completion-kit 0.8.0 → 0.9.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 4625f4e3f1afceb34f7603ee7a5025c78cb2499f6c2c287d83a9d02397f407b2
-  data.tar.gz: 740fecb69351c418aaececb19b9cfe3579bf9124a5df52a23d6db5851ce63fce
+  metadata.gz: ddf80d4e74705494435d5ae2d9f0ed5ce0dd927f32bffcb1a13819076f94bced
+  data.tar.gz: 74eadf6abc0f173d0047c961502aeaaab9b5b3de7dd155a00d5d054fb5b8f6e6
 SHA512:
-  metadata.gz: 548cc9666d2cf744babbe6a62047f2c4a44f79fe63490b795d84017742f3db4baff9313dfa4392e4b09c740645224493787d6af4208994d43a02a6610275e3c7
-  data.tar.gz: d8f405649cc42b70e4849f5e0fb9bfe733507baf91e47fdd3650c1ca1b3cab0099a337b038777799853a74cca8bc767baa6cf90107fd737eb519678d0dd4bdd0
+  metadata.gz: fa0c962d8282310584ff52a849eeb7efc3c66debe9246d8231e5e24e55c45e8566b4edf83a19d3a021dc4a41b6241c042e0af3059c1b47ba709412220628ed96
+  data.tar.gz: 20cdeabe363e212a572cbe6b1f08128a7aae88f5a2ef50a3b3d012e5fbef2a64c93571bc6ebb28e971dc819ac57d060a11a3006f1b7a60ebeb265957be221eab

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

@@ -216,6 +216,10 @@ form.button_to {
   justify-content: flex-end;
   flex-shrink: 0;
 }
+.ck-actions .ck-icon-btn {
+  width: 2rem;
+  height: 2rem;
+}
 
 .ck-toolbar__spacer {
   flex: 1;
@@ -665,6 +669,33 @@ tr:hover .ck-chip--publish {
   border-color: var(--ck-accent);
   color: var(--ck-accent);
 }
+.ck-chip--cta {
+  cursor: pointer;
+  background: var(--ck-accent-soft);
+  border-color: rgba(56, 189, 248, 0.4);
+  color: var(--ck-accent);
+  transition: border-color 0.15s, color 0.15s;
+}
+.ck-chip--cta:hover {
+  border-color: var(--ck-accent);
+  color: var(--ck-accent-hover);
+}
+.ck-version-created {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 10px;
+}
+.ck-version-state {
+  font-family: var(--ck-mono);
+  font-size: 0.66rem;
+  letter-spacing: 0.07em;
+  text-transform: uppercase;
+  color: var(--ck-dim);
+}
+.ck-version-state--live {
+  color: var(--ck-text);
+}
 
 .ck-chip--soft {
   background: var(--ck-accent-soft);
@@ -753,6 +784,11 @@ tr:hover .ck-chip--publish {
   flex-shrink: 0;
   color: var(--ck-warning);
 }
+.ck-kicker--icon {
+  display: inline-flex;
+  align-items: center;
+  gap: 0.4rem;
+}
 
 .ck-button:hover {
   transform: translateY(-1px);
@@ -2552,6 +2588,26 @@ select.ck-input {
   gap: 0.6rem;
   padding: 0.95rem 1.5rem 1.1rem;
 }
+.ck-modal__footer--split {
+  justify-content: space-between;
+  align-items: center;
+  gap: 1rem;
+  flex-wrap: wrap;
+  border-top: 1px solid var(--ck-line);
+}
+.ck-modal__foot-note {
+  color: var(--ck-dim);
+  font-size: 0.86rem;
+}
+.ck-modal__foot-actions {
+  display: inline-flex;
+  align-items: center;
+  gap: 0.6rem;
+}
+.ck-modal__foot-actions .ck-icon-btn {
+  width: 2rem;
+  height: 2rem;
+}
 
