OpenSlot — the on-sale infrastructure for high-demand drops

H0: Hack the Zero Stack submission · Track: Monetizable B2B App The on-sale/drop infrastructure that event businesses — artists, brands, promoters, ticket sellers — run to take a worldwide rush for the same scarce seat with zero oversell across regions and resale made a losing trade. Built on Amazon Aurora DSQL multi-region strong consistency (seat ledger)

• Amazon Aurora PostgreSQL PostGIS/pgvector (discovery). Frontend on Vercel / Next.js.

Who pays: the organizers (B2B — fees / subscription / anti-scalp value). Buyers transact free. The product surface that sells it is the organizer console (/org/console); the consumer flow shows the guarantee it delivers.

---

The one-sentence pitch

When a global drop goes on sale, the whole world hits the same limited inventory in the same second. Single-region or async-replicated databases either oversell, fall over, or serve stale "still available" reads. Aurora DSQL is the only engine that takes those writes active-active across regions with strong consistency — so OpenSlot can promise double-sell = 0 as a structural guarantee, not a hope.

Why two databases (deliberate — see PRD §5)

Plane	Engine	Job	Why it's forced
Seat ledger	Aurora DSQL	multi-region active-active strong-consistency claims	Only DSQL gives zero-oversell cross-region writes. Aurora PG (incl. Global DB) is async → structurally can't.
Discovery	Aurora PostgreSQL	"drops near me / by meaning" — PostGIS radius + pgvector semantic	DSQL doesn't support extensions → Postgres is forced.

A single database forces you to drop either geo+vector or multi-region strong-consistent writes. We need both → two planes, DSQL is the source of truth for availability.

What's in the box

Route	What it proves
/	Discovery — PostGIS radius + pgvector semantic search, live stock joined from the DSQL ledger
/demo	★ HERO — cross-region stampede console: fire N buyers at two regions, watch oversold stay 0
/event/[id]	Live seat map (relational render), countdown, capacity, on-sale state
/claim/[id]	OTP → device-bound session → OCC claim, surfacing OC000 honestly
/me	Identity+device-bound tickets with a rotating 30s barcode (screenshot expires)
/org/console	Live floor + fair re-release: cancel → locked offer to waitlist #1 (bots can't grab it)
/admin	Stub — but auth is enforced independently of NODE_ENV

Run it

npm install
npm run dev        # http://localhost:3000  (use PORT=3100 if 3000 is taken)

No AWS credentials are required to run the demo. OpenSlot ships with a deterministic in-memory simulation of the DSQL seat ledger that reproduces its exact semantics — row-per-seat contention, snapshot isolation, commit-time OC000 conflicts, cross-region commit latency, exactly-N capacity, fair re-release. Same seed → identical run (reproducible for the video).

Pointing at real Aurora (optional)

Set these and the real adapters activate with no call-site changes — the ledger SQL in lib/db.ts (DSQL_CLAIM_SQL) is a 1:1 translation of the simulation's check-and-set, wrapped in lib/occ.ts (withOccRetry, handles OC000/OC001):

DSQL_ENDPOINT_US_EAST_1=...     # us-east-1 regional endpoint
DSQL_ENDPOINT_US_EAST_2=...     # us-east-2 regional endpoint
DSQL_WITNESS_REGION=us-west-2   # quorum tiebreaker, no endpoint
AURORA_PG_URL=...               # PostGIS + pgvector discovery plane
OPENSLOT_DEMO_OTP=off           # hide the demo OTP hint in a real deploy

Architecture

 Vercel (Next.js, edge/functions)
   ├─ /demo, /event, /claim, /me, /org   ← UI
   └─ /api/* route handlers
        ├─ claim engine  → Aurora DSQL  (seat ledger · OCC · OC000 retry)   [source of truth]
        ├─ discovery     → Aurora PostgreSQL (PostGIS radius + pgvector)
        └─ otp / session → device-bound, sealed buyer_id

 Aurora DSQL  ── us-east-1 ⇄ us-east-2  (synchronous, strong-consistent)
              └─ witness us-west-2 (quorum tiebreaker)

Honesty fences (we don't overclaim)

• DSQL solves double-sell only. Resale/bots are solved by the other layers (identity+device binding, rotating barcode + gate scan, BotID, fair re-release). DSQL is necessary, not sufficient.
• "Resale impossible" is never claimed — accounts/devices can be transferred. We claim resale friction up, bot mass-claiming uneconomic.
• Cross-region commits cost ~2 RTT. That's the honest price of zero data loss and zero stale reads. It's visible in the demo (us-east-2 p95 > us-east-1).
• Multi-region pairing uses the documented US example (us-east-1 + us-east-2
  • witness us-west-2). Region/pair availability is re-checked at deploy time.

See CLAUDE.md for the full rules + judging criteria + AWS DSQL official facts, and docs/prd.md for the product spec (§17 = binding implementation decisions).

#H0Hackathon