113, 113, 0.05);
 }
-.ck-input--error:focus {
+.ck-input--error:focus,
+.ck-input[aria-invalid="true"]:focus {
   box-shadow: 0 0 0 3px rgba(248, 113, 113, 0.15);
   border-color: var(--ck-danger);
 }

data/app/controllers/completion_kit/onboarding_controller.rb

@@ -21,7 +21,7 @@ module CompletionKit
 
     def sample_data
       Onboarding::SampleData.install!
-      redirect_to onboarding_path, notice: "Loaded a sample dataset and prompt — edit or delete them whenever."
+      redirect_to onboarding_path, notice: "Loaded a sample dataset and prompt. Edit or delete them whenever."
     end
   end
 end

data/app/controllers/completion_kit/runs_controller.rb

@@ -40,6 +40,7 @@ module CompletionKit
     end
 
     def create
+      @judge_only = judge_only_chosen?
       @run = Run.new(run_attributes)
       if @run.save
         redirect_to run_path(@run), notice: "Run was successfully created."
@@ -50,10 +51,11 @@ module CompletionKit
     end
 
     def update
+      @judge_only = judge_only_chosen?
       if @run.responses.any? && run_generation_changed?
         attrs = run_attributes.to_h
         attrs.delete("name") if attrs["name"].to_s == @run.name.to_s
-        new_run = Run.new(attrs.merge(status: "pending"))
+        new_run = Run.new(attrs.merge(status: "pending", cloned_from_judge_model: @run.judge_model))
         if new_run.save
           redirect_to run_path(new_run), notice: "Saved as a new run. The previous run and its results are preserved."
         else
@@ -224,11 +226,16 @@ module CompletionKit
     end
 
     def run_params
-      params.require(:run).permit(:name, :prompt_id, :dataset_id, :judge_model, :temperature, :max_tokens, :judge_temperature, :output_column, :expected_column, metric_ids: [], tag_names: [])
+      params.require(:run).permit(:name, :prompt_id, :dataset_id, :judge_model, :temperature, :max_tokens, :judge_temperature, :output_column, :expected_column, :judge_only, metric_ids: [], tag_names: [])
     end
 
     def run_attributes
-      run_params.except(:metric_ids).merge(selected_metric_ids: run_params[:metric_ids])
+      attributes = run_params.except(:metric_ids, :judge_only).merge(selected_metric_ids: run_params[:metric_ids])
+      judge_only_chosen? ? attributes.merge(prompt_id: nil) : attributes
+    end
+
+    def judge_only_chosen?
+      run_params[:judge_only] == "1"
     end
 
     # Editing a run that already has results forks a new run — but only when a
@@ -237,9 +244,10 @@ module CompletionKit
     GENERATION_RUN_FIELDS = %i[prompt_id dataset_id judge_model temperature max_tokens judge_temperature output_column expected_column].freeze
 
     def run_generation_changed?
+      attributes = run_attributes
       GENERATION_RUN_FIELDS.each do |field|
-        next unless run_params.key?(field)
-        return true if normalize_run_field(field, run_params[field]) != normalize_run_field(field, @run.public_send(field))
+        next unless attributes.key?(field)
+        return true if normalize_run_field(field, attributes[field]) != normalize_run_field(field, @run.public_send(field))
       end
       return false unless params[:run].key?(:metric_ids)
       Array(params[:run][:metric_ids]).map(&:to_i).reject(&:zero?).sort != @run.metric_ids.sort
@@ -247,6 +255,7 @@ module CompletionKit
 
     def normalize_run_field(field, value)
       s = value.to_s.strip
+      return s.presence || "actual_output" if field == :output_column
       return nil if s.empty?
       case field
       when :temperature, :judge_temperature then s.to_f

data/app/helpers/completion_kit/application_helper.rb

@@ -151,18 +151,22 @@ module CompletionKit
       model[:name]
     end
 
-    def ck_grouped_models(models, selected = nil)
+    def ck_grouped_models(models, selected = nil, rejected: false)
+      kept = nil
       if selected.present? && models.none? { |m| m[:id] == selected }
-        retired = CompletionKit::Model.find_by(model_id: selected)
+        retired = CompletionKit::Model.find_by(model_id: selected) unless rejected
         if retired
           models = models + [{ id: retired.model_id, name: "#{retired.display_name || retired.model_id} (retired)", provider: retired.provider }]
+        else
+          kept = [["#{selected} (unavailable)", selected]]
         end
       end
 
       groups = models.group_by { |m| ck_model_optgroup_label(m) }
       ordered_keys = groups.keys.sort_by { |label| ck_model_optgroup_sort_key(label) }
       grouped = ordered_keys.map { |label| [label, groups[label].map { |m| [ck_model_option_label(m), m[:id]] }] }
-      grouped_options_for_select(grouped, selected)
+      options = grouped_options_for_select(grouped, selected)
+      kept ? options_for_select(kept, selected) + options : options
     end
 
