> ## 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.

# Architecture Overview

> The seven layers of Forge, and the three principles that shape all of them.

<Note>
  **In plain terms:** Forge is organized as a stack of layers, each with exactly one
  job. Lower layers handle trusted inputs and the model of the ship; higher layers
  handle reasoning, checking, and learning. This page is the map; each layer has its
  own page.
</Note>

## Three operating principles

Every layer below exists to serve one of these three principles:

1. **Sources have authority. Every claim cites.** Nothing the system asserts is
   ungrounded.
2. **Decisions are first-class objects, not log lines.** Every decision is recorded
   with its full reasoning context and is replayable.
3. **AI proposes; humans decide; the gate is structural.** Anything touching class,
   USCG, or safety always reaches a human.

See [Where Humans Stay in Control](/concepts/human-control) for the plain-language
version.

## The layer stack

Forge is organized as layers, each with one responsibility:

| Layer    | Name                                                                     | What it does                                                                      |
| -------- | ------------------------------------------------------------------------ | --------------------------------------------------------------------------------- |
| **L0**   | [Source & Authority Registry](/how-it-works/layers/l0-source-authority)  | Classifies every input by authority, version, access policy, and ownership        |
| **L1**   | [Ingestion & Normalization](/how-it-works/layers/l1-ingestion)           | Turns PDFs, CAD metadata, BOMs, rules, and prior files into traceable objects     |
| **L2**   | [Maritime Design Ontology](/how-it-works/layers/l2-ontology)             | The connected model of vessels, systems, rules, drawings, materials, and evidence |
| **L2.5** | [Engineering Decision Ledger](/how-it-works/layers/l2-5-decision-ledger) | The queryable record of *why* every decision was made                             |
| **L3**   | [CAD & Geometry World Model](/how-it-works/layers/l3-geometry)           | Deterministic geometry: measurements, clashes, clearances, tolerances             |
| **L4**   | [Rules & Applicability Engine](/how-it-works/layers/l4-rules)            | Applies ABS, USCG, IACS, IMO, owner, and yard rules with citations                |
| **L5**   | [Capability Catalog & Agents](/how-it-works/layers/l5-capabilities)      | The versioned menu of moves the AI is allowed to make                             |
| **L5.5** | (within L5) Confidence & Abstention                                      | Low-confidence outputs abstain, gather more context, or route to a human          |
| **L6**   | [Verification Ladder](/how-it-works/layers/l6-verification)              | Ordered checks from citation to class/USCG review and final sign-off              |
| **L7**   | [Outcome Observer](/how-it-works/layers/l7-outcomes)                     | Captures real outcomes, rework, supersession, and improvements                    |

<Tip>
  A useful way to read the stack: **L0 to L2.5** is about getting trustworthy data in
  and modeling it. **L3 and L4** compute facts and apply rules. **L5 to L7** govern
  what the AI does, check it, and learn from the results.
</Tip>

## The two clocks, mapped to layers

The [Two-Clock Problem](/concepts/two-clock-problem) shows up directly in the
architecture:

* The **state clock** ("what exists today?") lives across CAD and the ontology
  (L2, L3).
* The **event clock** ("why is it this way?") lives in the decision ledger (L2.5).

Both are required to answer questions like "are we still compliant under the new
rule version?"

## A note on the diagram

The full system has a detailed architecture diagram (Forge DesignOS v0.2). If the
visual diagram and this written spec ever disagree, the **written spec is
canonical**; the diagram is a simplification.