completion-kit 0.28.46 → 0.29.0

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

@@ -17,10 +17,11 @@
   <% pub = local_assigns[:published_metric_version] %>
   <% draft_instr_changed = pub && pub.instruction.to_s != edit_draft.instruction.to_s %>
   <% draft_rubric_changed = pub && pub.rubric_bands != edit_draft.rubric_bands %>
+  <% draft_subject = edit_draft.check? ? "check" : [("instruction" if draft_instr_changed), ("rubric" if draft_rubric_changed)].compact.join(" and ").presence || "version" %>
   <div class="ck-suggestion-banner" role="status">
     <div class="ck-suggestion-banner__body">
       <p class="ck-kicker">Draft pending</p>
-      <p class="ck-meta-copy">The form below shows your unpublished draft. Publish to replace the live<%= " instruction" if draft_instr_changed %><%= " and" if draft_instr_changed && draft_rubric_changed %><%= " rubric" if draft_rubric_changed %> for future runs, or keep editing.</p>
+      <p class="ck-meta-copy">The form below shows your unpublished draft. Publish to replace the live <%= draft_subject %> for future runs, or keep editing.</p>
     </div>
     <div class="ck-suggestion-banner__actions">
       <%= button_to "Discard draft", dismiss_suggestion_metric_path(metric, draft_id: edit_draft.id, back_to: "edit"),
@@ -164,99 +165,113 @@
 
     <% if show_check %>
     <% check = metric.check_config || {} %>
-    <div class="ck-field ck-field--spacious" data-ck-metric-editor="check" <%= "hidden" unless metric.check? %>>
+    <% kind = check["check_kind"] || CompletionKit::Checks::Registry.kinds.first %>
+    <div class="ck-field ck-field--spacious" data-ck-metric-editor="check" data-ck-check-form="<%= ck_check_form_data.to_json %>" <%= "hidden" unless metric.check? %>>
       <p class="ck-section-title">Check</p>
       <p class="ck-hint">A pass or fail rule with no AI and no cost. You only fill in the fields the chosen rule needs.</p>
 
       <div class="ck-field">
-        <p class="ck-label" id="metric_check_kind_label">Rule</p>
+        <p class="ck-label" id="metric_check_kind_label"><%= ck_check_field_label("check_kind") %></p>
         <%= render "completion_kit/shared/branded_select",
                    name: "metric[check_config][check_kind]",
-                   options: CompletionKit::Checks::Registry.kinds.map { |kind| [kind, ck_check_kind_label(kind)] },
-                   selected: check["check_kind"] || CompletionKit::Checks::Registry.kinds.first,
-                   labelledby: "metric_check_kind_label" %>
+                   options: CompletionKit::Checks::Registry.kinds.map { |option| [option, ck_check_kind_label(option)] },
+                   selected: kind,
+                   labelledby: "metric_check_kind_label",
+                   invalid: ck_check_field_invalid?(metric, "check_kind") %>
       </div>
 
       <div class="ck-field">
-        <p class="ck-label" id="metric_check_target_label">What to check</p>
+        <p class="ck-label" id="metric_check_target_label"><%= ck_check_field_label("target") %></p>
         <%= render "completion_kit/shared/branded_select",
                    name: "metric[check_config][target]",
                    options: CompletionKit::Checks::TargetResolver::TARGETS.map { |target| [target, ck_check_target_label(target)] },
                    selected: check["target"] || CompletionKit::Checks::TargetResolver::TARGETS.first,
-                   labelledby: "metric_check_target_label" %>
+                   labelledby: "metric_check_target_label",
+                   invalid: ck_check_field_invalid?(metric, "target") %>
       </div>
 
