completion-kit 0.11.0 → 0.12.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (35) hide show
  1. checksums.yaml +4 -4
  2. data/app/assets/stylesheets/completion_kit/application.css +51 -51
  3. data/app/controllers/completion_kit/{calibrations_controller.rb → agreements_controller.rb} +19 -19
  4. data/app/controllers/completion_kit/api/v1/{calibrations_controller.rb → agreements_controller.rb} +18 -18
  5. data/app/controllers/completion_kit/api/v1/metric_versions_controller.rb +2 -7
  6. data/app/controllers/completion_kit/api/v1/metrics_controller.rb +1 -1
  7. data/app/controllers/completion_kit/metrics_controller.rb +10 -11
  8. data/app/jobs/completion_kit/judge_review_job.rb +2 -2
  9. data/app/models/completion_kit/{calibration.rb → agreement.rb} +1 -1
  10. data/app/models/completion_kit/metric_version.rb +1 -17
  11. data/app/models/completion_kit/review.rb +1 -0
  12. data/app/services/completion_kit/{calibration_math.rb → agreement_math.rb} +1 -1
  13. data/app/services/completion_kit/mcp_dispatcher.rb +2 -2
  14. data/app/services/completion_kit/mcp_tools/{calibrations.rb → agreements.rb} +11 -11
  15. data/app/services/completion_kit/mcp_tools/judges.rb +3 -3
  16. data/app/services/completion_kit/mcp_tools/metric_versions.rb +2 -7
  17. data/app/services/completion_kit/{metric_calibration_examples.rb → metric_agreement_examples.rb} +6 -6
  18. data/app/services/completion_kit/{metric_calibration_stats.rb → metric_agreement_stats.rb} +6 -6
  19. data/app/services/completion_kit/metric_improvement_validator.rb +1 -1
  20. data/app/services/completion_kit/metric_variant_generator.rb +2 -2
  21. data/app/views/completion_kit/{calibrations → agreements}/_buttons.html.erb +33 -33
  22. data/app/views/completion_kit/{calibrations → agreements}/_trust_panel.html.erb +5 -5
  23. data/app/views/completion_kit/api_reference/_body.html.erb +15 -15
  24. data/app/views/completion_kit/metrics/_guiding_examples.html.erb +1 -1
  25. data/app/views/completion_kit/metrics/edit.html.erb +1 -1
  26. data/app/views/completion_kit/metrics/show.html.erb +6 -6
  27. data/app/views/completion_kit/responses/show.html.erb +4 -4
  28. data/app/views/completion_kit/runs/show.html.erb +1 -1
  29. data/config/routes.rb +3 -3
  30. data/db/migrate/20260531000002_backfill_review_metric_versions.rb +33 -0
  31. data/db/migrate/20260531000003_add_metric_version_fk_to_reviews.rb +6 -0
  32. data/db/migrate/20260531000004_rename_calibrations_to_agreements.rb +19 -0
  33. data/lib/completion_kit/version.rb +1 -1
  34. data/lib/completion_kit.rb +2 -2
  35. metadata +13 -10
@@ -1,11 +1,11 @@
1
1
  module CompletionKit
2
2
  module McpTools
3
- module Calibrations
3
+ module Agreements
4
4
  extend Base
5
5
 
6
6
  TOOLS = {
7
- "calibrations_list" => {
8
- description: "List calibrations. Filter by run_id, response_id, metric_id, or created_by.",
7
+ "agreements_list" => {
8
+ description: "List agreements. Filter by run_id, response_id, metric_id, or created_by.",
9
9
  inputSchema: {
10
10
  type: "object",
11
11
  properties: {
@@ -18,8 +18,8 @@ module CompletionKit
18
18
  },
19
19
  handler: :list
20
20
  },
21
- "calibrations_create" => {
22
- description: "Upsert a calibration for (run, response, metric, created_by). Verdict is one of agree, disagree, borderline. corrected_score (1..5) is required when verdict is 'disagree'.",
21
+ "agreements_create" => {
22
+ description: "Upsert an agreement for (run, response, metric, created_by). Verdict is one of agree, disagree, borderline. corrected_score (1..5) is required when verdict is 'disagree'.",
23
23
  inputSchema: {
24
24
  type: "object",
25
25
  properties: {
@@ -38,7 +38,7 @@ module CompletionKit
38
38
  }.freeze
39
39
 
40
40
  def self.list(args)
41
- scope = CompletionKit::Calibration.all
41
+ scope = CompletionKit::Agreement.all
42
42
  scope = scope.where(run_id: args["run_id"]) if args["run_id"]
43
43
  scope = scope.where(response_id: args["response_id"]) if args["response_id"]
44
44
  scope = scope.where(metric_id: args["metric_id"]) if args["metric_id"]
@@ -52,20 +52,20 @@ module CompletionKit
52
52
  metric = CompletionKit::Metric.find(args["metric_id"])
53
53
  created_by = args["created_by"].presence || "mcp"
54
54
 
55
- calibration = CompletionKit::Calibration.find_or_initialize_by(
55
+ agreement = CompletionKit::Agreement.find_or_initialize_by(
56
56
  run_id: run.id, response_id: response.id, metric_id: metric.id, created_by: created_by
57
57
  )
58
- calibration.assign_attributes(
58
+ agreement.assign_attributes(
59
59
  metric_version: CompletionKit::MetricVersion.ensure_current_for(metric),
60
60
  verdict: args["verdict"],
61
61
  corrected_score: args["corrected_score"],
62
62
  note: args["note"]
63
63
  )
64
64
 
