41.2M
Attempts per day
Peak observed burst of 1,840/s during a Black Friday drop. The mesh behind this page emits at that true ratio, scaled to what your device can draw.
Payment orchestration · 6 acquirers · 41M attempts/day
Your processor’s status page is a lagging indicator.
MERIDIAN scores every acquirer path on live latency and auth rate, continuously. When one degrades, traffic has already moved — a median of 340 milliseconds before anyone posts an incident.
- Attempts routed daily
- 41.2M
- Median failover
- 340ms
- Auth uptime, 90d
- 99.994%
- Acquirers in mesh
- 6live
The mesh behind this page is live. Scroll to take an acquirer offline.
01 — Mechanism
Every attempt is a routing decision, not a destination.
Most stacks pin a merchant to one acquirer and retry on failure. That retry costs you the customer. MERIDIAN treats the six acquirers as a graph and prices every edge on what it is doing right now.
-
Cost
Each path carries a live price
Edge cost is
latencyp95 × (1 − auth rate) × fee, recomputed on a 250ms tick from real settlement telemetry — not from a health check that only knows up from down. -
Solve
The cheapest path wins, per attempt
A weighted shortest-path solve runs against the current cost matrix. Two payments issued a second apart can legitimately take different acquirers. There is no sticky assignment left to go stale.
-
Shed
Degradation is a gradient, not a flag
An acquirer at 94% auth does not get cut off — it gets expensive, so it receives proportionally less traffic. Full evacuation happens only when the cost curve says it should.
-
Heal
Recovery is measured, never instant
A recovered acquirer earns traffic back on a ramp. Slamming full volume into a host that just came up is how a brownout becomes an outage.
02 — Watch it happen
ACQ-3 is about to go down. Keep your eyes up.
This is a replay of a real degradation on 14 March, driven by the same solver that runs in production. Nothing here is pre-rendered — the paths behind this text are being solved as you scroll.
Live solve — ACQ-3 highlighted
Routing topology — ACQ-3 evacuated
- Healthy acquirer carrying routed traffic
- Cost rising — receiving proportionally less
- Evacuated — no new attempts entering
- Each dot is one payment attempt in flight
Phase
Nominal
In flight
1,204 attempts
Median latency
96 ms
Auth rate
99.994 %
-
T+0.0s
ACQ-3 p95 crosses 800ms. It is still returning 200s, so every uptime monitor in the industry still calls it healthy.
-
T+0.25s
Cost tick lands. The ACQ-3 edge price triples. The solver starts preferring ACQ-1 and ACQ-5 on the next attempt, not the next deploy.
-
T+0.34s
Traffic has moved. Residual in-flight attempts drain through ACQ-3; new attempts no longer enter it. Auth rate never leaves five nines.
-
T+41s
Their status page updates. By then you have settled 338,000 payments through the other five.
03 — The mesh, healed
What five nines actually costs to hold.
340ms
Median failover
Measured from the first anomalous cost tick to zero new attempts entering the degraded acquirer. p99 is 1.1s.
99.994%
Auth uptime, 90 days
Across four acquirer degradations and one full regional outage. The un-routed control group over the same window: 99.71%.
+2.3pts
Auth-rate lift
Median across 40 merchants in their first quarter, from routing alone — before any retry or network-token tuning.
Send one payment through the mesh.
Sandbox keys, no card required. Point a test transaction at the router and watch it pick an acquirer in real time — then kill one and watch it pick another.
Sandbox is free forever. Production starts at 41M attempts/mo.
04 — Questions
The ones engineers ask first.
Does this add a hop to every payment?
Yes — 4 to 7ms at p95. The solver runs in-region against a cost matrix already in memory, so there is no network call to make a routing decision. Against a 340ms failover and 2.3 points of auth lift, that is a trade most teams take.
What happens if MERIDIAN itself goes down?
The SDK holds the last-known-good cost matrix and keeps routing against it locally. You degrade to stale routing, not to no routing. If it cannot solve at all, it falls through to your configured primary acquirer.
Can I pin certain traffic to one acquirer?
Yes. Constraints are first-class: pin by BIN range, currency, card brand, or merchant of record. Pinned traffic is excluded from the solve and reported separately, so it never flatters your routing numbers.
Where does the cost signal come from?
Your own settlement data, not a shared pool. Every attempt you route returns latency, response code, and settled amount, and that prices the next edge. Cold start uses acquirer-published SLAs for roughly the first 10,000 attempts.
Is the visualization on this page real?
It is the real solver logic — weighted shortest path over a live cost matrix — running against a seeded replay of the 14 March incident rather than against production traffic. Same algorithm, recorded inputs.