     # Optgroup label for the model select — mirrors the provider models table:
@@ -170,15 +174,15 @@ module CompletionKit
     # else is a single group.
     def ck_model_optgroup_label(model)
       case model[:provider]
-      when "openrouter" then "OpenRouter — #{model[:id].to_s.split("/", 2).first.delete_prefix("~")}"
-      when "openai"     then "OpenAI — #{ck_openai_model_family(model[:id])}"
+      when "openrouter" then "OpenRouter · #{model[:id].to_s.split("/", 2).first.delete_prefix("~")}"
+      when "openai"     then "OpenAI · #{ck_openai_model_family(model[:id])}"
       else ck_provider_label(model[:provider])
       end
     end
 
     def ck_model_optgroup_sort_key(label)
-      if label.start_with?("OpenAI — ")
-        [0, OPENAI_MODEL_FAMILY_ORDER.index(label.delete_prefix("OpenAI — ")), label]
+      if label.start_with?("OpenAI · ")
+        [0, OPENAI_MODEL_FAMILY_ORDER.index(label.delete_prefix("OpenAI · ")), label]
       elsif label.start_with?("OpenRouter")
         [2, 0, label]
       else

data/app/models/completion_kit/prompt.rb

@@ -44,7 +44,7 @@ module CompletionKit
     end
 
     def display_name
-      "#{name} — #{version_label}"
+      "#{name} · #{version_label}"
     end
 
     def llm_model_provider

data/app/models/completion_kit/run.rb

@@ -333,13 +333,10 @@ module CompletionKit
         unless judge_only_input_data_checks? || (dataset && dataset.headers.include?(column))
           return fail_with_summary!("Dataset has no \"#{column}\" column")
         end
-      else
-        generation_error = generation_client_error
-        return fail_with_summary!(generation_error) if generation_error
       end
 
-      judge_error = judge_model_error
-      return fail_with_summary!(judge_error) if judge_error
+      model_error = model_start_error
+      return fail_with_summary!(model_error) if model_error
 
       begin
         transaction do
@@ -490,7 +487,8 @@ module CompletionKit
     end
 
     def rerun!
-      ensure_models_resolve!
+      ensure_models_ready!
+
       Run.create!(
         prompt_id: prompt_id,
         dataset_id: dataset_id,
@@ -712,11 +710,20 @@ module CompletionKit
       %(No configured provider serves judge model "#{judge_model}")
     end
 
-    def ensure_models_resolve!
-      raise UnknownModelError, prompt.llm_model unless judge_only? || ApiConfig.provider_for_model(prompt.llm_model)
+    def model_start_error
+      (generation_client_error unless judge_only?) || judge_model_error
+    end
+
+    def ensure_models_ready!
+      unless judge_only?
+        raise UnknownModelError, prompt.llm_model unless ApiConfig.provider_for_model(prompt.llm_model)
+
+        client_error = generation_client_error
+        raise ConfigurationError, client_error if client_error
+      end
 
       judge_error = judge_model_error
-      raise ConfigurationError, judge_error if judge_error
+      raise UnknownModelError.new(judge_model, judge_error) if judge_error
     end
 
     def judge_model_newly_selected?
@@ -775,7 +782,7 @@ module CompletionKit
 
       if prompt.present?
         count = Run.where(prompt_id: prompt_id).count + 1
-        self.name = "#{prompt.name} — v#{prompt.version_number} ##{count}"
+        self.name = "#{prompt.name} · v#{prompt.version_number} ##{count}"
       elsif dataset.present?
         count = Run.where(prompt_id: nil, dataset_id: dataset.id).count + 1
         self.name = "#{dataset.name} scoring ##{count}"