65
- if calibration.save
66
- text_result(calibration.as_json)
65
+ if agreement.save
66
+ text_result(agreement.as_json)
67
67
  else
68
- error_result(calibration.errors.full_messages.join(", "))
68
+ error_result(agreement.errors.full_messages.join(", "))
69
69
  end
70
70
  end
71
71
  end
@@ -33,7 +33,7 @@ module CompletionKit
33
33
  handler: :replay
34
34
  },
35
35
  "judges_compare" => {
36
- description: "Compare two metric versions' calibration stats side by side. Pass either two metric_version_ids or one metric_id with metric_version_a_id / metric_version_b_id.",
36
+ description: "Compare two metric versions' agreement stats side by side. Pass either two metric_version_ids or one metric_id with metric_version_a_id / metric_version_b_id.",
37
37
  inputSchema: {
38
38
  type: "object",
39
39
  properties: {
@@ -77,8 +77,8 @@ module CompletionKit
77
77
  metric = CompletionKit::Metric.find(args["metric_id"])
78
78
  a = CompletionKit::MetricVersion.where(metric_id: metric.id).find(args["metric_version_a_id"])
79
79
  b = CompletionKit::MetricVersion.where(metric_id: metric.id).find(args["metric_version_b_id"])
80
- stats_a = CompletionKit::MetricCalibrationStats.for(metric, metric_version: a)
81
- stats_b = CompletionKit::MetricCalibrationStats.for(metric, metric_version: b)
80
+ stats_a = CompletionKit::MetricAgreementStats.for(metric, metric_version: a)
81
+ stats_b = CompletionKit::MetricAgreementStats.for(metric, metric_version: b)
82
82
  text_result({
83
83
  metric_id: metric.id,
84
84
  a: metric_version_payload(a, stats_a),
@@ -47,13 +47,8 @@ module CompletionKit
47
47
 
48
48
  def self.publish(args)
49
49
  version = CompletionKit::MetricVersion.find(args["metric_version_id"])
50
- if version.published? && !version.current?
51
- audit = version.revert!
52
- text_result(audit.as_json)
53
- else
54
- version.publish!
55
- text_result(version.reload.as_json)
56
- end
50
+ version.publish!
51
+ text_result(version.reload.as_json)
57
52
  end
58
53
 
59
54
  def self.dismiss(args)
@@ -1,5 +1,5 @@
1
1
  module CompletionKit
2
- module MetricCalibrationExamples
2
+ module MetricAgreementExamples
3
3
  DEFAULT_JUDGE_EXAMPLE_LIMIT = 5
4
4
 
5
5
  module_function
@@ -9,18 +9,18 @@ module CompletionKit
9
9
  end
10
10
 
11
11
  def disagreements_for(metric, limit: 8)
12
- calibrations_for(metric, verdict: "disagree", limit: limit)
12
+ agreements_for(metric, verdict: "disagree", limit: limit)
13
13
  end
14
14
 
15
15
  def borderlines_for(metric, limit: 6)
16
- calibrations_for(metric, verdict: "borderline", limit: limit)
16
+ agreements_for(metric, verdict: "borderline", limit: limit)
17
17
  end
18
18
 
19
19
  def judge_examples_for(metric, exclude_response_id: nil, limit: DEFAULT_JUDGE_EXAMPLE_LIMIT)
20
20
  current_version = MetricVersion.current.find_by(metric_id: metric.id)
21
21
  return [] unless current_version
22
22
 
23
- relation = Calibration
23
+ relation = Agreement
24
24
  .where(metric_id: metric.id, metric_version_id: current_version.id, excluded_from_examples: false)
25
25
  .where.not(corrected_score: nil)
26
26
  relation = relation.where.not(response_id: exclude_response_id) if exclude_response_id
@@ -28,8 +28,8 @@ module CompletionKit
28
28
  .reject { |example| example[:judge_score].nil? }
29
29
  end
30
30
 
31
- def calibrations_for(metric, verdict:, limit:)
32
- base = Calibration.where(metric_id: metric.id, verdict: verdict)
31
+ def agreements_for(metric, verdict:, limit:)
32
+ base = Agreement.where(metric_id: metric.id, verdict: verdict)
33
33
  current_version = MetricVersion.current.find_by(metric_id: metric.id)
34
34
  scoped = current_version ? base.where(metric_version_id: current_version.id) : base
35
35
  effective = scoped.exists? ? scoped : base
@@ -1,5 +1,5 @@
1
1
  module CompletionKit
2
- class MetricCalibrationStats
2
+ class MetricAgreementStats
3
3
  PROVISIONAL_MIN = 10
4
4
  FIRM_MIN = 30
5
5
 
@@ -49,7 +49,7 @@ module CompletionKit
49
49
  end
50
50
 
51
51
  def call
52
- scope = Calibration.where(metric_id: @metric.id)
52
+ scope = Agreement.where(metric_id: @metric.id)
53
53
  if @metric_version
54
54
  scope = scope.where(metric_version_id: @metric_version.id)
55
55
  elsif !@all_versions
@@ -62,12 +62,12 @@ module CompletionKit
62
62
  disagrees = verdicts.count { |v, _, _| v == "disagree" }
63
63
  borderlines = verdicts.count { |v, _, _| v == "borderline" }
64
64
 
