I compared committed HEAD (e8b4177ea) against origin/main, excluding all uncommitted Nirnay work. No stash was needed.

One important finding: this committed change does not modify the parent/co-applicant workflow implementations. It changes how Pravesh starts and locates workflows, and how backend callbacks resolve customers. So the biggest QA risk is an old workflow successfully calling the new backend but receiving results that its saved state handles incorrectly.

For QA, prepare these three starting points:

Keep the old applications’ temporal_workflow_id blank. That is how their mobile-based workflow fallback is exercised.

Highest-priority compatibility checks

# What to test What must happen
1 Open the old early application and continue OTP, KYC and personal information. Requests reach its original loan-application-{mobile} workflow. No replacement parent workflow is created.
2 Create a fresh application. Complete the first step, close the app, and reopen it. Workflow ID is loan-application-{app_id} and is saved against the application. Subsequent steps still reach that workflow.
3 Keep an old application live. Start a new application using the same mobile but a different person and verified PAN. Both applications work independently. Updating either must not change the other’s workflow, documents, consent or customer details.
4 Create two applications with the same mobile on the same day. Also try simultaneous submissions from two devices. Successful applications have distinct application IDs and workflow IDs. A collision must never open or update somebody else’s application.
5 Start a new application, then simulate a timeout after Temporal starts it but before the API returns successfully. Retry and reopen it. There is a recoverable application with the correct stored workflow ID. It must not accidentally fall back to an old workflow sharing the mobile. This is a particularly important failure-window test.
6 An old workflow started before deployment but has not yet created its application row. Allow its callback to finish after deployment. Its row remains compatible with the old mobile-based workflow ID; it must not acquire an application-based ID for a workflow that never existed.
7 On the old progressed application, edit both co-applicants; delete the first; edit the remaining one; add another. Each request reaches the correct child workflow. Index changes must not make one person’s PAN, bureau or documents appear against another.
8 Retry a co-applicant deletion after the backend already deleted it—for example, simulate a lost response. Retry is safe, the remaining co-applicant stays intact, and the deleted person’s identity participation is inactive.
9 Restart the worker and continue the old applications. Include an application created before earlier workflow patches. Workflows resume without nondeterminism errors, repeated customer creation or duplicated child workflows. Use genuinely old histories, not just freshly created applications with old-looking IDs.

Identity and dedupe checks

# What to test What must happen
10 Two different people share one mobile, including a primary applicant and co-applicant in the same application. Give each a different verified PAN. Shared mobile alone does not block or merge them. They remain separate people.
11 A returning customer uses the same verified PAN and matching name, but a different mobile. Their old application is still open. The committed default policy allows reuse across applications. The same identity/customer is used; the older application’s captured information remains intact.
12 Repeat #11 using an old customer with no existing identity-resolution records. The legacy customer is discovered and linked when verified identity evidence is submitted. It must not require the customer to have been created after this release.
13 Add the primary applicant again as a co-applicant using the same verified PAN. Repeat with the same co-applicant twice. The second participation is blocked with a readable message. Changing mobile must not bypass this. Repeat against an old application without identity-participation rows.
14 Delete a co-applicant, then add that same person again. Also use that person as primary in another application. Re-addition is allowed once the old participation is inactive. Roles across different applications are supported. Historical evidence remains available.
15 Submit the same personal-information step twice, then change an ordinary field and save again. A retry or edit does not treat the person as a newly added duplicate, create extra active participation, or corrupt their existing identity link.
16 Use a PAN already associated with one name, but submit a substantially different name. A failed PAN name/DOB check must remain a verification failure. If accepted verification evidence still conflicts with historical captures, the case must stop for review. It must not silently merge or overwrite either person.
17 Use the no-PAN/Voter ID path: first upload or enter it, then complete validation. Repeat for an existing Voter ID customer. Unverified evidence is not treated as verified ownership. After successful validation, returning-person matching works. Check this through the actual old worker flow—not only a direct backend callback.
18 Provide conflicting verified identifiers—for example, PAN linked to person A and alternate ID linked to person B. No automatic merge or silent reassignment. The user gets a review-required response and cannot complete the affected step.

Archive and downstream checks

# What to test What must happen
19 Archive an abandoned old application containing KYC, documents, verification results, bureau and co-applicants. Then apply again as the same person. Snapshot and existing evidence remain accessible; old participation becomes inactive; the returning customer resolves to the existing identity. Repeat the archive operation to check it is safe.
20 Archive an application, then let an already-running callback finish. Separately terminate its workflow and reopen the application. A late callback must not silently reactivate or overwrite the archived application. The UI must handle the terminated workflow without endless loading. Archiving alone does not terminate Temporal in this implementation.
21 After dedupe links a new provisional customer to an old customer, continue bureau, banking/AA, document submission and Nucleus push. Repeat for a co-applicant. Every step remains attached to the correct person and application. Bureau reuse/refresh follows its existing validity rules. No duplicate LMS customer should be created.

I would give #5 and #21 extra attention. The workflow ID is saved only after the initial Temporal update succeeds. Also, the backend can repoint an application participant to an existing customer while the old workflow still holds its original customer_id. Those are concrete places where a happy-path UI check can miss a failure.

For every test, record the application ID, parent/child workflow IDs and run IDs, participant customer ID, identity ID, and any CIF. Check both the UI and those associations. A successful screen transition alone does not prove correct routing.

Finally, the committed Nucleus change records returned customer/CIF associations; it is not the later Nirnay global CIF-creation control. Treat existing-CIF reuse as something to verify end to end, rather than assuming dedupe alone guarantees it.