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

# Dreams view

# Dreams view

The **Dreams** screen in the [Control UI v2](https://docs.wednesdayai.dev/admin/web/control-ui-v2.md) console (sidebar group
**Learning**, path `/dreams`) shows what the background task passes did: recent runs with their
status and measurement, the budget window, the [Dream Cycle](https://docs.wednesdayai.dev/developers/plugins/dream-cycle) state, and
the health of [Learning Core](https://docs.wednesdayai.dev/developers/plugins/learning-core.md) lessons and skills. It is read-only:
nothing on the screen changes Gateway state.

The sidebar lists **Dreams** only when the gateway provides `dream.status` (Dream Cycle is
loaded). Without it the screen stays out of the sidebar and the command palette, and a direct
visit to `/dreams` shows "This feature is not enabled on this gateway".

The data comes from the durable task run store through `tasks.status` and `tasks.runEvents`
(see [Task operations](https://docs.wednesdayai.dev/admin/gateway/tasks.md)), from `dream.status`, `dream.weights`, and
`dream.candidates`, and from `learning.overview`.

## Panels

### Runs

A table of the 50 newest task-pass runs: run id, task, status, mode (a **dry-run** chip marks
dry runs), a measurement badge, and when the run started. Above the table, the budget line
shows admission units used in the current window against the cap, the window length and scope,
and the number of budget denials among the runs in view. These are governor admission units,
not model spend.

The measurement badge says what can be shown for a run:

| Badge | Meaning |
| - | - |
| complete | The run has a complete `pass.metrics` row. |
| unavailable | The run's handler never settled its measurement (for example a timeout). |
| missing | The run finished without a metrics row (for example a crash, or a pass that ignored its abort signal). |
| truncated | The bounded `tasks.runEvents` response dropped this run's rows, so the metrics row is not in view. |
| pruned | The run's events are older than the retention period the gateway reports and were deleted. |
| denied | A budget gate deferred the run before it started. |
| in progress | The run has not finished. |
| not loaded | The run's events could not be loaded, so there is nothing to show. |

### Dream state

Three cards, each from its own call:

* **Dream status** (`dream.status`): whether the cycle is enabled or paused, the queued
  question and candidate counts (a trailing `+` means the count is a lower bound), each
  configured stage with its last completion, and the active cycles with their current stage,
  pending passes and failed stages.
* **Weights** (`dream.weights`): per-agent adaptation weights - success rate, completed and
  failed totals, and when each was last updated.
* **Candidates** (`dream.candidates`): the 50 most recent crystallisation candidates with
  signature, task, and first-seen time.

### Consolidation health

* **Counts:** lesson and proposal counts from `learning.overview`, including pending proposals.
* **Quality trend:** the `qualityScore` sequence of recent consolidation runs, newest last.
  Crash-recovered runs without metrics cannot be classified and are counted separately.
* **Preemptions:** consolidation runs preempted (lease, reclaim, or ceiling) out of the
  consolidation runs in view. Shutdown preemptions are not counted as preempted.
* **Distill authoring:** the authoring path of the newest completed distill run that recorded
  one.

## Availability

Each panel checks its own gateway method, so one missing plugin never blanks the screen. A
panel whose method is missing - Dream Cycle or Learning Core is not loaded, for example - shows
"`<method>` is not enabled on this gateway."; enable the plugin to fill the panel. A failed call
shows the gateway's error message.

## Refresh

Every panel reloads itself when the gateway reports a change (`dreams.update`) and polls every
30 seconds as a fallback while the browser tab is visible. Opening the screen always loads
fresh data.

## Per-run cost

The Dreams screen shows no per-run model cost; the budget window counts governor admission units,
never model spend. The gateway's pricing still decides the `costSource` value on every run:

* A model declared under `models.providers` **without** a `cost` block has no published price.
  Its runs show `costSource: unpriced`, like a model with no config entry at all.
* A model whose `cost` block is all zeros is free: its runs show `costSource: priced` and
  `$0.0000`.
* Set `cost` (`input`, `output`, `cacheRead`, `cacheWrite`) on a model to see real spend.

See [Run measurement](https://docs.wednesdayai.dev/admin/gateway/tasks.md#run-measurement) for every
`costSource` value.

## Related

* [Control UI v2](https://docs.wednesdayai.dev/admin/web/control-ui-v2.md)
* [Task operations](https://docs.wednesdayai.dev/admin/gateway/tasks.md)
* [Dream Cycle](https://docs.wednesdayai.dev/developers/plugins/dream-cycle)
* [Learning Core](https://docs.wednesdayai.dev/developers/plugins/learning-core.md)


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