65
- ci = CalibrationMath.wilson_interval(successes: agrees, n: n)
65
+ ci = AgreementMath.wilson_interval(successes: agrees, n: n)
66
66
 
67
67
  pairs = score_pairs(verdicts)
68
- mae_value = CalibrationMath.mae(pairs)
69
- pearson_value = CalibrationMath.pearson(pairs)
70
- kappa_value = CalibrationMath.quadratic_weighted_kappa(pairs, categories: 1..5)
68
+ mae_value = AgreementMath.mae(pairs)
69
+ pearson_value = AgreementMath.pearson(pairs)
70
+ kappa_value = AgreementMath.quadratic_weighted_kappa(pairs, categories: 1..5)
71
71
 
72
72
  Result.new(
73
73
  sample_size: n,
@@ -28,7 +28,7 @@ module CompletionKit
28
28
  current = MetricVersion.current.find_by(metric_id: @metric.id)
29
29
  return [] unless current
30
30
 
31
- base = Calibration.where(metric_id: @metric.id, metric_version_id: current.id, verdict: %w[agree disagree])
31
+ base = Agreement.where(metric_id: @metric.id, metric_version_id: current.id, verdict: %w[agree disagree])
32
32
  @key_size_before_cap = base.count
33
33
  base.includes(response: :reviews)
34
34
  .order(created_at: :desc)
@@ -41,8 +41,8 @@ module CompletionKit
41
41
  private
42
42
 
43
43
  def build_meta_prompt
44
- disagreements = MetricCalibrationExamples.disagreements_for(@metric)
45
- borderlines = MetricCalibrationExamples.borderlines_for(@metric)
44
+ disagreements = MetricAgreementExamples.disagreements_for(@metric)
45
+ borderlines = MetricAgreementExamples.borderlines_for(@metric)
46
46
  sections = []
47
47
  sections << "You are an expert evaluator. The judge below is misaligned with humans. Propose #{@count == 1 ? "a single" : "#{@count}"} concrete rewrite that closes the gap."
48
48
  sections << ""
@@ -1,34 +1,34 @@
1
- <div id="calibration_<%= response_row.id %>_<%= metric.id %>" class="ck-calibration">
2
- <% current_verdict = calibration&.verdict %>
1
+ <div id="agreement_<%= response_row.id %>_<%= metric.id %>" class="ck-agreement">
2
+ <% current_verdict = agreement&.verdict %>
3
3
  <% pending_verdict = local_assigns[:pending_verdict] %>
4
4
  <% active_verdict = pending_verdict || current_verdict %>
5
5
  <% error = local_assigns[:error] %>
6
6
  <% me = CompletionKit.config.username.presence || "operator" %>
7
- <% other_calibrations = CompletionKit::Calibration
7
+ <% other_agreements = CompletionKit::Agreement
8
8
  .where(response_id: response_row.id, metric_id: metric.id)
9
9
  .where.not(created_by: me)
10
10
  .order(created_at: :asc).to_a %>
11
11
  <% verdict_icons = { "agree" => "hand-thumb-up", "disagree" => "hand-thumb-down", "borderline" => "scale" } %>
12
- <p class="ck-calibration__prompt">
13
- <span class="ck-calibration__label">Your verdict</span>
14
- <% if other_calibrations.any? %>
15
- <span class="ck-calibration__meta"><%= pluralize(other_calibrations.size, "other verdict") %> on this score</span>
16
- <span class="ck-calibration__sep">·</span>
12
+ <p class="ck-agreement__prompt">
13
+ <span class="ck-agreement__label">Your verdict</span>
14
+ <% if other_agreements.any? %>
15
+ <span class="ck-agreement__meta"><%= pluralize(other_agreements.size, "other verdict") %> on this score</span>
16
+ <span class="ck-agreement__sep">·</span>
17
17
  <% end %>
18
- <%= link_to metric_path(metric), class: "ck-calibration__meta-link" do %><%= heroicon_tag "adjustments-horizontal", variant: :outline, "aria-hidden": "true" %>Calibration →<% end %>
18
+ <%= link_to metric_path(metric, anchor: "agreement"), class: "ck-agreement__meta-link" do %><%= heroicon_tag "adjustments-horizontal", variant: :outline, "aria-hidden": "true" %>Agreement →<% end %>
19
19
  </p>
20
- <div class="ck-calibration__buttons">
20
+ <div class="ck-agreement__buttons">
21
21
  <% verdict_icons = { "agree" => "hand-thumb-up", "disagree" => "hand-thumb-down", "borderline" => "scale" } %>
22
22
  <% verdict_hints = {
23
23
  "agree" => "The score looks right.",
24
24
  "disagree" => "The score is wrong — you'll pick the right one.",
25
25
  "borderline" => "The rubric is unclear here; either score could be defensible."
26
26
  } %>
27
- <% CompletionKit::Calibration::VERDICTS.each do |verdict| %>
28
- <%= button_to run_response_calibrations_path(run, response_row, metric_id: metric.id, verdict: verdict),
27
+ <% CompletionKit::Agreement::VERDICTS.each do |verdict| %>
28
+ <%= button_to run_response_agreements_path(run, response_row, metric_id: metric.id, verdict: verdict),
29
29
  method: :post,
30
30
  form: { data: { turbo: "true" } },
31
- class: "ck-calibration__pill ck-calibration__pill--#{verdict}#{' is-active' if verdict == active_verdict}",
31
+ class: "ck-agreement__pill ck-agreement__pill--#{verdict}#{' is-active' if verdict == active_verdict}",
32
32
  "aria-pressed": (verdict == active_verdict).to_s,
33
33
  title: verdict_hints[verdict] do %>
34
34
  <%= heroicon_tag verdict_icons[verdict], variant: :outline, size: 14, "aria-hidden": "true" %>
@@ -38,26 +38,26 @@
38
38
  </div>
39
39
 