data/app/services/completion_kit/mcp_tools/metric_versions.rb

@@ -27,7 +27,7 @@ module CompletionKit
           handler: :publish
         },
         "metric_versions_dismiss" => {
-          description: "Destroy a draft MetricVersion (use for either source: 'edit' or source: 'suggestion'). Published versions are refused — to demote a published version, publish a different one as current instead.",
+          description: "Destroy a draft MetricVersion (use for either source: 'edit' or source: 'suggestion'). Published versions are refused. To demote a published version, publish a different one as current instead.",
           inputSchema: {
             type: "object",
             properties: {

data/app/services/completion_kit/mcp_tools/prompts.rb

@@ -28,7 +28,7 @@ module CompletionKit
           handler: :create
         },
         "prompts_update" => {
-          description: "Update a prompt. If the prompt already has runs, this creates a new DRAFT version (current=false) rather than editing in place or publishing — promote it with prompts_publish — so an agent's edits don't go live without a gate. If it has no runs, it is updated in place.",
+          description: "Update a prompt. If the prompt already has runs, this creates a new DRAFT version (current=false) rather than editing in place or publishing. Promote it with prompts_publish, so an agent's edits don't go live without a gate. If it has no runs, it is updated in place.",
           inputSchema: {
             type: "object",
             properties: {

data/app/services/completion_kit/prompt_improvement_service.rb

@@ -40,7 +40,7 @@ module CompletionKit
           if review.check?
             sections << "  #{review.metric_name}: #{review.passed ? "PASS" : "FAIL"}"
           else
-            sections << "  #{review.metric_name}: #{review.ai_score}/5 — #{review.ai_feedback}"
+            sections << "  #{review.metric_name}: #{review.ai_score}/5. #{review.ai_feedback}"
           end
         end
         sections << ""

data/app/views/completion_kit/agreements/_buttons.html.erb

@@ -21,7 +21,7 @@
     <% verdict_icons = { "agree" => "hand-thumb-up", "disagree" => "hand-thumb-down", "borderline" => "scale" } %>
     <% verdict_hints = {
          "agree" => "The score looks right.",
-         "disagree" => "The score is wrong — you'll pick the right one.",
+         "disagree" => "The score is wrong. You'll pick the right one.",
          "borderline" => "The rubric is unclear here; either score could be defensible."
        } %>
     <% CompletionKit::Agreement::VERDICTS.each do |verdict| %>

data/app/views/completion_kit/api_reference/_body.html.erb

@@ -196,7 +196,7 @@
       </div>
       <%= render "completion_kit/api_reference/resource_list", title: "Your responses",
             items: recent_runs.first(1).map { |r|
-              { name: "#{r.name} — responses",
+              { name: "#{r.name} · responses",
                 url: "#{base_url}/api/v1/runs/#{r.id}/responses", dom_id: "responses_ep_#{r.id}" }
             } %>
     </div>
@@ -329,7 +329,7 @@
       </div>
       <div class="ck-api-endpoint">
         <p class="ck-api-method"><span class="ck-chip ck-chip--soft">GET</span> /api/v1/runs/:run_id/responses/:response_id/metrics/:metric_id/agreements</p>
-        <p class="ck-meta-copy">List agreements scoped to a specific run, response, and metric &mdash; the nested equivalent of the top-level index filtered by those three ids.</p>
+        <p class="ck-meta-copy">List agreements scoped to a specific run, response, and metric. It is the nested equivalent of the top-level index filtered by those three ids.</p>
       </div>
       <div class="ck-api-endpoint">
         <p class="ck-api-method"><span class="ck-chip" style="color: var(--ck-danger);">DELETE</span> /api/v1/agreements/:id</p>

data/app/views/completion_kit/metric_groups/_form.html.erb

@@ -2,10 +2,21 @@
   <% metric_group_delete_id = "ck_delete_metric_group_#{metric_group.id}" %>
   <% members_n = metric_group.metrics.count %>
   <% metric_group_delete_confirm = members_n.zero? ?
-       "Delete \"#{metric_group.name}\"? It has no members." :
-       "Delete \"#{metric_group.name}\"? Removes this grouping. #{pluralize(members_n, 'member metric')} #{members_n == 1 ? 'is' : 'are'} kept." %>
+       "Delete \"#{metric_group.name_was}\"? It has no members." :
+       "Delete \"#{metric_group.name_was}\"? Removes this grouping. #{pluralize(members_n, 'member metric')} #{members_n == 1 ? 'is' : 'are'} kept." %>
 <% end %>
 <%= form_with(model: metric_group, url: metric_group.persisted? ? metric_group_path(metric_group) : metric_groups_path, local: true) do |form| %>
+  <% if metric_group.errors.any? %>
+    <div class="ck-flash ck-flash--alert" role="alert">
+      <p class="ck-flash__title"><%= pluralize(metric_group.errors.count, "problem") %> prevented this metric group from being saved.</p>
+      <ul class="ck-error-list">
+        <% metric_group.errors.full_messages.each do |message| %>
+          <li><%= message %></li>
+        <% end %>
+      </ul>
+    </div>
+  <% end %>
+
   <div class="ck-card ck-form-card">
     <div class="ck-field">
       <%= form.label :name, "Metric group name", class: "ck-label" %>