       <div class="ck-field" data-ck-check-field="target_path">
-        <label class="ck-label" for="metric_check_target_path">Path into the JSON</label>
-        <p class="ck-hint">Where to look inside the response, like data.items.0.name.</p>
-        <input type="text" name="metric[check_config][target_path]" id="metric_check_target_path" class="ck-input" value="<%= check["target_path"] %>">
+        <label class="ck-label" for="metric_check_target_path"><%= ck_check_field_label("target_path") %></label>
+        <p class="ck-hint">Where to look inside the response, like data.items.0.name. Use a number to pick an item out of a list.</p>
+        <input type="text" name="metric[check_config][target_path]" id="metric_check_target_path" class="ck-input" value="<%= check["target_path"] %>" <%= tag.attributes(ck_check_field_aria(metric, "target_path")) %>>
       </div>
 
       <div class="ck-field" data-ck-check-field="compare_to">
-        <p class="ck-label">Compare against</p>
+        <p class="ck-label"><%= ck_check_field_label("compare_to") %></p>
         <p class="ck-hint">Check every output against one fixed value, or against each row's own expected value (its answer key).</p>
         <label class="ck-radio">
-          <input type="radio" name="metric[check_config][compare_to]" value="constant" <%= "checked" unless check["compare_to"] == "expected" %>>
-          <span>A value I type</span>
+          <input type="radio" name="metric[check_config][compare_to]" value="constant" <%= "checked" unless check["compare_to"] == "expected" %> <%= tag.attributes(ck_check_field_aria(metric, "compare_to")) %>>
+          <span><%= ck_check_field_value("compare_to", "constant") %></span>
         </label>
         <label class="ck-radio">
-          <input type="radio" name="metric[check_config][compare_to]" value="expected" <%= "checked" if check["compare_to"] == "expected" %>>
-          <span>Each row's expected value</span>
+          <input type="radio" name="metric[check_config][compare_to]" value="expected" <%= "checked" if check["compare_to"] == "expected" %> <%= tag.attributes(ck_check_field_aria(metric, "compare_to")) %>>
+          <span><%= ck_check_field_value("compare_to", "expected") %></span>
         </label>
       </div>
 
       <div class="ck-field" data-ck-check-field="value">
-        <label class="ck-label" for="metric_check_value">Text to look for</label>
-        <input type="text" name="metric[check_config][value]" id="metric_check_value" class="ck-input" value="<%= check["value"] %>">
+        <label class="ck-label" for="metric_check_value"><span data-ck-check-label="value"><%= ck_check_field_label("value", kind) %></span></label>
+        <input type="text" name="metric[check_config][value]" id="metric_check_value" class="ck-input" value="<%= ck_check_input_value(check["value"]) %>" <%= tag.attributes(ck_check_field_aria(metric, "value")) %>>
       </div>
 
       <div class="ck-field" data-ck-check-field="pattern">
-        <label class="ck-label" for="metric_check_pattern">Pattern</label>
+        <label class="ck-label" for="metric_check_pattern"><%= ck_check_field_label("pattern") %></label>
         <p class="ck-hint">A regular expression to match against.</p>
-        <input type="text" name="metric[check_config][pattern]" id="metric_check_pattern" class="ck-input" value="<%= check["pattern"] %>">
+        <input type="text" name="metric[check_config][pattern]" id="metric_check_pattern" class="ck-input" value="<%= check["pattern"] %>" <%= tag.attributes(ck_check_field_aria(metric, "pattern")) %>>
       </div>
 
       <div class="ck-field" data-ck-check-field="json_path">
-        <label class="ck-label" for="metric_check_json_path">JSON path</label>
+        <label class="ck-label" for="metric_check_json_path"><%= ck_check_field_label("json_path") %></label>
         <p class="ck-hint">The path to the value inside the response, like data.status.</p>
-        <input type="text" name="metric[check_config][json_path]" id="metric_check_json_path" class="ck-input" value="<%= check["json_path"] %>">
+        <input type="text" name="metric[check_config][json_path]" id="metric_check_json_path" class="ck-input" value="<%= check["json_path"] %>" <%= tag.attributes(ck_check_field_aria(metric, "json_path")) %>>
       </div>
 
