completion-kit 0.5.42 → 0.5.43

data/app/controllers/completion_kit/metrics_controller.rb

@@ -1,7 +1,7 @@
 module CompletionKit
   class MetricsController < ApplicationController
     include CompletionKit::TagFiltering
-    before_action :set_metric, only: [:show, :edit, :update, :destroy, :add_few_shot, :publish_draft, :suggest_variants, :dismiss_suggestion]
+    before_action :set_metric, only: [:show, :edit, :update, :destroy, :add_few_shot, :remove_few_shot, :publish_draft, :suggest_variants, :dismiss_suggestion]
 
     def index
       @metrics = apply_tag_filter(Metric.includes(:metric_groups, :tags).order(:name))
@@ -35,14 +35,16 @@ module CompletionKit
     end
 
     def show
+      @published_judge_version = JudgeVersion.ensure_current_for(@metric)
       @disagreements = Calibration.where(metric_id: @metric.id, verdict: "disagree")
-                                  .includes(response: [:reviews, :run])
+                                  .includes(:judge_version, response: [:reviews, :run])
                                   .order(created_at: :desc)
                                   .limit(50)
       @edit_draft = JudgeVersion.drafts.where(metric_id: @metric.id, source: "edit").order(created_at: :desc).first
-      @published_judge_version = JudgeVersion.published.where(metric_id: @metric.id, current: true).first
       @suggestion_draft = JudgeVersion.drafts.where(metric_id: @metric.id, source: "suggestion").order(created_at: :desc).first
-      @improve_disagreement_count = @disagreements.size
+      @improve_disagreement_count = Calibration.where(metric_id: @metric.id, verdict: "disagree",
+                                                      judge_version_id: @published_judge_version.id).count
+      @versions = JudgeVersion.where(metric_id: @metric.id).order(version_number: :desc).to_a
     end
 
     def new
@@ -50,6 +52,9 @@ module CompletionKit
     end
 
     def edit
+      @suggestion_draft = JudgeVersion.drafts.where(metric_id: @metric.id, source: "suggestion").order(created_at: :desc).first
+      @edit_draft = JudgeVersion.drafts.where(metric_id: @metric.id, source: "edit").order(created_at: :desc).first
+      @published_judge_version = JudgeVersion.published.where(metric_id: @metric.id, current: true).first
     end
 
     def create
@@ -76,9 +81,10 @@ module CompletionKit
     end
 
     def suggest_variants
+      target = params[:back_to] == "edit" ? edit_metric_path(@metric) : metric_path(@metric)
       disagreement_count = Calibration.where(metric_id: @metric.id, verdict: "disagree").count
       if disagreement_count.zero?
-        redirect_to metric_path(@metric), alert: "Mark at least one row as Disagree before asking the model to suggest a change."
+        redirect_to target, alert: "Mark at least one row as Disagree before asking the model to suggest a change."
         return
       end
 
@@ -87,38 +93,36 @@ module CompletionKit
       generator = JudgeVariantGenerator.new(@metric, count: 1)
       variants = generator.call
       if variants.empty?
-        redirect_to metric_path(@metric), alert: "The model returned no usable variants. Try again with a different model."
+        redirect_to target, alert: "The model returned no usable variants. Try again with a different model."
         return
       end
       generator.persist!(variants)
-      redirect_to metric_path(@metric), notice: "Drafted a new version. Review it below."
+      redirect_to target, notice: "Drafted a new version. Review it below."
     end
 
     def dismiss_suggestion
-      draft = JudgeVersion.drafts.where(metric_id: @metric.id, source: "suggestion").find_by(id: params[:draft_id])
+      draft = JudgeVersion.drafts.where(metric_id: @metric.id).find_by(id: params[:draft_id])
       draft&.destroy
-      redirect_to metric_path(@metric), notice: "Dismissed."
+      target = params[:back_to] == "edit" ? edit_metric_path(@metric) : metric_path(@metric)
+      redirect_to target, notice: "Dismissed."
     end
 
     def publish_draft
-      scope = JudgeVersion.drafts.where(metric_id: @metric.id)
-      draft = params[:draft_id].present? ? scope.find_by(id: params[:draft_id]) : scope.order(created_at: :desc).first
-
-      if draft.nil?
-        redirect_to metric_path(@metric), alert: "No draft to publish."
+      scope = JudgeVersion.where(metric_id: @metric.id)
+      version = if params[:draft_id].present?
+                  scope.find_by(id: params[:draft_id])
+                else
+                  JudgeVersion.drafts.where(metric_id: @metric.id).order(created_at: :desc).first
+                end
+
+      if version.nil?
+        redirect_to metric_path(@metric), alert: "No version to publish."
         return
       end
 