data/app/views/completion_kit/metric_groups/edit.html.erb

@@ -1,7 +1,7 @@
 <% content_for :ck_title, "Edit metric group" %>
 <ol class="ck-breadcrumb">
   <li><%= link_to "Metrics", metrics_path %></li>
-  <li><%= link_to @metric_group.name, metric_group_path(@metric_group) %></li>
+  <li><%= link_to @metric_group.name_was, metric_group_path(@metric_group) %></li>
   <li>Edit</li>
 </ol>
 

data/app/views/completion_kit/onboarding/show.html.erb

@@ -72,7 +72,7 @@
 
       <% if @checklist.sample_loadable? %>
         <div class="ck-launch__sample">
-          <p class="ck-launch__sample-copy">Just exploring? Drop in a sample dataset and prompt to poke around — they're labelled <code>Sample:</code>, and you can edit or delete them whenever.</p>
+          <p class="ck-launch__sample-copy">Just exploring? Drop in a sample dataset and prompt to poke around. They're labelled <code>Sample:</code>, and you can edit or delete them whenever.</p>
           <%= button_to onboarding_sample_data_path, method: :post, class: "ck-launch__sample-cta", form: { style: "display:inline" } do %>
             Load sample data <span aria-hidden="true">&rarr;</span>
           <% end %>
@@ -81,7 +81,7 @@
 
       <div class="ck-launch__panel-footer">
         <%= button_to dismiss_onboarding_path, method: :post, class: "ck-launch__dismiss", form: { style: "display:inline" } do %>
-          Skip setup &mdash; go to the app <span aria-hidden="true">&rarr;</span>
+          Skip setup and go to the app <span aria-hidden="true">&rarr;</span>
         <% end %>
         <p class="ck-launch__dismiss-hint">You can come back to this from <strong>Settings &rarr; Getting started</strong> any time.</p>
       </div>

data/app/views/completion_kit/prompts/_form.html.erb

@@ -47,13 +47,14 @@
       <% available = CompletionKit::ApiConfig.available_models(scope: :generation) %>
       <% if available.any? %>
         <div class="ck-select-with-action">
-          <%= form.select :llm_model, ck_grouped_models(available, prompt.llm_model), { include_blank: "Select a model" }, { class: "ck-input", id: "prompt_llm_model" } %>
+          <%= form.select :llm_model, ck_grouped_models(available, prompt.llm_model, rejected: prompt.errors[:llm_model].any?), { include_blank: "Select a model" }, { class: "ck-input", id: "prompt_llm_model", **ck_field_aria(form, :llm_model) } %>
           <% ck_refreshing = CompletionKit::ProviderCredential.discovery_in_progress? %>
           <button type="button" class="ck-icon-btn<%= ' ck-icon-btn--spinning' if ck_refreshing %>" title="Refresh models" <%= 'disabled' if ck_refreshing %> onclick="ckRefreshModels()"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 16 16" fill="currentColor" width="16" height="16"><path fill-rule="evenodd" d="M13.836 2.477a.75.75 0 0 1 .75.75v3.182a.75.75 0 0 1-.75.75h-3.182a.75.75 0 0 1 0-1.5h1.37l-.84-.841a4.5 4.5 0 0 0-7.08.681.75.75 0 0 1-1.264-.808 6 6 0 0 1 9.44-.908l.84.84V3.227a.75.75 0 0 1 .75-.75Zm-.911 7.5A.75.75 0 0 1 13.199 11a6 6 0 0 1-9.44.908l-.84-.84v1.68a.75.75 0 0 1-1.5 0V9.567a.75.75 0 0 1 .75-.75h3.182a.75.75 0 0 1 0 1.5h-1.37l.84.841a4.5 4.5 0 0 0 7.08-.681.75.75 0 0 1 1.024-.274Z" clip-rule="evenodd"/></svg></button>
         </div>
       <% else %>
         <p class="ck-meta-copy">No models available. <%= link_to "Add a provider", provider_credentials_path, class: "ck-link" %> or click refresh after configuring a provider.</p>
       <% end %>
+      <%= ck_field_error(form, :llm_model) %>
       <div hidden data-refresh-progress-carriers data-ck-refresh-url="<%= refresh_models_path %>" data-ck-statuses-url="<%= statuses_provider_credentials_path %>">
         <% CompletionKit::ProviderCredential.find_each do |pc| %>
           <%= turbo_stream_from "completion_kit_provider_#{pc.id}" %>

