> ## Documentation Index
> Fetch the complete documentation index at: https://deepline.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Score accounts against labeled examples

> Ask your agent to test account scores against labeled examples.

[Agent guide: CLI and SDK instructions →](/docs/guides/account-scoring-workflows)

## 1. Have your context ready

Have positive and negative account examples, your scoring criteria, and evidence requirements.

## 2. Ask your agent

If the evidence or criteria are unclear, start with source research:

```deepline-prompt theme={null}
/deepline-research Identify evidence sources for the scoring or prioritization decision I describe. Separate observed facts, candidate proxies, and missing information. Check whether the evidence can be available at the decision date. Recommend a source plan and explain its coverage before collecting additional data. Do not invent scoring weights or treat source availability as proof of predictive value.
```

Once the criteria and evidence plan are clear, ask for scoring. Skip the research step when you already have reviewed sources and approved rules:

```deepline-prompt theme={null}
/deepline-scoring Build an account-scoring workflow with Deepline. Express the criteria explicitly and preserve the evidence behind each contribution. Separate unknown, false, and unavailable inputs. Compare a representative labeled sample and show disagreements rather than changing labels to match the score. Inspect source coverage and usage before a pilot. Return scores for review without changing CRM fields.
```

Complete any requested sign-in yourself. Give the agent the missing information before approving execution. Provider-backed work can incur usage.

## 3. Review the result

Disagreements can be traced to the rule, source, or label. A consistent computation does not establish that the scoring policy predicts business outcomes.

Ask your agent:

```deepline-prompt theme={null}
/deepline-scoring Compare the result with the acceptance rule we agreed. Show the evidence for accepted and rejected outcomes, unresolved questions, and recorded Deepline usage where work ran. If work is still running or a response is uncertain, inspect the original saved execution before proposing another request. Explain the next decision I need to make.
```

Approve a larger job, recurring work, or external changes only after reviewing that decision.

## Implement the reviewed policy

After reviewing the scoring rules and evaluation limits, hand the agreed policy to GTM execution:

```deepline-prompt theme={null}
/deepline-gtm Implement the reviewed scoring or prioritization policy from this conversation. Preserve the agreed model or rule version, input definitions, evidence requirements, missing-data behavior, and output contract. Reuse collected evidence where it fits. Check the Play and propose a representative pilot before expanding the job or changing CRM records. Do not tune the scoring policy silently during implementation.
```

## Check your understanding

Does an eloquent score explanation prove that the calculation used the correct inputs?

> **Show the answer**
>
> No. Compare the inputs, rule version, and computed result. The explanation is
> not independent proof.

## Related guides

* [Write Plays](/docs/human/sdk-v2/plays)
* [Batch CSV enrichment](/docs/human/sdk-v2/batch-csv)
* [Prioritize accounts with reviewable rules](/docs/human/guides/account-prioritization)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.