-      JudgeVersion.transaction do
-        JudgeVersion.where(metric_id: @metric.id, state: "published").update_all(current: false)
-        draft.update!(state: "published", current: true)
-        @metric.update_columns(
-          instruction: draft.instruction,
-          rubric_bands: Array(draft.rubric_bands).to_json
-        )
-      end
-
-      redirect_to metric_path(@metric), notice: "This judge version is now live."
+      version.publish!
+      redirect_to metric_path(@metric),
+                  notice: "#{@metric.name} #{version.version_label} is now the published version."
     end
 
     def add_few_shot
@@ -139,6 +143,13 @@ module CompletionKit
       redirect_to metric_path(@metric), notice: "Got it. The judge will remember this next time it grades."
     end
 
+    def remove_few_shot
+      cal_id = params[:calibration_id].to_i
+      remaining = Array(@metric.few_shot_examples).reject { |fs| fs["calibration_id"].to_i == cal_id }
+      @metric.update!(few_shot_examples: remaining)
+      redirect_to metric_path(@metric), notice: "Forgotten. The judge won't see this case next time."
+    end
+
     private
 
     def set_metric

data/app/jobs/completion_kit/judge_review_job.rb

@@ -57,6 +57,7 @@ module CompletionKit
         run.prompt&.template,
         criteria: metric.instruction.to_s,
         rubric_text: metric.display_rubric_text,
+        human_examples: few_shot_payload(metric),
         input_data: response.input_data
       )
 
@@ -119,5 +120,15 @@ module CompletionKit
       response = Response.find_by(id: response_id)
       RunCompletionCheckJob.perform_later(response.run_id) if response
     end
+
+    def few_shot_payload(metric)
+      Array(metric.few_shot_examples).map do |fs|
+        {
+          human_score: fs["human_score"],
+          response_text: fs["response"].to_s,
+          human_note: fs["human_note"].to_s
+        }
+      end
+    end
   end
 end

data/app/models/completion_kit/judge_version.rb

@@ -7,8 +7,11 @@ module CompletionKit
 
     serialize :rubric_bands, coder: JSON
 
+    before_validation :assign_version_number, on: :create
+
     validates :metric_id, presence: true
     validates :state, inclusion: { in: STATES }
+    validates :version_number, presence: true, uniqueness: { scope: :metric_id }
 
     scope :current, -> { where(current: true) }
     scope :published, -> { where(state: "published") }
@@ -20,7 +23,8 @@ module CompletionKit
         instruction: metric.instruction,
         rubric_bands: metric.rubric_bands,
         current: true,
-        state: "published"
+        state: "published",
+        published_at: Time.current
       )
     end
 
@@ -32,17 +36,44 @@ module CompletionKit
       state == "published"
     end
 
+    def version_label
+      "v#{version_number}"
+    end
+
+    def publish!
+      JudgeVersion.transaction do
+        self.class.where(metric_id: metric_id).where.not(id: id).update_all(current: false)
+        reload
+        update!(state: "published", current: true, published_at: published_at || Time.current)
+        metric.update_columns(
+          instruction: instruction,
+          rubric_bands: Array(rubric_bands).to_json
+        )
+      end
+      self
+    end
+
     def as_json(options = {})
       {
         id: id,
         metric_id: metric_id,
+        version_number: version_number,
         instruction: instruction,
         rubric_bands: rubric_bands,
         current: current,
         state: state,
         source: source,
+        published_at: published_at,
         created_at: created_at
       }
     end
+
+    private
+
+    def assign_version_number
+      return if version_number.present?
+      max = self.class.where(metric_id: metric_id).maximum(:version_number).to_i
+      self.version_number = max + 1
+    end
   end
 end

