ROLLING THUNDER
demand-verified ordering for Banglador's Ultimate · reference architecture
LAYER 1 · MOBILE CLIENT
Crave button"I NEED THE ULTIMATE" — one tap, one signal
Thunder meterlive zone progress toward threshold
Order & checkoutcart, payment, only during open window
Live trackingcourier GPS, wave status
⇅ signals up · meter + window pushes down
LAYER 2 · API GATEWAY
Auth (JWT)who is signaling
Rate limitingone signal per user per window, enforced
Device attestationfirst line of anti-gaming
↓ validated signals
LAYER 3 · DEMAND ENGINE
Signal aggregationpool by geozone + time window
Threshold rulesX signals in Y minutes = meter full
Demand heatmapfeeds the meter UI in real time
↓ "threshold met" event
LAYER 4 · AI VERIFICATION (THE BOUNCER)
Legitimacy scoringbot / multi-phone / farm detection
Demand forecastwill signals convert to paid orders?
Capacity checkcan the kitchen handle this wave size?
Approval decisionapprove → open window · reject → meter resets
↓ approved wave (size, zone, window expiry)
LAYER 5 · ORDERING
Cart & checkoutlive only while window is open
Payments (Stripe)auth now, capture on kitchen accept
Order state machineplaced → accepted → cooking → dispatched → delivered
Instant refundswave fails → money back before complaints
↓ batched order wave
LAYER 6 · FULFILLMENT
Kitchen display (KDS)one batched ticket wave, not 40 singles
Capacity APIthe hard ceiling the AI respects
Courier dispatchbatched routing, DoorDash-style
Live GPS trackingback up to the client
↓ everything learns from
LAYER 7 · DATA PLATFORM
Event logevery signal, approval, order, refund
Demand forecasting modelpredict the next thunder before it forms
User reputationfeeds legitimacy scoring
DESIGN PRINCIPLES
- Scarcity is the product. The thunder meter isn't a loading screen — it's the reason to open the app. Always-on ordering is DoorDash; this is a drop.
- The AI gates, it doesn't serve. Verification is a separate layer from ordering on purpose. If the model ever starts taking orders, you've rebuilt DoorDash with extra steps.
- Thresholds are public, models are not. Everyone sees the meter and the number. Nobody sees the legitimacy score. Transparency where it builds trust, opacity where it prevents gaming.
- Batch the kitchen. The whole economic argument: verified waves mean predictable prep. A wave of 40 beats 40 random orders.
- Fail loud and refund instantly. A rejected wave or a kitchen failure must resolve in seconds with money back — trust is the only inventory.