Payment route health

PAYMENTS0
DELIVERED0
FAILED OVER0
DELIVERY RATE--

Ready for payment traffic

LIVE TOPOLOGY

PCI authorization

ACTIVE
PCI payment route map Payments enter through the gateway. Particle swarm candidates allocate traffic across five acquirers: Adyen, Chase EU, Worldpay EU, Stripe, and Braintree. ENTRY POINT PCI gateway ACQUIRER 01 Adyen API HEALTHY ACQUIRER 02 Chase EU HEALTHY ACQUIRER 03 Worldpay EU HEALTHY ACQUIRER 04 Stripe API HEALTHY ACQUIRER 05 Braintree API HEALTHY
Available route Payment in flight PSO particles
PSO split pending

POLICY SEARCH · STANDARD

Particle population

PSO MIX
Particle swarm policy search One row per provider. Each dot is a candidate policy's traffic share for that provider; the outlined marker shows the share the swarm has settled on. TRAFFIC SHARE PER PROVIDER
Candidate policy Best found

SYNTHETIC LOAD TEST

PSO vs single-provider rule

The workload is 55% Standard, 25% Priority, and 20% Fee-sensitive. Each gets its own allocation but shares provider capacity. Ten paired seeds use 1,200 requests each; greedy moves 1% between providers by marginal utility, while the single-provider rule sends everything to the best provider at full load.

BEST SINGLE-PROVIDER IF/ELSE

Not run
FIRST-ATTEMPT APPROVALS
--
MEAN LATENCY
--
MEAN FEE
--
UTILITY / REQUEST
--
FITNESS EVALUATIONS
--

CAPACITY-AWARE IF/THEN

Not run
FIRST-ATTEMPT APPROVALS
--
MEAN LATENCY
--
MEAN FEE
--
UTILITY / REQUEST
--
FITNESS EVALUATIONS
--

PARTICLE SWARM

Not run
FIRST-ATTEMPT APPROVALS
--
MEAN LATENCY
--
MEAN FEE
--
UTILITY / REQUEST
--
FITNESS EVALUATIONS
--

Not benchmarked yet

WHERE DOES A SWARM ACTUALLY WIN?

Algorithm stress tests

The load test above rewards pure exploitation on a smooth, low-dimensional landscape — exactly where a direct search wins. These two tests change the landscape's shape instead of just re-running the same one: one adds a reward you can only reach by passing through a worse patch first, the other scales up the number of routing choices.

TEST 1 · MULTI-MODAL LANDSCAPE

Volume-discount cliff

One provider offers a volume rebate once it carries at least 55% of traffic. Reaching that share means passing through allocations that are worse than staying put, so a local optimum sits between the start and the reward. 30 random provider setups, each scored against a fine grid-search optimum.

GREEDY (1% MOVES)

FOUND THE REBATE WHEN IT MATTERED
--
MEAN GAP TO OPTIMUM
--
MEAN FITNESS EVALUATIONS
--

PARTICLE SWARM

FOUND THE REBATE WHEN IT MATTERED
--
MEAN GAP TO OPTIMUM
--
MEAN FITNESS EVALUATIONS
--

Not run yet

TEST 2 · DIMENSIONAL SCALING

More routing choices

Same style of smooth landscape as the load test above, but with 3 to 25 synthetic providers instead of 3. Greedy re-checks every from/to provider pair on each move; the swarm's cost per iteration is set by its fixed particle count, not by the number of providers.

PROVIDERS GREEDY EVALS PSO EVALS GREEDY UTILITY PSO UTILITY GREEDY TIME PSO TIME
Not run yet

TEST 3 · NOISY EVALUATIONS

Sampled instead of exact fitness

Real traffic doesn't hand you a closed-form expected utility — it hands you the realized outcome of however many transactions you've actually routed so far. Here each fitness call only sees 15 simulated transactions instead of the exact expectation. Greedy commits to a move the moment one noisy comparison looks best and never revisits it; the swarm keeps re-sampling across many particles every iteration. 40 random provider setups, each scored against the true (noise-free) optimum.

GREEDY (1% MOVES)

MEAN GAP TO TRUE OPTIMUM
--
GAP STD. DEVIATION
--
WORST-CASE GAP
--

PARTICLE SWARM

MEAN GAP TO TRUE OPTIMUM
--
GAP STD. DEVIATION
--
WORST-CASE GAP
--

Not run yet

TRANSACTION TRACE

Recent activity

0 EVENTS
  1. No payment activity yet