completion-kit 0.10.0 → 0.12.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 4772b264a668a86e004f78c7bc2397d93f5266d8dc287b422728d951ab24fbcc
-  data.tar.gz: 63b221e144e930df9978607a78533d354f659ca2bf6141046291168af50d3cd7
+  metadata.gz: cdcc8d4cdaf4b7aa4b3cff7cb0dd3fe65ce213bbce7b8ab1ba52cba304bff19a
+  data.tar.gz: d2b25e3b12b187b3df15e9b8347668a3dcf529b765fd5427a1dc2579665ee664
 SHA512:
-  metadata.gz: a3249ae1c734dcee0c6f9410baf0400f4b16e091b220d86e8417dec91ff9943a165bbc6c8368629cc14054c3af7946bdb693f006a5756de40a809c41db5bbe3a
-  data.tar.gz: 541323c93b08f08f32c2f024e3709ee5f3e4e48144cd8e9cb2ba8891312fd9b5712e4ca39cbcc739b64ba61ff0ac890be3528e5bfd9c238e073d640b37b47e90
+  metadata.gz: 81921860b28c13076623a462e9cc82569a3721f035f5eb8dbe236c177fa02f9277aa7425659ae583da6713dfcc9467fc03683b3305f65fd8eaf29525cc93e143
+  data.tar.gz: c03aa7c4ad395228e4268ee166a908779a6c50e614c1a0591cf52569712236ffb4ccd9b6d8dc918d40e2ebb253fc48760fdd4e95d3e374e4cf68fad7b1b7ee19

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

@@ -3158,7 +3158,7 @@ select.ck-input {
 #ck-tab-datasets:checked ~ .ck-api-tabs__nav label[for="ck-tab-datasets"],
 #ck-tab-metrics:checked ~ .ck-api-tabs__nav label[for="ck-tab-metrics"],
 #ck-tab-metric-groups:checked ~ .ck-api-tabs__nav label[for="ck-tab-metric-groups"],
-#ck-tab-calibrations:checked ~ .ck-api-tabs__nav label[for="ck-tab-calibrations"],
+#ck-tab-agreements:checked ~ .ck-api-tabs__nav label[for="ck-tab-agreements"],
 #ck-tab-tags:checked ~ .ck-api-tabs__nav label[for="ck-tab-tags"],
 #ck-tab-providers:checked ~ .ck-api-tabs__nav label[for="ck-tab-providers"] {
   color: var(--ck-accent);
@@ -3173,7 +3173,7 @@ select.ck-input {
 #ck-tab-datasets:checked ~ .ck-api-tabs__panels .ck-api-tabs__panel:nth-child(5),
 #ck-tab-metrics:checked ~ .ck-api-tabs__panels .ck-api-tabs__panel:nth-child(6),
 #ck-tab-metric-groups:checked ~ .ck-api-tabs__panels .ck-api-tabs__panel:nth-child(7),
-#ck-tab-calibrations:checked ~ .ck-api-tabs__panels .ck-api-tabs__panel:nth-child(8),
+#ck-tab-agreements:checked ~ .ck-api-tabs__panels .ck-api-tabs__panel:nth-child(8),
 #ck-tab-tags:checked ~ .ck-api-tabs__panels .ck-api-tabs__panel:nth-child(9),
 #ck-tab-providers:checked ~ .ck-api-tabs__panels .ck-api-tabs__panel:nth-child(10) {
   display: block;
@@ -3215,7 +3215,7 @@ select.ck-input {
   #ck-tab-datasets:checked ~ .ck-api-tabs__nav label[for="ck-tab-datasets"],
   #ck-tab-metrics:checked ~ .ck-api-tabs__nav label[for="ck-tab-metrics"],
   #ck-tab-metric-groups:checked ~ .ck-api-tabs__nav label[for="ck-tab-metric-groups"],
-  #ck-tab-calibrations:checked ~ .ck-api-tabs__nav label[for="ck-tab-calibrations"],
+  #ck-tab-agreements:checked ~ .ck-api-tabs__nav label[for="ck-tab-agreements"],
   #ck-tab-tags:checked ~ .ck-api-tabs__nav label[for="ck-tab-tags"],
   #ck-tab-providers:checked ~ .ck-api-tabs__nav label[for="ck-tab-providers"] {
     border-left-color: transparent;
@@ -3619,10 +3619,9 @@ select.ck-input {
 .ck-prompt-versions-table th:nth-child(3), .ck-prompt-versions-table td:nth-child(3) { width: 8rem; white-space: nowrap; }
 .ck-prompt-versions-table th:nth-child(4), .ck-prompt-versions-table td:nth-child(4) { width: auto; }
 