       <div class="ck-field" data-ck-check-field="expected">
-        <label class="ck-label" for="metric_check_expected">Expected value</label>
-        <p class="ck-hint">The value it should equal.</p>
-        <input type="text" name="metric[check_config][expected]" id="metric_check_expected" class="ck-input" value="<%= check["expected"] %>">
+        <label class="ck-label" for="metric_check_expected"><%= ck_check_field_label("expected") %></label>
+        <p class="ck-hint">The value it should equal. A number, true or false matches that JSON value. Put it in quotes, like "200", to match it as text.</p>
+        <input type="text" name="metric[check_config][expected]" id="metric_check_expected" class="ck-input" value="<%= ck_check_expected_input_value(check["expected"]) %>" <%= tag.attributes(ck_check_field_aria(metric, "expected")) %>>
       </div>
 
       <div class="ck-field" data-ck-check-field="expected_path">
-        <label class="ck-label" for="metric_check_expected_path">Field in the answer key</label>
+        <label class="ck-label" for="metric_check_expected_path"><%= ck_check_field_label("expected_path") %></label>
         <p class="ck-hint">Optional. If the expected value is JSON, name the field to read from it, like data.answer. Leave blank to compare against the whole value.</p>
-        <input type="text" name="metric[check_config][expected_path]" id="metric_check_expected_path" class="ck-input" value="<%= check["expected_path"] %>">
+        <input type="text" name="metric[check_config][expected_path]" id="metric_check_expected_path" class="ck-input" value="<%= check["expected_path"] %>" <%= tag.attributes(ck_check_field_aria(metric, "expected_path")) %>>
+      </div>
+
+      <div class="ck-field" data-ck-check-field="score_by">
+        <p class="ck-label" id="metric_check_score_by_label"><%= ck_check_field_label("score_by") %></p>
+        <p class="ck-hint"><%= ck_check_field_hint("score_by") %></p>
+        <%= render "completion_kit/shared/branded_select",
+                   name: "metric[check_config][score_by]",
+                   options: ck_check_score_by_options,
+                   selected: check["score_by"] || CompletionKit::Checks::ListOverlap::DEFAULT_MEASURE,
+                   labelledby: "metric_check_score_by_label",
+                   invalid: ck_check_field_invalid?(metric, "score_by") %>
       </div>
 
       <div class="ck-field-row">
-        <div class="ck-field" data-ck-check-field="min">
-          <label class="ck-label" for="metric_check_min">Shortest allowed</label>
-          <input type="number" name="metric[check_config][min]" id="metric_check_min" class="ck-input" value="<%= check["min"] %>">
-        </div>
-        <div class="ck-field" data-ck-check-field="max">
-          <label class="ck-label" for="metric_check_max">Longest allowed</label>
-          <input type="number" name="metric[check_config][max]" id="metric_check_max" class="ck-input" value="<%= check["max"] %>">
-        </div>
+        <% %w[min max].each do |bound| %>
+          <div class="ck-field" data-ck-check-field="<%= bound %>">
+            <label class="ck-label" for="metric_check_<%= bound %>"><span data-ck-check-label="<%= bound %>"><%= ck_check_field_label(bound, kind) %></span></label>
+            <p class="ck-hint" data-ck-check-hint="<%= bound %>"><%= ck_check_field_hint(bound, kind) %></p>
+            <input type="number" step="any" name="metric[check_config][<%= bound %>]" id="metric_check_<%= bound %>" class="ck-input" value="<%= check[bound] %>" <%= tag.attributes(ck_check_field_aria(metric, bound)) %>>
+          </div>
+        <% end %>
+      </div>
+
+      <div class="ck-field" data-ck-check-field="tolerance">
+        <label class="ck-label" for="metric_check_tolerance"><%= ck_check_field_label("tolerance") %></label>
+        <p class="ck-hint">How far a number can be out and still count as correct. Write a plain number for a fixed amount, or add a percent sign, like 2%, for a share of the expected number. Leave blank to require an exact match.</p>
+        <input type="text" inputmode="decimal" name="metric[check_config][tolerance]" id="metric_check_tolerance" class="ck-input" value="<%= check["tolerance"] %>" <%= tag.attributes(ck_check_field_aria(metric, "tolerance")) %>>
       </div>
 