40
40
  <% if error.present? %>
41
- <p class="ck-calibration__error" role="alert"><%= error %></p>
41
+ <p class="ck-agreement__error" role="alert"><%= error %></p>
42
42
  <% end %>
43
43
 
44
- <% if other_calibrations.any? %>
45
- <details class="ck-calibration__others">
46
- <summary class="ck-calibration__others-summary">
44
+ <% if other_agreements.any? %>
45
+ <details class="ck-agreement__others">
46
+ <summary class="ck-agreement__others-summary">
47
47
  <%= heroicon_tag "chevron-right", variant: :outline, size: 14, "aria-hidden": "true" %>
48
- <span>What others said (<%= other_calibrations.size %>)</span>
48
+ <span>What others said (<%= other_agreements.size %>)</span>
49
49
  </summary>
50
- <ul class="ck-calibration__others-list">
51
- <% other_calibrations.each do |other| %>
52
- <li class="ck-calibration__others-item ck-calibration__others-item--<%= other.verdict %>">
53
- <div class="ck-calibration__others-row">
54
- <span class="ck-calibration__others-verdict">
50
+ <ul class="ck-agreement__others-list">
51
+ <% other_agreements.each do |other| %>
52
+ <li class="ck-agreement__others-item ck-agreement__others-item--<%= other.verdict %>">
53
+ <div class="ck-agreement__others-row">
54
+ <span class="ck-agreement__others-verdict">
55
55
  <%= heroicon_tag verdict_icons[other.verdict], variant: :outline, size: 14, "aria-hidden": "true" %>
56
56
  <%= other.verdict %>
57
57
  </span>
58
- <span class="ck-calibration__others-by"><%= other.created_by %></span>
58
+ <span class="ck-agreement__others-by"><%= other.created_by %></span>
59
59
  <% if other.corrected_score %>
60
- <span class="ck-calibration__others-stars" aria-label="<%= pluralize(other.corrected_score.to_i, 'star') %>" title="<%= pluralize(other.corrected_score.to_i, 'star') %>">
60
+ <span class="ck-agreement__others-stars" aria-label="<%= pluralize(other.corrected_score.to_i, 'star') %>" title="<%= pluralize(other.corrected_score.to_i, 'star') %>">
61
61
  <% 5.times do |i| %>
62
62
  <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>
63
63
  <% end %>
@@ -65,7 +65,7 @@
65
65
  <% end %>
66
66
  </div>
67
67
  <% if other.note.to_s.present? %>
68
- <p class="ck-calibration__others-note">"<%= other.note.to_s.truncate(140) %>"</p>
68
+ <p class="ck-agreement__others-note">"<%= other.note.to_s.truncate(140) %>"</p>
69
69
  <% end %>
70
70
  </li>
71
71
  <% end %>
@@ -74,10 +74,10 @@
74
74
  <% end %>
75
75
 
76
76
  <% if active_verdict == "disagree" %>
77
- <% existing_score = (calibration&.corrected_score || review&.ai_score)&.round %>
78
- <%= form_with url: run_response_calibrations_path(run, response_row),
77
+ <% existing_score = (agreement&.corrected_score || review&.ai_score)&.round %>
78
+ <%= form_with url: run_response_agreements_path(run, response_row),
79
79
  method: :post, local: false,
80
- class: "ck-calibration__detail" do |f| %>
80
+ class: "ck-agreement__detail" do |f| %>
81
81
  <%= hidden_field_tag :metric_id, metric.id %>
82
82
  <%= hidden_field_tag :verdict, "disagree" %>
83
83
  <p class="ck-label">What should the score have been?</p>
@@ -93,16 +93,16 @@
93
93
  <% end %>
94
94
  </div>
95
95
  </fieldset>
96
- <textarea name="note" rows="2" class="ck-input ck-input--area" placeholder="Why? (optional)"><%= calibration&.note %></textarea>
96
+ <textarea name="note" rows="2" class="ck-input ck-input--area" placeholder="Why? (optional)"><%= agreement&.note %></textarea>
97
97
  <%= f.submit "Save", class: "#{ck_button_classes(:dark)}#{' ck-button--just-saved' if local_assigns[:just_saved]}" %>
98
98
  <% end %>
99
99
  <% elsif active_verdict == "borderline" %>
100
- <%= form_with url: run_response_calibrations_path(run, response_row),
100
+ <%= form_with url: run_response_agreements_path(run, response_row),
101
101
  method: :post, local: false,
102
- class: "ck-calibration__detail" do |f| %>
102
+ class: "ck-agreement__detail" do |f| %>
103
103
  <%= hidden_field_tag :metric_id, metric.id %>
104
104
  <%= hidden_field_tag :verdict, "borderline" %>
105
- <textarea name="note" rows="2" class="ck-input ck-input--area" placeholder="What made this borderline? (optional)"><%= calibration&.note %></textarea>
105
+ <textarea name="note" rows="2" class="ck-input ck-input--area" placeholder="What made this borderline? (optional)"><%= agreement&.note %></textarea>
106
106
  <%= f.submit "Save", class: "#{ck_button_classes(:dark)}#{' ck-button--just-saved' if local_assigns[:just_saved]}" %>
107
107
  <% end %>
108
108
  <% end %>
@@ -4,7 +4,7 @@
4
4
  <% current_metric_version = metric && CompletionKit::MetricVersion.current.find_by(metric_id: metric.id) %>
5
5
  <% target_response = if (stats.sample_size.zero? || stats.counter_only?) && metric && current_metric_version
6
6
  created_by = CompletionKit.config.username.presence || "operator"
7
- verdicted_ids = CompletionKit::Calibration.where(metric_id: metric.id, created_by: created_by, metric_version_id: current_metric_version.id).pluck(:response_id)
7
+ verdicted_ids = CompletionKit::Agreement.where(metric_id: metric.id, created_by: created_by, metric_version_id: current_metric_version.id).pluck(:response_id)
8
8
  CompletionKit::Response.joins(:reviews)
9
9
  .where(reviews: { metric_id: metric.id, metric_version_id: current_metric_version.id })
10
10
  .where.not(reviews: { ai_score: nil })
@@ -12,20 +12,20 @@
12
12
  .order(created_at: :desc).first
13
13
  end %>
14
14
  <% prior_version_verdicts = if stats.sample_size.zero? && metric && current_metric_version
15
- CompletionKit::Calibration.where(metric_id: metric.id).where.not(metric_version_id: current_metric_version.id).count
15
+ CompletionKit::Agreement.where(metric_id: metric.id).where.not(metric_version_id: current_metric_version.id).count
16
16
  else
17
17
  0
18
18
  end %>
19
19
 
