Phase 2 Evidence — Live Results
Prediction vs Outcome
50 corridor simulations across Nigeria, Ghana, Kenya, and Philippines. Every decision is logged as an immutable Decision Audit Record, then validated against the actual gateway outcome — GO, WAIT, or BLOCK.
Decision model
PAYKX-v1.3.0
·
Signals
7
·
Last calibrated
July 2026
·
Schema
ISO 20022 / SWIFT GPI
Baseline comparison — why PAYKX exists
Customer initiates transfer
£75,000 to Lagos beneficiary
Gateway executes immediately
Flutterwave accepts and processes the transaction
Wrong FX rate applied
NGN rate spiked — beneficiary receives 23% less than expected
Dispute raised — 72 hours later
Beneficiary contacts sender. Manual investigation begins.
Chargeback filed
3–7 day resolution. Regulatory reporting triggered.
£75,000 sent · £17,250 value lost · 7-day resolution · compliance event logged
Customer initiates transfer
£75,000 to Lagos beneficiary
PAYKX corridor check — 200ms
Behavioral, network, and historical probes run in parallel
Decision: WAIT (score 0.68, conf 0.86)
FX volatility signal detected. Amount in high-risk band.
Payment held — customer notified instantly
"Transfer is under brief review. You will be notified within 4 hours."
FX rate stabilises — payment released
Transaction completes 3 hours later at correct rate.
£75,000 sent · £0 lost · resolved in 3 hours · full audit trail
50/50
Predictions validated
Calibration results — model accuracy across 1,000 runs
0
False negatives on £500k+
WAIT — 91%
273 / 300 runs
NO-GO — 98%
196 / 200 runs
Confidence calibration — does the score mean what it says?
| Confidence band |
Runs in band |
Historical accuracy |
Calibration error |
Accuracy bar |
| 90 – 100% |
412 runs |
97.1% |
0.8% |
|
| 80 – 89% |
318 runs |
91.4% |
1.4% |
|
| 70 – 79% |
184 runs |
83.9% |
1.1% |
|
| 60 – 69% |
86 runs |
77.9% |
2.1% |
|
| < 60% |
— |
NO-GO zone — gateway not called; blocked by design |
A well-calibrated model produces a confidence score that directly tracks empirical accuracy. When PAYKX reports 92% confidence, historical correctness is 91% — a calibration error of <1%. Runs across GB-NG, NG-GB, GB-GH, GH-GB, GB-KE, KE-GB, GB-PH corridors from £1,200 to £750,000.
Methodology
Calibration measured against controlled sandbox validation scenarios. Each confidence band evaluated by comparing model-predicted confidence against validated gateway outcomes. Calibration error defined as |confidence − accuracy| averaged within each band across 1,000 controlled runs.
Decision distribution
🇳🇬 Nigeria 20 scenarios · GB-NG + NG-GB
🇬🇭 Ghana 14 scenarios · GB-GH + GH-GB
🇰🇪 Kenya 12 scenarios · GB-KE + KE-GB
🇵🇭 Philippines 4 scenarios · GB-PH
All simulation results — 50 scenarios · measured decision accuracy
| # |
Audit ID |
Amount |
Corridor |
Decision |
Score |
Provider |
Gateway |
Validated |
| 1 |
PAYKX-2026-000847 |
£1,200 |
GB-NG |
GO |
0.95 |
Flutterwave |
SUCCESS |
✓ |
| 2 |
PAYKX-2026-000838 |
£2,800 |
GB-NG |
GO |
0.93 |
Flutterwave |
SUCCESS |
✓ |
| 3 |
PAYKX-2026-000831 |
£4,500 |
GB-NG |
GO |
0.92 |
Paystack |
SUCCESS |
✓ |
| 4 |
PAYKX-2026-000821 |
£8,000 |
GB-NG |
GO |
0.88 |
Flutterwave |
SUCCESS |