-      <label class="ck-checkbox" data-ck-check-field="case_sensitive">
-        <input type="checkbox" name="metric[check_config][case_sensitive]" value="true"<%= " checked" if check["case_sensitive"] %>>
-        <span>Case sensitive</span>
-      </label>
-      <label class="ck-checkbox" data-ck-check-field="multiline">
-        <input type="checkbox" name="metric[check_config][multiline]" value="true"<%= " checked" if check["multiline"] %>>
-        <span>Multiline</span>
-      </label>
-      <label class="ck-checkbox" data-ck-check-field="trim">
-        <input type="checkbox" name="metric[check_config][trim]" value="true"<%= " checked" if check["trim"] %>>
-        <span>Trim whitespace</span>
-      </label>
+      <% CompletionKit::Checks::Registry::FLAG_KEYS.each do |flag| %>
+        <label class="ck-checkbox" data-ck-check-field="<%= flag %>">
+          <input type="hidden" name="metric[check_config][<%= flag %>]" value="false">
+          <input type="checkbox" name="metric[check_config][<%= flag %>]" value="true"<%= " checked" if CompletionKit::Checks::Flag.on?(check, flag) %>>
+          <span><%= ck_check_field_label(flag) %></span>
+        </label>
+      <% end %>
     </div>
     <% end %>
 

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

@@ -125,7 +125,7 @@
                   <svg viewBox="0 0 24 24" width="16" height="16" stroke-width="1.75" class="ck-star <%= i < review.ai_score.to_i ? "ck-star--filled" : "ck-star--empty" %>"><polygon points="12 2 15.09 8.26 22 9.27 17 14.14 18.18 21.02 12 17.77 5.82 21.02 7 14.14 2 9.27 8.91 8.26 12 2"/></svg>
                 <% end %>
               <% elsif review.check? && !review.passed.nil? %>
-                <%= ck_check_badge(review.passed) %>
+                <%= ck_check_badge(review.passed, review.check_score) %>
               <% elsif review.status == "failed" %>
                 <span class="<%= ck_badge_classes(:failed) %>" title="<%= review.error_message.presence || (review.check? ? "Check failed" : "Judge failed") %>">Failed</span>
               <% else %>

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

@@ -2,7 +2,7 @@
 <% if has_value && response %>
   <%= link_to run_response_path(run, response), class: "ck-cmp-score__link", title: "#{side}: open this response (#{run.name})" do %>
     <% if kind == "check" %>
-      <%= ck_check_badge(passed) %>
+      <%= ck_check_badge(passed, local_assigns[:check_score]) %>
     <% else %>
       <span class="<%= ck_badge_classes(ck_score_kind(score.to_f)) %>"><%= score %></span>
     <% end %>

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

@@ -161,7 +161,7 @@
         <%= 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" %>">
+            <label class="ck-checkbox-label" data-metric-tags="<%= metric.tag_names.join(",") %>" data-needs-judge="<%= metric.llm_judge? ? "1" : "0" %>" data-compare-expected="<%= metric.check? && metric.check_config.to_h["compare_to"] == "expected" ? "1" : "0" %>">
               <%= check_box_tag "run[metric_ids][]", metric.id, run.selected_metric_ids.include?(metric.id), class: "ck-checkbox", id: "run_metric_#{metric.id}" %>
               <span class="ck-checkbox-label__box" aria-hidden="true"></span>
               <span class="ck-checkbox-label__body">
@@ -209,9 +209,15 @@ function updateRunForm() {
   if (judgeHint) judgeHint.textContent = '';
   if (metricsHint) metricsHint.textContent = '';
 
