completion-kit 0.5.18 → 0.5.20

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 25e795a8d94b2f0984e65d312eb152ffabcda8e7929a52529906ff1fab6cfa4f
-  data.tar.gz: 91110f1c5736d0c6d3c12ef5aa5a75b66ffe0d565cf5777bd9f2d322936e01d5
+  metadata.gz: b8563167a68aab93caf0e05022c5f1216d096eaae162173e39c5061ba253f3e0
+  data.tar.gz: 7de1eb04e23d658744195330b7d3f0383d9c02c5e3d3a78788aaf96bc9a0d1b0
 SHA512:
-  metadata.gz: 87678468d49193e088d4b1c6c28abe5a63ef67de679cfeb894dd64238810d456e7505bae78224e2c330860f177deff8bf834fab0a1c9abdca4e2cafd81e5463b
-  data.tar.gz: 2755a9885f3242017dce04a91680f4499b5cd8907f1043f9bab41d46651ec234ebe4c0f03886b72a0dac421a8787f56923da884b6d5112f85c409392dbfea969
+  metadata.gz: 0ad34abea12f77b56701f313aa471193e384698a0a2add21724510f210018ee86de287d7fc449df5574a0d9ca12c90883c16d8170b53f254852659b9c4e76ca1
+  data.tar.gz: 3b4118ce8416f45bf42dff0a88dbe20f0d2061cf66f456f5070c3dc51c80dc81e400329ea342b476a77fc44aa461ad161d784e8fe3c50e6d8cfff9df3331b042

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

@@ -345,6 +345,185 @@ form.button_to {
   grid-template-columns: repeat(2, minmax(0, 1fr));
 }
 