data/app/services/completion_kit/judge_variant_generator.rb

@@ -86,7 +86,7 @@ module CompletionKit
       sections << "REASONING: <one short sentence: what changes and why>"
       sections << "INSTRUCTION:"
       sections << "<the rewritten instruction>"
-      sections << "RUBRIC:                  # optional — omit this block if the rubric is unchanged"
+      sections << "RUBRIC:                  # optional. Omit this block if the rubric is unchanged."
       sections << "5: <description for 5 stars>"
       sections << "4: <description for 4 stars>"
       sections << "3: <description for 3 stars>"
@@ -133,11 +133,13 @@ module CompletionKit
     end
 
     def calibrations_for(metric, verdict:, limit:)
-      Calibration.where(metric_id: metric.id, verdict: verdict)
-                 .includes(response: :reviews)
-                 .order(created_at: :desc)
-                 .limit(limit)
-                 .map do |cal|
+      scope = Calibration.where(metric_id: metric.id, verdict: verdict)
+      current_version = JudgeVersion.current.find_by(metric_id: metric.id)
+      scope = scope.where(judge_version_id: current_version.id) if current_version
+      scope.includes(response: :reviews)
+           .order(created_at: :desc)
+           .limit(limit)
+           .map do |cal|
         review = cal.response.reviews.find { |r| r.metric_id == metric.id }
         {
           input: cal.response.input_data,

data/app/services/completion_kit/metric_calibration_stats.rb

@@ -31,18 +31,30 @@ module CompletionKit
       end
     end
 
-    def self.for(metric, judge_version: nil)
-      new(metric: metric, judge_version: judge_version).call
+    CURRENT = :current
+
+    def self.for(metric, judge_version: CURRENT)
+      resolved = case judge_version
+                 when CURRENT then JudgeVersion.current.find_by(metric_id: metric.id)
+                 when nil then nil
+                 else judge_version
+                 end
+      new(metric: metric, judge_version: resolved, all_versions: judge_version.nil?).call
     end
 
-    def initialize(metric:, judge_version: nil)
+    def initialize(metric:, judge_version: nil, all_versions: false)
       @metric = metric
       @judge_version = judge_version
+      @all_versions = all_versions
     end
 
     def call
       scope = Calibration.where(metric_id: @metric.id)
-      scope = scope.where(judge_version_id: @judge_version.id) if @judge_version
+      if @judge_version
+        scope = scope.where(judge_version_id: @judge_version.id)
+      elsif !@all_versions
+        scope = scope.none
+      end
 
       verdicts = scope.pluck(:verdict, :corrected_score, :response_id)
       n = verdicts.length

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

@@ -187,7 +187,7 @@
       </div>
       <%= render "completion_kit/api_reference/resource_list", title: "Your datasets",
             items: datasets.map { |d|
-              { name: d.name, meta: pluralize([d.csv_data.to_s.lines.count - 1, 0].max, "row"),
+              { name: d.name, meta: pluralize([d.csv_data.to_s.lines.count - 1, 0].max, "entry"),
                 url: "#{base_url}/api/v1/datasets/#{d.id}", dom_id: "dataset_ep_#{d.id}" }
             } %>
     </div>

