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

# Choose a probabilistic futures workflow

> Choose how to forecast joint outcomes, revise estimates as evidence arrives, and use the returned probabilities in application rules or planning.

Adapt-1 learns Probabilistic Futures from completed observations in a Domain. Each forecast assigns probabilities to joint outcomes, such as demand and wind generation for the same delivery interval. Issue a forecast with the evidence you have, revise it as more evidence arrives, and submit the measured outcome once to score the retained revisions.

## When to use futures

| Your application needs | Configure the Domain to |
| - | - |
| The probability that several outcomes occur together, such as high demand and low generation | Forecast the targets jointly and evaluate the event across the same probability atoms. |
| A forecast before all feeds arrive, such as a shipment forecast that changes as scans arrive | Send the inputs available at each issue time and keep the same `future_id` for revisions of one outcome. |
| A first forecast for a new site with little or no local history | Pool compatible sites on common feature and target scales so the new site can draw on their completed observations. |
| A decision that respects a modeled probability cap, such as holding an inventory shortfall below a set probability | Use the forecast with an application rule, or use probabilistic planning if the action changes the distribution of the outcome. |
| An online regression or classification distribution | Start with a compact set of informative signals and compare bandwidth, conditioning, and target scores as outcomes arrive. |

For a seasonal series, include calendar or lag features available at issue time and check that retained history represents the cycles you need. Compare a single-target point forecast with a seasonal or online regression baseline when your application only consumes one expected value. For continuous command sequences, see [Machina](https://docs.reilabs.org/docs/machina/index.md).

## Choose how to act on a forecast

| How the action affects the result | Workflow |
| - | - |
| The outcome is exogenous and you already know the cost of each action for each outcome. For example, the capacity you reserve has no effect on demand. | Forecast the outcome with `action_path` omitted. Score permitted actions in your application using the forecast's joint probabilities and your cost or constraint formula. [Issue and revise a forecast](https://docs.reilabs.org/docs/neuroadapt/futures/forecast.md). |
| Each discrete action has a different observed outcome distribution. For example, each storage dispatch command can leave a different reserve. | Start with a one-step plan. Collect transitions for the available actions and declare the objective and the maximum modeled failure probability. [Plan with uncertain outcomes](https://docs.reilabs.org/docs/neuroadapt/futures/planning.md). |
| The action changes what states and choices you can face later. For example, discharging storage changes the reserve available next interval. | Start with a short planning horizon and verify that reachable intermediate states have observations for the actions you want to evaluate. [Plan across multiple steps](https://docs.reilabs.org/docs/neuroadapt/futures/planning.md#plan-across-multiple-steps). |

## Start with a small run

1. Define one outcome horizon. Record the evidence available at issue time and the time at which each outcome becomes available.
2. Pick a Domain template for joint targets, missing feeds, pooled entities, or planning. [Run the examples](https://docs.reilabs.org/docs/neuroadapt/futures/examples.md).
3. Probe the setup with a small set of representative observations. Check returned probability mass, forecast resolution, and the outcomes or costs your application will measure. [Check a Domain before streaming](https://docs.reilabs.org/docs/neuroadapt/futures/diagnostics.md#check-a-domain-before-streaming).


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