+.ck-grid--cards-3 {
+  grid-template-columns: repeat(3, minmax(0, 1fr));
+}
+
+/* ── Dashboard: workspace stat ribbon ─────────────────────────────────
+   Replaces the old oversized count cards. One thin instrument strip,
+   four navigable segments split by hairline dividers. */
+.ck-statbar {
+  display: grid;
+  grid-template-columns: repeat(4, minmax(0, 1fr));
+  margin-top: 1.5rem;
+  border: 1px solid var(--ck-line);
+  border-radius: var(--ck-radius-lg);
+  background: var(--ck-surface);
+  overflow: hidden;
+}
+.ck-statbar__item {
+  display: flex;
+  flex-direction: column;
+  gap: 0.25rem;
+  padding: 0.95rem 1.3rem;
+  text-decoration: none;
+  border-left: 1px solid var(--ck-line);
+  transition: background 0.15s ease;
+}
+.ck-statbar__item:first-child { border-left: 0; }
+.ck-statbar__item:hover { background: var(--ck-surface-hover); }
+.ck-statbar__label {
+  font-family: var(--ck-mono);
+  font-size: 0.68rem;
+  letter-spacing: 0.14em;
+  text-transform: uppercase;
+  color: var(--ck-dim);
+}
+.ck-statbar__value {
+  font-family: var(--ck-mono);
+  font-size: 1.65rem;
+  line-height: 1;
+  color: var(--ck-text);
+  transition: color 0.15s ease;
+}
+.ck-statbar__item:hover .ck-statbar__value { color: var(--ck-accent); }
+
+/* ── Dashboard: pulse cards (activity / worst metric / failed reviews) ── */
+.ck-pulse-grid { margin-top: 0.75rem; }
+
+.ck-stat-card {
+  display: flex;
+  flex-direction: column;
+  min-height: 10.5rem;
+}
+.ck-stat-card .ck-kicker { margin-bottom: auto; }
+.ck-stat-card__foot {
+  margin-top: auto;
+  padding-top: 0.85rem;
+  font-family: var(--ck-mono);
+  font-size: 0.78rem;
+  color: var(--ck-muted);
+}
+.ck-stat-card__foot--split {
+  display: flex;
+  align-items: center;
+  gap: 0.55rem;
+}
+.ck-stat-card__foot--split > :first-child {
+  margin-right: auto;
+  min-width: 0;
+}
+.ck-stat-card__figure { color: var(--ck-text); }
+
+.ck-stat-card__body {
+  display: flex;
+  align-items: baseline;
+  gap: 0.6rem;
+  margin: 0.9rem 0 0.3rem;
+}
+.ck-stat-card__metric {
+  font-size: 1.2rem;
+  line-height: 1.25;
+  color: var(--ck-text);
+}
+.ck-stat-card__metric--empty { color: var(--ck-dim); }
+.ck-stat-card__score { align-self: center; }
+.ck-stat-card__count {
+  font-family: var(--ck-mono);
+  font-size: 2.6rem;
+  line-height: 1;
+}
+.ck-stat-card__count.is-clean { color: var(--ck-success, #4ade80); }
+.ck-stat-card__count.is-danger { color: var(--ck-danger, #f87171); }
+
+/* Dashboard activity sparkline — fixed-height row of bars, height set
+   inline per day. Bars hug the bottom; quiet days collapse to a sliver.
+   The busiest day(s) get the bright accent so the peak reads instantly. */
+.ck-sparkline {
+  display: flex;
+  align-items: flex-end;
+  gap: 3px;
+  height: 3.5rem;
+  margin: 0.9rem 0 0.3rem;
+  border-bottom: 1px solid var(--ck-line);
+  padding-bottom: 1px;
+}
+.ck-sparkline__bar {
+  flex: 1;
+  min-height: 2px;
+  background: var(--ck-line-strong);
+  border-radius: 2px 2px 0 0;
+  transition: background 0.15s ease;
+}
+.ck-sparkline__bar.is-peak { background: var(--ck-accent); }
+.ck-sparkline__bar:hover { background: var(--ck-accent-hover); }
+
+/* One orchestrated page-load reveal — the dashboard "boots up". */
+@keyframes ck-rise {
+  from { opacity: 0; transform: translateY(8px); }
+  to   { opacity: 1; transform: translateY(0); }
+}
+.ck-rise {
+  animation: ck-rise 0.32s ease both;
+  animation-delay: var(--rise-delay, 0ms);
+}
+@media (prefers-reduced-motion: reduce) {
+  .ck-rise { animation: none; }
+}
+
+/* ── Dashboard: prompt improvements list ──────────────────────────────
+   One row per family that improved version-over-version. Name takes the
+   slack; version transition, score transition, and delta sit to the right. */
+.ck-improvements {
+  list-style: none;
+  margin: 0.85rem 0 0;
+  padding: 0;
+}
+.ck-improvement {
+  display: flex;
+  align-items: baseline;
+  gap: 1.1rem;
+  padding: 0.7rem 0;
+  border-top: 1px solid var(--ck-line);
+  font-family: var(--ck-mono);
+  font-size: 0.85rem;
+}
+.ck-improvement:first-child { border-top: 0; }
+.ck-improvement__name {
+  flex: 1;
+  min-width: 0;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+.ck-improvement__versions {
+  color: var(--ck-dim);
+  font-size: 0.78rem;
+  white-space: nowrap;
+}
+.ck-improvement__scores {
+  display: inline-flex;
+  align-items: baseline;
+  gap: 0.4rem;
+  color: var(--ck-muted);
+  white-space: nowrap;
+}
+.ck-improvement__arrow { color: var(--ck-dim); }
+.ck-improvement__to { color: var(--ck-text); }
+.ck-improvement__delta {
+  min-width: 5rem;
+  text-align: right;
+}
+.ck-improvement__delta.is-gain { color: var(--ck-success); }
+.ck-improvement__delta.is-loss { color: var(--ck-danger); }
+.ck-improvements__empty {
+  margin: 0.85rem 0 0;
+  max-width: 46rem;
+  color: var(--ck-muted);
+  font-size: 0.85rem;
+  line-height: 1.6;
+}
+
 .ck-grid--spaced,
 .ck-card--spaced,
 .ck-empty--spaced {
@@ -608,15 +787,18 @@ tr:hover .ck-chip--publish {
   max-width: 380px;
 }
 
+/* Login brand lockup — the standard puzzle logo + two-tone wordmark,
+   stacked and centred above the sign-in form. */
 .ck-login__brand {
-  font-family: var(--ck-mono);
-  font-size: 1.1rem;
-  font-weight: 700;
-  color: var(--ck-accent);
-  text-transform: uppercase;
-  letter-spacing: 0.04em;
-  text-align: center;
-  margin: 0 0 2rem;
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  gap: 0.45rem;
+  margin: 0 auto 2rem;
+}
+.ck-login__brand .ck-brand__name {
+  padding-top: 0;
+  font-size: 1.4rem;
 }
 