-.ck-metric-versions-table th:nth-child(1), .ck-metric-versions-table td:nth-child(1) { width: 34%; }
-.ck-metric-versions-table th:nth-child(2), .ck-metric-versions-table td:nth-child(2) { width: 33%; white-space: nowrap; }
-.ck-metric-versions-table th:nth-child(3), .ck-metric-versions-table td:nth-child(3) { width: 33%; white-space: nowrap; }
-.ck-metric-versions-table .ck-version-cell { justify-content: flex-start; gap: 0.75rem; }
+.ck-metric-versions-table th:nth-child(1), .ck-metric-versions-table td:nth-child(1) { width: 18rem; }
+.ck-metric-versions-table th:nth-child(2), .ck-metric-versions-table td:nth-child(2) { width: 16rem; white-space: nowrap; }
+.ck-metric-versions-table th:nth-child(3), .ck-metric-versions-table td:nth-child(3) { width: auto; white-space: nowrap; }
 
 
 .ck-source-chip {
@@ -5378,33 +5377,33 @@ a.tag-mark {
   outline-offset: 2px;
 }
 
-.ck-calibration {
+.ck-agreement {
   margin-top: 12px;
   padding-top: 12px;
   border-top: 1px dashed var(--ck-line);
 }
-.ck-calibration__prompt {
+.ck-agreement__prompt {
   margin: 0 0 10px;
   font-family: var(--ck-mono);
   font-size: 0.72rem;
   letter-spacing: 0.04em;
   color: var(--ck-dim);
 }
-.ck-calibration__prompt > * + * {
+.ck-agreement__prompt > * + * {
   margin-left: 8px;
 }
-.ck-calibration__label {
+.ck-agreement__label {
   letter-spacing: 0.08em;
   text-transform: uppercase;
   color: var(--ck-dim);
 }
-.ck-calibration__meta {
+.ck-agreement__meta {
   color: var(--ck-muted);
 }
-.ck-calibration__sep {
+.ck-agreement__sep {
   color: var(--ck-line-strong);
 }
-.ck-calibration__meta-link {
+.ck-agreement__meta-link {
   color: var(--ck-accent);
   text-decoration: none;
   white-space: nowrap;
@@ -5412,12 +5411,12 @@ a.tag-mark {
   letter-spacing: 0.08em;
 }
 
