> ## Documentation Index
> Fetch the complete documentation index at: https://helloluciallc.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Adjudication Doctrine
Adjudication is the senior-review doctrine and metadata layer that helps ambiguous or high-value Eval Labs cases become usable canonical signal.
***
## Definition
Adjudication is the process of making a final call when a response needs deeper interpretation.
In the current Eval Labs app, adjudication is supported through schema, routing state, metadata, and exports.
It should not be described as a shipped dedicated senior-adjudication editing screen unless that UI path is confirmed in source.
It answers:
```text theme={"dark"}
What did this case really mean?
What should Lucia learn from it?
What label should become canonical?
Should this become a canon candidate?
```
***
## What adjudication is not
Adjudication is not casual reviewing.
It is not employee preference.
It is not a place to rewrite every response.
It is not a dumping ground for vague notes.
***
## When a case needs adjudication
A case should move toward adjudication when:
* employee marks senior review
* reviewer flags risk or confusion
* Lucia may have overclaimed
* intent is ambiguous
* the response could teach Lucia something reusable
* two reviewers disagree materially
* the case touches trust, safety, money, maintenance, guest distress, or owner overwhelm
* the case is a canon candidate
In the app workflow, these cases are represented through review state, canon-candidate state, and exported adjudication metadata.
***
## Final label fields
Adjudication metadata may preserve final structured labels such as:
```text theme={"dark"}
guestIntent
followThroughRequired
actionType
emotionalRead
ownerStressLevel
```
These fields are not employee-facing taxonomy homework.
They belong to the senior interpretation layer.
***
## Adjudication metadata
Supported adjudication metadata fields are:
```text theme={"dark"}
finalLabels
reason
adjudicator
adjudicatedAt
```
The reason should be short, specific, and useful for future training.
Good:
```text theme={"dark"}
The operator was disoriented, not asking for a dashboard summary. Lucia should orient first, then give one next move.
```
Bad:
```text theme={"dark"}
Bad vibes.
```
***
## Canon candidates
A canon candidate is a case that may teach Lucia something durable.
Examples:
* excellent containment pattern
* repeated failure pattern
* high-trust wording lesson
* new intent family
* reusable escalation rule
* important edge-case boundary
Canon candidates should be reviewed by a senior reviewer before they become doctrine.
***
## Adjudication rule
The adjudicator owns canonical meaning. The employee reviewer owns observed reaction.
Do not collapse those roles.