 .ck-modal__close-btn {
   appearance: none;
@@ -3582,52 +3638,59 @@ 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: 16rem; }
-.ck-metric-versions-table th:nth-child(2), .ck-metric-versions-table td:nth-child(2) { width: 12rem; white-space: nowrap; }
-.ck-metric-versions-table th:nth-child(3), .ck-metric-versions-table td:nth-child(3) { width: auto; }
+.ck-metric-versions-table th:nth-child(1), .ck-metric-versions-table td:nth-child(1) { width: 14rem; }
+.ck-metric-versions-table th:nth-child(2), .ck-metric-versions-table td:nth-child(2) { width: auto; }
+.ck-metric-versions-table th:nth-child(3), .ck-metric-versions-table td:nth-child(3) { width: 9rem; white-space: nowrap; }
+.ck-metric-versions-table th:nth-child(4), .ck-metric-versions-table td:nth-child(4) { width: 9rem; white-space: nowrap; }
+
+.ck-change-link {
+  background: none;
+  border: 0;
+  padding: 0;
+  cursor: pointer;
+  font-family: inherit;
+  font-size: 0.86rem;
+  text-align: left;
+  color: var(--ck-text);
+}
+.ck-change-link:hover,
+.ck-change-link:focus-visible {
+  color: var(--ck-accent);
+  text-decoration: underline;
+}
+.ck-change-link--trivial {
+  color: var(--ck-dim);
+}
+.ck-change-link--major {
+  color: rgb(217, 119, 6);
+}
 
 .ck-source-chip {
   display: inline-block;
-  padding: 2px 8px;
-  border-radius: 4px;
   font-family: var(--ck-mono);
-  font-size: 0.7rem;
+  font-size: 0.68rem;
   letter-spacing: 0.06em;
   text-transform: uppercase;
-  border: 1px solid var(--ck-line-strong);
-  background: var(--ck-surface-soft);
   color: var(--ck-muted);
 }
 .ck-source-chip--ai {
-  border-color: rgba(56, 189, 248, 0.35);
-  background: rgba(56, 189, 248, 0.08);
   color: var(--ck-accent);
 }
 .ck-source-chip--manual {
-  border-color: var(--ck-line-strong);
   color: var(--ck-text);
 }
 .ck-source-chip--initial {
-  border-color: var(--ck-line);
   color: var(--ck-dim);
 }
 .ck-source-chip--revert {
-  border-color: rgba(245, 158, 11, 0.35);
-  background: rgba(245, 158, 11, 0.08);
   color: rgb(217, 119, 6);
 }
 .ck-source-chip--current {
-  border-color: var(--ck-line-strong);
   color: var(--ck-text);
 }
 .ck-source-chip--past {
-  border-color: var(--ck-line);
   color: var(--ck-dim);
 }
-.ck-disagreement--stale {
-  opacity: 0.7;
-}
-
 .ck-suggestions-table th:nth-child(1), .ck-suggestions-table td:nth-child(1) { width: 16rem; white-space: nowrap; }
 .ck-suggestions-table th:nth-child(2), .ck-suggestions-table td:nth-child(2) { width: auto; }
 .ck-suggestions-table th:nth-child(3), .ck-suggestions-table td:nth-child(3) { width: 6rem; white-space: nowrap; }
@@ -5552,224 +5615,80 @@ a.tag-mark {
   font-weight: 600;
 }
 
