RL — Reinforcement Learning Experiments

You are managing A/B testing experiments powered by Thompson Sampling. The RL module tracks impressions (turns) and conversions (rewards) across content variants and automatically optimizes traffic distribution.

Preamble — Auto-discover Current State

# List all active experiments
drush rl:list --format=yaml

# Get current module configuration
drush rl:config:list

# If experiments exist, get status of the most active one
# drush rl:status <experiment_id> --format=yaml

Commands Reference

Discovery & Analysis (read-only)

Command	Alias	Purpose
rl:list	rll	List all experiments with summary stats
rl:status <id>	rlst	Detailed status: phase, confidence, traffic distribution
rl:performance <id>	rlp	Arm-level performance with human-readable labels
rl:trends <id>	rlt	Historical trends with period aggregation
rl:export <id>	rle	Export complete experiment data
rl:analyze <id>	rla	Full analysis with actionable recommendations

Experiment Lifecycle

Command	Alias	Purpose
rl:experiment:create <id>	rl-ec	Create experiment (--module, --name, --dry-run)
rl:experiment:update <id>	rl-eu	Update name/module (--module, --name, --dry-run)
rl:experiment:delete <id>	rl-ed	Delete experiment + all data (--dry-run)

Configuration

Command	Alias	Purpose
rl:config:get [key]	rl-cg	Get one or all settings
rl:config:set <key> <value>	rl-cs	Set a setting with validation (--dry-run)
rl:config:list	rl-cl	List all settings with current values
rl:config:reset	rl-cr	Reset all settings to defaults (--dry-run)
rl:event-log:clear	rl-elc	Clear all snapshots (--dry-run)

Setup

Command	Alias	Purpose
rl:setup-ai	rl-sa	Install AI skill files (--host, --check)

Configuration Keys

Key	Type	Default	Description
debug_mode	boolean	false	Log Thompson Sampling scores
enable_event_log	boolean	true	Record snapshots for historical charts
event_log_max_rows	integer	100000	Max snapshot rows (1000–10000000)
chart_line_threshold	integer	9	Max variants for 2D line chart (2–50)

Workflow Examples

Create and monitor a new A/B test

# Create experiment
drush rl:experiment:create hero_cta_test \
  --module=my_module \
  --name="Hero CTA Button Color Test"

# Check status after data collection
drush rl:status hero_cta_test

# Get detailed performance breakdown
drush rl:performance hero_cta_test --format=yaml

# Full analysis with recommendations
drush rl:analyze hero_cta_test

Configure the module

# Enable debug logging temporarily
drush rl:config:set debug_mode true

# Increase snapshot storage
drush rl:config:set event_log_max_rows 500000

# Check all settings
drush rl:config:list

Key Concepts

• Experiment: A multi-armed bandit test comparing content variants
• Arm: A single variant (identified by arm_id, often a node ID)
• Turn/Impression: One display of a variant to a visitor
• Reward/Conversion: A successful outcome (click, form submit, etc.)
• Thompson Sampling: Bayesian algorithm that balances exploration vs exploitation
• Phase: exploration, learning, exploitation, conclusive
• Confidence: Statistical confidence that the top performer is truly best

Notes

• Experiments are tracked in custom database tables, not Drupal config entities
• The rl.php endpoint provides high-performance tracking with minimal Drupal bootstrap
• Arms are typically node IDs; the analyzer resolves them to node titles automatically
• Views management is handled by drush_webmaster (wm:view:* commands) — do not duplicate