# Notification Service Implementation Plan

This plan has been already implemented and stays here for historical reasons.

It should NOT be threated as source of truth for service functionality.

## Summary

This plan builds `Notification Service` as a durable asynchronous orchestration
service between domain producers, `Gateway`, `Mail Service`, and `User Service`.
The implementation must keep business-audience resolution in the producer,
contact enrichment in `Notification Service`, client push delivery in
`Gateway`, and email execution in `Mail Service`.

## Global Rules

- Keep `Notification Service` orchestration-only.
- Preserve direct auth-code email flow from `Auth / Session Service` to `Mail Service`.
- Use one dedicated Redis Stream for normalized notification intents.
- Keep route retries independent per channel.
- Do not make notification delivery a correctness dependency for gameplay or
  geo review state.
- Keep user-facing push payloads lightweight.

## ~~Stage 01.~~ Freeze Vocabulary And Cross-Service Ownership

Status: implemented.

Note:

- Later-stage artifacts may already exist in the repository as draft or
  pre-staged documentation.
- Their presence does not mark the corresponding later stages as implemented.

Goal:

- remove ambiguity before runtime work starts

Tasks:

- Freeze `notification:intents` as the dedicated ingress stream.
- Freeze that producers publish concrete `recipient_user_id` values for
  user-targeted intents.
- Freeze that `Notification Service` resolves user email and locale from
  `User Service`.
- Freeze that admin-only notifications use type-specific configured email
  lists.
- Freeze that `template_id == notification_type`.
- Freeze that private-game invites in v1 are user-bound by internal `user_id`.

Exit criteria:

- `ARCHITECTURE.md`, `TESTING.md`, and service READMEs no longer contradict the
  agreed notification model

## ~~Stage 02.~~ Define The Intent Contract

Status: implemented.

Goal:

- publish one stable producer-to-notification contract

Tasks:

- Add `notification/api/intents-asyncapi.yaml`.
- Freeze envelope fields:
  - `notification_type`
  - `producer`
  - `audience_kind`
  - `recipient_user_ids_json`
  - `idempotency_key`
  - `occurred_at_ms`
  - `request_id`
  - `trace_id`
  - `payload_json`
- Freeze duplicate and conflict rules on `(producer, idempotency_key)`.
- Freeze `audience_kind=user|admin_email`.

Exit criteria:

- every producer can publish normalized intents without service-specific side
  agreements

## ~~Stage 03.~~ Freeze The Notification Catalog

Status: implemented.

Goal:

- turn product decisions into one exact type catalog

Tasks:

- Freeze v1 types and channel matrix.
- Freeze which types are user-targeted versus admin-only.
- Freeze that `lobby.application.submitted` is user-targeted for private games
  and admin-email-only for public games.
- Freeze that `lobby.invite.revoked` produces no notification.
- Freeze payload requirements per type.

Exit criteria:

- no notification type remains partially specified

## ~~Stage 04.~~ Define Push Payload Schemas

Status: implemented.

Goal:

- freeze lightweight client-facing payloads

Tasks:

- Add `pkg/schema/fbs/notification.fbs`.
- Define one table per user-facing push type.
- Generate Go bindings under `pkg/schema/fbs/notification`.
- Document the mapping from `notification_type` to FlatBuffers table.

Exit criteria:

- `Gateway` and future client code have one stable schema file for
  user-facing notification payloads

## ~~Stage 05.~~ Freeze Mail Template Contracts

Status: implemented.

Goal:

- make email handoff deterministic

Tasks:

- Freeze `payload_mode=template` for notification-generated email.
- Add initial `en` templates for all supported email types in
  `mail/templates/<template_id>/en`.
- Update `mail` documentation so notification template IDs align with
  `notification_type`.
- Keep `Auth / Session Service` auth-code mail unchanged.

Exit criteria:

- every supported email notification type has a documented template directory

## ~~Stage 06.~~ Define Redis State And Retry Model

Status: implemented.

Goal:

- freeze durable service-local storage before runtime code

Tasks:

- Define `notification_record`, `notification_route`,
  `notification_idempotency_record`, `notification_dead_letter_entry`, and
  malformed-intent storage.
- Freeze Redis keys and schedule structures.
- Freeze route status vocabulary:
  - `pending`
  - `published`
  - `failed`
  - `dead_letter`
  - `skipped`
- Freeze retry budgets:
  - `push=3`
  - `email=7`

Exit criteria:

- the runtime can restart without losing accepted-or-retryable work

## ~~Stage 07.~~ Build The Runnable Service Skeleton

Status: implemented.

Goal:

- create the initial process shape

Tasks:

- Add `cmd/notification`.
- Add `internal/app`, `internal/config`, `internal/api`, `internal/service`,
  and `internal/adapters` packages.
- Wire Redis startup checks, graceful shutdown, logger setup, and telemetry.
- Do not add an operator REST API in v1.