 .ck-login__form {
@@ -4457,3 +4639,121 @@ a.tag-mark {
   .ck-launch__ready-panel,
   .ck-launch__step { animation: none; }
 }
+
+.ck-icon-btn {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  flex: none;
+  width: 1.65rem;
+  height: 1.65rem;
+  padding: 0;
+  color: var(--ck-dim);
+  background: transparent;
+  border: 1px solid var(--ck-line-strong);
+  border-radius: 6px;
+  cursor: pointer;
+  transition: color 0.15s ease, border-color 0.15s ease;
+}
+.ck-icon-btn:hover {
+  color: var(--ck-text);
+  border-color: var(--ck-dim);
+}
+.ck-icon-btn svg { display: block; }
+.ck-icon-btn form,
+.ck-failure-list__item form,
+.ck-flyout__item form { display: inline-flex; margin: 0; }
+
+.ck-failure-list {
+  list-style: none;
+  margin: 0.6rem 0 0;
+  padding: 0;
+  display: flex;
+  flex-direction: column;
+  gap: 0.35rem;
+}
+.ck-failure-list__item {
+  display: flex;
+  align-items: center;
+  gap: 0.5rem;
+  font-size: 0.8rem;
+}
+.ck-failure-list__surface {
+  flex: none;
+  padding: 0.06rem 0.45rem;
+  font-size: 0.64rem;
+  font-weight: 700;
+  text-transform: uppercase;
+  letter-spacing: 0.04em;
+  border-radius: 4px;
+  background: var(--ck-surface-soft);
+  color: var(--ck-dim);
+}
+.ck-failure-list__surface--run { color: var(--ck-warning); }
+.ck-failure-list__surface--generation { color: var(--ck-danger); }
+.ck-failure-list__surface--judge { color: var(--ck-info); }
+.ck-failure-list__cause {
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+.ck-failure-list__item .ck-icon-btn { margin-left: auto; }
+
+/* Flyout: lives inline in the card's always-present footer row, so the
+   toggle appearing or disappearing never changes the card's height. The
+   panel is an absolute popover that opens upward over the card body. */
+.ck-flyout {
+  position: relative;
+  flex: none;
+}
+.ck-flyout__toggle {
+  display: inline-flex;
+  align-items: center;
+  gap: 0.3rem;
+  font-family: var(--ck-mono);
+  font-size: 0.66rem;
+  letter-spacing: 0.05em;
+  text-transform: uppercase;
+  color: var(--ck-dim);
+  cursor: pointer;
+  list-style: none;
+}
+.ck-flyout__toggle::-webkit-details-marker { display: none; }
+.ck-flyout__toggle::marker { content: ""; }
+.ck-flyout__toggle:hover { color: var(--ck-text); }
+.ck-flyout[open] .ck-flyout__toggle { color: var(--ck-text); }
+.ck-flyout__panel {
+  position: absolute;
+  bottom: calc(100% + 0.5rem);
+  right: 0;
+  z-index: 30;
+  width: 16rem;
+  max-height: 13rem;
+  overflow-y: auto;
+  margin: 0;
+  padding: 0.45rem;
+  list-style: none;
+  display: flex;
+  flex-direction: column;
+  gap: 0.2rem;
+  background: var(--ck-bg-strong);
+  border: 1px solid var(--ck-line-strong);
+  border-radius: var(--ck-radius);
+  box-shadow: 0 -14px 30px rgba(0, 0, 0, 0.5);
+}
+.ck-flyout__item {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 0.6rem;
+  padding: 0.3rem 0.4rem;
+  font-size: 0.78rem;
+  border-radius: 5px;
+}
+.ck-flyout__item:hover { background: var(--ck-surface-hover); }
+.ck-flyout__label { color: var(--ck-text); }
+.ck-flyout__meta {
+  margin-left: 0.35rem;
+  color: var(--ck-dim);
+  font-size: 0.7rem;
+}

