Independent prototype — synthetic data

Honesty

What works, and what is simulated.

This is an independent prototype built for a hackathon. It is not a government product and is not affiliated with any government body.

Status derivation

real

Every status on the track page is computed from the event stream by pure functions in lib/trail/derive.ts. There is no status column anywhere in the data model. Deliberately.

The instrumented fetch

real

A document is only rendered at the end of a logged fetch. Opening one in the officer view writes a DOC_OPENED event and stamps the step, which changes the citizen view. Nobody marks anything as done.

Waiting vs handling split

real

Queue time and handling time are computed separately from enteredAt, firstOpenedAt and completedAt. No current system separates these.

Flag and comment

real

An officer flags one document with a reason and a comment. The citizen sees it verbatim under the same share code, and does not restart the application.

Access log

real

The /access page lists every open, re-open, unusual access and blocked attempt for the citizen, and the revoke button really revokes the code. The rules surface unusual access; they do not prevent a first-time misuse, and we do not claim they do.

Document connections

real

The /locker page derives links between documents at runtime by comparing normalised fields — shared address, shared phone — and flags the ones that disagree. Nothing is precomputed.

Languages

mocked

The track screen translates the status block, status labels and step names into Hindi and Kannada (server-rendered, works with JavaScript off). Office names, the access log and the other screens stay in English. This is partial translation for the demo, not a full localisation.

Consistency checks

real

The pre-submission checks on /apply run real string comparison on the selected document fields in lib/trail/consistency.ts. The seeded PIN mismatch between the ID and ration card is caught, with the step it would fail at named. No AI is used here.

Insights & turnaround

real

Every median on /insights is computed from the timeline of 20 applications (2 seeded plus 18 deterministic synthetic ones), by the same timing functions the citizen page uses. Nothing is hardcoded — change the applications and the medians move.

Application creation

real

Generating a code on /apply creates a real trackable application and share code in the store, writes the events, and the new code works on both the citizen and officer views immediately.

DigiLocker

mocked

No real DigiLocker is contacted. Requester integration requires being an onboarded organisation with a signed agreement, and there is no tier for a citizen-chosen tool. Documents here are local synthetic fixtures.

Identity and reference numbers

mocked

All people are fictional. No 12-digit numbers exist anywhere in the UI, the code or the seed data. References use the format MOCK-ID-•••• 4417.

Officer identity

mocked

The officer view is a demo affordance so a reviewer can see the architecture work. There is no authentication and no real officer account.

Storage

mocked

Demo state lives in server memory, not a database. It resets on a cold start, which is why the demo ships pre-seeded and why the reset control below exists.

Guided journey

mocked

The routing on /journey is deterministic so it never breaks. If an OPENAI_API_KEY is set, an OpenAI model rewrites the recommendation in plainer language on request — never making the decision, and only ever seeing the broad band answers, never document contents. Without a key it falls back silently.

Submissions

mocked

Nothing is submitted to any department. No live government system is contacted, called or tested.

Back to tracking