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

# L7 · Outcome Observer

> Closes the loop. Compares decisions to what actually happened, and learns.

<Note>
  **In plain terms:** L7 watches what happens *after* a decision. Did the build go as
  predicted? Did the clearance match the as-built measurement? Those real outcomes
  become signals that improve the system over time.
</Note>

## What it does

L7 closes the loop from approved decisions to later outcomes. Approved decisions get
tracked against actual results:

* Was the rework predicted?
* Did the material spec hold?
* Did the build-readiness assessment match production reality?

Supersession events, rework events, and partner feedback all become training signals
for future capability improvements.

## The learning flywheel, with a brake

L7 is the learning flywheel, but it has a deliberate brake on it:

> The flywheel only turns once governance is stable, and it never turns on
> tenant-private data without explicit permission.

In other words, the system learns from outcomes, but a yard's private data is never
used to train anything without that yard's explicit consent. See
[Keeping Your Data Yours](/concepts/keeping-data-yours).

## What L7 outputs

* `outcome_event`
* `eval_result`
* `capability_update_proposal`
* `training_example`

## How it connects to capability promotion

A capability does not get more autonomy just because someone wants it to. It earns
promotion through **observed performance**. A capability that misses its quality
targets stays in observe-only or draft-only mode; it is not promoted to
propose-with-review or execute-after-approval until L7's evidence shows it has earned
it. See [How We Measure Quality](/how-it-works/evaluation-targets) for the targets
that gate this.