-.ck-calibration__others {
+.ck-agreement__others {
   margin: 10px 0 0;
   font-family: var(--ck-mono);
   font-size: 0.78rem;
 }
-.ck-calibration__others-summary {
+.ck-agreement__others-summary {
   display: inline-flex;
   align-items: center;
   gap: 6px;
@@ -5427,20 +5426,20 @@ a.tag-mark {
   user-select: none;
   list-style: none;
 }
-.ck-calibration__others-summary:hover,
-.ck-calibration__others-summary:focus-visible {
+.ck-agreement__others-summary:hover,
+.ck-agreement__others-summary:focus-visible {
   color: var(--ck-accent-hover);
 }
-.ck-calibration__others-summary::-webkit-details-marker { display: none; }
-.ck-calibration__others-summary svg {
+.ck-agreement__others-summary::-webkit-details-marker { display: none; }
+.ck-agreement__others-summary svg {
   width: 12px;
   height: 12px;
   transition: transform 0.15s;
 }
-.ck-calibration__others[open] .ck-calibration__others-summary svg {
+.ck-agreement__others[open] .ck-agreement__others-summary svg {
   transform: rotate(90deg);
 }
-.ck-calibration__others-list {
+.ck-agreement__others-list {
   list-style: none;
   padding: 8px 0 0;
   margin: 0;
@@ -5448,24 +5447,24 @@ a.tag-mark {
   flex-direction: column;
   gap: 6px;
 }
-.ck-calibration__others-item {
+.ck-agreement__others-item {
   padding: 8px 10px;
   background: var(--ck-surface-soft);
   border: 1px solid var(--ck-line);
   border-radius: 4px;
   color: var(--ck-dim);
 }
-.ck-calibration__others-item--agree { border-left: 2px solid var(--ck-success); }
-.ck-calibration__others-item--disagree { border-left: 2px solid var(--ck-danger); }
-.ck-calibration__others-item--borderline { border-left: 2px solid var(--ck-warning); }
-.ck-calibration__others-row {
+.ck-agreement__others-item--agree { border-left: 2px solid var(--ck-success); }
+.ck-agreement__others-item--disagree { border-left: 2px solid var(--ck-danger); }
+.ck-agreement__others-item--borderline { border-left: 2px solid var(--ck-warning); }
+.ck-agreement__others-row {
   display: flex;
   flex-wrap: wrap;
   align-items: center;
   gap: 10px;
   line-height: 1;
 }
-.ck-calibration__others-verdict {
+.ck-agreement__others-verdict {
   display: inline-flex;
   align-items: center;
   gap: 4px;
@@ -5474,24 +5473,24 @@ a.tag-mark {
   font-weight: 500;
   color: var(--ck-text);
 }
-.ck-calibration__others-item--agree .ck-calibration__others-verdict { color: var(--ck-success); }
-.ck-calibration__others-item--disagree .ck-calibration__others-verdict { color: var(--ck-danger); }
-.ck-calibration__others-item--borderline .ck-calibration__others-verdict { color: var(--ck-warning); }
-.ck-calibration__others-by {
+.ck-agreement__others-item--agree .ck-agreement__others-verdict { color: var(--ck-success); }
+.ck-agreement__others-item--disagree .ck-agreement__others-verdict { color: var(--ck-danger); }
+.ck-agreement__others-item--borderline .ck-agreement__others-verdict { color: var(--ck-warning); }
+.ck-agreement__others-by {
   color: var(--ck-muted);
 }
-.ck-calibration__others-stars {
+.ck-agreement__others-stars {
   display: inline-flex;
   align-items: center;
   gap: 2px;
 }
-.ck-calibration__others-stars svg { display: block; }
-.ck-calibration__others-note {
+.ck-agreement__others-stars svg { display: block; }
+.ck-agreement__others-note {
   margin: 6px 0 0;
   color: var(--ck-dim);
   line-height: 1.5;
 }
-.ck-calibration__meta-link svg {
+.ck-agreement__meta-link svg {
   display: inline-block;
   width: 12px;
   height: 12px;
@@ -5500,16 +5499,16 @@ a.tag-mark {
   position: relative;
   top: -1px;
 }
-.ck-calibration__meta-link:hover,
-.ck-calibration__meta-link:focus-visible {
+.ck-agreement__meta-link:hover,
+.ck-agreement__meta-link:focus-visible {
   color: var(--ck-accent-hover);
 }
-.ck-calibration__buttons {
+.ck-agreement__buttons {
   display: flex;
   gap: 6px;
   flex-wrap: wrap;
 }
-.ck-calibration__pill {
+.ck-agreement__pill {
   display: inline-flex;
   align-items: center;
   gap: 0.4rem;
@@ -5526,50 +5525,50 @@ a.tag-mark {
   cursor: pointer;
   transition: background 0.12s, border-color 0.12s, color 0.12s;
 }
-.ck-calibration__pill svg {
+.ck-agreement__pill svg {
   width: 14px;
   height: 14px;
 }
-.ck-calibration__pill:hover,
-.ck-calibration__pill:focus-visible {
+.ck-agreement__pill:hover,
+.ck-agreement__pill:focus-visible {
   color: var(--ck-text);
   border-color: var(--ck-dim);
 }
-.ck-calibration__pill--agree.is-active {
+.ck-agreement__pill--agree.is-active {
   background: var(--ck-success-soft);
   border-color: rgba(45, 212, 168, 0.35);
   color: var(--ck-success);
 }
-.ck-calibration__pill--disagree.is-active {
+.ck-agreement__pill--disagree.is-active {
   background: var(--ck-danger-soft);
   border-color: rgba(248, 113, 113, 0.35);
   color: var(--ck-danger);
 }
-.ck-calibration__pill--borderline.is-active {
+.ck-agreement__pill--borderline.is-active {
   background: var(--ck-warning-soft);
   border-color: rgba(224, 164, 88, 0.35);
   color: var(--ck-warning);
 }
-.ck-calibration__pill--agree:hover { border-color: rgba(45, 212, 168, 0.45); color: var(--ck-success); }
-.ck-calibration__pill--disagree:hover { border-color: rgba(248, 113, 113, 0.45); color: var(--ck-danger); }
-.ck-calibration__pill--borderline:hover { border-color: rgba(224, 164, 88, 0.45); color: var(--ck-warning); }
-.ck-calibration__detail {
+.ck-agreement__pill--agree:hover { border-color: rgba(45, 212, 168, 0.45); color: var(--ck-success); }
+.ck-agreement__pill--disagree:hover { border-color: rgba(248, 113, 113, 0.45); color: var(--ck-danger); }
+.ck-agreement__pill--borderline:hover { border-color: rgba(224, 164, 88, 0.45); color: var(--ck-warning); }
+.ck-agreement__detail {
   margin-top: 12px;
   display: flex;
   flex-direction: column;
   gap: 12px;
 }
-.ck-calibration__detail > * {
+.ck-agreement__detail > * {
   margin: 0;
 }
-.ck-calibration__detail .ck-button {
+.ck-agreement__detail .ck-button {
   align-self: flex-start;
 }
-.ck-calibration__detail textarea {
+.ck-agreement__detail textarea {
   font-family: var(--ck-mono);
   font-size: 0.82rem;
 }
-.ck-calibration__value {
+.ck-agreement__value {
   color: var(--ck-accent);
   font-family: var(--ck-mono);
   font-weight: 600;
@@ -5661,7 +5660,7 @@ a.tag-mark {
   background: linear-gradient(180deg, var(--ck-accent-soft), var(--ck-surface));
 }
 