data/app/controllers/completion_kit/dashboard_dismissals_controller.rb

@@ -0,0 +1,41 @@
+module CompletionKit
+  class DashboardDismissalsController < ApplicationController
+    WINDOW = 7.days
+
+    def create
+      record = resolve_dismissable
+      DashboardDismissal.create(dismissable: record, baseline_score: baseline_for(record))
+      render_cards
+    end
+
+    def destroy
+      DashboardDismissal.find(params[:id]).destroy
+      render_cards
+    end
+
+    private
+
+    def dismissal_params
+      params.require(:dashboard_dismissal).permit(:dismissable_type, :dismissable_id)
+    end
+
+    def resolve_dismissable
+      type = dismissal_params[:dismissable_type]
+      raise ActiveRecord::RecordNotFound unless DashboardDismissal::DISMISSABLE_TYPES.include?(type)
+      type.constantize.find(dismissal_params[:dismissable_id])
+    end
+
+    def baseline_for(record)
+      return nil unless record.is_a?(Metric)
+      DashboardStats.metric_average(record.id, since: WINDOW.ago)
+    end
+
+    def render_cards
+      @worst_metric = DashboardStats.worst_metric(since: WINDOW.ago)
+      @failures = DashboardStats.failures(since: WINDOW.ago)
+      @ignored_metrics = DashboardDismissal.metrics
+      @ignored_failures = DashboardDismissal.failures
+      render :refresh, formats: [:turbo_stream]
+    end
+  end
+end

data/app/models/completion_kit/dashboard_dismissal.rb

@@ -0,0 +1,18 @@
+module CompletionKit
+  class DashboardDismissal < ApplicationRecord
+    FAILURE_TYPES = %w[
+      CompletionKit::Run
+      CompletionKit::Response
+      CompletionKit::Review
+    ].freeze
+    DISMISSABLE_TYPES = (["CompletionKit::Metric"] + FAILURE_TYPES).freeze
+
+    belongs_to :dismissable, polymorphic: true
+
+    validates :dismissable_type, inclusion: { in: DISMISSABLE_TYPES }
+    validates :dismissable_id, uniqueness: { scope: :dismissable_type }
+
+    scope :metrics, -> { where(dismissable_type: "CompletionKit::Metric").includes(:dismissable) }
+    scope :failures, -> { where(dismissable_type: FAILURE_TYPES).includes(:dismissable) }
+  end
+end

data/app/models/completion_kit/metric.rb

@@ -13,6 +13,7 @@ module CompletionKit
     has_many :metric_group_memberships, dependent: :destroy
     has_many :metric_groups, through: :metric_group_memberships, source: :metric_group
     has_many :reviews, dependent: :nullify
+    has_many :dashboard_dismissals, as: :dismissable, dependent: :destroy
 
     serialize :rubric_bands, coder: JSON
 

data/app/models/completion_kit/response.rb

@@ -5,6 +5,7 @@ module CompletionKit
 
     belongs_to :run
     has_many :reviews, dependent: :destroy
+    has_many :dashboard_dismissals, as: :dismissable, dependent: :destroy
 
     delegate :prompt, to: :run
 

data/app/models/completion_kit/review.rb

@@ -5,6 +5,7 @@ module CompletionKit
 
     belongs_to :response
     belongs_to :metric, optional: true
+    has_many :dashboard_dismissals, as: :dismissable, dependent: :destroy
 
     validates :metric_name, presence: true
     validates :status, inclusion: { in: STATUSES }

data/app/models/completion_kit/run.rb

@@ -11,6 +11,7 @@ module CompletionKit
     has_many :run_metrics, -> { order(:position) }, dependent: :destroy
     has_many :metrics, through: :run_metrics
     has_many :suggestions, dependent: :destroy
+    has_many :dashboard_dismissals, as: :dismissable, dependent: :destroy
 
     validates :name, presence: true
     validates :status, inclusion: { in: STATUSES }