+  var judgedMetrics = Array.prototype.filter.call(metrics, function(cb) {
+    var lbl = cb.closest('[data-needs-judge]');
+    return !!lbl && lbl.dataset.needsJudge === '1';
+  });
   var judgeUnpaired = judgeListed && !!judge && metrics.length === 0;
-  var metricsUnpaired = judgeListed && !judge && metrics.length > 0;
-  if (judgeListed && judge && metrics.length > 0) {
+  var metricsUnpaired = judgeListed && !judge && judgedMetrics.length > 0;
+  if (judgeListed && judge && metrics.length > 0 && judgedMetrics.length === 0) {
+    if (judgeHint) judgeHint.textContent = 'Checks need no judge, so this run will not use it.';
+  } else 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';

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

@@ -21,9 +21,10 @@
               <span class="ck-metric-pip__label" aria-hidden="true"><%= r.metric_name %> <strong><%= r.ai_score %></strong></span>
             </span>
           <% else %>
-            <span class="ck-metric-pip ck-metric-pip--<%= ck_pass_rate_kind(r.passed ? 1.0 : 0.0) %>" role="img" aria-label="<%= r.metric_name %> <%= r.passed ? "Pass" : "Fail" %>">
+            <% verdict = ck_check_verdict(r.passed, r.check_score) %>
+            <span class="ck-metric-pip ck-metric-pip--<%= ck_pass_rate_kind(r.passed ? 1.0 : 0.0) %>" role="img" aria-label="<%= r.metric_name %> <%= verdict %>">
               <span class="ck-metric-pip__bar"></span>
-              <span class="ck-metric-pip__label" aria-hidden="true"><%= r.metric_name %> <strong><%= r.passed ? "Pass" : "Fail" %></strong></span>
+              <span class="ck-metric-pip__label" aria-hidden="true"><%= r.metric_name %> <strong><%= verdict %></strong></span>
             </span>
           <% end %>
         <% end %>

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

@@ -36,9 +36,9 @@
       <div class="ck-metric-bar ck-metric-bar--compact">
         <% metrics.each do |m| %>
           <% if m[:kind] == "check" %>
-            <span class="ck-metric-pip ck-metric-pip--<%= ck_pass_rate_kind(m[:pass_rate]) %>" role="img" aria-label="<%= m[:name] %> <%= (m[:pass_rate] * 100).round %>% pass rate">
+            <span class="ck-metric-pip ck-metric-pip--<%= ck_pass_rate_kind(m[:pass_rate]) %>" role="img" aria-label="<%= ck_check_average_label(m[:name], m) %>">
               <span class="ck-metric-pip__bar"></span>
-              <span class="ck-metric-pip__label" aria-hidden="true"><%= m[:name] %> <strong><%= (m[:pass_rate] * 100).round %>%</strong></span>
+              <span class="ck-metric-pip__label" aria-hidden="true"><%= m[:name] %> <strong><%= ck_check_average(m) %></strong></span>
             </span>
           <% else %>
             <span class="ck-metric-pip ck-metric-pip--<%= ck_score_kind(m[:avg]) %>" role="img" aria-label="<%= m[:name] %> average score <%= m[:avg] %>">
@@ -56,12 +56,14 @@
     <% avg = run.avg_score %>
     <% pass_rate = run.check_pass_rate %>
     <% if avg || pass_rate %>
-      <% if avg %>
-        <span class="<%= ck_badge_classes(ck_score_kind(avg)) %>"><%= avg %></span>
-      <% end %>
-      <% if pass_rate %>
-        <span class="<%= ck_badge_classes(ck_pass_rate_kind(pass_rate)) %>"><%= (pass_rate * 100).round %>%</span>
-      <% end %>
+      <div class="ck-chip-row">
+        <% if avg %>
+          <span class="<%= ck_badge_classes(ck_score_kind(avg)) %>"><%= avg %></span>
+        <% end %>
+        <% if pass_rate %>
+          <span class="<%= ck_badge_classes(ck_pass_rate_kind(pass_rate)) %>"><%= (pass_rate * 100).round %>%</span>
+        <% end %>
+      </div>
     <% else %>
       <span class="ck-runs-table__dim">—</span>
     <% end %>

