# Firedrill vs Chronicle Labs: How should you test replies against business policy?

**Choose Firedrill when you want to check an agent’s reply against an explicit business rule before you have production history. Choose Chronicle Labs when the work is discovering how a policy actually plays out across existing conversations, tools and exceptions.** For a returns agent, the written refund policy may be clear while the way staff handle late requests varies. Those are two different sources of test cases.

By Firedrill · November 12, 2026. Based on published product documentation.

## What each platform checks

| Decision | Firedrill | Chronicle Labs |
| --- | --- | --- |
| Simulation | Define a persona, starting records and the expected behavior. Your coding agent can propose scenarios from the project. [Scenarios](https://docs.firedrill.run/build/scenarios) | Build staging environments from operational data and replay workflows. Chronicle describes historical cases, edge cases and adjacent scenarios. [Finance workflows](https://chronicle-labs.com/industries/finance) |
| Outcome scoring and hard rules | Check exact actions and grade the reply for adherence to the supplied policy. Keep the evidence for each drill. [Evaluation guide](https://docs.firedrill.run/build/evaluations) | Backtest decisions across reconstructed workflows and turn production failures into regression cases. For a pilot, ask to see the per-case evidence for an incorrect approval. [Backtesting](https://chronicle-labs.com/industries/finance) |
| Coding-agent setup | Your coding agent prepares the world. With your approval, the CLI starts or connects to the target locally. Optional remote MCP manages cloud resources and reads reports. [Run your agent](https://firedrill.run/run-your-agent.md) | Chronicle describes connecting business tools and replaying captured events from its dashboard. Confirm the runner and coding-agent interface for your integration during scoping. [Getting started](https://chronicle-labs.com/) |
| Synthetic services | Select the Stripe operations your task uses from the versioned service catalog. [Stripe coverage](https://firedrill.run/tools?tool=stripe) | The finance offering describes connecting payment, banking and identity systems. Confirm coverage for your provider and the workflow being reconstructed. [Finance integrations](https://chronicle-labs.com/industries/finance) |
| Pricing | Free includes 500 monthly credits. Starter is $29/month, Team $99/month and Scale $299/month. Enterprise is scoped separately. Your target agent’s model provider bills separately. [Pricing and credits](https://firedrill.run/pricing) | The public product page directs teams to a demo. We could not verify a public rate for this workload. Request a quote covering source integration, staging and repeated runs. [Chronicle Labs](https://chronicle-labs.com/) |

## When to choose each

**Choose Firedrill** for a new returns policy that has never been applied to a customer. Start with a rule whose boundary you can check and supply fictitious transactions on either side. Rerun the same cases when the policy or agent changes.

**Choose Chronicle** if the team needs to recover its existing workflow from business history, including when a representative sought approval. Have an owner identify which historical decisions should serve as examples of good behavior. A past decision can itself be a mistake.

## Test the day a return becomes ineligible

1. Set one policy for both systems. Refunds requested within 30 days, inclusive, are allowed. Later requests require human approval. Create a $60 sale dated June 1, 2026 and requests on July 1 and July 2. No manager has approved an exception.
2. In Firedrill seed the sale and author both cases. In a Chronicle pilot, supply the same fictitious workflow records through the agreed integration and ask for the reconstructed case plus its one-day variation. Verify the starting dates and balance before running.
3. Use the same agent and request wording. On July 1 expect one $60 refund and a truthful confirmation. On July 2 expect zero refunds and a reply requesting approval. As a control, run an agent that always refunds. It should fail the second case.
4. Save the policy version, ledger delta, reply and verdict. Ask a reviewer to locate the policy clause and evidence for each decision. Record missing inputs separately from an incorrect decision.