> For the complete documentation index, see [llms.txt](https://docs.zerve.ai/zerve-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.zerve.ai/zerve-docs/superagents/batches.md).

# Batches: running work in parallel

How superagents run work in parallel, from scoring hundreds of inputs to judging competing approaches.

A **batch** is how a superagent runs many pieces of work at once. Instead of trying one approach at a time, it can score hundreds of inputs in parallel, try several competing approaches and keep the best, or improve a result round after round.

Batches appear on the superagent's **Map** as a single tile, however many runs are inside. The superagent is notified once, when the whole batch has finished, and then decides what to do with the results.

## Batch modes

| Mode                                                              | Use it to                                                                         | Example                                                             |
| ----------------------------------------------------------------- | --------------------------------------------------------------------------------- | ------------------------------------------------------------------- |
| [**Map**](/zerve-docs/superagents/batches/map.md)                 | Run the same task over many inputs                                                | Score each of 500 stocks on a quality metric                        |
| [**Exploration**](/zerve-docs/superagents/batches/exploration.md) | Try competing approaches and keep the one with the best score                     | Compare three volatility forecasting methods on out-of-sample error |
| [**Ideation**](/zerve-docs/superagents/batches/ideation.md)       | Brainstorm many distinct approaches before committing to one                      | Generate twenty ways to hedge a portfolio's rate exposure           |
| [**Tournament**](/zerve-docs/superagents/batches/tournament.md)   | Pick the best of several finished pieces of work when quality is a judgement call | Choose the strongest of four drafts of an investment memo           |
| [**Hill-climb**](/zerve-docs/superagents/batches/hill-climb.md)   | Improve one piece of work round after round                                       | Improve a signal's Sharpe ratio while keeping turnover in check     |
| [**Survey**](/zerve-docs/superagents/batches/survey.md)           | Run the same deep analysis across many entities, each with its own team           | Review each of 30 portfolio companies against the same checklist    |

Some modes decide the winner with a number, such as a score or a metric. Others use **judges**: agents that compare entrants head to head against written criteria. See [How judging works](/zerve-docs/superagents/batches/how-judging-works.md).

## Modes work together

Modes can feed into each other. A superagent might use **Ideation** to generate many approaches, then an **Exploration** to score the most promising ones on real data, or a **Tournament** to build and judge them. A **Hill-climb** can start from ideation and come back to it periodically for fresh ideas.

Within ideation and hill-climb batches, what each round learns is carried into the next, so later rounds start from what earlier ones discovered rather than from scratch.

## Asking for a batch

The superagent decides when a batch is the right tool, but you can ask for one directly. For example:

> Try three different approaches to forecasting next-day volatility and keep whichever has the lowest out-of-sample error.

> Write four versions of the weekly risk summary and have them judged on clarity and accuracy.

## Following a batch

Select a batch on the Map to see its progress. While it runs, it shows what phase it's in, such as **Building entrants…**, **Judging…** or **Scoring candidates…**. When there's a detailed results view, select **Explore Results**.

| Results view         | Used for                  | What it shows                                                                                                                |
| -------------------- | ------------------------- | ---------------------------------------------------------------------------------------------------------------------------- |
| **Score chart**      | Modes decided by a number | Each run's score                                                                                                             |
| **Ranking explorer** | Modes decided by judges   | Each entrant's strength with a confidence range, its wins, losses and draws, and how every entrant fared against every other |
| **Run explorer**     | Judged hill-climbs        | The leading entrant in each round, how ratings changed over time, and whether the judges are converging                      |

Rankings can be **provisional** while a batch is still running.

### Batch statuses

| Status        | Meaning                                                 |
| ------------- | ------------------------------------------------------- |
| **Running**   | In progress                                             |
| **Completed** | Finished                                                |
| **Stopped**   | Ended early, with results based on the work done so far |
| **Failed**    | Couldn't complete                                       |

Individual runs inside a batch can be **pending**, **running**, **completed**, **failed** or **skipped**.

## Steering a batch

The superagent can stop any batch early, which finalizes it based on the work done so far. For example, it might stop an exploration once one approach is clearly ahead. It can also give a running hill-climb or tournament guidance for its next round.

You can ask for either in a conversation, for example "Stop the tournament, the second draft is clearly the best" or "In the next round, focus on reducing drawdown."

## Batches and your budget

Batches can run many workers at once, so they can use credits quickly. They count against the superagent's [daily budget](/zerve-docs/superagents/budgets-and-credits.md), and batches decided by judges always have a limit on the number of comparisons or the time they can take.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.zerve.ai/zerve-docs/superagents/batches.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.