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

@@ -3,14 +3,19 @@
   <% pending_verdict = local_assigns[:pending_verdict] %>
   <% active_verdict = pending_verdict || current_verdict %>
   <% error = local_assigns[:error] %>
-  <% verdict_count = CompletionKit::Calibration.where(response_id: response_row.id, metric_id: metric.id).count %>
+  <% me = CompletionKit.config.username.presence || "operator" %>
+  <% other_calibrations = CompletionKit::Calibration
+        .where(response_id: response_row.id, metric_id: metric.id)
+        .where.not(created_by: me)
+        .order(created_at: :asc).to_a %>
+  <% verdict_icons = { "agree" => "hand-thumb-up", "disagree" => "hand-thumb-down", "borderline" => "scale" } %>
   <p class="ck-calibration__prompt">
     <span class="ck-calibration__label">Your verdict</span>
-    <% if verdict_count > 0 %>
-      <span class="ck-calibration__count"><%= pluralize(verdict_count, "verdict") %> on this score · <%= link_to "trust level →", metric_path(metric), class: "ck-link" %></span>
-    <% else %>
-      <span class="ck-calibration__hint">Tell us what you think — was the score right? Verdicts roll up into the metric's <%= link_to "trust level", metric_path(metric), class: "ck-link" %>.</span>
+    <% if other_calibrations.any? %>
+      <span class="ck-calibration__meta"><%= pluralize(other_calibrations.size, "other verdict") %> on this score</span>
+      <span class="ck-calibration__sep">·</span>
     <% end %>
+    <%= link_to metric_path(metric), class: "ck-calibration__meta-link" do %><%= heroicon_tag "adjustments-horizontal", variant: :outline, "aria-hidden": "true" %>Calibration →<% end %>
   </p>
   <div class="ck-calibration__buttons">
     <% verdict_icons = { "agree" => "hand-thumb-up", "disagree" => "hand-thumb-down", "borderline" => "scale" } %>
@@ -36,6 +41,38 @@
     <p class="ck-calibration__error" role="alert"><%= error %></p>
   <% end %>
 
+  <% if other_calibrations.any? %>
+    <details class="ck-calibration__others">
+      <summary class="ck-calibration__others-summary">
+        <%= heroicon_tag "chevron-right", variant: :outline, size: 14, "aria-hidden": "true" %>
+        <span>What others said (<%= other_calibrations.size %>)</span>
+      </summary>
+      <ul class="ck-calibration__others-list">
+        <% other_calibrations.each do |other| %>
+          <li class="ck-calibration__others-item ck-calibration__others-item--<%= other.verdict %>">
+            <div class="ck-calibration__others-row">
+              <span class="ck-calibration__others-verdict">
+                <%= heroicon_tag verdict_icons[other.verdict], variant: :outline, size: 14, "aria-hidden": "true" %>
+                <%= other.verdict %>
+              </span>
+              <span class="ck-calibration__others-by"><%= other.created_by %></span>
+              <% if other.corrected_score %>
+                <span class="ck-calibration__others-stars" aria-label="<%= pluralize(other.corrected_score.to_i, 'star') %>" title="<%= pluralize(other.corrected_score.to_i, 'star') %>">
+                  <% 5.times do |i| %>
+                    <svg viewBox="0 0 24 24" width="12" height="12" stroke-width="1.75" class="ck-star <%= i < other.corrected_score.to_i ? "ck-star--filled" : "ck-star--empty" %>" aria-hidden="true"><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 %>
+                </span>
+              <% end %>
+            </div>
+            <% if other.note.to_s.present? %>
+              <p class="ck-calibration__others-note">"<%= other.note.to_s.truncate(140) %>"</p>
+            <% end %>
+          </li>
+        <% end %>
+      </ul>
+    </details>
+  <% end %>
+
   <% if active_verdict == "disagree" %>
     <% existing_score = (calibration&.corrected_score || review&.ai_score)&.round %>
     <%= form_with url: run_response_calibrations_path(run, response_row),
@@ -51,7 +88,7 @@
             <% radio_id = "ck-star-#{response_row.id}-#{metric.id}-#{n}" %>
             <input type="radio" name="corrected_score" id="<%= radio_id %>" value="<%= n %>" <%= "checked" if existing_score == n %> required>
             <label for="<%= radio_id %>" title="<%= pluralize(n, 'star') %>" aria-label="<%= pluralize(n, 'star') %>">
-              <svg viewBox="0 0 24 24" width="28" height="28" stroke-width="1.5" aria-hidden="true"><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>
+              <svg viewBox="0 0 24 24" stroke-width="1.5" aria-hidden="true"><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>
             </label>
           <% end %>
         </div>