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

@@ -19,7 +19,7 @@
     <div class="ck-run-config__row">
       <span class="ck-run-config__key">Judge</span>
       <span style="text-transform: none;"><%= run.judge_model %></span>
-      <% unless run.judge_configured? %>
+      <% if run.llm_metrics.any? && !run.llm_judge_configured? %>
         <span class="ck-run-config__warn">provider not configured</span>
       <% end %>
     </div>

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

@@ -41,9 +41,9 @@
                   <% run.metrics.order(:name).each do |metric| %>
                     <% avg_for_metric = metric_lookup[metric.name.to_s.downcase] %>
                     <% if avg_for_metric && avg_for_metric[:kind] == "check" %>
-                      <span class="ck-metric-pip ck-metric-pip--<%= ck_pass_rate_kind(avg_for_metric[:pass_rate]) %>">
+                      <span class="ck-metric-pip ck-metric-pip--<%= ck_pass_rate_kind(avg_for_metric[:pass_rate]) %>" title="<%= ck_check_average_label(metric.name, avg_for_metric) %>">
                         <span class="ck-metric-pip__bar"></span>
-                        <span class="ck-metric-pip__label"><%= metric.name %> <strong><%= (avg_for_metric[:pass_rate] * 100).round %>%</strong></span>
+                        <span class="ck-metric-pip__label"><%= metric.name %> <strong><%= ck_check_average(avg_for_metric) %></strong></span>
                       </span>
                     <% elsif avg_for_metric %>
                       <span class="ck-metric-pip ck-metric-pip--<%= ck_score_kind(avg_for_metric[:avg]) %>">

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

@@ -56,18 +56,22 @@
             <td class="ck-compare-metric" data-label="Metric" title="<%= pm[:metric_name] %>"><%= pm[:metric_name] %></td>
             <td class="ck-cmp-score" data-label="A → B">
               <span class="ck-cmp-score__pair">
-                <%= render "compare_score", kind: pm[:kind], passed: pm[:left_passed], score: pm[:left_score], response: row[:left_response], run: @run, side: "A" %>
+                <%= render "compare_score", kind: pm[:kind], passed: pm[:left_passed], score: pm[:left_score], check_score: pm[:left_check_score], response: row[:left_response], run: @run, side: "A" %>
                 <span class="ck-cmp-score__arrow" aria-hidden="true">→</span>
-                <%= render "compare_score", kind: pm[:kind], passed: pm[:right_passed], score: pm[:right_score], response: row[:right_response], run: @other_run, side: "B" %>
+                <%= render "compare_score", kind: pm[:kind], passed: pm[:right_passed], score: pm[:right_score], check_score: pm[:right_check_score], response: row[:right_response], run: @other_run, side: "B" %>
               </span>
             </td>
             <td class="ck-num ck-cmp-delta" data-label="Δ">
               <% if pm[:kind] == "check" %>
-                <%= ck_result_change_badge(pm[:result_change]) || content_tag(:span, "—", class: "ck-cmp-score__none") %>
+                <% points = ck_check_score_change(pm[:left_check_score], pm[:right_check_score]) %>
+                <% badge = ck_result_change_badge(pm[:result_change]) unless points && pm[:result_change] == "same" %>
+                <% if badge || points %>
+                  <span class="ck-cmp-delta__pair"><%= badge %><%= ck_delta_chip(points, "%") if points %></span>
+                <% else %>
+                  <span class="ck-cmp-score__none">—</span>
+                <% end %>
               <% elsif pm[:delta] %>
-                <% d = pm[:delta] %>
-                <% dcls = d > 0 ? "ck-delta--positive" : d < 0 ? "ck-delta--negative" : "ck-delta--zero" %>
-                <span class="ck-delta <%= dcls %>"><span class="ck-delta__arrow" aria-hidden="true"><%= d > 0 ? "▲" : d < 0 ? "▼" : "=" %></span><%= d > 0 ? "+#{d}" : d.to_s %></span>
+                <%= ck_delta_chip(pm[:delta]) %>
               <% else %>
                 <span class="ck-cmp-score__none">—</span>
               <% end %>

