[#648] Add Career Discord settings in Blade

.forge/features/career-discord-config/spec.md

@@ -0,0 +1,62 @@
+# Career Discord Configuration Spec
+
+Status: implemented and locally verified; [PR #664](https://github.com/KnightHacks/forge/pull/664) closed unmerged. Not republished.
+
+## Product promise
+
+Officers can configure Career posting destinations and digest notification roles
+in Blade's existing `/admin/roles/config` screen. Reorganizing Discord should
+not require a developer to change destination IDs in source code.
+
+This implements configuration for [#648](https://github.com/KnightHacks/forge/issues/648),
+part of [Career #633](https://github.com/KnightHacks/forge/issues/633).
+
+## Settings and people
+
+| Setting                     | Purpose                                         |
+| --------------------------- | ----------------------------------------------- |
+| Internship posting channel  | Internship posts and digests                    |
+| New-grad posting channel    | New-grad posts and digests                      |
+| Internship digest role      | Opt-in audience mentioned by internship digests |
+| New-grad digest role        | Opt-in audience mentioned by new-grad digests   |
+| Source-health alert channel | Optional destination for source-failure alerts  |
+
+Officers edit settings. Notification roles choose recipients and do not grant
+access to officer settings. The existing Officers role remains separate.
+Members' role subscriptions and actual delivery are owned by later tickets.
+
+## Settled behavior
+
+- Use the existing editor, with production and development ID fields and Save.
+- Digests share their track's posting channel; no extra digest channel is added.
+- The failure channel starts **Not configured**. An officer can set or clear it
+  later. Other production IDs remain required.
+- All five settings display **Unused** until real consumers are added.
+- Ordinary edits preserve the existing officer gate, audit history, and cache
+  invalidation. Reloading shows the persisted value.
+- Adding configuration does not send messages or assign Discord roles.
+
+The user supplied the four production destinations/audiences and explicitly
+requested leaving the failure destination unset for officer setup later.
+Consolidated-channel IDs were recorded as context, not extra settings.
+
+## Acceptance
+
+1. Officers see all five settings under the correct channel/role groups.
+2. Valid edits persist; non-officers cannot read or edit officer configuration.
+3. Existing settings and officer edits survive migration and seed replay.
+4. The unset failure entry is visible, editable, and accurately labelled.
+5. Cron/API callers can resolve required IDs or explicitly read an optional ID.
+6. Missing migration rows remain errors rather than silently disabling delivery.
+
+## Scope limits and remaining setup
+
+No job polling, posting, digest scheduling, subscription management, new Discord
+objects, channel picker, or new settings page. Downstream #651/#653/#654 own
+consumers. They must skip only an unset failure notification while retaining
+normal work and error logging.
+
+Development destinations remain unknown. Resolve them or explicitly accept the
+existing production fallback before real non-production delivery is connected.
+Coordinate the five-key contract with downstream owners and confirm deployment
+sequencing before merge.