-.ck-trust-panel {
-  display: inline-flex;
-  flex-direction: column;
-  gap: 6px;
-  margin-top: 12px;
-  padding: 10px 14px;
-  background: var(--ck-surface-soft);
-  border: 1px solid var(--ck-line);
-  border-radius: 6px;
-}
-.ck-trust-panel__label {
-  margin: 0;
-  font-family: var(--ck-mono);
-  font-size: 0.7rem;
-  letter-spacing: 0.08em;
-  text-transform: uppercase;
-  color: var(--ck-dim);
-}
-.ck-trust-panel__body {
+.ck-trust-line {
+  margin: 14px 0 0;
   display: flex;
+  flex-wrap: wrap;
   align-items: baseline;
-  gap: 10px;
-}
-.ck-trust-panel__counter {
-  font-family: var(--ck-mono);
-  font-size: 1.6rem;
-  font-weight: 600;
-  color: var(--ck-accent);
-}
-.ck-trust-panel__counter-of {
-  font-size: 0.9rem;
-  color: var(--ck-dim);
-  margin-left: 4px;
-}
-.ck-trust-panel__hint {
-  font-family: var(--ck-mono);
-  font-size: 0.72rem;
-  color: var(--ck-dim);
-  letter-spacing: 0.04em;
-}
-.ck-trust-panel__score {
-  font-family: var(--ck-mono);
-  font-size: 1.6rem;
-  font-weight: 600;
-  color: var(--ck-success);
-}
-.ck-trust-panel__score-pct {
-  font-size: 0.9rem;
-  color: var(--ck-dim);
-  margin-left: 2px;
-}
-.ck-trust-panel__margin {
+  gap: 6px 20px;
   font-family: var(--ck-mono);
   font-size: 0.8rem;
-  color: var(--ck-dim);
-}
-.ck-trust-panel__gate {
-  font-family: var(--ck-mono);
-  font-size: 0.66rem;
-  letter-spacing: 0.08em;
-  text-transform: uppercase;
-  padding: 2px 6px;
-  border-radius: 3px;
-  background: var(--ck-surface);
-  border: 1px solid var(--ck-line);
-  color: var(--ck-dim);
-}
-.ck-trust-panel--firm .ck-trust-panel__gate {
-  color: var(--ck-success);
-  border-color: rgba(45, 212, 168, 0.35);
-}
-.ck-trust-panel__details {
-  display: flex;
-  flex-wrap: wrap;
-  gap: 14px;
-  font-family: var(--ck-mono);
-  font-size: 0.72rem;
-  color: var(--ck-dim);
+  color: var(--ck-text);
+  line-height: 1.5;
 }
-.ck-trust-panel__borderline {
-  color: var(--ck-warning);
+.ck-trust-line__lead {
+  font-weight: 600;
 }
-
-.ck-trust-panel__borderline--ok { color: var(--ck-dim); }
-.ck-trust-panel__borderline--warning { color: var(--ck-warning); }
-.ck-trust-panel__borderline--danger { color: var(--ck-danger); }
-
-.ck-trust-line {
-  margin: 10px 0 0;
-  font-family: var(--ck-mono);
-  font-size: 0.82rem;
+.ck-trust-line__hint {
   color: var(--ck-dim);
-  line-height: 1.55;
 }
-.ck-trust-line > * + * {
-  margin-left: 6px;
+.ck-cal-stat {
+  display: inline-flex;
+  align-items: baseline;
+  gap: 6px;
 }
-.ck-trust-line__label {
-  font-size: 0.82rem;
-  letter-spacing: 0.08em;
+.ck-cal-stat__label {
+  font-size: 0.64rem;
+  letter-spacing: 0.09em;
   text-transform: uppercase;
   color: var(--ck-muted);
-  white-space: nowrap;
-}
-.ck-trust-line__label svg {
-  display: inline-block;
-  width: 14px;
-  height: 14px;
-  vertical-align: middle;
-  margin-right: 2px;
-  position: relative;
-  top: -1px;
-}
-.ck-trust-line__state {
-  color: var(--ck-text);
-  font-weight: 500;
-}
-.ck-trust-icon {
-  display: inline-block;
-  width: 11px;
-  height: 11px;
-  vertical-align: -1px;
-  margin-right: 4px;
-  color: var(--ck-muted);
 }
-.ck-trust-line__state {
-  color: var(--ck-text);
-}
-.ck-trust-line__sep {
-  color: var(--ck-line-strong);
-}
-.ck-trust-line__counter {
-  font-weight: 600;
+.ck-trust-line__figure {
   color: var(--ck-accent);
-}
-.ck-trust-line__score {
   font-weight: 600;
-  color: var(--ck-success);
-}
-.ck-trust-line--early .ck-trust-line__score,
-.ck-trust-line--provisional .ck-trust-line__score {
-  color: var(--ck-accent);
-}
-.ck-trust-line__margin {
-  color: var(--ck-dim);
-}
-.ck-trust-line__gate {
-  font-size: 0.68rem;
-  letter-spacing: 0.06em;
-  text-transform: uppercase;
-  padding: 1px 6px;
-  border-radius: 3px;
-  border: 1px solid var(--ck-line);
-  color: var(--ck-dim);
 }
-.ck-trust-line--firm .ck-trust-line__gate {
+.ck-trust-line--firm .ck-trust-line__figure {
   color: var(--ck-success);
-  border-color: rgba(45, 212, 168, 0.35);
-}
-.ck-trust-line__hint {
-  color: var(--ck-dim);
 }
-.ck-trust-line__link {
+.ck-trust-line__borderline--ok { color: var(--ck-dim); }
+.ck-trust-line__borderline--warning { color: var(--ck-warning); }
+.ck-trust-line__borderline--danger { color: var(--ck-danger); }
+
+.ck-cal-link {
+  font-family: var(--ck-mono);
+  font-size: 0.78rem;
+  font-weight: 500;
+  letter-spacing: 0.02em;
   color: var(--ck-accent);
   text-decoration: none;
-  font-weight: 500;
   white-space: nowrap;
 }
-.ck-trust-line__link:hover,
-.ck-trust-line__link:focus-visible {
+.ck-cal-link:hover,
+.ck-cal-link:focus-visible {
   color: var(--ck-accent-hover);
 }
-.ck-trust-line__borderline--ok { color: var(--ck-dim); }
-.ck-trust-line__borderline--warning { color: var(--ck-warning); }
-.ck-trust-line__borderline--danger { color: var(--ck-danger); }
-
-@media (max-width: 640px) {
-  .ck-trust-panel {
-    display: flex;
-    width: 100%;
-    box-sizing: border-box;
-  }
-  .ck-trust-panel__body {
-    flex-direction: column;
-    align-items: flex-start;
-    gap: 4px;
-  }
-  .ck-trust-panel__hint {
-    line-height: 1.4;
-    white-space: normal;
-  }
-}
-
-.ck-disagreements-table td .ck-meta-copy {
-  font-size: 0.78rem;
+.ck-cal-link--button {
+  background: none;
+  border: 0;
+  padding: 0;
+  cursor: pointer;
 }
-.ck-few-shot-list {
-  list-style: decimal;
-  padding-left: 1.4rem;
-  margin: 0;
+.ck-cal-foot {
+  margin-top: 14px;
+  padding-top: 12px;
+  border-top: 1px solid var(--ck-line);
   display: flex;
-  flex-direction: column;
-  gap: 12px;
+  align-items: baseline;
+  justify-content: space-between;
+  gap: 12px 20px;
+  flex-wrap: wrap;
 }
-.ck-few-shot-item {
-  padding: 10px 12px;
-  background: var(--ck-surface-soft);
-  border: 1px solid var(--ck-line);
-  border-radius: 6px;
+.ck-cal-foot__note {
+  color: var(--ck-dim);
+  font-size: 0.86rem;
 }
-.ck-few-shot-item__scores {
-  display: flex;
-  align-items: center;
-  gap: 8px;
-  font-family: var(--ck-mono);
-  font-size: 0.75rem;
-  letter-spacing: 0.04em;
-  text-transform: uppercase;
+.ck-cal-foot__form {
+  display: inline;
 }
 
