Explain End to End Persistent TaskFlow without relying on framework buzzwords.
Connect the lesson to the phase outcome: Build persistent task CRUD through controller → service → Prisma with explicit contracts and verified database effects.
Reproduce the relevant behavior and verify it with evidence.
Diagnose at least one representative failure instead of guessing.
Figure 11.9 — TaskFlow boundary map
HTTP route
→
controller
→
service
→
Prisma query
→
PostgreSQL
→
mapped response
This lesson belongs to Phase 11. Keep the full request/data path visible even when today's code changes only one box.
Instructor workflow: predict first → implement the smallest slice → run it → inspect evidence → break one assumption → diagnose → explain it back. Do not copy the whole code block before understanding the first failing boundary.
End-to-End Persistent TaskFlow
Reconnect the thin Next.js view and debug across layers
Phase 11 replaces process memory with durable PostgreSQL-backed application behavior.
Reconnect the thin Next.js view
The backend is the focus of Phase 11, but a short integration check keeps the full-stack mental model alive.
Next.js can now read the durable /tasks resource instead of the old in-memory /todos endpoint.
if (!response.ok) {
throw new Error(`Task API returned ${response.status}.`);
}
return response.json();
}
The API base URL remains server-only in this example because the Server Component performs the request.
No database URL or database credential is ever exposed to the browser.
Task page
import { getTasks } from "@/lib/api/tasks";
The browser still talks to an application layer. It never connects directly to PostgreSQL.
End-to-end test
• Create a Task with PowerShell against NestJS.
• Refresh the Next.js /tasks page.
• Confirm the new Task title and status render.
• Restart NestJS, refresh Next.js again, and confirm the Task still renders.
Debug by layer
Observation Likely layer to inspect first
NestJS /tasks fails in PowerShell Backend / Prisma / database
PowerShell succeeds, Next.js fails Next.js API client / environment URL
API returns [] but SQL shows rows Prisma query / wrong database connection
SQL has no row after POST Create path / migration / DB connection
Rows disappear only after restart You are still hitting memory somewhere
Still intentionally simple
No mutation UI, optimistic update, TanStack Query cache, authentication, or authorization is added yet. Those later
phases should solve real problems one at a time.
Executable lesson studio · crud
Now turn the explanation into engineering skill.
End to End Persistent TaskFlow matters because it changes a real boundary in TaskFlow. The goal is to understand the inputs, owner, observable output, and failure modes before adding more abstraction.
By the end, you should be able to
Explain End to End Persistent TaskFlow in terms of concrete inputs, ownership, outputs, and evidence.
Implement or reproduce the smallest TaskFlow slice that demonstrates this crud concept.
Diagnose one representative crud failure using observable evidence instead of trial-and-error.
Figure 11.9B — Mental model
HTTP contract
→
controller
→
service rule
→
Prisma operation
→
database row
Read left-to-right while implementing. When debugging, walk right-to-left until the first observed fact differs from your prediction.
Code-along sequence
One coherent change at a time
01
Restate the goal of “End to End Persistent TaskFlow” as one observable behavior. Write the success signal before touching code.
02
Trace the boundary map: HTTP contract → controller → service rule → Prisma operation → database row. Mark the one boundary this lesson changes.
03
Inspect the current repository state and record the exact files, route, table, or runtime involved. Avoid creating an abstraction before you can name the existing owner.
04
Implement the smallest vertical slice. Add one coherent change, run the narrowest verification command, and read the output before continuing.
05
Connect the slice to TaskFlow’s phase artifact: Working /tasks API supporting create/list/read/update/delete against PostgreSQL with consistent error behavior.. Verify behavior from the outside as a user or API client would.
06
Create one controlled failure that is safe to reproduce. Predict the symptom first, capture evidence, locate the earliest broken boundary, then fix only that cause.
07
Close the reference and rebuild or explain the core change from memory. Record what you would monitor or test before trusting the same change in production.
Do not copy this as a finished answer. Predict what each boundary does, type the smallest useful slice, run it, then compare the result with your prediction.
Verification
Prove the happy path
Run the narrowest relevant command or request and require a clean exit/status.
Inspect the boundary-specific evidence: terminal output, Network panel, database row/query plan, test report, container health, or deployment logs.
Repeat the behavior once after the change so a one-off cached or stale result cannot masquerade as success.
Break it on purpose
Build debugging memory
Controllers contain Prisma calls
Change one assumption in the crud example—identity, input, configuration, ordering, connectivity, or state—and predict the exact symptom before running it.
SymptomInvestigation path
Expected result does not appear
Return to the nearest observable boundary: input, network, process log, database, or rendered UI. Compare actual evidence with the predicted path.
A shortcut appears to fix it
Change one variable at a time and reproduce the original failure after the fix; otherwise you have correlation, not diagnosis.
Closed-reference challenge
Can you transfer the idea?
Without copying the final source, create a second small scenario that applies “End to End Persistent TaskFlow” to a different TaskFlow resource or input. Write the acceptance criteria first, implement it, and provide the evidence that each criterion is satisfied.
Explain-back
Answer these without reopening the solution
What problem does End to End Persistent TaskFlow solve in TaskFlow, and what would be observably wrong without it?
Which part of the crud boundary owns the decision, and which neighboring layer must not own it?
What evidence proves the happy path instead of merely suggesting it worked?
Name one realistic failure or abuse case and the first place you would inspect.
If requirements doubled tomorrow, which part of this design should stay stable and which part is allowed to change?
Lesson studio
Turn the reading into skill
01
Do it
Implement list/read with deterministic ordering and missing-record behavior.
Use the smallest change that demonstrates this lesson. Stop after it works and capture evidence.02
Break it safely
Controllers contain Prisma calls
Predict the symptom before causing the failure. Then locate the earliest boundary where reality diverges from expectation.03
Explain it back
Why does this responsibility live here, what input does it trust, what output proves success, and what would go wrong if another layer owned it?
If your explanation is vague, repeat the boundary trace before moving on.
Exit ticket
Before marking complete
Close the reference. Reproduce the core idea from memory or explain the exact request/data flow aloud. Then check your diff or notes for accidental shortcuts that bypass validation, authorization, error handling, or verification earned in earlier phases.