Move report generation out of the request while preserving ownership, status, and reliable dispatch.

POST /reports creates a durable queued job and returns 202 with a Location header. A protected status endpoint exposes queued, running, completed, or failed without leaking another user's job.

The API process restarts immediately after returning 202; the persisted outbox still dispatches the report command and status remains queryable.

• Persist job and dispatch intent atomically.
• Return a status resource.
• Authorize status by owner.
• Make the worker retry-safe.

I model the job as an API resource rather than an in-memory task. One SaveChanges stores both the queued job and outbox message, then a separate dispatcher delivers work after commit.

app.MapPost("/reports", async (CreateReportRequest request,
    ClaimsPrincipal user, AppDbContext db, CancellationToken ct) =>
{
    var userId = user.RequireUserId();
    var job = ReportJob.Queue(userId, request.Filters, TimeProvider.System.GetUtcNow());
    db.ReportJobs.Add(job);
    db.OutboxMessages.Add(OutboxMessage.Create(
        "report.requested", new { job.Id, job.UserId, job.Filters }));
    await db.SaveChangesAsync(ct); // Job and dispatch intent commit together.

    return Results.AcceptedAtRoute("GetReportJob", new { id = job.Id },
        new ReportJobResponse(job.Id, "queued", null));
}).RequireAuthorization();

app.MapGet("/reports/{id:guid}", async Task<IResult> (Guid id,
    ClaimsPrincipal user, AppDbContext db, CancellationToken ct) =>
{
    var userId = user.RequireUserId();
    var job = await db.ReportJobs.AsNoTracking()
        .SingleOrDefaultAsync(x => x.Id == id && x.UserId == userId, ct);
    return job is null ? Results.NotFound() : Results.Ok(ReportJobResponse.From(job));
}).WithName("GetReportJob").RequireAuthorization();

202 means accepted for processing, not completed. Durable state and outbox delivery survive process failure; the worker still needs idempotent claiming, bounded retries, and a terminal failure policy.

• Dispatch is temporarily unavailable.
• The same job is delivered twice.
• The user requests another user's status.
• The job completes before the first poll.

• Restart between API commit and dispatch.
• Deliver the outbox message twice.
• Assert Location and ownership filtering.
• Exercise failed terminal status.

• Using Task.Run inside the endpoint.
• Returning 200 before work completes with no status model.
• Publishing before the database commit.
• Leaving jobs queued forever without monitoring.