 .ck-draft-pending {
@@ -5777,47 +5696,6 @@ a.tag-mark {
   background: linear-gradient(180deg, var(--ck-accent-soft), var(--ck-surface));
 }
 
-.ck-suggestion-banner {
-  display: inline-flex;
-  align-items: center;
-  gap: 10px;
-  margin-top: 10px;
-  padding: 8px 14px;
-  background: var(--ck-accent-soft);
-  border: 1px solid rgba(6, 182, 212, 0.35);
-  border-radius: 6px;
-  color: var(--ck-accent);
-  font-family: var(--ck-mono);
-  font-size: 0.82rem;
-  text-decoration: none;
-}
-.ck-suggestion-banner:hover,
-.ck-suggestion-banner:focus-visible {
-  border-color: var(--ck-accent);
-}
-
-.ck-metrics-table__trust {
-  margin: 4px 0 0;
-  font-family: var(--ck-mono);
-  font-size: 0.72rem;
-  letter-spacing: 0.03em;
-  color: var(--ck-dim);
-}
-.ck-metrics-table__trust-label {
-  text-transform: uppercase;
-  letter-spacing: 0.08em;
-  color: var(--ck-muted);
-  margin-right: 4px;
-}
-.ck-metrics-table__trust-state {
-  color: var(--ck-text);
-}
-.ck-metrics-table__trust-rate {
-  font-weight: 600;
-  color: var(--ck-success);
-  margin-right: 6px;
-}
-
 .ck-calibration__error {
   margin: 8px 0 0;
   padding: 8px 10px;
@@ -5883,30 +5761,9 @@ a.tag-mark {
   0% { background: var(--ck-success); border-color: var(--ck-success); }
 }
 