data/app/views/completion_kit/calibrations/_trust_panel.html.erb

@@ -1,31 +1,30 @@
 <% stats = local_assigns[:stats] %>
-<div class="ck-trust-panel ck-trust-panel--<%= stats.gate %>">
-  <p class="ck-trust-panel__label" title="How often this metric's scores match the humans who reviewed them.">Trust level</p>
-  <% if stats.counter_only? %>
-    <div class="ck-trust-panel__body">
-      <span class="ck-trust-panel__counter"><%= stats.sample_size %><span class="ck-trust-panel__counter-of">/ <%= CompletionKit::MetricCalibrationStats::PROVISIONAL_MIN %></span></span>
-      <span class="ck-trust-panel__hint">verdicts so far<% if stats.short_to_target > 0 %> · <%= stats.short_to_target %> more before we can score the judge<% end %></span>
-    </div>
+<% metric = local_assigns[:metric] %>
+<% anchor = metric&.name&.parameterize %>
+<% target_response = if (stats.sample_size.zero? || stats.counter_only?) && metric
+     created_by = CompletionKit.config.username.presence || "operator"
+     verdicted_ids = CompletionKit::Calibration.where(metric_id: metric.id, created_by: created_by).pluck(:response_id)
+     CompletionKit::Response.joins(:reviews)
+       .where(reviews: { metric_id: metric.id })
+       .where.not(reviews: { ai_score: nil })
+       .where.not(id: verdicted_ids)
+       .order(created_at: :desc).first
+   end %>
+
+<p class="ck-trust-line ck-trust-line--<%= stats.gate %>">
+  <span class="ck-trust-line__label"><%= heroicon_tag "adjustments-horizontal", variant: :outline, "aria-hidden": "true" %>Calibration</span>
+  <% if stats.sample_size.zero? %>
+    <span class="ck-trust-line__state">Not measured yet.</span>
+    <span class="ck-trust-line__hint">Needs <%= CompletionKit::MetricCalibrationStats::PROVISIONAL_MIN %> verdicts on the judge's scores.<% if target_response %>
+      <%= link_to "Give a verdict →", run_response_path(target_response.run, target_response, anchor: anchor), class: "ck-trust-line__link" %>
+    <% else %> Run this metric on a dataset, then give a verdict.<% end %></span>
+  <% elsif stats.counter_only? %>
+    <span class="ck-trust-line__counter"><%= stats.sample_size %>/<%= CompletionKit::MetricCalibrationStats::PROVISIONAL_MIN %></span>
+    <span class="ck-trust-line__hint"><%= pluralize(stats.sample_size, "verdict") %><% if stats.short_to_target > 0 %> · <%= stats.short_to_target %> more before this can be measured<% end %><% if target_response %> · <%= link_to "Give another verdict →", run_response_path(target_response.run, target_response, anchor: anchor), class: "ck-trust-line__link" %><% end %></span>
   <% else %>
-    <div class="ck-trust-panel__body">
-      <span class="ck-trust-panel__score"
-            title="Roughly how often the judge and the humans agreed.">~<%= (stats.agreement_point * 100).round %><span class="ck-trust-panel__score-pct">%</span></span>
-      <span class="ck-trust-panel__margin"
-            title="The range we're confident the true rate sits in, given how few verdicts we have so far.">±<%= (stats.margin * 100).round %> pt</span>
-      <span class="ck-trust-panel__gate"
-            title="<%= stats.firm? ? 'Enough verdicts for a reliable read.' : 'Early read. Keep giving verdicts to tighten the margin.' %>"><%= stats.firm? ? "settled" : "early read" %></span>
-    </div>
-    <div class="ck-trust-panel__details">
-      <span><%= pluralize(stats.sample_size, "verdict") %></span>
-      <% if stats.borderline_rate && stats.borderline_rate > 0 %>
-        <% level = if stats.borderline_rate > 0.30 then "danger"
-                   elsif stats.borderline_rate > 0.15 then "warning"
-                   else "ok" end %>
-        <span class="ck-trust-panel__borderline ck-trust-panel__borderline--<%= level %>"
-              title="<%= level == 'ok' ? 'Some reviewers said the rubric was unclear here.' : 'A lot of reviewers say the rubric is unclear here. Consider splitting the metric or rewriting the rubric.' %>">
-          <%= (stats.borderline_rate * 100).round %>% said "unclear"
-        </span>
-      <% end %>
-    </div>
+    <span class="ck-trust-line__score" title="Roughly how often the judge and the humans agreed.">~<%= (stats.agreement_point * 100).round %>%</span>
+    <span class="ck-trust-line__margin" title="The range we're confident the true rate sits in.">±<%= (stats.margin * 100).round %> pt</span>
+    <span class="ck-trust-line__gate" title="<%= stats.firm? ? 'Enough verdicts for a reliable read.' : 'Early read. Keep giving verdicts.' %>"><%= stats.firm? ? "settled" : "early" %></span>
+    <span class="ck-trust-line__hint"><%= pluralize(stats.sample_size, "verdict") %><% if stats.borderline_rate && stats.borderline_rate > 0 %><% level = stats.borderline_rate > 0.30 ? "danger" : stats.borderline_rate > 0.15 ? "warning" : "ok" %> · <span class="ck-trust-line__borderline ck-trust-line__borderline--<%= level %>" title="<%= level == 'ok' ? '' : 'Reviewers said the rubric was unclear here.' %>"><%= (stats.borderline_rate * 100).round %>% unclear</span><% end %></span>
   <% end %>
-</div>
+</p>

