# Core concepts

A quick tour of the vocabulary used throughout LoopDesk. Skim this once and the rest of the docs will read faster.

## Item

A single AI-produced decision waiting on (or having passed through) the loop. Examples: a drafted support reply, a content moderation verdict, an invoice category, a generated email. Items are immutable once decided.

## Project

A logical bucket of items sharing the same routing config: thresholds, risk-flag rules, reviewer pool, guidelines, and webhook destination. Most teams start with one project per AI use case (e.g. "Support replies", "Invoice tagging").

## Suggestion

The AI's proposed answer for an item. Can be a string, structured JSON, or a multi-field object. LoopDesk does not run the model — your pipeline produces the suggestion and submits it.

## Confidence

A `0.0–1.0` score your pipeline attaches to each suggestion. It is the primary input to the routing decision. Confidence is *your* number — LoopDesk does not recompute it. Calibrate it carefully (see [Tuning thresholds](/hitl-docs/guides/tuning-thresholds.md)).

## Risk flags

A free-form array of strings your pipeline attaches when something deserves attention regardless of confidence: `pii`, `high_value`, `legal`, `medical`, `policy_unclear`, `new_user`, etc. A single flag can force human review or escalation.

## Decision

The terminal state of an item: `approved`, `rejected`, or `escalated_final`. A decision carries the final answer (which may differ from the AI suggestion), the reviewer's id, and a timestamp. Decisions trigger outbound webhooks.

## Override

When a reviewer's decision differs from the AI suggestion. Overrides are the training signal: they tell you where your model is wrong, where guidelines need to evolve, and where thresholds need tuning.

## Learning

A captured override, stored with full context (input, suggestion, correction, reason, project, guideline). Learnings flow into `/learnings` and can be exported to fine-tune or improve prompts.

## The loop

```
INPUT → AI pipeline → SUGGESTION → routing policy → ┬─ auto-approve → DECISION → webhook
                                                    ├─ queue → human → DECISION → webhook
                                                    └─ escalate → senior → DECISION → webhook
                                                                       │
                                                                       └─→ override → LEARNING
```

Everything in this documentation maps back to a step in that diagram.


---

# Agent Instructions: Querying This Documentation

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

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://hitl-01.gitbook.io/hitl-docs/getting-started/core-concepts.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
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.