data/app/services/completion_kit/dashboard_stats.rb

@@ -0,0 +1,167 @@
+module CompletionKit
+  # Read-only aggregate queries powering the standalone dashboard cards.
+  # Each method is a small, scoped query — nothing here writes or caches.
+  class DashboardStats
+    # Runs per calendar day for the trailing `days` window, oldest first.
+    # Always returns one entry per day (count 0 for quiet days) so callers
+    # can render a fixed-width sparkline.
+    def self.activity(days: 14)
+      since = (days - 1).days.ago.to_date
+      counts = Run.where("created_at >= ?", since.beginning_of_day)
+                  .group("DATE(created_at)")
+                  .count
+      (0...days).map do |offset|
+        date = since + offset
+        { date: date, count: counts[date] || counts[date.to_s] || 0 }
+      end
+    end
+
+    # The metric with the lowest average judge score across succeeded reviews
+    # in the window — the prompt-engineering target. Dismissed metrics are
+    # skipped while their average holds at or above the score snapshotted when
+    # they were dismissed; a metric that regresses below that baseline
+    # resurfaces and its stale dismissal is cleared. Returns nil when nothing
+    # qualifies. `response` is the single worst-scoring response, for a deep
+    # link.
+    def self.worst_metric(since:)
+      averages = scored_reviews_since(since)
+                 .joins(:metric)
+                 .group("completion_kit_metrics.id")
+                 .average(:ai_score)
+      return nil if averages.empty?
+
+      dismissals = metric_dismissals
+      metrics = Metric.where(id: averages.keys).index_by(&:id)
+
+      averages.sort_by { |_id, avg| avg }.each do |metric_id, avg|
+        rounded = avg.to_f.round(2)
+        dismissal = dismissals[metric_id]
+        next if dismissal && rounded >= dismissal.baseline_score.to_f
+
+        dismissal&.destroy
+        worst = scored_reviews_since(since).where(metric_id: metric_id).order(:ai_score).first
+        metric = metrics[metric_id]
+        return {
+          metric: metric,
+          name: metric.name,
+          avg: rounded,
+          response: worst.response,
+          score: worst.ai_score.to_f
+        }
+      end
+      nil
+    end
+
+    # The rounded average judge score for one metric across the window, or nil
+    # when it has no scored reviews. Used to snapshot a dismissal's baseline.
+    def self.metric_average(metric_id, since:)
+      scored_reviews_since(since).where(metric_id: metric_id).average(:ai_score)&.to_f&.round(2)
+    end
+
+    # Everything that terminally failed in the window across all three
+    # surfaces — failed runs, failed generations, failed judge reviews —
+    # excluding any the user has dismissed. Returns a count and an items list
+    # ordered most-recent-first; each item carries its surface, the failing
+    # record, the run it belongs to (for a deep link), and a cause string.
+    def self.failures(since:)
+      dismissed = failure_dismissal_keys
+      items = []
+
+      Run.where(status: "failed").where("created_at >= ?", since).find_each do |run|
+        next if dismissed.include?(["CompletionKit::Run", run.id])
+        items << {
+          surface: "run", record: run, run: run,
+          cause: run.failure_summary.presence || "Run failed", at: run.updated_at
+        }
+      end
+
+      Response.where(status: "failed").where("created_at >= ?", since)
+              .includes(:run).find_each do |response|
+        next if dismissed.include?(["CompletionKit::Response", response.id])
+        items << {
+          surface: "generation", record: response, run: response.run,
+          cause: failure_cause(response), at: response.updated_at
+        }
+      end
+
+      Review.where(status: "failed").where("completion_kit_reviews.created_at >= ?", since)
+            .includes(response: :run).find_each do |review|
+        next if dismissed.include?(["CompletionKit::Review", review.id])
+        items << {
+          surface: "judge", record: review, run: review.response.run,
+          cause: failure_cause(review), at: review.updated_at
+        }
+      end
+
+      items.sort_by! { |item| item[:at] }
+      items.reverse!
+      { count: items.size, items: items }
+    end
+
+    # The most recent measurable change per prompt family — gains and
+    # regressions both. For each family the comparison is:
+    #   * latest scored version vs the published version, when a draft sits
+    #     ahead of what's live ("is my work-in-progress better?")
+    #   * published vs the previous scored version, when the latest version
+    #     IS the published one ("did my last publish help?")
+    # Biggest movement first. Empty until something has been iterated and
+    # re-judged on both sides of the comparison.
+    def self.prompt_changes(limit: 5)
+      scores = Review.joins(response: :run)
+                     .where(status: "succeeded")
+                     .where.not(ai_score: nil)
+                     .group("completion_kit_runs.prompt_id")
+                     .average(:ai_score)
+      return [] if scores.empty?
+
+      Prompt.where(id: scores.keys).group_by(&:family_key).filter_map do |_key, versions|
+        scored = versions.select { |v| scores[v.id] }.sort_by(&:version_number)
+        next if scored.size < 2
+
+        candidate = scored.last
+        published = versions.find(&:current?)
+        baseline =
+          if published && published != candidate && scores[published.id]
+            published
+          else
+            scored[-2]
+          end
+
+        delta = (scores[candidate.id] - scores[baseline.id]).to_f.round(2)
+        next if delta.zero?
+
+        {
+          prompt: candidate,
+          from_version: baseline.version_number,
+          to_version: candidate.version_number,
+          from_score: scores[baseline.id].to_f.round(2),
+          to_score: scores[candidate.id].to_f.round(2),
+          delta: delta
+        }
+      end.sort_by { |row| -row[:delta].abs }.first(limit)
+    end
+
+    def self.scored_reviews_since(since)
+      Review.joins(:response)
+            .where(status: "succeeded")
+            .where("completion_kit_reviews.created_at >= ?", since)
+            .where.not(ai_score: nil)
+    end
+    private_class_method :scored_reviews_since
+
+    def self.metric_dismissals
+      DashboardDismissal.where(dismissable_type: "CompletionKit::Metric").index_by(&:dismissable_id)
+    end
+    private_class_method :metric_dismissals
+
+    def self.failure_dismissal_keys
+      DashboardDismissal.failures.map { |d| [d.dismissable_type, d.dismissable_id] }.to_set
+    end
+    private_class_method :failure_dismissal_keys
+
+    def self.failure_cause(record)
+      record.error_class.presence || "Unknown error"
+    end
+    private_class_method :failure_cause
+  end
+end