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

@@ -1,3 +1,9 @@
+<% suggestion = local_assigns[:suggestion_draft] %>
+<% edit_draft = local_assigns[:edit_draft] %>
+<% suggestion_bands = suggestion ? Array(suggestion.rubric_bands).each_with_object({}) { |b, h| h[b["stars"].to_i] = b["description"].to_s } : {} %>
+<% suggested_instruction = suggestion&.instruction.to_s %>
+<% instruction_changed = suggestion && suggested_instruction.present? && suggested_instruction != metric.instruction.to_s %>
+
 <%= form_with(model: metric, local: true) do |form| %>
   <% if metric.errors.any? %>
     <div class="ck-flash ck-flash--alert" role="alert">
@@ -10,6 +16,50 @@
     </div>
   <% end %>
 
+  <% if edit_draft && !suggestion %>
+    <% pub = local_assigns[:published_judge_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 %>
+    <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">An unpublished draft of this metric is saved. 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>
+      </div>
+      <div class="ck-suggestion-banner__actions">
+        <%= button_to "Discard draft", dismiss_suggestion_metric_path(metric, draft_id: edit_draft.id, back_to: "edit"),
+              method: :delete, form_class: "inline-block",
+              class: ck_button_classes(:light, variant: :outline),
+              data: { turbo_confirm: "Drop this draft?" } %>
+        <%= button_to "Publish this version", publish_draft_metric_path(metric, draft_id: edit_draft.id),
+              method: :post, form_class: "inline-block",
+              class: ck_button_classes(:dark) %>
+      </div>
+    </div>
+  <% end %>
+
+  <% if suggestion %>
+    <div class="ck-suggestion-banner" role="status">
+      <div class="ck-suggestion-banner__body">
+        <p class="ck-kicker">Proposed improvements</p>
+        <p class="ck-meta-copy">Based on your disagreements, the model proposed these changes to the instruction and rubric. Apply pieces inline below, take everything at once, try again, or discard.</p>
+      </div>
+      <div class="ck-suggestion-banner__actions">
+        <%= button_to "Try again", suggest_variants_metric_path(metric, back_to: "edit"),
+              method: :post, form_class: "inline-block",
+              class: ck_button_classes(:light, variant: :outline),
+              title: "Discard these improvements and ask the model for fresh ones.",
+              data: { turbo_confirm: "Replace these improvements with fresh ones from the model?" } %>
+        <%= button_to "Discard", dismiss_suggestion_metric_path(metric, draft_id: suggestion.id, back_to: "edit"),
+              method: :delete, form_class: "inline-block",
+              class: ck_button_classes(:light, variant: :outline),
+              data: { turbo_confirm: "Drop these improvements?" } %>
+        <%= button_to "Take everything", publish_draft_metric_path(metric, draft_id: suggestion.id),
+              method: :post, form_class: "inline-block",
+              class: ck_button_classes(:dark) %>
+      </div>
+    </div>
+  <% end %>
+
   <div class="ck-card ck-form-card">
     <div class="ck-field">
       <%= form.label :name, "Metric name", class: "ck-label" %>
@@ -22,14 +72,34 @@
       <p class="ck-hint">What should the judge assess? This instruction is sent to the LLM judge when scoring outputs.</p>
       <%= form.text_area :instruction, rows: 8, class: "ck-input ck-input--area", placeholder: "Evaluate whether the output...", **ck_field_aria(form, :instruction) %>
       <%= ck_field_error(form, :instruction) %>
+
+      <% if instruction_changed %>
+        <div class="ck-inline-suggestion">
+          <div class="ck-inline-suggestion__head">
+            <p class="ck-kicker">Suggested wording</p>
+            <button type="button"
+                    class="<%= ck_button_classes(:light, variant: :outline) %> ck-inline-suggestion__apply"
+                    data-ck-apply
+                    data-target="metric[instruction]"
+                    data-value="<%= h(suggested_instruction) %>">Use this wording</button>
+          </div>
+          <div class="ck-inline-suggestion__diff">
+            <pre class="ck-inline-suggestion__pane ck-inline-suggestion__pane--before"><%= ck_word_diff_old(metric.instruction.to_s, suggested_instruction) %></pre>
+            <pre class="ck-inline-suggestion__pane ck-inline-suggestion__pane--after"><%= ck_word_diff_new(metric.instruction.to_s, suggested_instruction) %></pre>
+          </div>
+        </div>
+      <% end %>
     </div>
 