data/app/views/completion_kit/provider_credentials/_models_card.html.erb

@@ -35,7 +35,7 @@
                     Gen<span class="ck-info-toggle" tabindex="0">?</span><span class="ck-info-popup ck-info-popup--right">Generation models produce the responses your prompts ask for. Pick one when creating a prompt.</span>
                   </th>
                   <th scope="col" class="ck-model-table__cap">
-                    Judge<span class="ck-info-toggle" tabindex="0">?</span><span class="ck-info-popup ck-info-popup--right">Judge models score generated responses against your metrics. Pick one when configuring a run. A <strong>?</strong> means we haven't confirmed this model works as a judge — it's still selectable, and a successful run promotes it to ✓.</span>
+                    Judge<span class="ck-info-toggle" tabindex="0">?</span><span class="ck-info-popup ck-info-popup--right">Judge models score generated responses against your metrics. Pick one when configuring a run. A <strong>?</strong> means we haven't confirmed this model works as a judge. It's still selectable, and a successful run promotes it to ✓.</span>
                   </th>
                 </tr>
               </thead>
@@ -58,7 +58,7 @@
                         <% if m.supports_judging %>
                           <span class="ck-model-table__tick" aria-label="Supports judging">✓</span>
                         <% elsif m.supports_judging.nil? %>
-                          <span class="ck-model-table__unknown" aria-label="Untested as judge" title="Untested as a judge — selectable; a successful run confirms it">?</span>
+                          <span class="ck-model-table__unknown" aria-label="Untested as judge" title="Untested as a judge but selectable. A successful run confirms it.">?</span>
                         <% else %>
                           <span class="ck-model-table__dash" aria-label="Not usable as judge">—</span>
                         <% end %>

data/app/views/completion_kit/runs/_form.html.erb

@@ -19,7 +19,7 @@
 
     <div class="ck-field">
       <label class="ck-checkbox-label">
-        <%= check_box_tag "run[judge_only]", "1", run.persisted? && run.judge_only?, id: "run_judge_only", class: "ck-checkbox" %>
+        <%= check_box_tag "run[judge_only]", "1", local_assigns[:judge_only] || (run.persisted? && run.judge_only?), id: "run_judge_only", class: "ck-checkbox" %>
         <span class="ck-checkbox-label__box" aria-hidden="true"></span>
         <span class="ck-checkbox-label__body">
           <span class="ck-checkbox-label__text">Score existing outputs</span>
@@ -98,10 +98,11 @@
       <% available = CompletionKit::ApiConfig.available_models(scope: :judging) %>
       <% if available.any? %>
         <div class="ck-select-with-action">
-          <%= form.select :judge_model, ck_grouped_models(available, run.judge_model), { include_blank: "None" }, { class: "ck-input", id: "run_judge_model" } %>
+          <%= form.select :judge_model, ck_grouped_models(available, run.judge_model, rejected: run.errors[:judge_model].any?), { include_blank: "None" }, { class: "ck-input", id: "run_judge_model", **ck_field_aria(form, :judge_model) } %>
           <% ck_refreshing = CompletionKit::ProviderCredential.discovery_in_progress? %>
           <button type="button" class="ck-icon-btn<%= ' ck-icon-btn--spinning' if ck_refreshing %>" title="Refresh models" <%= 'disabled' if ck_refreshing %> onclick="ckRefreshModels()"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 16 16" fill="currentColor" width="16" height="16"><path fill-rule="evenodd" d="M13.836 2.477a.75.75 0 0 1 .75.75v3.182a.75.75 0 0 1-.75.75h-3.182a.75.75 0 0 1 0-1.5h1.37l-.84-.841a4.5 4.5 0 0 0-7.08.681.75.75 0 0 1-1.264-.808 6 6 0 0 1 9.44-.908l.84.84V3.227a.75.75 0 0 1 .75-.75Zm-.911 7.5A.75.75 0 0 1 13.199 11a6 6 0 0 1-9.44.908l-.84-.84v1.68a.75.75 0 0 1-1.5 0V9.567a.75.75 0 0 1 .75-.75h3.182a.75.75 0 0 1 0 1.5h-1.37l.84.841a4.5 4.5 0 0 0 7.08-.681.75.75 0 0 1 1.024-.274Z" clip-rule="evenodd"/></svg></button>
         </div>
+        <%= ck_field_error(form, :judge_model) %>
         <p class="ck-field-hint" id="judge-hint"></p>
         <div hidden data-refresh-progress-carriers data-ck-refresh-url="<%= refresh_models_path %>" data-ck-statuses-url="<%= statuses_provider_credentials_path %>">
           <% CompletionKit::ProviderCredential.find_each do |pc| %>
@@ -112,7 +113,7 @@
         <p class="ck-field-hint" id="refresh-status" role="status" aria-live="polite" style="min-height: 1.2em; margin-top: -0.25rem; font-size: 0.75rem;">&nbsp;</p>
       <% else %>
         <p class="ck-field-hint" style="color: var(--ck-warning);">No judge models available.&ensp;<%= link_to "Add a provider", provider_credentials_path, class: "ck-link" %></p>
-        <%= form.hidden_field :judge_model, value: nil, id: "run_judge_model" %>
+        <%= form.hidden_field :judge_model, id: "run_judge_model" %>
       <% end %>
     </div>
 