data/app/views/completion_kit/dashboard/_eye_icon.html.erb

@@ -0,0 +1 @@
+<svg viewBox="0 0 24 24" width="14" height="14" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M1 12s4-8 11-8 11 8 11 8-4 8-11 8-11-8-11-8z"/><circle cx="12" cy="12" r="3"/></svg>

data/app/views/completion_kit/dashboard/_eye_off_icon.html.erb

@@ -0,0 +1 @@
+<svg viewBox="0 0 24 24" width="14" height="14" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true"><path d="M17.94 17.94A10.07 10.07 0 0 1 12 20c-7 0-11-8-11-8a18.45 18.45 0 0 1 5.06-5.94M9.9 4.24A9.12 9.12 0 0 1 12 4c7 0 11 8 11 8a18.5 18.5 0 0 1-2.16 3.19m-6.72-1.07a3 3 0 1 1-4.24-4.24"/><line x1="1" y1="1" x2="23" y2="23"/></svg>

data/app/views/completion_kit/dashboard/_failures_card.html.erb

@@ -0,0 +1,47 @@
+<div class="ck-card ck-stat-card ck-rise" id="ck-failures-card" style="--rise-delay: 180ms;">
+  <p class="ck-kicker">Failures · last 7 days</p>
+  <div class="ck-stat-card__body">
+    <span class="ck-stat-card__count<%= failures[:count].positive? ? ' is-danger' : ' is-clean' %>"><%= failures[:count] %></span>
+  </div>
+
+  <% if failures[:items].any? %>
+    <ul class="ck-failure-list">
+      <% failures[:items].each do |item| %>
+        <li class="ck-failure-list__item">
+          <span class="ck-failure-list__surface ck-failure-list__surface--<%= item[:surface] %>"><%= item[:surface] %></span>
+          <% if item[:run] %>
+            <%= link_to item[:cause], completion_kit.run_path(item[:run]), class: "ck-link ck-failure-list__cause" %>
+          <% else %>
+            <span class="ck-failure-list__cause"><%= item[:cause] %></span>
+          <% end %>
+          <%= button_to completion_kit.dashboard_dismissals_path,
+                params: { dashboard_dismissal: { dismissable_type: item[:record].class.name,
+                                                 dismissable_id: item[:record].id } },
+                class: "ck-icon-btn", title: "Ignore this failure",
+                "aria-label": "Ignore #{item[:surface]} failure" do %><%= render "completion_kit/dashboard/eye_off_icon" %><% end %>
+        </li>
+      <% end %>
+    </ul>
+  <% end %>
+
+  <% if failures[:items].empty? || ignored_failures.any? %>
+    <div class="ck-stat-card__foot ck-stat-card__foot--split">
+      <span><% if failures[:items].empty? %>All clear — nothing failed this week.<% end %></span>
+      <% if ignored_failures.any? %>
+        <details class="ck-flyout">
+          <summary class="ck-flyout__toggle"><%= ignored_failures.size %> ignored</summary>
+          <ul class="ck-flyout__panel">
+            <% ignored_failures.each do |dismissal| %>
+              <li class="ck-flyout__item">
+                <span class="ck-flyout__label"><%= dismissal.dismissable_type.demodulize %> #<%= dismissal.dismissable_id %></span>
+                <%= button_to completion_kit.dashboard_dismissal_path(dismissal),
+                      method: :delete, class: "ck-icon-btn", title: "Un-ignore",
+                      "aria-label": "Un-ignore #{dismissal.dismissable_type.demodulize} #{dismissal.dismissable_id}" do %><%= render "completion_kit/dashboard/eye_icon" %><% end %>
+              </li>
+            <% end %>
+          </ul>
+        </details>
+      <% end %>
+    </div>
+  <% end %>
+</div>

