intelligems-test-verdict（智力测试结论）

/test-verdict

Analyze any Intelligems A/B test and deliver a clear verdict: WINNER, LOSER, FLAT, KEEP RUNNING, or TOO EARLY. Includes risk framing, revenue-vs-conversion analysis, and a device segment quick-check.

Works with both active and ended tests.

Step 0: Workspace Check

Check if the shared workspace exists and is ready:

test -d ~/intelligems-analytics/venv && test -f ~/intelligems-analytics/ig_client.py && echo "READY" || echo "NEEDS_SETUP"

If NEEDS_SETUP: Run the /intelligems-core skill first to set up the workspace. This installs the shared Python libraries and virtual environment at ~/intelligems-analytics/.

Specifically:

1. Copy setup_workspace.sh from the intelligems-core skill's references/ folder and run it.
2. Copy all four core Python files (ig_client.py, ig_metrics.py, ig_helpers.py, ig_config.py) from intelligems-core/references/ into ~/intelligems-analytics/.

Step 1: Get API Key

Check if an API key is already configured:

source ~/intelligems-analytics/venv/bin/activate && python3 -c "
from dotenv import load_dotenv; import os
load_dotenv(os.path.expanduser('~/intelligems-analytics/.env'))
key = os.getenv('INTELLIGEMS_API_KEY', '')
print('HAS_KEY' if key and key != 'your_api_key_here' else 'NO_KEY')
"

If NO_KEY: Ask the user:

"What's your Intelligems API key? You can get one by contacting support@intelligems.io"

Then save it:

echo "INTELLIGEMS_API_KEY=<user's key>" > ~/intelligems-analytics/.env

Never hardcode or assume an API key.

Step 2: Copy Verdict Script

Copy the verdict script from this skill's references into the workspace:

cp references/verdict.py ~/intelligems-analytics/verdict.py

Step 3: Select Test

The script accepts an optional test ID as an argument.

If the user already specified a test ID or name:

• Pass the ID directly: python3 verdict.py <test_id>

If no test was specified:

• Run the script without arguments — it will list all active experiments and ask the user to pick one.
• If the user wants to analyze an ended test, they need to provide the test ID directly.

Step 4: Run Analysis

Execute the verdict script with the workspace virtual environment:

cd ~/intelligems-analytics && source venv/bin/activate && python3 verdict.py [optional_test_id]

The script will:

1. Fetch overview analytics for the selected test
2. Run a maturity check (enough time, visitors, orders?)
3. Compute the verdict based on confidence and uplift
4. Check revenue vs. conversion alignment
5. Fetch device-type segments for a quick contradiction check
6. Print all results in a structured format

Step 5: Present Verdict

Read the script output and present results conversationally. Structure your response like this:

1. Big Verdict Headline

Lead with the verdict in bold. Make it unmistakable:

• WINNER — "This test is a winner. Roll it out."
• LOSER — "This test is hurting performance. Kill it."
• FLAT — "No meaningful difference. Your call."
• KEEP RUNNING — "Promising signals, but not enough data yet."
• TOO EARLY — "Way too early to tell. Check back later."

2. Plain-English Reasoning

Explain WHY in one or two sentences. No jargon. Reference the actual numbers.

3. Risk Framing

Always frame confidence as risk:

"At 82% confidence, there's an 18% chance the control was actually better."

If COGS data exists, note whether profit and revenue agree.

4. Revenue vs. Conversion Note

If they diverge, explain what it means in plain English:

"Revenue is up but conversion is down — fewer people are buying, but those who do spend more. That's worth watching."

If aligned, just say: "Revenue and conversion are moving together — clean signal."

5. Segment Quick-Check

Highlight the most interesting finding from the device segment check. Especially flag contradictions:

"Overall this is a winner, but it's actually losing on mobile. Worth investigating before rolling out."

6. What to Test Next

Based on the test type and outcome, suggest what to explore next. Examples:

• Pricing winner: "Try testing a slightly higher price point to find the ceiling."
• Shipping loser: "Consider testing free shipping with a minimum order threshold."
• Content flat: "The messaging isn't moving the needle. Try a completely different angle."
• Too early: "Just wait. Check back in X days when you'll have enough orders."

Step 6: Set Up Slack Automation (Optional)

If the user wants verdicts delivered to Slack automatically, set up a Slack integration:

1. Get a Slack webhook URL

Ask the user:

"To send verdicts to Slack, you need an Incoming Webhook URL. Go to https://api.slack.com/apps → Create New App → Incoming Webhooks → Add New Webhook. Which channel should verdicts go to?"

2. Copy the Slack module

cp ../intelligems-core/references/ig_slack.py ~/intelligems-analytics/ig_slack.py

3. Test it

Run the verdict with the --slack flag:

cd ~/intelligems-analytics && source venv/bin/activate && python3 verdict.py <test_id> --slack "<webhook_url>"

Verify the message appears in Slack with the verdict emoji, reasoning, risk, and segments.

4. Schedule it (optional)

Copy and run the automation setup script:

cp ../intelligems-core/references/setup_automation.sh ~/intelligems-analytics/
bash ~/intelligems-analytics/setup_automation.sh verdict.py "<webhook_url>" 9 0

This creates a macOS LaunchAgent that runs the verdict daily at 9:00 AM and posts to Slack.

Notes

• Ended tests: The skill works with ended tests — just pass the test ID directly. The script handles both active and ended tests.
• Multi-variation tests: If a test has multiple variants, the script analyzes each and highlights the best one.
• COGS data: When available, the script automatically uses Gross Profit per Visitor (GPV) instead of Revenue per Visitor (RPV) as the primary metric.
• Slack output: When using --slack, the script formats results as Slack Block Kit messages with verdict emoji, reasoning, risk assessment, and segment insights.