20
- <p class="ck-trust-line ck-trust-line--<%= stats.gate %>">
20
+ <p id="agreement" class="ck-trust-line ck-trust-line--<%= stats.gate %>">
21
21
  <% if stats.sample_size.zero? %>
22
22
  <span class="ck-trust-line__lead">Not measured yet.</span>
23
- <span class="ck-trust-line__hint"><%= current_metric_version ? "#{current_metric_version.version_label} needs" : "Needs" %> <%= CompletionKit::MetricCalibrationStats::PROVISIONAL_MIN %> human reviews of the judge's scores.</span>
23
+ <span class="ck-trust-line__hint"><%= current_metric_version ? "#{current_metric_version.version_label} needs" : "Needs" %> <%= CompletionKit::MetricAgreementStats::PROVISIONAL_MIN %> human reviews of the judge's scores.</span>
24
24
  <% if target_response %>
25
25
  <%= link_to "Review a judge's score →", run_response_path(target_response.run, target_response, anchor: anchor), class: "ck-cal-link" %>
26
26
  <% end %>
27
27
  <% elsif stats.counter_only? %>
28
- <span class="ck-cal-stat"><span class="ck-cal-stat__label">Human reviews</span> <strong><%= stats.sample_size %> / <%= CompletionKit::MetricCalibrationStats::PROVISIONAL_MIN %></strong></span>
28
+ <span class="ck-cal-stat"><span class="ck-cal-stat__label">Human reviews</span> <strong><%= stats.sample_size %> / <%= CompletionKit::MetricAgreementStats::PROVISIONAL_MIN %></strong></span>
29
29
  <% if stats.short_to_target > 0 %><span class="ck-trust-line__hint"><%= stats.short_to_target %> more to report a rate</span><% end %>
30
30
  <% if target_response %>
31
31
  <%= link_to "Review another score →", run_response_path(target_response.run, target_response, anchor: anchor), class: "ck-cal-link" %>
@@ -17,7 +17,7 @@
17
17
  <input type="radio" name="ck-api-tab" id="ck-tab-datasets" class="ck-api-tabs__radio">
18
18
  <input type="radio" name="ck-api-tab" id="ck-tab-metrics" class="ck-api-tabs__radio">
19
19
  <input type="radio" name="ck-api-tab" id="ck-tab-metric-groups" class="ck-api-tabs__radio">
20
- <input type="radio" name="ck-api-tab" id="ck-tab-calibrations" class="ck-api-tabs__radio">
20
+ <input type="radio" name="ck-api-tab" id="ck-tab-agreements" class="ck-api-tabs__radio">
21
21
  <input type="radio" name="ck-api-tab" id="ck-tab-tags" class="ck-api-tabs__radio">
22
22
  <input type="radio" name="ck-api-tab" id="ck-tab-providers" class="ck-api-tabs__radio">
23
23
 
@@ -29,7 +29,7 @@
29
29
  <label for="ck-tab-datasets" class="ck-api-tabs__label">Datasets <span class="ck-api-tabs__count">5</span></label>
30
30
  <label for="ck-tab-metrics" class="ck-api-tabs__label">Metrics <span class="ck-api-tabs__count">10</span></label>
31
31
  <label for="ck-tab-metric-groups" class="ck-api-tabs__label">Metric Groups <span class="ck-api-tabs__count">5</span></label>
32
- <label for="ck-tab-calibrations" class="ck-api-tabs__label">Calibrations <span class="ck-api-tabs__count">3</span></label>
32
+ <label for="ck-tab-agreements" class="ck-api-tabs__label">Agreements <span class="ck-api-tabs__count">3</span></label>
33
33
  <label for="ck-tab-tags" class="ck-api-tabs__label">Tags <span class="ck-api-tabs__count">5</span></label>
34
34
  <label for="ck-tab-providers" class="ck-api-tabs__label">Providers <span class="ck-api-tabs__count">5</span></label>
35
35
  </nav>
@@ -238,8 +238,8 @@
238
238
  } %>
239
239
 