@@ -157,6 +158,7 @@
           </div>
         <% end %>
 
+        <%= hidden_field_tag "run[metric_ids][]", "", id: nil %>
         <div class="ck-metric-checkboxes" data-metric-checkboxes role="group" aria-labelledby="run_metrics_label">
           <% @all_metrics.each do |metric| %>
             <label class="ck-checkbox-label" data-metric-tags="<%= metric.tag_names.join(",") %>" data-compare-expected="<%= metric.check? && metric.check_config.to_h["compare_to"] == "expected" ? "1" : "0" %>">
@@ -194,6 +196,7 @@ function updateRunForm() {
 
   var prompt = promptEl ? promptEl.value : '';
   var judge = judgeEl ? judgeEl.value : '';
+  var judgeListed = !!judgeEl && judgeEl.type !== 'hidden';
   var metrics = document.querySelectorAll('input[name="run[metric_ids][]"]:checked');
   var judgeField = document.getElementById('judge-field');
   var metricsField = document.getElementById('metrics-field');
@@ -206,13 +209,15 @@ function updateRunForm() {
   if (judgeHint) judgeHint.textContent = '';
   if (metricsHint) metricsHint.textContent = '';
 
-  if (judge && metrics.length > 0) {
-    if (judgeHint) judgeHint.textContent = 'Responses will be generated then judged automatically.';
-  } else if (judge && metrics.length === 0) {
+  var judgeUnpaired = judgeListed && !!judge && metrics.length === 0;
+  var metricsUnpaired = judgeListed && !judge && metrics.length > 0;
+  if (judgeListed && judge && metrics.length > 0) {
+    if (judgeHint) judgeHint.textContent = judgeOnly ? 'Existing outputs will be judged automatically.' : 'Responses will be generated then judged automatically.';
+  } else if (judgeUnpaired) {
     if (judgeField) judgeField.className = 'ck-field ck-field--info';
     if (metricsField) metricsField.className = 'ck-field ck-field--info';
     if (metricsHint) metricsHint.textContent = 'Select at least one metric or group to enable judging.';
-  } else if (!judge && metrics.length > 0) {
+  } else if (metricsUnpaired) {
     if (judgeField) judgeField.className = 'ck-field ck-field--info';
     if (judgeHint) judgeHint.textContent = 'Select a judge model to enable judging.';
   }
@@ -286,7 +291,7 @@ function updateRunForm() {
       if (col === '' || headersJudge.indexOf(col) === -1) {
         valid = false;
         if (datasetField) datasetField.className = 'ck-field ck-field--error';
-        if (datasetHint) datasetHint.textContent = "Dataset has no \"" + col + "\" column — pick a different output column or dataset.";
+        if (datasetHint) datasetHint.textContent = "Dataset has no \"" + col + "\" column. Pick a different output column or dataset.";
       }
     } else if (!dataset) {
       if (datasetField) datasetField.className = 'ck-field ck-field--info';
@@ -297,8 +302,7 @@ function updateRunForm() {
     if (hasVars && !dataset) valid = false;
     if (missingVars.length > 0) valid = false;
   }
-  if (judge && metrics.length === 0) valid = false;
-  if (!judge && metrics.length > 0) valid = false;
+  if (judgeUnpaired || metricsUnpaired) valid = false;
   if (submitBtn) submitBtn.disabled = !valid;
 
