refactor: extract CooldownFollowUpRunner from CooldownSession (Wave 2A, #375)

src/acceptance/setup.ts

@@ -2,5 +2,6 @@ import '@domain/rules/cycle-rules.steps.js';
 import '@domain/services/cycle-manager.steps.js';
 import '@features/cycle-management/bridge-run-syncer.steps.js';
 import '@features/cycle-management/cooldown-belt-computer.steps.js';
+import '@features/cycle-management/cooldown-follow-up-runner.steps.js';
 import '@features/cycle-management/cycle-activation-name-resolver.steps.js';
 import '@infra/execution/session-bridge.steps.js';

src/features/cycle-management/cooldown-follow-up-runner.feature

@@ -0,0 +1,116 @@
+Feature: Follow-up analysis pipeline during cooldown
+  After a cooldown phase completes its core processing, a series of follow-up
+  analyses run to improve the knowledge system: predictions are matched to
+  outcomes, calibration biases are detected, learnings are promoted through
+  the knowledge hierarchy, expired learnings are flagged, and friction
+  points are resolved.
+
+  Background:
+    Given the follow-up pipeline environment is ready
+
+  # ── Prediction matching ──────────────────────────────────────
+
+  Scenario: predictions are matched to outcomes for each bet run
+    Given prediction matching is enabled
+    And the cycle has bets with runs "run-1" and "run-2"
+    When the follow-up pipeline runs
+    Then predictions are matched for run "run-1"
+    And predictions are matched for run "run-2"
+
+  Scenario: prediction matching is skipped when not enabled
+    Given prediction matching is not enabled
+    And the cycle has bets with runs "run-1" and "run-2"
+    When the follow-up pipeline runs
+    Then no prediction matching occurs
+
+  # ── Calibration detection ────────────────────────────────────
+
+  Scenario: calibration biases are detected for each bet run
+    Given calibration detection is enabled
+    And the cycle has bets with runs "run-1" and "run-2"
+    When the follow-up pipeline runs
+    Then calibration is checked for run "run-1"
+    And calibration is checked for run "run-2"
+
+  Scenario: calibration detection is skipped when not enabled
+    Given calibration detection is not enabled
+    And the cycle has bets with runs "run-1" and "run-2"
+    When the follow-up pipeline runs
+    Then no calibration detection occurs
+
+  # ── Hierarchical promotion ───────────────────────────────────
+
+  Scenario: learnings are promoted through the knowledge hierarchy
+    Given hierarchical promotion is enabled
+    And the knowledge store contains step-tier learnings
+    When the follow-up pipeline runs
+    Then step learnings are promoted to flavor tier
+    And flavor learnings are promoted to stage tier
+    And stage learnings are promoted to category tier
+
+  Scenario: hierarchical promotion failure does not abort the pipeline
+    Given hierarchical promotion is enabled
+    And hierarchical promotion will fail with an internal error
+    When the follow-up pipeline runs
+    Then a warning is logged about hierarchical promotion failure
+    And cooldown continues normally
+
+  # ── Expiry check ─────────────────────────────────────────────
+
+  Scenario: expired learnings are flagged during cooldown
+    Given expiry checking is available
+    And learnings have expired
+    When the follow-up pipeline runs
+    Then expired learnings are flagged
+
+  Scenario: expiry check is skipped when the knowledge store lacks the capability
+    Given expiry checking is not available
+    When the follow-up pipeline runs
+    Then no expiry check occurs
+
+  Scenario: expiry check failure does not abort the pipeline
+    Given expiry checking is available
+    And the expiry check will fail with an internal error
+    When the follow-up pipeline runs
+    Then a warning is logged about expiry check failure
+    And cooldown continues normally
+
+  # ── Friction analysis ────────────────────────────────────────
+
+  Scenario: friction points are resolved for each bet run
+    Given friction analysis is enabled
+    And the cycle has bets with runs "run-1" and "run-2"
+    When the follow-up pipeline runs
+    Then friction is analyzed for run "run-1"
+    And friction is analyzed for run "run-2"
+
+  Scenario: friction analysis is skipped when not enabled
+    Given friction analysis is not enabled
+    And the cycle has bets with runs "run-1" and "run-2"
+    When the follow-up pipeline runs
+    Then no friction analysis occurs
+
+  # ── Pipeline ordering ─────────────────────────────────────────
+
+  Scenario: calibration runs after prediction matching to use its results
+    Given prediction matching is enabled
+    And calibration detection is enabled
+    And the cycle has bets with runs "run-1"
+    When the follow-up pipeline runs
+    Then predictions are matched before calibration is checked for run "run-1"
+
+  # ── Per-run error isolation ──────────────────────────────────
+
+  Scenario: a failing run does not prevent other runs from being analyzed
+    Given prediction matching is enabled
+    And the cycle has bets with runs "run-1" and "run-2"
+    And prediction matching will fail for run "run-1"
+    When the follow-up pipeline runs
+    Then predictions are matched for run "run-2"
+    And a warning is logged about the run failure
+
+  Scenario: bets without a run are silently skipped
+    Given prediction matching is enabled
+    And the cycle has a bet without a run
+    When the follow-up pipeline runs
+    Then no prediction matching occurs