Exit criteria:

- the process boots with Redis and configuration validation only

## ~~Stage 08.~~ Implement Intent Acceptance And Idempotency

Status: implemented.

Goal:

- durably accept valid intents and reject invalid or conflicting duplicates

Tasks:

- Consume `notification:intents`.
- Validate the envelope and normalized payload.
- Persist idempotency records and accepted notification records.
- Record malformed intents separately.
- Materialize channel routes according to the type catalog and `audience_kind`.

Exit criteria:

- valid intents are durable and replay-safe before downstream publication begins

## ~~Stage 09.~~ Implement User Enrichment And Locale Resolution

Status: implemented.

Goal:

- make user-targeted routes self-sufficient for later publication

Tasks:

- Read users by `user_id` from `User Service`.
- Extract `email` and `preferred_language`.
- Apply `en` fallback when locale is missing or unsupported.
- Keep admin-email routes independent from `User Service`.

Exit criteria:

- every user-targeted route can be published without additional producer input

## ~~Stage 10.~~ Implement Push Publication

Status: implemented.

Goal:

- hand off user-facing notification events to `Gateway`

Tasks:

- Encode the correct FlatBuffers table per `notification_type`.
- Publish client events into the configured `Gateway` stream with `user_id`
  targeting only.
- Apply independent `push` retry policy and route-level dead-letter handling.

Exit criteria:

- user-targeted push notifications survive temporary `Gateway` stream failures

## ~~Stage 11.~~ Implement Mail Publication

Status: implemented.

Goal:

- hand off non-auth email notifications to `Mail Service`

Tasks:

- Build template-mode generic mail commands.
- Set `template_id == notification_type`.
- Pass through normalized template variables from `payload_json`.
- Apply independent `email` retry policy and route-level dead-letter handling.

Exit criteria:

- user and admin email notifications are durably handed off to `Mail Service`

## ~~Stage 12.~~ Integrate Producers

Status: implemented.

Note:

- Implemented as the shared Go producer contract module
  `galaxy/notificationintent` because `Game Lobby` and `Geo Profile Service`
  code modules are not present in this repository yet.

Goal:

- move upstream services onto the new notification contract

Tasks:

- `Game Master` publishes:
  - `game.turn.ready`
  - `game.finished`
  - `game.generation_failed`
- `Game Lobby` publishes:
  - `lobby.runtime_paused_after_start`
  - `lobby.application.submitted`
  - `lobby.membership.approved`
  - `lobby.membership.rejected`
  - `lobby.invite.created`
  - `lobby.invite.redeemed`
  - `lobby.invite.expired`
- `Geo Profile Service` publishes:
  - `geo.review_recommended`
- Update `Game Lobby` architecture and later implementation plan to use
  user-bound private invites by `user_id`.

Exit criteria:

- producers no longer rely on ad hoc notification-side audience inference

## ~~Stage 13.~~ Add Observability And Recovery Coverage

Status: implemented.

Goal:

- make the async runtime supportable in operations

Tasks:

- Add metrics for intake, duplicates, enrichment, publish attempts, retries,
  dead letters, and lag.
- Add structured logging fields shared across intake and route publishers.
- Document manual recovery steps for dead-letter inspection and replay.

Exit criteria:

- the runtime exposes enough signals to detect stuck, noisy, or broken delivery

## ~~Stage 14.~~ Complete Test Coverage And Documentation Alignment

Status: implemented.

Goal:

- close the loop across service tests, boundary tests, and docs

Tasks:

- Add service tests for malformed intents, duplicates, locale fallback, retry
  budgets, and route isolation.
- Add inter-service tests with `Gateway`, `Mail Service`, `Game Master`,
  `Game Lobby`, and `Geo Profile Service`.
- Update `TESTING.md`.
- Update `ARCHITECTURE.md`, `mail/README.md`, `geoprofile/README.md`, and
  gateway examples.
- Verify docs still state that auth-code mail bypasses `Notification Service`.

Exit criteria:

- the implementation and the cross-service documentation describe the same
  contracts

## Final Acceptance Criteria

The implementation is complete only when all of the following hold:

- valid intents are consumed from `notification:intents`
- duplicates are idempotent and conflicting duplicates are rejected
- user enrichment resolves email and locale from `User Service`
- `push` and `email` routes are persisted and retried independently
- route dead letters are isolated per channel and per recipient
- `Gateway` fan-out remains user-wide, not session-specific
- `Mail Service` receives template-mode commands whose template IDs match
  notification types
- admin notifications remain `email`-only
- auth-code email still bypasses `Notification Service`