data/app/views/completion_kit/shared/_branded_select.html.erb

@@ -2,7 +2,7 @@
 <div class="ck-select" data-ck-select>
   <input type="hidden" name="<%= name %>" value="<%= chosen && chosen.first %>" data-ck-select-value>
   <button type="button" class="ck-select__trigger" data-ck-select-trigger
-          aria-haspopup="listbox" aria-expanded="false" aria-labelledby="<%= labelledby %>">
+          aria-haspopup="listbox" aria-expanded="false" aria-labelledby="<%= labelledby %>" <%= tag.attributes("aria-invalid": ("true" if local_assigns[:invalid])) %>>
     <span class="ck-select__label" data-ck-select-label><%= chosen && chosen.last %></span>
     <span class="ck-select__chevron" aria-hidden="true"><%= heroicon_tag "chevron-down" %></span>
   </button>

data/config/locales/en.yml

@@ -6,6 +6,8 @@ en:
       completion_kit/run:
         judge_model: Judge model
         metric_ids: Metrics
+      completion_kit/metric:
+        check_config: Check
       completion_kit/metric_group:
         metric_ids: Metrics
       completion_kit/provider_credential:

data/db/migrate/20260903000001_add_check_score_to_completion_kit_reviews.rb

@@ -0,0 +1,5 @@
+class AddCheckScoreToCompletionKitReviews < ActiveRecord::Migration[8.1]
+  def change
+    add_column :completion_kit_reviews, :check_score, :decimal, precision: 5, scale: 4
+  end
+end

data/lib/completion_kit/version.rb

@@ -1,3 +1,3 @@
 module CompletionKit
-  VERSION = "0.28.46"
+  VERSION = "0.29.0"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: completion-kit
 version: !ruby/object:Gem::Version
-  version: 0.28.46
+  version: 0.29.0
 platform: ruby
 authors:
 - Damien Bastin
@@ -264,13 +264,20 @@ files:
 - app/services/completion_kit/checks/contains.rb
 - app/services/completion_kit/checks/equals.rb
 - app/services/completion_kit/checks/expected_resolver.rb
+- app/services/completion_kit/checks/flag.rb
 - app/services/completion_kit/checks/json_path_equals.rb
+- app/services/completion_kit/checks/labels.rb
 - app/services/completion_kit/checks/length_bounds.rb
+- app/services/completion_kit/checks/list_overlap.rb
 - app/services/completion_kit/checks/not_contains.rb
+- app/services/completion_kit/checks/numeric_bounds.rb
+- app/services/completion_kit/checks/numeric_equals.rb
+- app/services/completion_kit/checks/numeric_value.rb
 - app/services/completion_kit/checks/regex.rb
 - app/services/completion_kit/checks/registry.rb
 - app/services/completion_kit/checks/result.rb
 - app/services/completion_kit/checks/target_resolver.rb
+- app/services/completion_kit/checks/textual.rb
 - app/services/completion_kit/checks/valid_json.rb
 - app/services/completion_kit/csv_processor.rb
 - app/services/completion_kit/dashboard_stats.rb
@@ -452,6 +459,7 @@ 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/20260903000001_add_check_score_to_completion_kit_reviews.rb
 - db/migrate/20261013000001_drop_completion_kit_mcp_sessions.rb
 - lib/completion-kit.rb
 - lib/completion_kit.rb