-.ck-calibration__error {
+.ck-agreement__error {
   margin: 8px 0 0;
   padding: 8px 10px;
   background: var(--ck-danger-soft);
@@ -6001,3 +6000,67 @@ a.tag-mark {
   width: 2rem;
   height: 2rem;
 }
+
+.ck-suggestion-status:empty { display: none; }
+.ck-suggestion-status {
+  margin-top: 10px;
+  display: flex;
+  align-items: baseline;
+  gap: 10px;
+  flex-wrap: wrap;
+}
+
+.ck-scoreboard {
+  margin-bottom: 16px;
+  padding-bottom: 14px;
+  border-bottom: 1px solid var(--ck-line);
+}
+.ck-scoreboard__headline {
+  margin: 0 0 8px;
+  font-size: 0.95rem;
+  color: var(--ck-text);
+}
+.ck-scoreboard__was {
+  font-family: var(--ck-mono);
+  font-size: 0.74rem;
+  color: var(--ck-muted);
+  margin-left: 6px;
+}
+.ck-scoreboard__tally {
+  list-style: none;
+  margin: 0;
+  padding: 0;
+  display: flex;
+  gap: 18px;
+}
+.ck-scoreboard__stat {
+  font-family: var(--ck-mono);
+  font-size: 0.72rem;
+  letter-spacing: 0.06em;
+  text-transform: uppercase;
+  color: var(--ck-muted);
+}
+.ck-scoreboard__stat strong { color: var(--ck-text); }
+.ck-scoreboard__stat--break strong { color: var(--ck-warning); }
+.ck-scoreboard__note {
+  margin: 8px 0 0;
+  font-size: 0.78rem;
+  color: var(--ck-muted);
+}
+.ck-version-change {
+  display: inline-flex;
+  align-items: baseline;
+  gap: 0.6rem;
+}
+.ck-version-score {
+  font-family: var(--ck-mono);
+  font-size: 0.74rem;
+  color: var(--ck-dim);
+}
+.ck-version-score__label {
+  font-size: 0.6rem;
+  letter-spacing: 0.08em;
+  text-transform: uppercase;
+  color: var(--ck-muted);
+  margin-right: 0.2rem;
+}

data/app/controllers/completion_kit/{calibrations_controller.rb → agreements_controller.rb}