   ckUpdateMetricGroupsState();

data/app/views/completion_kit/runs/_response_row.html.erb

@@ -4,7 +4,8 @@
   <td class="ck-response-cell__text">
     <% if response.status == "failed" %>
       <% err = response.error_payload %>
-      <span class="ck-response-cell__preview ck-response-cell__error"><%= err && err[:provider]&.titleize %><%= " #{err[:status]}" if err && err[:status] %> — <%= truncate(err && err[:message].to_s, length: 600) %></span>
+      <% err_prefix = [err && err[:provider]&.titleize, err && err[:status]].compact.join(" ") %>
+      <span class="ck-response-cell__preview ck-response-cell__error"><%= "#{err_prefix}: " if err_prefix.present? %><%= truncate(err && err[:message].to_s, length: 600) %></span>
     <% else %>
       <span class="ck-response-cell__preview"><%= truncate(response.response_text.to_s, length: 600) %></span>
     <% end %>

data/app/views/completion_kit/runs/edit.html.erb

@@ -16,4 +16,4 @@
   </div>
 </section>
 
-<%= render "form", run: @run %>
+<%= render "form", run: @run, judge_only: @judge_only %>

data/app/views/completion_kit/runs/new.html.erb

@@ -10,4 +10,4 @@
   </div>
 </section>
 
-<%= render "form", run: @run %>
+<%= render "form", run: @run, judge_only: @judge_only %>

data/app/views/layouts/completion_kit/application.html.erb

@@ -1,7 +1,7 @@
 <!DOCTYPE html>
 <html lang="en">
 <head>
-  <title><%= [content_for(:ck_title).presence, "CompletionKit"].compact.join(" — ") %></title>
+  <title><%= [content_for(:ck_title).presence, "CompletionKit"].compact.join(" · ") %></title>
   <meta name="viewport" content="width=device-width, initial-scale=1">
   <%= csrf_meta_tags %>
   <%= csp_meta_tag %>

data/db/migrate/20261013000001_drop_completion_kit_mcp_sessions.rb

@@ -0,0 +1,16 @@
+class DropCompletionKitMcpSessions < ActiveRecord::Migration[8.1]
+  def up
+    drop_table :completion_kit_mcp_sessions, if_exists: true
+  end
+
+  def down
+    create_table :completion_kit_mcp_sessions do |t|
+      t.string :session_id, null: false
+      t.datetime :expires_at, null: false
+      t.timestamps
+    end
+
+    add_index :completion_kit_mcp_sessions, :session_id, unique: true
+    add_index :completion_kit_mcp_sessions, :expires_at
+  end
+end

data/lib/completion_kit/errors.rb

@@ -9,9 +9,9 @@ module CompletionKit
   class UnknownModelError < ConfigurationError
     attr_reader :model_name
 
-    def initialize(model_name = nil)
+    def initialize(model_name = nil, message = nil)
       @model_name = model_name
-      super(%(No configured provider serves model "#{model_name}"))
+      super(message || %(No configured provider serves model "#{model_name}"))
     end
   end
 

data/lib/completion_kit/version.rb

@@ -1,3 +1,3 @@
 module CompletionKit
-  VERSION = "0.28.45"
+  VERSION = "0.28.46"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: completion-kit
 version: !ruby/object:Gem::Version
-  version: 0.28.45
+  version: 0.28.46
 platform: ruby
 authors:
 - Damien Bastin
@@ -452,6 +452,7 @@ files:
 - db/migrate/20260730000002_create_completion_kit_prompt_serves.rb
 - db/migrate/20260804000001_add_judge_temperature_ignored_to_runs.rb
 - db/migrate/20260804000002_default_runs_to_no_temperature.rb
+- db/migrate/20261013000001_drop_completion_kit_mcp_sessions.rb
 - lib/completion-kit.rb
 - lib/completion_kit.rb
 - lib/completion_kit/concurrency_check.rb