* fix(dashboard): cache org stats in Redis with versioned invalidation Replace the never-expiring process-local cache in the org dashboard page load with API-side Redis caching keyed per org. A version stamp prevents slow reads from writing pre-purge data back into the cache. Invalidate on course create/update/delete, student enroll/remove, certificate claim, and progress reset. Remove the dashboard SSR module cache so every page load reads through the API cache layer. Fixes #852 Co-authored-by: Rotimi Best <[email protected]> * fix(dashboard): address PR review on stats cache invalidation - Validate bust via ZDashStats instead of raw query access - Use import type for OrganisationAnalytics - Resolve orgId before course delete so invalidation is reliable - Invalidate immediately after persisted writes, before fallible side effects - Invalidate on member role changes and program course access enrollment Co-authored-by: Rotimi Best <[email protected]> * fix(dashboard): complete stats cache invalidation coverage Wire org stats cache purge on lesson completion (manual toggle and video-watch auto-complete) so topCourses completion percentages stay fresh. Harden bulk member adds with sequential inserts and a finally invalidation boundary, and only purge org invite enrollments when new group memberships were actually inserted. Co-authored-by: Rotimi Best <[email protected]> * fix(core): make lesson completion stats purge best-effort Wrap org stats cache invalidation in try/catch so a lookup or Redis failure cannot turn a successful lesson completion into a 500. Co-authored-by: Rotimi Best <[email protected]> * fix(dashboard): keep org stats cache through lesson completion Lesson complete is a high-frequency event and usually does not change the rounded dashboard percentages. Rely on a 10 minute TTL instead of purging the snapshot on every completion. Co-authored-by: Rotimi Best <[email protected]> * docs: explain how admin dashboard numbers are calculated Add a user-facing guide covering headline counts, top courses, certificates, login activity, and when the 10 minute snapshot refreshes. Co-authored-by: Rotimi Best <[email protected]> ---------
@cio/core — shared business services
This package holds the business-logic layer that both apps/api (HTTP) and apps/jobs (BullMQ worker) need to call. Anything in here is transport-agnostic: no Hono Context, no ContentfulStatusCode, no c.json, no zValidator middleware. Just functions that compose database queries, apply business rules, and raise AppError on failure.
The name is intentional: this is the core of the platform. The HTTP routes in apps/api and the worker processors in apps/jobs are thin transports over this layer. Anything they share has to live here — otherwise one app ends up depending on the other (apps/jobs → apps/api was the original sin that motivated extracting this package).
What lives here
| Folder | Contents |
|---|---|
src/services/agent |
The durable agent runner, chat-mode tool implementations, planning helpers. |
src/services/course |
Course / section / content / landing-page business logic. |
src/services/exercise |
Exercise + question CRUD and orchestration. |
src/services/lesson |
Lesson read/write helpers. |
src/services/assets |
Asset upload, attach, detach, presign — used by both API routes and worker. |
src/services/jobs |
BullMQ enqueue helpers (e.g. startMediaJob) callable from any service. |
src/services/lesson-language |
Localized lesson content helpers. |
src/utils |
Storage, presigning, sanitization, tinybird, redis helpers. |
src/config |
Env schema (env), storage config, dashboard URL helpers. |
Placement rule
┌─────────────────────────────────────────────────────┐
│ Does the function need a Hono Context, or call │
│ `c.json`, or validate a request shape? │
│ │
│ YES → it's a route handler, stays in │
│ `apps/api/src/routes/{domain}/…` │
│ │
│ NO → it's a business service, goes in │
│ `packages/core/src/services/{domain}/…`│
└─────────────────────────────────────────────────────┘
If the function only touches the database and has no business logic, it doesn't belong here either — it goes in packages/db/src/queries/{domain}/. The split is:
- Query layer (
@cio/db/queries): pure SQL/Drizzle helpers, no business rules. - Service layer (
@cio/core/services): composes queries, applies rules, throwsAppError. - Route layer (
apps/api/src/routes): HTTP transport — auth, validation, status codes, response shapes. - Worker layer (
apps/jobs/src/workers): BullMQ transport — pulls a job, calls services, persists outcome.
What does NOT belong here
- HTTP-shaped code (Hono types, response helpers, content-type handling) — keep in
apps/api. - BullMQ worker bootstrapping (Redis connection, worker lifecycle) — keep in
apps/jobs. - Frontend / Svelte / UI primitives — those live in
apps/dashboardorpackages/ui. - Raw Drizzle query functions with no business logic — those go in
packages/db/queries.
Why it isn't called @cio/services
@cio/core is shorter and reads cleaner at import sites (from '@cio/core/services/course/content'). The folder structure makes it obvious that it's a services layer; the package name doesn't need to repeat that.
History
This package was originally extracted as @cio/agent-runtime to hold the durable agent runner and its transitive closure. That closure turned out to be ~80% generic business services that apps/api and the worker both need, so the package was renamed to @cio/core to match its contents. The agent runner (services/agent/course-generation-runner.ts) remains the top-level export for apps/jobs, but it's just one of many services living here now.