240
240
  <div class="ck-api-endpoint" style="padding-top: 1.5rem;">
241
- <p class="ck-kicker" style="margin-bottom: 0.5rem;">Calibration loop</p>
242
- <p class="ck-meta-copy">Drive metric improvement from disagree-flagged calibrations: ask the model to rewrite the instruction and rubric into a new draft version.</p>
241
+ <p class="ck-kicker" style="margin-bottom: 0.5rem;">Agreement loop</p>
242
+ <p class="ck-meta-copy">Drive metric improvement from disagree-flagged agreements: ask the model to rewrite the instruction and rubric into a new draft version.</p>
243
243
  </div>
244
244
  <div class="ck-api-endpoint">
245
245
  <p class="ck-api-method"><span class="ck-chip ck-chip--soft">POST</span> /api/v1/metrics/:id/suggest_variants</p>
@@ -250,7 +250,7 @@
250
250
 
251
251
  <div class="ck-api-endpoint" style="padding-top: 1.5rem;">
252
252
  <p class="ck-kicker" style="margin-bottom: 0.5rem;">Metric versions</p>
253
- <p class="ck-meta-copy">Every metric carries a history of versions (the current published one, prior published ones, and any draft suggestions). Reviews and calibrations record the version they ran against, so the API can surface stale state and let you revert.</p>
253
+ <p class="ck-meta-copy">Every metric carries a history of versions (the current published one, prior published ones, and any draft suggestions). Reviews and agreements record the version they ran against, so the API can surface stale state and let you revert.</p>
254
254
  </div>
255
255
  <div class="ck-api-endpoint">
256
256
  <p class="ck-api-method"><span class="ck-chip ck-chip--soft">GET</span> /api/v1/metrics/:metric_id/metric_versions</p>
@@ -294,23 +294,23 @@
294
294
  </div>
295
295
 
296
296
  <div class="ck-api-tabs__panel">
297
- <h2 class="ck-section-title">Calibrations</h2>
298
- <p class="ck-copy">Per-verdict feedback events on a response/metric pair: agree, disagree (with a corrected score and note), or borderline. Calibrations capture the metric version that was current when the verdict was cast, which is what drives the trust signal and the "stale" indicators across the rest of the API.</p>
297
+ <h2 class="ck-section-title">Agreements</h2>
298
+ <p class="ck-copy">Per-verdict feedback events on a response/metric pair: agree, disagree (with a corrected score and note), or borderline. Agreements capture the metric version that was current when the verdict was cast, which is what drives the trust signal and the "stale" indicators across the rest of the API.</p>
299
299
  <div class="ck-api-endpoint">
300
- <p class="ck-api-method"><span class="ck-chip ck-chip--soft">GET</span> /api/v1/calibrations</p>
301
- <p class="ck-meta-copy">List calibrations across all runs. Supports filtering by any combination of the query params below.</p>
300
+ <p class="ck-api-method"><span class="ck-chip ck-chip--soft">GET</span> /api/v1/agreements</p>
301
+ <p class="ck-meta-copy">List agreements across all runs. Supports filtering by any combination of the query params below.</p>
302
302
  <p class="ck-api-params"><strong>Optional filters:</strong>&ensp;<code>run_id</code>, <code>response_id</code>, <code>metric_id</code>, <code>metric_version_id</code>, <code>created_by</code>, <code>verdict</code> (<code>agree</code>, <code>disagree</code>, or <code>borderline</code>)</p>
303
- <%= render "completion_kit/api_reference/example", base_url: base_url, token: token, real_token: real_token, cmd: "curl \"#{base_url}/api/v1/calibrations?metric_id=1&verdict=disagree\" \\\n -H \"Authorization: Bearer #{token}\"" %>
303
+ <%= render "completion_kit/api_reference/example", base_url: base_url, token: token, real_token: real_token, cmd: "curl \"#{base_url}/api/v1/agreements?metric_id=1&verdict=disagree\" \\\n -H \"Authorization: Bearer #{token}\"" %>
304
304
  </div>
305
305
  <div class="ck-api-endpoint">
306
- <p class="ck-api-method"><span class="ck-chip ck-chip--soft">POST</span> /api/v1/runs/:run_id/responses/:response_id/metrics/:metric_id/calibrations</p>
307
- <p class="ck-meta-copy">Cast a calibration on a specific response/metric pair. The metric version on the record is set automatically from the run's review.</p>
306
+ <p class="ck-api-method"><span class="ck-chip ck-chip--soft">POST</span> /api/v1/runs/:run_id/responses/:response_id/metrics/:metric_id/agreements</p>
307
+ <p class="ck-meta-copy">Cast an agreement on a specific response/metric pair. The metric version on the record is set automatically from the run's review.</p>
308
308
  <p class="ck-api-params"><strong>Required:</strong>&ensp;<code>verdict</code>, <code>created_by</code>&emsp;<strong>Optional:</strong>&ensp;<code>corrected_score</code>, <code>note</code></p>
309
- <%= render "completion_kit/api_reference/example", base_url: base_url, token: token, real_token: real_token, cmd: "curl -X POST #{base_url}/api/v1/runs/1/responses/42/metrics/3/calibrations \\\n -H \"Authorization: Bearer #{token}\" \\\n -H \"Content-Type: application/json\" \\\n -d '{\"verdict\": \"disagree\", \"corrected_score\": 3, \"note\": \"too generous\", \"created_by\": \"alice\"}'" %>
309
+ <%= render "completion_kit/api_reference/example", base_url: base_url, token: token, real_token: real_token, cmd: "curl -X POST #{base_url}/api/v1/runs/1/responses/42/metrics/3/agreements \\\n -H \"Authorization: Bearer #{token}\" \\\n -H \"Content-Type: application/json\" \\\n -d '{\"verdict\": \"disagree\", \"corrected_score\": 3, \"note\": \"too generous\", \"created_by\": \"alice\"}'" %>
310
310
  </div>
311
311
  <div class="ck-api-endpoint">
312
- <p class="ck-api-method"><span class="ck-chip" style="color: var(--ck-danger);">DELETE</span> /api/v1/calibrations/:id</p>
313
- <p class="ck-meta-copy">Delete a calibration. Returns 204 No Content.</p>
312
+ <p class="ck-api-method"><span class="ck-chip" style="color: var(--ck-danger);">DELETE</span> /api/v1/agreements/:id</p>
313
+ <p class="ck-meta-copy">Delete an agreement. Returns 204 No Content.</p>
314
314
  </div>
315
315
  </div>
316
316
 
