Download the recorded evidence
← Engineering

Trust needs
clear boundaries.

Controls are tied to specific actions and failure paths. Tests provide evidence for those paths; deployment configuration and release review remain part of the work.

The implementation approach.

Identity before action

Protected endpoints validate authentication, role and resource ownership. OTP attempts are bounded, stored codes are hashed, and refresh tokens rotate. Invalid or unavailable verification does not grant access.

Money belongs to the server

Order prices and totals are calculated from server-side catalog data. Amounts use integer paise. Client-submitted prices and a successful checkout callback are not accepted as payment proof.

Explicit handover and payment

Handover QR tokens are short-lived and single-use, with server-side participant and proximity checks. Payment handling verifies webhooks and checks order amounts and event identity.

Database-enforced contention

Claiming uses database transactions, row locks and guarded state updates. A uniqueness constraint prevents more than one active order for a vendor. Events commit alongside their state changes.

A narrow public waitlist

The public waitlist validates a bounded input schema, requires consent, suppresses honeypot submissions and uses an atomic Redis attempt limit. Normalized contact-and-role uniqueness handles duplicate submissions.

Keep personal data out of diagnostics

Waitlist request logs exclude submitted contact and location details. Its requests receive server-generated IDs. Dependency errors return fixed public messages rather than raw exception text.

Built in reviewable steps.

Implementation sequence. These stages do not imply that every field test or release gate has passed.

  1. 01

    Foundations

    Shared design tokens, strict TypeScript, database migrations and two Android apps.

  2. 02

    Identity and ordering

    OTP, sessions, ownership checks, server pricing, idempotency and a transactional outbox.

  3. 03

    Claim and tracking

    Order-state guards, concurrent claim tests, authorized realtime rooms and shared GPS smoothing.

  4. 04

    Handover and payment

    Arrival checks, single-use QR verification, payment confirmation and delivery history.

  5. 05

    Failure and load evidence

    Recorded local dependency-failure tests and nine measured HTTP workloads.

  6. 06

    Website and early access

    Static product pages, the shared-engine browser simulator and an end-to-end tested waitlist.

What the evidence leaves open.

Local tests do not establish production capacity, handset notification deduplication or real-world GPS performance. The recorded M8.1/M8.2 harnesses do not close the deferred full clean-database suite. Public waitlist operation still needs durable storage and reviewed proxy identity, retention and deployment settings.

Read the recovery evidenceSource repository