data/app/views/completion_kit/dashboard/_worst_metric_card.html.erb

@@ -0,0 +1,54 @@
+<div class="ck-card ck-stat-card ck-rise" id="ck-worst-metric-card" style="--rise-delay: 120ms;">
+  <p class="ck-kicker">Worst metric · last 7 days</p>
+  <div class="ck-stat-card__body">
+    <% if worst_metric %>
+      <span class="ck-stat-card__metric"><%= worst_metric[:name] %></span>
+      <span class="<%= ck_badge_classes(ck_score_kind(worst_metric[:avg])) %> ck-stat-card__score"><%= worst_metric[:avg] %></span>
+    <% else %>
+      <span class="ck-stat-card__metric ck-stat-card__metric--empty">No scored reviews</span>
+    <% end %>
+  </div>
+
+  <div class="ck-stat-card__foot ck-stat-card__foot--split">
+    <span>
+      <% if worst_metric && worst_metric[:response] %>
+        <%= link_to "Worst response →",
+              completion_kit.run_response_path(worst_metric[:response].run, worst_metric[:response]),
+              class: "ck-link" %>
+      <% elsif worst_metric %>
+        Lowest score this week.
+      <% else %>
+        Run a judge to populate this.
+      <% end %>
+    </span>
+
+    <% if ignored_metrics.any? %>
+      <details class="ck-flyout">
+        <summary class="ck-flyout__toggle"><%= ignored_metrics.size %> ignored</summary>
+        <ul class="ck-flyout__panel">
+          <% ignored_metrics.each do |dismissal| %>
+            <li class="ck-flyout__item">
+              <span class="ck-flyout__label">
+                <%= dismissal.dismissable.name %>
+                <% if dismissal.baseline_score %>
+                  <span class="ck-flyout__meta">baseline <%= dismissal.baseline_score %></span>
+                <% end %>
+              </span>
+              <%= button_to completion_kit.dashboard_dismissal_path(dismissal),
+                    method: :delete, class: "ck-icon-btn", title: "Un-ignore",
+                    "aria-label": "Un-ignore #{dismissal.dismissable.name}" do %><%= render "completion_kit/dashboard/eye_icon" %><% end %>
+            </li>
+          <% end %>
+        </ul>
+      </details>
+    <% end %>
+
+    <% if worst_metric %>
+      <%= button_to completion_kit.dashboard_dismissals_path,
+            params: { dashboard_dismissal: { dismissable_type: "CompletionKit::Metric",
+                                             dismissable_id: worst_metric[:metric].id } },
+            class: "ck-icon-btn", title: "Ignore this metric",
+            "aria-label": "Ignore #{worst_metric[:name]}" do %><%= render "completion_kit/dashboard/eye_off_icon" %><% end %>
+    <% end %>
+  </div>
+</div>

data/app/views/completion_kit/dashboard_dismissals/refresh.turbo_stream.erb

@@ -0,0 +1,8 @@
+<%= turbo_stream.replace "ck-worst-metric-card" do %>
+  <%= render "completion_kit/dashboard/worst_metric_card",
+             worst_metric: @worst_metric, ignored_metrics: @ignored_metrics %>
+<% end %>
+<%= turbo_stream.replace "ck-failures-card" do %>
+  <%= render "completion_kit/dashboard/failures_card",
+             failures: @failures, ignored_failures: @ignored_failures %>
+<% end %>

