Add Adr

Add an ADR

An ADR captures one architecturally significant decision — what was decided, why, and what was rejected. The set lives in docs/adr/, is managed by Log4brains (.log4brains.yml), and is published as a static site from main.

ADRs are immutable. Once merged, the only thing that changes is the status. A decision that gets reversed doesn't get edited — it gets a new ADR, and the old one is marked superseded by [link]. Write it as a record of what was true on the date, not as living documentation.

When it earns an ADR

Write one when the decision constrains future work and someone will later ask "why is it like this?" — a framework or library choice, a persistence or hosting pattern, a convention every slice has to follow. Skip it for reversible implementation details, bug fixes, and anything a code comment covers.

1. Filename

YYYYMMDD-<kebab-slug>.md, where the date is when the decision was made, not when the record was written.

The slug usually leads with the category, but it isn't mandatory — both shapes exist in the repo:

20251018-api-use-fastendpoints-instead-of-minimal-apis.md
20260515-vertical-slices-use-a-folder-per-slice.md
20260612-use-aggregate-specification-classes-with-factory-methods.md

2. Start from the template

Copy docs/adr/template.md. It carries the full section structure with inline <!-- --> guidance on what each part is for and which are optional. Don't hand-roll the skeleton — the template is the source of truth for the shape.

docs/adr/index.md is the Log4brains site homepage, not a manual index. There is nothing to register — dropping the file in docs/adr/ is enough.

3. Title

H1, pattern [Category] - [Decision], or just the decision when no category fits:

• API - Use FastEndpoints instead of Minimal APIs
• Database - Use SQL Temporal Tables when data auditing is required
• Use Aggregate Specification Classes with Factory Methods

Categories in use: API, Project, Vertical Slices. Introduce a new one when the decision genuinely doesn't fit — don't force it into an existing category.

4. Metadata

- Status: accepted
- Deciders: Daniel Mackay, Anton Polkanov
- Date: 2026-06-12
- Tags: domain, specifications

Technical Story: <GitHub issue link>   <!-- optional -->

Status is one of draft, proposed, accepted, rejected, deprecated, or superseded by [xxx](yyyymmdd-xxx.md). Don't invent a deciders list — ask who was involved, or leave the placeholder for the PR author to fill in.

Tags are free-form and lowercase. Existing ones cluster around technology (dotnet, sql, azure), architecture (vsa, clean-architecture), domain (api, database, security), and process (testing, deployment, observability).

5. Sections

Required: Context and Problem Statement, Considered Options, Decision Outcome. Everything else in the template is optional — include it when it adds something.

• Context and Problem Statement — the situation that forced a choice. Enough for a reader who wasn't there. Often lands best as a question.
• Considered Options — every option genuinely evaluated, including the status quo when "do nothing" was on the table. One that lists a single option isn't a decision record, it's an announcement.
• Decision Outcome — lead with Chosen option: "X", because <reasoning>, then the consequences.
• Pros and Cons / Consequences — ✅ for positive, ❌ for negative. Every option gets honest cons; an option with no downsides means the analysis is thin.

6. Writing style

• Write for a team member who joins in two years and wasn't in the room.
• Active voice, concrete nouns, no jargon you haven't defined.
• Use the vocabulary from CONTEXT.md and avoid the synonyms it lists under Avoid — a slice is a slice, not a module or a handler.
• Link to real code (src/WebApi/Features/Heroes/CreateHero/) and to related ADRs by relative path. Images go in docs/adr/l4b-static/ — Log4brains serves that folder at the site root, so reference them as /l4b-static/<file> with alt text. The folder doesn't exist yet; create it with the first image.
• Keep code examples minimal and focused on the decision, not the implementation.

Repo context worth referencing

The decisions that keep recurring here, and where the existing patterns live:

• Slice organisation — one folder per use case under src/WebApi/Features/{Feature}/, with CreateHero as the reference shape. See ADR 20260515-vertical-slices-use-a-folder-per-slice.
• API surface — FastEndpoints with typed request/response, FluentValidation running automatically, OpenAPI via Summary classes. See ADR 20251018-api-use-fastendpoints-instead-of-minimal-apis.
• Persistence — EF Core on SQL Server, strongly typed IDs via Vogen, queries through Ardalis.Specification factory methods. See ADR 20260612-use-aggregate-specification-classes-with-factory-methods.
• Domain modelling — aggregate roots, domain events, value objects, Guid.CreateVersion7().
• Testing — unit, integration (Testcontainers + Respawn), architecture (NetArchTest).
• Orchestration — .NET 10 with Aspire: AppHost, ServiceDefaults, an AddEFMigrations migrations resource, and a run-mode-only Seeder.

If the new ADR contradicts an existing one, say so explicitly rather than quietly overriding it — name the ADR and argue why it's worth reopening.