@@ -1,36 +1,36 @@
 module CompletionKit
-  class CalibrationsController < ApplicationController
-    before_action :ensure_calibration_enabled
+  class AgreementsController < ApplicationController
+    before_action :ensure_agreement_enabled
     before_action :set_scope
 
     def create
-      created_by = calibration_creator
-      existing = Calibration.find_by(
+      created_by = agreement_creator
+      existing = Agreement.find_by(
         run_id: @run.id, response_id: @response.id, metric_id: @metric.id, created_by: created_by
       )
 
       if params[:verdict] == "disagree" && params[:corrected_score].blank?
-        render_calibration(calibration: existing, pending_verdict: "disagree")
+        render_agreement(agreement: existing, pending_verdict: "disagree")
         return
       end
 
-      calibration = existing || Calibration.new(
+      agreement = existing || Agreement.new(
         run: @run, response: @response, metric: @metric, created_by: created_by
       )
-      calibration.assign_attributes(
+      agreement.assign_attributes(
         metric_version: MetricVersion.ensure_current_for(@metric),
         verdict: params[:verdict],
         corrected_score: params[:corrected_score].presence,
         note: params[:note].presence
       )
 
-      if calibration.save
-        render_calibration(calibration: calibration, just_saved: true)
+      if agreement.save
+        render_agreement(agreement: agreement, just_saved: true)
       else
-        render_calibration(
-          calibration: existing,
+        render_agreement(
+          agreement: existing,
           pending_verdict: params[:verdict],
-          error: calibration.errors.full_messages.to_sentence,
+          error: agreement.errors.full_messages.to_sentence,
           status: :unprocessable_entity
         )
       end
@@ -38,10 +38,10 @@ module CompletionKit
 
     private
 
-    def render_calibration(calibration:, pending_verdict: nil, error: nil, just_saved: false, status: :ok)
+    def render_agreement(agreement:, pending_verdict: nil, error: nil, just_saved: false, status: :ok)
       locals = {
         review: review_for_metric,
-        calibration: calibration,
+        agreement: agreement,
         run: @run,
         response_row: @response,
         metric: @metric,
@@ -50,14 +50,14 @@ module CompletionKit
         just_saved: just_saved
       }
       render turbo_stream: turbo_stream.replace(
-        "calibration_#{@response.id}_#{@metric.id}",
-        partial: "completion_kit/calibrations/buttons",
+        "agreement_#{@response.id}_#{@metric.id}",
+        partial: "completion_kit/agreements/buttons",
         locals: locals
       ), status: status
     end
 
-    def ensure_calibration_enabled
-      head :not_found unless CompletionKit.config.judge_calibration_enabled
+    def ensure_agreement_enabled
+      head :not_found unless CompletionKit.config.judge_agreement_enabled
     end
 
     def set_scope
@@ -70,7 +70,7 @@ module CompletionKit
       @response.reviews.find_by(metric_id: @metric.id)
     end
 
-    def calibration_creator
+    def agreement_creator
       request.env["HTTP_X_REMOTE_USER"].presence || CompletionKit.config.username.presence || "operator"
     end
   end

data/app/controllers/completion_kit/api/v1/{calibrations_controller.rb → agreements_controller.rb}

@@ -1,13 +1,13 @@
 module CompletionKit
   module Api
     module V1
-      class CalibrationsController < BaseController
-        before_action :ensure_calibration_enabled
+      class AgreementsController < BaseController
+        before_action :ensure_agreement_enabled
         before_action :set_nested_scope, only: [:create]
-        before_action :load_calibration, only: [:destroy]
+        before_action :load_agreement, only: [:destroy]
 
         def index
-          scope = Calibration.all
+          scope = Agreement.all
           scope = scope.where(run_id: params[:run_id]) if params[:run_id].present?
           scope = scope.where(response_id: params[:response_id]) if params[:response_id].present?
           scope = scope.where(metric_id: params[:metric_id]) if params[:metric_id].present?
@@ -18,31 +18,31 @@ module CompletionKit
         end
 
         def create
-          calibration = scope_calibrations.find_or_initialize_by(created_by: created_by_param)
-          calibration.assign_attributes(
+          agreement = scope_agreements.find_or_initialize_by(created_by: created_by_param)
+          agreement.assign_attributes(
             run: @run,
             response: @response,
             metric: @metric,
             metric_version: MetricVersion.ensure_current_for(@metric),
-            **calibration_params
+            **agreement_params
           )
 