     <div class="ck-field ck-field--spacious">
-      <p class="ck-section-title">Rubric</p>
+      <p class="ck-section-title">Rubric<%= render "completion_kit/metrics/rubric_hint" %></p>
       <p class="ck-hint">What each star rating means for this metric.</p>
 
       <div class="ck-rubric-builder">
         <% metric.rubric_bands_for_form.each_with_index do |band, index| %>
+          <% suggested_band = suggestion_bands[band["stars"].to_i].to_s %>
+          <% band_changed = suggestion && suggested_band.present? && suggested_band != band["description"].to_s %>
+          <% target_name = "metric[rubric_bands][#{index}][description]" %>
           <div class="ck-rubric-row">
             <div class="ck-rubric-row__stars">
               <% 5.times do |i| %>
@@ -38,7 +108,23 @@
               <input type="hidden" name="metric[rubric_bands][<%= index %>][stars]" value="<%= band["stars"] %>">
             </div>
             <div class="ck-rubric-row__fields">
-              <textarea name="metric[rubric_bands][<%= index %>][description]" rows="2" class="ck-input ck-input--area"><%= band["description"] %></textarea>
+              <textarea name="<%= target_name %>" rows="2" class="ck-input ck-input--area"><%= band["description"] %></textarea>
+              <% if band_changed %>
+                <div class="ck-inline-suggestion ck-inline-suggestion--band">
+                  <div class="ck-inline-suggestion__head">
+                    <p class="ck-kicker">Suggested band</p>
+                    <button type="button"
+                            class="<%= ck_button_classes(:light, variant: :outline) %> ck-inline-suggestion__apply"
+                            data-ck-apply
+                            data-target="<%= target_name %>"
+                            data-value="<%= h(suggested_band) %>">Use this band</button>
+                  </div>
+                  <div class="ck-inline-suggestion__diff">
+                    <pre class="ck-inline-suggestion__pane ck-inline-suggestion__pane--before"><%= ck_word_diff_old(band["description"].to_s, suggested_band) %></pre>
+                    <pre class="ck-inline-suggestion__pane ck-inline-suggestion__pane--after"><%= ck_word_diff_new(band["description"].to_s, suggested_band) %></pre>
+                  </div>
+                </div>
+              <% end %>
             </div>
           </div>
         <% end %>
@@ -57,11 +143,11 @@
         <% confirm = parts.empty? ? "Delete \"#{metric.name}\"? It's not in use." : "Delete \"#{metric.name}\"? It's #{parts.to_sentence}." %>
         <%= button_to metric_path(metric), method: :delete,
               form_class: "inline-block",
-              class: "ck-icon-btn",
+              class: "ck-icon-btn ck-icon-btn--form",
               title: "Delete metric",
               "aria-label": "Delete metric",
               data: { turbo_confirm: confirm } do %>
-          <%= heroicon_tag "trash", variant: :outline, size: 16, "aria-hidden": "true" %>
+          <%= heroicon_tag "trash", variant: :outline, size: 24, "aria-hidden": "true" %>
         <% end %>
       <% end %>
       <%= link_to "Cancel", metrics_path, class: ck_button_classes(:light, variant: :outline), tabindex: "0" %>

