completion-kit 0.5.18 → 0.5.19

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 25e795a8d94b2f0984e65d312eb152ffabcda8e7929a52529906ff1fab6cfa4f
-  data.tar.gz: 91110f1c5736d0c6d3c12ef5aa5a75b66ffe0d565cf5777bd9f2d322936e01d5
+  metadata.gz: fe8be5a5838f6a270f5ed934e250f0dd55d673b51ba15ed29579429dadcaaa03
+  data.tar.gz: 6044e3c6805e697c1e8e80b28760ab5f749e2ba14bd976b67225f0d3cffb6eff
 SHA512:
-  metadata.gz: 87678468d49193e088d4b1c6c28abe5a63ef67de679cfeb894dd64238810d456e7505bae78224e2c330860f177deff8bf834fab0a1c9abdca4e2cafd81e5463b
-  data.tar.gz: 2755a9885f3242017dce04a91680f4499b5cd8907f1043f9bab41d46651ec234ebe4c0f03886b72a0dac421a8787f56923da884b6d5112f85c409392dbfea969
+  metadata.gz: efc0e7b9a1b23eeec3b3ba91ab65d94d2454bec0554cdd8e36ab6d1286b9c208de2e1ea24fec1c19f73d01dc044097df6e2ac004a0de2565bc6e6b4965af1d85
+  data.tar.gz: e9fca34118e8a1b10107807e1243f946ad6148f7af8e92d23c3c9bfc20f10f2d91c6db0c4b285996f8267b6511593417a11a7f6c1002b9d8dd2a9f32f5793860

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

@@ -345,6 +345,176 @@ 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__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 +778,18 @@ 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 {

data/app/services/completion_kit/dashboard_stats.rb

@@ -0,0 +1,99 @@
+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. Returns nil when there
+    # are no scored reviews. `response` is the single worst-scoring response
+    # for that metric, for a deep link.
+    def self.worst_metric(since:)
+      averages = scored_reviews_since(since).group(:metric_name).average(:ai_score)
+      return nil if averages.empty?
+
+      name, avg = averages.min_by { |_, value| value }
+      # averages is non-empty, so at least one review carries this
+      # metric_name — worst is always present here.
+      worst = scored_reviews_since(since)
+              .where(metric_name: name)
+              .order(:ai_score)
+              .first
+      {
+        name: name,
+        avg: avg.to_f.round(2),
+        response: worst.response,
+        score: worst.ai_score.to_f
+      }
+    end
+
+    # Reviews that terminally failed in the window — parse failures, judge
+    # truncations, provider errors. Invisible on the dashboard otherwise.
+    def self.failed_review_count(since:)
+      Review.where(status: "failed").where("created_at >= ?", since).count
+    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
+  end
+end

data/lib/completion_kit/version.rb

@@ -1,3 +1,3 @@
 module CompletionKit
-  VERSION = "0.5.18"
+  VERSION = "0.5.19"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: completion-kit
 version: !ruby/object:Gem::Version
-  version: 0.5.18
+  version: 0.5.19
 platform: ruby
 authors:
 - Damien Bastin
@@ -281,6 +281,7 @@ 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