-          if calibration.save
-            render json: calibration, status: calibration.previously_new_record? ? :created : :ok
+          if agreement.save
+            render json: agreement, status: agreement.previously_new_record? ? :created : :ok
           else
-            render_validation_errors(calibration)
+            render_validation_errors(agreement)
           end
         end
 
         def destroy
-          @calibration.destroy!
+          @agreement.destroy!
           head :no_content
         end
 
         private
 
-        def ensure_calibration_enabled
-          render_error("Calibration disabled", status: :not_found) unless CompletionKit.config.judge_calibration_enabled
+        def ensure_agreement_enabled
+          render_error("Agreement disabled", status: :not_found) unless CompletionKit.config.judge_agreement_enabled
         end
 
         def set_nested_scope
@@ -53,17 +53,17 @@ module CompletionKit
           not_found
         end
 
-        def load_calibration
-          @calibration = Calibration.find(params[:id])
+        def load_agreement
+          @agreement = Agreement.find(params[:id])
         rescue ActiveRecord::RecordNotFound
           not_found
         end
 
-        def scope_calibrations
-          Calibration.where(run_id: @run.id, response_id: @response.id, metric_id: @metric.id)
+        def scope_agreements
+          Agreement.where(run_id: @run.id, response_id: @response.id, metric_id: @metric.id)
         end
 
-        def calibration_params
+        def agreement_params
           params.permit(:verdict, :corrected_score, :note).to_h.symbolize_keys
         end
 

data/app/controllers/completion_kit/api/v1/metric_versions_controller.rb

@@ -14,13 +14,8 @@ module CompletionKit
         end
 
         def publish
-          if @version.published? && !@version.current?
-            audit = @version.revert!
-            render json: audit
-          else
-            @version.publish!
-            render json: @version.reload
-          end
+          @version.publish!
+          render json: @version.reload
         end
 
         def destroy

data/app/controllers/completion_kit/api/v1/metrics_controller.rb

@@ -37,7 +37,7 @@ module CompletionKit
         end
 
         def suggest_variants
-          disagreement_count = Calibration.where(metric_id: @metric.id, verdict: "disagree").count
+          disagreement_count = Agreement.where(metric_id: @metric.id, verdict: "disagree").count
           if disagreement_count.zero?
             render_error("Mark at least one case as Disagree before asking the model to suggest a change.", status: :unprocessable_entity)
             return

data/app/controllers/completion_kit/metrics_controller.rb

@@ -39,9 +39,9 @@ module CompletionKit
     def show
       @edit_draft = MetricVersion.drafts.where(metric_id: @metric.id, source: "edit").order(created_at: :desc).first
       @suggestion_draft = MetricVersion.drafts.where(metric_id: @metric.id, source: "suggestion").order(created_at: :desc).first
-      @improve_disagreement_count = Calibration.where(metric_id: @metric.id, verdict: "disagree").count
+      @improve_disagreement_count = Agreement.where(metric_id: @metric.id, verdict: "disagree").count
       @versions = MetricVersion.where(metric_id: @metric.id).order(version_number: :desc).to_a
-      @guiding_examples = CompletionKit.config.judge_examples_from_reviews ? MetricCalibrationExamples.judge_examples_for(@metric) : []
+      @guiding_examples = CompletionKit.config.judge_examples_from_reviews ? MetricAgreementExamples.judge_examples_for(@metric) : []
     end
 
     def new
@@ -52,7 +52,7 @@ module CompletionKit
       @suggestion_draft = MetricVersion.drafts.where(metric_id: @metric.id, source: "suggestion").order(created_at: :desc).first
       @edit_draft = MetricVersion.drafts.where(metric_id: @metric.id, source: "edit").order(created_at: :desc).first
       @published_metric_version = MetricVersion.published.where(metric_id: @metric.id, current: true).first
-      @improve_disagreement_count = Calibration.where(metric_id: @metric.id, verdict: "disagree").count
+      @improve_disagreement_count = Agreement.where(metric_id: @metric.id, verdict: "disagree").count
 
       if @edit_draft
         @metric.instruction = @edit_draft.instruction
@@ -117,26 +117,22 @@ module CompletionKit
 