@@ -12,7 +12,7 @@
12
12
  <span class="ck-guiding__output"><%= truncate(example[:output].to_s, length: 90) %></span>
13
13
  <span class="ck-guiding__scores"><span class="ck-guiding__judge"><%= example[:judge_score].to_i %></span> &rarr; <span class="ck-guiding__human"><%= example[:human_score].to_i %></span></span>
14
14
  <% end %>
15
- <%= button_to exclude_example_metric_path(metric, calibration_id: example[:id]),
15
+ <%= button_to exclude_example_metric_path(metric, agreement_id: example[:id]),
16
16
  method: :post, form_class: "inline-block", class: "ck-icon-btn",
17
17
  title: "Stop using this case", "aria-label": "Stop using this case",
18
18
  data: { turbo_confirm: "Stop using this corrected case to guide the judge?" } do %><%= heroicon_tag "x-mark", variant: :outline, size: 16, "aria-hidden": "true" %><% end %>
@@ -10,7 +10,7 @@
10
10
  </div>
11
11
  </section>
12
12
 
13
- <% if CompletionKit.config.judge_calibration_enabled && @suggestion_draft.nil? && @edit_draft.nil? && @improve_disagreement_count.to_i.positive? %>
13
+ <% if CompletionKit.config.judge_agreement_enabled && @suggestion_draft.nil? && @edit_draft.nil? && @improve_disagreement_count.to_i.positive? %>
14
14
  <div class="ck-suggestion-banner" role="status">
15
15
  <div class="ck-suggestion-banner__body">
16
16
  <p class="ck-kicker">Improve from reviews</p>
@@ -42,7 +42,7 @@
42
42
  </div>
43
43
  </section>
44
44
 
45
- <% if CompletionKit.config.judge_calibration_enabled && @versions.any? %>
45
+ <% if CompletionKit.config.judge_agreement_enabled && @versions.any? %>
46
46
  <% predecessor_of = @versions.index_with { |v| @versions.detect { |o| o.version_number < v.version_number } } %>
47
47
  <section class="ck-card ck-card--spaced">
48
48
  <p class="ck-kicker">Versions</p>
@@ -73,7 +73,7 @@
73
73
  <%= button_to "Make current", publish_draft_metric_path(@metric, draft_id: v.id),
74
74
  method: :post, form_class: "inline-block",
75
75
  class: "ck-chip ck-chip--publish",
76
- data: { turbo_confirm: "Make #{v.version_label} the version to use? #{v.version_label} will be used in test runs using this metric now. Reviews you have already given stay with the version they were made against." } %>
76
+ data: { turbo_confirm: "Make #{v.version_label} the version to use? It becomes the version used in test runs, and the reviews you gave on it count again. Reviews on the version you're leaving stay with it." } %>
77
77
  <% end %>
78
78
  </div>
79
79
  <% vs = v.validation_summary %>
@@ -171,7 +171,7 @@
171
171
  <span class="ck-modal__foot-note">Roll this metric back to this version.</span>
172
172
  <%= button_to "Make #{v.version_label} current →", publish_draft_metric_path(@metric, draft_id: v.id),
173
173
  method: :post, form_class: "inline-block", class: ck_button_classes(:dark),
174
- data: { turbo_confirm: "Make #{v.version_label} the version to use? #{v.version_label} will be used in test runs using this metric now. Reviews you have already given stay with the version they were made against." } %>
174
+ data: { turbo_confirm: "Make #{v.version_label} the version to use? It becomes the version used in test runs, and the reviews you gave on it count again. Reviews on the version you're leaving stay with it." } %>
175
175
  <% end %>
176
176
  </footer>
177
177
  </article>
@@ -179,7 +179,7 @@
179
179
  <% end %>
180
180
  <% end %>
181
181
 