-.ck-disagreement-list {
-  list-style: none;
-  padding: 0;
-  margin: 12px 0 0;
-  display: flex;
-  flex-direction: column;
-  gap: 12px;
-}
-.ck-disagreement {
-  padding: 14px;
-  background: var(--ck-surface-soft);
-  border: 1px solid var(--ck-line);
-  border-radius: 6px;
-  display: flex;
-  flex-direction: column;
-  gap: 8px;
-}
-.ck-disagreement--remembered {
-  border-color: rgba(45, 212, 168, 0.35);
-  background: rgba(45, 212, 168, 0.04);
-}
-
 .ck-suggestion-banner {
   margin: 0 0 1rem;
+  max-width: 36rem;
   padding: 0.9rem 1rem;
   border: 1px solid rgba(56, 189, 248, 0.35);
   background: rgba(56, 189, 248, 0.05);
@@ -5917,6 +5774,10 @@ a.tag-mark {
   gap: 1rem;
   flex-wrap: wrap;
 }
+.ck-suggestion-banner__actions .ck-icon-btn {
+  width: 2rem;
+  height: 2rem;
+}
 .ck-suggestion-banner__body { min-width: 0; flex: 1 1 320px; }
 .ck-suggestion-banner__actions {
   display: inline-flex;
@@ -5965,54 +5826,6 @@ a.tag-mark {
   color: var(--ck-success);
   border-color: rgba(45, 212, 168, 0.45);
 }
-.ck-disagreement__action {
-  display: inline-flex;
-  align-items: center;
-  gap: 8px;
-  flex-wrap: wrap;
-}
-.ck-disagreement__head {
-  display: flex;
-  align-items: center;
-  justify-content: space-between;
-  gap: 12px;
-  flex-wrap: wrap;
-}
-.ck-disagreement__scores {
-  display: inline-flex;
-  align-items: center;
-  gap: 8px;
-  flex-wrap: wrap;
-}
-.ck-disagreement__scores-label {
-  font-family: var(--ck-mono);
-  font-size: 0.7rem;
-  letter-spacing: 0.08em;
-  text-transform: uppercase;
-  color: var(--ck-dim);
-}
-.ck-disagreement__scores-arrow {
-  color: var(--ck-dim);
-}
-.ck-disagreement__note {
-  margin: 0;
-  color: var(--ck-text);
-  font-size: 0.92rem;
-  line-height: 1.45;
-}
-.ck-disagreement__source {
-  margin: 0;
-  font-size: 0.78rem;
-}
-.ck-disagreement__source-link {
-  color: var(--ck-accent);
-  text-decoration: none;
-  font-weight: 500;
-}
-.ck-disagreement__source-link:hover,
-.ck-disagreement__source-link:focus-visible {
-  color: var(--ck-accent-hover);
-}
 
 .ck-starter-row {
   margin-top: 2rem;

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

@@ -2,7 +2,7 @@ module CompletionKit
   module Api
     module V1
       class MetricsController < BaseController
-        before_action :set_metric, only: [:show, :update, :destroy, :suggest_variants, :add_few_shot, :remove_few_shot]
+        before_action :set_metric, only: [:show, :update, :destroy, :suggest_variants]
 
         def index
           scope = Metric.includes(:tags)
@@ -54,33 +54,6 @@ module CompletionKit
           render json: versions, status: :created
         end
 
-        def add_few_shot
-          calibration = Calibration.where(metric_id: @metric.id, verdict: "disagree").find(params[:calibration_id])
-          review = calibration.response.reviews.find_by(metric_id: @metric.id)
-          examples = Array(@metric.few_shot_examples)
-          examples << {
-            "input" => calibration.response.input_data.to_s.truncate(2000),
-            "response" => calibration.response.response_text.to_s.truncate(2000),
-            "judge_score" => review&.ai_score&.to_f,
-            "judge_feedback" => review&.ai_feedback.to_s.truncate(1000),
-            "human_score" => calibration.corrected_score&.to_f,
-            "human_note" => calibration.note.to_s.truncate(1000),
-            "calibration_id" => calibration.id,
-            "added_at" => Time.current.utc.iso8601
-          }
-          @metric.update!(few_shot_examples: examples)
-          render json: @metric.reload
-        rescue ActiveRecord::RecordNotFound
-          render_error("Calibration not found or not a disagree on this metric.", status: :not_found)
-        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)
-          render json: @metric.reload
-        end
-
         private
 
         def set_metric