     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?
+      counts = Agreement.where(metric_id: @metric.id, verdict: %w[agree disagree]).group(:verdict).count
+      if counts["disagree"].to_i.zero?
         redirect_to target, alert: "Mark at least one case as Disagree before asking the model to suggest a change."
         return
       end
 
-      MetricVersion.drafts.where(metric_id: @metric.id, source: "suggestion").destroy_all
+      MetricSuggestionJob.perform_later(@metric.id)
 
-      generator = MetricVariantGenerator.new(@metric, count: 1)
-      variants = generator.call
-      if variants.empty?
-        redirect_to target, alert: "The model returned no usable variants. Try again with a different model."
-        return
-      end
-      versions = generator.persist!(variants)
-      new_version = versions.max_by(&:version_number)
       if params[:back_to] == "edit"
-        redirect_to edit_metric_path(@metric), notice: "Drafted #{new_version.version_label} from your reviews. Review the proposed changes below, then Publish to use it."
+        redirect_to metric_path(@metric), notice: "Drafting a change from your reviews. It will appear here once it's tested."
       else
-        redirect_to metric_path(@metric, show_change: new_version.id), notice: "Drafted #{new_version.version_label} from your reviews."
+        render turbo_stream: turbo_stream.replace(
+          "ck-suggestion-status-#{@metric.id}",
+          partial: "completion_kit/metrics/suggestion_pending",
+          locals: { metric: @metric, count: counts.values.sum }
+        )
       end
     end
 
@@ -149,12 +145,12 @@ module CompletionKit
     end
 
     def exclude_example
-      calibration = Calibration.where(metric_id: @metric.id).find(params[:calibration_id])
-      calibration.update!(excluded_from_examples: true)
+      agreement = Agreement.where(metric_id: @metric.id).find(params[:agreement_id])
+      agreement.update!(excluded_from_examples: true)
       render turbo_stream: turbo_stream.replace(
         "ck-guiding-#{@metric.id}",
         partial: "completion_kit/metrics/guiding_examples",
-        locals: { metric: @metric, examples: MetricCalibrationExamples.judge_examples_for(@metric) }
+        locals: { metric: @metric, examples: MetricAgreementExamples.judge_examples_for(@metric) }
       )
     end
 
@@ -175,13 +171,12 @@ module CompletionKit
       reverting = was_published_already && !version.current?
       previously_current = MetricVersion.current.find_by(metric_id: @metric.id)
 
+      version.publish!
+
       if reverting
-        audit = version.revert!
-        prior_label = previously_current.version_label
         redirect_to metric_path(@metric),
-                    notice: "Reverted #{@metric.name} to #{version.version_label} (logged as #{audit.version_label}). Human reviews collected against #{prior_label} stay tied to it."
+                    notice: "#{@metric.name} is back on #{version.version_label}. Its reviews count again; the ones you gave on #{previously_current.version_label} stay with that version."
       else
-        version.publish!
         redirect_to metric_path(@metric),
                     notice: "#{@metric.name} #{version.version_label} is now the published version."
       end

data/app/jobs/completion_kit/judge_review_job.rb

@@ -82,10 +82,10 @@ module CompletionKit
     private
 
     def review_examples_for(metric, response)
-      return nil unless CompletionKit.config.judge_calibration_enabled
+      return nil unless CompletionKit.config.judge_agreement_enabled
       return nil unless CompletionKit.config.judge_examples_from_reviews
 
-      MetricCalibrationExamples.judge_examples_for(metric, exclude_response_id: response.id)
+      MetricAgreementExamples.judge_examples_for(metric, exclude_response_id: response.id)
     end
 
     def confirm_judging_capability(judge_model_id)