182
- <% if CompletionKit.config.judge_calibration_enabled %>
182
+ <% if CompletionKit.config.judge_agreement_enabled %>
183
183
  <% draft = @suggestion_draft || @edit_draft %>
184
184
  <section class="ck-card ck-card--spaced">
185
185
  <div class="ck-prompt-preview__header">
@@ -197,8 +197,8 @@
197
197
  <%= turbo_stream_from "metric_#{@metric.id}_suggestion" %>
198
198
  <div id="ck-suggestion-status-<%= @metric.id %>" class="ck-suggestion-status"></div>
199
199
  <p class="ck-meta-copy">How often the judge lands on the same score you would. Review its scores to build that signal, and improve the metric to raise it.</p>
200
- <%= render "completion_kit/calibrations/trust_panel",
201
- stats: CompletionKit::MetricCalibrationStats.for(@metric),
200
+ <%= render "completion_kit/agreements/trust_panel",
201
+ stats: CompletionKit::MetricAgreementStats.for(@metric),
202
202
  metric: @metric %>
203
203
  <% if CompletionKit.config.judge_examples_from_reviews %>
204
204
  <%= render "completion_kit/metrics/guiding_examples", metric: @metric, examples: @guiding_examples %>
@@ -124,13 +124,13 @@
124
124
  <% if review.ai_feedback.present? %>
125
125
  <p class="ck-review-card__feedback"><%= review.ai_feedback %></p>
126
126
  <% end %>
127
- <% if CompletionKit.config.judge_calibration_enabled && review.metric && review.ai_score %>
128
- <% existing = CompletionKit::Calibration.find_by(
127
+ <% if CompletionKit.config.judge_agreement_enabled && review.metric && review.ai_score %>
128
+ <% existing = CompletionKit::Agreement.find_by(
129
129
  response_id: @response.id, metric_id: review.metric_id,
130
130
  created_by: CompletionKit.config.username.presence || "operator"
131
131
  ) %>
132
- <%= render "completion_kit/calibrations/buttons",
133
- review: review, calibration: existing, run: @run,
132
+ <%= render "completion_kit/agreements/buttons",
133
+ review: review, agreement: existing, run: @run,
134
134
  response_row: @response, metric: review.metric %>
135
135
  <% end %>
136
136
  </div>
@@ -18,7 +18,7 @@
18
18
  <% dataset_preview_lines = dataset_lines.first(50) %>
19
19
  <% end %>
20
20
 
21
- <% if CompletionKit.config.judge_calibration_enabled %>
21
+ <% if CompletionKit.config.judge_agreement_enabled %>
22
22
  <% stale_summary = @run.stale_review_summary %>
23
23
  <% if stale_summary.any? %>
24
24
  <div class="ck-stale-versions-banner" role="status">
data/config/routes.rb CHANGED
@@ -41,7 +41,7 @@ CompletionKit::Engine.routes.draw do
41
41
  get :compare
42
42
  end
43
43
  resources :responses, only: [:show] do
44
- resources :calibrations, only: [:create]
44
+ resources :agreements, only: [:create]
45
45
  end
46
46
  end
47
47
 
@@ -75,7 +75,7 @@ CompletionKit::Engine.routes.draw do
75
75
  end
76
76
  resources :responses, only: [:index, :show] do
77
77
  resources :metrics, only: [] do
78
- resources :calibrations, only: [:index, :create]
78
+ resources :agreements, only: [:index, :create]
79
79
  end
80
80
  end
81
81
  end
@@ -93,7 +93,7 @@ CompletionKit::Engine.routes.draw do
93
93
  resources :metric_groups
94
94
  resources :tags
95
95
  resources :provider_credentials
96
- resources :calibrations, only: [:index, :destroy]
96
+ resources :agreements, only: [:index, :destroy]
97
97
  end
98
98
  end
99
99
 
@@ -0,0 +1,33 @@
1
+ class BackfillReviewMetricVersions < ActiveRecord::Migration[8.1]
2
+ def up
3
+ quoted_true = ActiveRecord::Base.connection.quote(true)
4
+ now = ActiveRecord::Base.connection.quote(Time.current)
5
+
6
+ execute <<~SQL
7
+ INSERT INTO completion_kit_metric_versions
8
+ (metric_id, instruction, rubric_bands, current, state, version_number, published_at, created_at, updated_at)
9
+ SELECT m.id, m.instruction, m.rubric_bands, #{quoted_true}, 'published', 1, #{now}, #{now}, #{now}
10
+ FROM completion_kit_metrics m
11
+ WHERE NOT EXISTS (
12
+ SELECT 1 FROM completion_kit_metric_versions mv WHERE mv.metric_id = m.id
13
+ )
14
+ SQL
15
+
16
+ execute <<~SQL
17
+ UPDATE completion_kit_reviews
18
+ SET metric_version_id = (
19
+ SELECT mv.id FROM completion_kit_metric_versions mv
20
+ WHERE mv.metric_id = completion_kit_reviews.metric_id AND mv.current = #{quoted_true}
21
+ LIMIT 1
22
+ )
23
+ WHERE metric_id IS NOT NULL
24
+ AND (
25
+ metric_version_id IS NULL
26
+ OR metric_version_id NOT IN (SELECT id FROM completion_kit_metric_versions)
27
+ )
28
+ SQL
29
+ end
30
+
31
+ def down
32
+ end
33
+ end
@@ -0,0 +1,6 @@
1
+ class AddMetricVersionFkToReviews < ActiveRecord::Migration[8.1]
2
+ def change
3
+ add_foreign_key :completion_kit_reviews, :completion_kit_metric_versions,
4
+ column: :metric_version_id, on_delete: :nullify
5
+ end
6
+ end