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

@@ -0,0 +1,25 @@
+<% current_bands = local_assigns[:current_bands] || [] %>
+<% draft_bands = local_assigns[:draft_bands] || [] %>
+<% lookup = ->(bands, stars) { bands.find { |b| b["stars"].to_i == stars }&.dig("description").to_s } %>
+<div class="ck-rubric-diff">
+  <% 5.downto(1) do |stars| %>
+    <% old_band = lookup.call(current_bands, stars) %>
+    <% new_band = lookup.call(draft_bands, stars) %>
+    <% changed = old_band != new_band %>
+    <div class="ck-rubric-diff__row ck-rubric-diff__row--<%= changed ? "changed" : "unchanged" %>">
+      <div class="ck-rubric-diff__stars">
+        <% 5.times do |i| %>
+          <svg viewBox="0 0 24 24" width="14" height="14" stroke-width="1.75" class="ck-star <%= i < stars ? "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 %>
+      </div>
+      <% if changed %>
+        <div class="ck-rubric-diff__panes">
+          <pre class="ck-rubric-diff__pane ck-rubric-diff__pane--before"><%= ck_word_diff_old(old_band, new_band) %></pre>
+          <pre class="ck-rubric-diff__pane ck-rubric-diff__pane--after"><%= ck_word_diff_new(old_band, new_band) %></pre>
+        </div>
+      <% else %>
+        <p class="ck-rubric-diff__unchanged"><%= old_band.presence || "—" %></p>
+      <% end %>
+    </div>
+  <% end %>
+</div>

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

@@ -0,0 +1,4 @@
+<span class="ck-info-hint" tabindex="0" role="button" aria-label="What is a rubric?">
+  <%= heroicon_tag "information-circle", variant: :outline, "aria-hidden": "true" %>
+  <span class="ck-info-popup">How the judge picks 1 to 5. Each row says what an output has to look like to earn that many stars. The judge reads these descriptions when it scores, so clearer rows give you more consistent scoring.</span>
+</span>

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

@@ -1,11 +1,15 @@
-<article class="ck-starter-card">
-  <h3 class="ck-starter-card__name"><%= link_to starter.name, starter_preview_metrics_path(key: starter.key), class: "ck-link" %></h3>
-  <p class="ck-starter-card__desc"><%= starter.description %></p>
-  <div class="ck-starter-card__actions">
-    <%= link_to "Preview", starter_preview_metrics_path(key: starter.key), class: ck_button_classes(:dark) + " ck-button--sm" %>
-    <%= button_to "Don't show this one", dismiss_starter_metrics_path(key: starter.key),
-          method: :post, form_class: "inline-block",
-          class: "ck-link ck-starter-card__dismiss",
+<%= link_to starter_preview_metrics_path(key: starter.key), class: "ck-starter-card" do %>
+  <div class="ck-starter-card__body">
+    <p class="ck-starter-card__name"><strong><%= starter.name %></strong></p>
+    <p class="ck-starter-card__desc"><%= starter.description %></p>
+  </div>
+  <div class="ck-starter-card__foot">
+    <span class="ck-starter-card__cta">Preview →</span>
+    <%= button_to "dismiss", dismiss_starter_metrics_path(key: starter.key),
+          method: :post,
+          form: { onclick: "event.stopPropagation();" },
+          form_class: "inline-block ck-starter-card__dismiss-form",
+          class: "ck-starter-card__dismiss",
           data: { turbo_confirm: "Hide \"#{starter.name}\" from this list?" } %>
   </div>
-</article>
+<% end %>

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

@@ -10,4 +10,8 @@
   </div>
 </section>
 
-<%= render "form", metric: @metric %>
+<%= render "form",
+      metric: @metric,
+      suggestion_draft: @suggestion_draft,
+      edit_draft: @edit_draft,
+      published_judge_version: @published_judge_version %>