Best Practices

Get the most out of AskVerdict with these tips for writing questions, choosing modes, and interpreting verdicts.

Writing Good Questions

Be Specific

Include Context

Context dramatically improves verdict quality:

# Without context — generic answer
askverdict debate "Should we hire remotely?"
 
# With context — specific, actionable answer
askverdict debate "Should we hire remotely?" \
  --context "Series A startup, 12 people in SF office, hiring 5 engineers, roles are backend and ML. Current team values in-person collaboration. Budget is tight."

Frame as Decisions

AskVerdict works best with clear decision points:

• "Should we X or Y?" — comparison
• "Should we do X?" — go/no-go
• "Is X the right approach for Y?" — evaluation

Choosing Modes

Fast (1 credit)

Use for:

• Quick sanity checks
• Binary yes/no decisions
• High-volume batch processing
• Screening before deeper analysis

Balanced (3 credits)

Use for:

• Standard business decisions
• Feature prioritization
• Vendor selection
• Most everyday decisions

Thorough (8 credits)

Use for:

• Strategic decisions with long-term impact
• Decisions involving significant investment
• When you need to present analysis to stakeholders
• Complex multi-factor decisions

Interpreting Verdicts

Confidence Scores

• 80+ — Strong signal. One side is clearly stronger.
• 60-79 — Good direction but with caveats. Read the reasoning carefully.
• 40-59 — Genuinely close call. Consider running a Thorough mode or adding more context.
• Below 40 — The question may need reframing, or both options are equally valid.

When to Disagree

AskVerdict provides analysis, not orders. Disagree when:

• You have information the AI doesn't (internal politics, personal relationships)
• The confidence is low and your intuition is strong
• Domain expertise gives you insight the model lacks

Workflow Tips

Chain Debates

Break complex decisions into parts:

1. "Should we enter market X?" → If yes:
2. "Should we build or acquire to enter market X?" → If build:
3. "Should we hire a team or use contractors to build for market X?"

Track Outcomes

Use outcome tracking to learn which types of questions get the best results:

# Record what you actually decided
askverdict outcomes record dbt_abc123 --result "Chose PostgreSQL, deployed in March"

Batch Analysis

For portfolio or screening use cases:

# Analyze multiple options
for option in "React" "Vue" "Svelte"; do
  askverdict debate "Should we use $option for our dashboard?" \
    --mode fast --json >> analysis.jsonl
done