.forge/features/career-discord-config/srd.md

@@ -0,0 +1,132 @@
+# Career Discord Configuration SRD
+
+Status: implementation complete; full local checks and focused review passed.
+Base: upstream main `b88d4de35ac47f7cb1555658741778a792351c90`.
+
+## Contract and provenance
+
+| Key                             | Kind    | Production ID         | Development ID |
+| ------------------------------- | ------- | --------------------- | -------------- |
+| `career_internship_channel`     | channel | `1537949209867583499` | null           |
+| `career_new_grad_channel`       | channel | `1537949230742900827` | null           |
+| `career_internship_digest_role` | role    | `1537949002031562862` | null           |
+| `career_new_grad_digest_role`   | role    | `1537948941348503683` | null           |
+| `career_source_health_channel`  | channel | null                  | null           |
+
+Source: user-supplied production IDs on 2026-11-02. These were not independently
+verified against Discord. The user confirmed the failure channel stays unset
+and an officer fills it later. Null development values retain the existing
+production-fallback behavior; no Career delivery consumer is enabled here.
+Development routing must be resolved before live non-production delivery.
+
+The five names follow the parent #633 automated proposal and the user's accepted
+track-specific plan. The child #648 automated proposal suggested three generic
+keys instead; it is not the implemented contract. Coordinate with downstream
+owners before merge. Schema changes were explained before the user authorized
+Phase 4; Phase 5 was explicitly requested for optional editing/reading behavior.
+
+Additional user inventory, excluded from new routing keys:
+
+- Internship consolidated: `1546636298188488704`.
+- New-grad consolidated: `1546636299891384390`.
+- Officers role: `486629374758748180`, matching the existing officer-role seed;
+  it is not a digest audience.
+
+## Ownership
+
+- `@forge/consts`: supported keys, optional-key metadata, consumer labels.
+- `@forge/db`: table, environment resolver, migrations and storage constraints.
+- `@forge/validators`: trimmed ID input and key-specific required-value checks.
+- `@forge/utils/discord-config`: shared reads and process-local snapshot cache.
+- Existing `@forge/api` router: officer access, writes, auditing and invalidation.
+- Blade: existing console, field display and controlled edit dialog.
+
+No procedure, authorization rule, app, dependency, secret, deployment workflow,
+or actual delivery consumer is added. All new consumer arrays stay empty.
+The source scanner recognizes optional as well as strict reader calls so future
+consumers must update usage metadata honestly.
+
+## Storage and reads
+
+One row is one supported setting; officers edit its values, not its key/kind.
+Production uses productionId. Other environments use developmentId, falling
+back to productionId when developmentId is null.
+
+Only career_source_health_channel may have a null productionId. A named
+key-specific database check keeps every other setting required. Existing
+17–20 digit snowflake checks remain intact. There are no placeholder IDs.
+
+DISCORD.OPTIONAL_CONFIG_KEYS contains only the failure key. Shared input
+validation normalizes blank/whitespace to null, then rejects a missing production
+ID for other keys with a productionId field error. Unknown keys/fields and
+malformed snowflakes remain invalid. Validation does not verify Discord object
+existence, guild membership or bot permissions.
+
+getDiscordConfigId retains Promise<string>: missing rows and unresolved nulls
+throw. getOptionalDiscordConfigId accepts only OptionalConfigKey and returns
+string or null for intentional unset state. Missing rows and database errors
+still throw. Both share the existing whole-table cache and 60-second TTL.
+
+The unchanged update procedure locks the stored row, writes the update and audit
+in one transaction, then invalidates the writing process's cache after commit.
+Other processes converge within the existing TTL. No row creation is exposed.
+
+## Blade behavior
+
+Null production/resolved fields show Not configured. A null development field
+shows Reuses production only if production exists; otherwise Not configured.
+The same rules apply to desktop tables and mobile cards.
+
+The optional editor permits configuring, clearing, and label-only edits while
+unset. Other production fields remain required and disable Save when blank.
+Existing guild-repoint confirmation, mutation feedback and officer gates remain.
+
+## Migrations and rollout
+
+- 0064_career_discord_config: relax production_id NOT NULL, add
+  knight_hacks_discord_config_production_id_required_check, then seed the five
+  rows with their final descriptions using ON CONFLICT (key) DO NOTHING to
+  preserve officer edits.
+- One generated snapshot and journal entry. Historical 0026 SQL, provenance
+  fixture and locked migration hashes remain unchanged.
+
+Deploy compatible code, drain old Blade instances, then apply the migrations.
+Old code cannot map new keys to consumer labels and cannot correctly present
+an unset production ID. Confirm this sequence with the maintainer: CI's optional
+migrate_prod job waits for build, but has no code-deployment/drain dependency.
+The repository variable and external deployment timing remain unverified.
+
+Prefer a forward fix. Do not roll code back across registry support while the
+new rows remain visible. Restoring NOT NULL also requires first configuring or
+removing the unset failure entry. Never rewrite an applied migration.
+
+## Proof and limits
+
+The migration suites cover fresh installs, the previous schema, preservation of
+existing edits, repeated seeds, exact IDs/kinds, required-key null rejection,
+snowflake checks, and historical lineage. API tests cover null/value audit diffs,
+SQL persistence, list results and immediate read invalidation. UI tests cover
+both responsive renderings and editor guards; utils tests cover reads and cache.
+
+Real Chromium verified officer save/reload/clear against a disposable database,
+SQL confirmation, required-field Save behavior, desktop and 320px mobile display,
+and dialog bounds without document overflow. Screenshots were inspected.
+Existing E2E auth mode skips audit writes; separate API integration tests run
+with that mode off and supply real audit proof. Test databases were dropped.
+No live Discord call, private environment, running Blade database or production
+deployment was used. Final check receipts live in status.md and the local review
+packet, not a published PR.
+
+## PR setup note
+
+An officer must configure the source-health channel in Blade before Discord
+failure alerts are enabled. Missing failure routing must not disable ordinary
+Career work or error logging. Resolve development routing before live testing.
+
+## Description copy
+
+Match the existing short purpose statements: channels for internship/new-grad
+posts and digests, roles pinged for each digest, and an optional channel for job
+source failure alerts. Remove repeated change warnings and officer-access
+explanations from row descriptions. Keep setup and rollout guidance in the
+editor/help and PR notes.