---
title: Measure a run
description: >-
  Read pass rate, tokens, tool calls, turns, cost, and time for each agent and
  environment.
---
> ## Documentation Index
> Fetch the complete documentation index at: https://docs.reeval.ai/llms.txt
> Use this file to discover all available pages before exploring further.

Every attempt records what it took. `reeval run` prints the pass rate and the medians, and `reeval query` reads every measure.

```bash
npx reeval query "SELECT cell_key, outcome, turns, tools, input_tokens, output_tokens, agent_ms FROM attempts"
```

## Read the measures

| Measure | What it counts | Where to read it |
| :--- | :--- | :--- |
| Pass rate | Passed attempts out of counted attempts | The `pass` column of the summary |
| Tokens | Input, output, cache, and reasoning tokens, as the harness reports them | The `tokens` column of the `effort` table, and `input_tokens` and `output_tokens` on `attempts` |
| Tool calls | The tools the agent called | The `tools` column of the `effort` table, and `tools` on `attempts` |
| Turns | The prompts the evaluation sent, one per `c.send()` | `turns` on `attempts` |
| Cost | The tokens priced with models.dev or with custom prices, and what a gateway reported | `reeval run --dry-run` before a run, and `reportedCostUsd` in the record |
| Time | Agent time per attempt | The `time` column of the `effort` table, and `agent_ms` on `attempts` |

## Read the assertion and `effort` tables

With several cells, the block for an evaluation adds an assertion table, headed `check`. It has one column per cell and lists each assertion that failed or did not run in some cell, as passed attempts out of counted attempts. Passed assertions print as `+N more passed`, and `couldn't check` means the assertion raised an error. For a comparison's table, see [Read the result](/results/compare/lift#read-the-result).

The `effort` table prints the median tool calls, time, and tokens of each cell. `tokens` appears when a cell reported them:

```text
  effort           tools  time
  math             1      1.0s
  project-layout   1      <0.1s
```

The `effort` table prints only when every cell has at least four counted attempts, and either the evaluation failed or a variant used fewer or more tokens, tool calls, or time than its baseline.

## Compare the measures

In a comparison, `npx reeval query effort` lists each variant's tool calls, time, and tokens against its baseline. For more information about how reeval calls a difference, see [Tokens, tool calls and time](/reference/statistics#tokens-tool-calls-and-time).

## Reference

For more information about every column, see [Query views](/reference/query-views). For more information about the record's token fields, see [Run record](/reference/run-record).