data/config/routes.rb

@@ -15,6 +15,7 @@ CompletionKit::Engine.routes.draw do
   resources :metrics
   resources :metric_groups
   resources :tags
+  resources :dashboard_dismissals, only: [:create, :destroy]
 
   resources :runs do
     member do

data/db/migrate/20260516000001_create_completion_kit_dashboard_dismissals.rb

@@ -0,0 +1,15 @@
+class CreateCompletionKitDashboardDismissals < ActiveRecord::Migration[8.1]
+  def change
+    create_table :completion_kit_dashboard_dismissals do |t|
+      t.string :dismissable_type, null: false
+      t.bigint :dismissable_id, null: false
+      t.decimal :baseline_score, precision: 4, scale: 1
+      t.timestamps
+    end
+
+    add_index :completion_kit_dashboard_dismissals,
+              [:dismissable_type, :dismissable_id],
+              unique: true,
+              name: "index_ck_dashboard_dismissals_on_dismissable"
+  end
+end

data/lib/completion_kit/version.rb

@@ -1,3 +1,3 @@
 module CompletionKit
-  VERSION = "0.5.18"
+  VERSION = "0.5.20"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: completion-kit
 version: !ruby/object:Gem::Version
-  version: 0.5.18
+  version: 0.5.20
 platform: ruby
 authors:
 - Damien Bastin
@@ -242,6 +242,7 @@ files:
 - app/controllers/completion_kit/api/v1/tags_controller.rb
 - app/controllers/completion_kit/api_reference_controller.rb
 - app/controllers/completion_kit/application_controller.rb
+- app/controllers/completion_kit/dashboard_dismissals_controller.rb
 - app/controllers/completion_kit/datasets_controller.rb
 - app/controllers/completion_kit/mcp_controller.rb
 - app/controllers/completion_kit/metric_groups_controller.rb
@@ -262,6 +263,7 @@ files:
 - app/jobs/completion_kit/run_completion_check_job.rb
 - app/mailers/completion_kit/application_mailer.rb
 - app/models/completion_kit/application_record.rb
+- app/models/completion_kit/dashboard_dismissal.rb
 - app/models/completion_kit/dataset.rb
 - app/models/completion_kit/mcp_session.rb
 - app/models/completion_kit/metric.rb
@@ -281,6 +283,7 @@ files:
 - app/services/completion_kit/anthropic_client.rb
 - app/services/completion_kit/api_config.rb
 - app/services/completion_kit/csv_processor.rb
+- app/services/completion_kit/dashboard_stats.rb
 - app/services/completion_kit/judge_service.rb
 - app/services/completion_kit/llm_client.rb
 - app/services/completion_kit/mcp_dispatcher.rb
@@ -308,6 +311,11 @@ files:
 - app/views/completion_kit/api_reference/_resource_card.html.erb
 - app/views/completion_kit/api_reference/_resource_list.html.erb
 - app/views/completion_kit/api_reference/index.html.erb
+- app/views/completion_kit/dashboard/_eye_icon.html.erb
+- app/views/completion_kit/dashboard/_eye_off_icon.html.erb
+- app/views/completion_kit/dashboard/_failures_card.html.erb
+- app/views/completion_kit/dashboard/_worst_metric_card.html.erb
+- app/views/completion_kit/dashboard_dismissals/refresh.turbo_stream.erb
 - app/views/completion_kit/datasets/_form.html.erb
 - app/views/completion_kit/datasets/edit.html.erb
 - app/views/completion_kit/datasets/index.html.erb
@@ -382,6 +390,7 @@ files:
 - db/migrate/20260509000002_create_completion_kit_taggings.rb
 - db/migrate/20260513000001_create_completion_kit_mcp_sessions.rb
 - db/migrate/20260514000001_allow_judge_only_runs.rb
+- db/migrate/20260516000001_create_completion_kit_dashboard_dismissals.rb
 - lib/completion-kit.rb
 - lib/completion_kit.rb
 - lib/completion_kit/concurrency_check.rb