Under the hood · provably fair
Provable, not promised. Every claim on this page is a live query from Aurora DSQL — two active-active regions, strong consistency.
Settling a proof rally — this updates automatically.
Loading region data…
Real numbers from a load test on Tokyo Aurora DSQL — not simulated, not claimed. The correctness properties are identical at any N.
2,500
join attempts
398
OCC retries → success
$100→$52
price melt
1
distinct price paid
$0.00
ledger drift
0
double-joins / charges
3.6 s
settle 2,000 users
Live price is a derived read
tier(sum(count_shards))
The price is computed from the authoritative count on every read — never stored, never raced. Crossing a tier threshold lowers the price for everyone atomically. No stale tier value; no transition race to lose.
128-shard counter per rally
random_shard() → count++
A single count++ row is the classic serializable-database contention killer. Random shard selection spreads the hot write across 128 rows, keeping OCC conflicts rare and all retrying cleanly to success.
Idempotent joins by design
INSERT … ON CONFLICT DO NOTHING
A unique (rally_id, user_id) constraint means a double-tap, a browser retry, or a race between two windows never double-counts. The ledger only charges you once — enforced at the database level.
DynamoDB handles the fan-out
liveFrame → DDB → SSE
The live frame is projected to DynamoDB on every committed join. Watchers subscribe via SSE — never via DSQL aggregate scans. Correctness lives in DSQL; million-scale fan-out lives in DynamoDB.