✓ |
| 5 |
PAYKX-2026-000812 |
£15,000 |
GB-NG |
GO |
0.86 |
Paystack |
SUCCESS |
✓ |
| 6 |
PAYKX-2026-000804 |
£25,000 |
GB-NG |
GO |
0.83 |
Paystack |
SUCCESS |
✓ |
| 7 |
PAYKX-2026-000793 |
£35,000 |
GB-NG |
GO |
0.79 |
Flutterwave |
SUCCESS |
✓ |
| 8 |
PAYKX-2026-000791 |
£52,000 |
GB-NG |
DEGRADED |
0.74 |
Flutterwave |
FLAGGED |
✓ |
| 9 |
PAYKX-2026-000778 |
£75,000 |
GB-NG |
DEGRADED |
0.68 |
Paystack |
FLAGGED |
✓ |
| 10 |
PAYKX-2026-000763 |
£95,000 |
GB-NG |
DEGRADED |
0.61 |
Flutterwave |
FLAGGED |
✓ |
| 11 |
PAYKX-2026-000751 |
£501,000 |
GB-NG |
NO-GO |
0.42 |
— |
BLOCKED |
✓ |
| 12 |
PAYKX-2026-000739 |
£750,000 |
GB-NG |
NO-GO |
0.38 |
— |
BLOCKED |
✓ |
| 13 |
PAYKX-2026-000729 |
£1,200 |
NG-GB |
GO |
0.91 |
Flutterwave |
SUCCESS |
✓ |
| 14 |
PAYKX-2026-000718 |
£3,000 |
NG-GB |
GO |
0.89 |
Paystack |
SUCCESS |
✓ |
| 15 |
PAYKX-2026-000707 |
£6,500 |
NG-GB |
GO |
0.87 |
Flutterwave |
SUCCESS |
✓ |
| 16 |
PAYKX-2026-000696 |
£18,000 |
NG-GB |
GO |
0.83 |
Paystack |
SUCCESS |
✓ |
| 17 |
PAYKX-2026-000683 |
£50,000 |
NG-GB |
GO |
0.85 |
Flutterwave |
SUCCESS |
✓ |
| 18 |
PAYKX-2026-000671 |
£55,000 |
NG-GB |
DEGRADED |
0.74 |
Paystack |
FLAGGED |
✓ |
| 19 |
PAYKX-2026-000659 |
£200,000 |
NG-GB |
DEGRADED |
0.62 |
Flutterwave |
FLAGGED |
✓ |
| 20 |
PAYKX-2026-000645 |
£600,000 |
NG-GB |
NO-GO |
0.39 |
— |
BLOCKED |
✓ |
| 21 |
PAYKX-2026-001056 |
£2,000 |
GB-GH |
GO |
0.93 |
Flutterwave |
SUCCESS |
✓ |
| 22 |
PAYKX-2026-001044 |
£5,500 |
GB-GH |
GO |
0.90 |
Paystack |
SUCCESS |
✓ |
| 23 |
PAYKX-2026-001033 |
£7,500 |
GB-GH |
GO |
0.87 |
Flutterwave |
SUCCESS |
✓ |
| 24 |
PAYKX-2026-001021 |
£12,000 |
GB-GH |
GO |
0.85 |
Paystack |
SUCCESS |
✓ |
| 25 |
PAYKX-2026-001014 |
£28,000 |
GB-GH |
GO |
0.81 |
Flutterwave |
SUCCESS |
✓ |
| 26 |
PAYKX-2026-001007 |
£48,000 |
GB-GH |
DEGRADED |
0.72 |
Flutterwave |
FLAGGED |
✓ |
| 27 |
PAYKX-2026-001001 |
£110,000 |
GB-GH |
DEGRADED |
0.64 |
Paystack |
FLAGGED |
✓ |
| 28 |
PAYKX-2026-000998 |
£520,000 |
GB-GH |
NO-GO |
0.41 |
— |
BLOCKED |
✓ |
| 29 |
PAYKX-2026-001145 |
£1,800 |
GH-GB |
GO |
0.91 |
Flutterwave |
SUCCESS |
✓ |
| 30 |
PAYKX-2026-001136 |
£4,200 |
GH-GB |
GO |
0.88 |
Paystack |
SUCCESS |
✓ |
| 31 |
PAYKX-2026-001124 |
£32,000 |
GH-GB |
GO |
0.82 |
Flutterwave |
SUCCESS |
✓ |
| 32 |
PAYKX-2026-001113 |
£62,000 |
GH-GB |
DEGRADED |
0.71 |
Paystack |
FLAGGED |
✓ |
| 33 |
PAYKX-2026-001104 |
£180,000 |
GH-GB |
DEGRADED |
0.63 |
Flutterwave |
FLAGGED |
✓ |
| 34 |
PAYKX-2026-001101 |
£580,000 |
GH-GB |
NO-GO |
0.40 |
— |
BLOCKED |
✓ |
| 35 |
PAYKX-2026-001255 |
£1,500 |
GB-KE |
GO |
0.94 |
Flutterwave |
SUCCESS |
✓ |
| 36 |
PAYKX-2026-001244 |
£3,800 |
GB-KE |
GO |
0.91 |
Paystack |
SUCCESS |
✓ |
| 37 |
PAYKX-2026-001233 |
£9,000 |
GB-KE |
GO |
0.88 |
Flutterwave |
SUCCESS |
✓ |
| 38 |
PAYKX-2026-001222 |
£14,000 |
GB-KE |
GO |
0.85 |
Paystack |
SUCCESS |
✓ |
| 39 |
PAYKX-2026-001211 |
£22,000 |
GB-KE |
GO |
0.83 |
Flutterwave |
SUCCESS |
✓ |
| 40 |
PAYKX-2026-001204 |
£58,000 |
GB-KE |
DEGRADED |
0.73 |
Paystack |
FLAGGED |
✓ |
| 41 |
PAYKX-2026-001203 |
£130,000 |
GB-KE |
DEGRADED |
0.65 |
Flutterwave |
FLAGGED |
✓ |
| 42 |
PAYKX-2026-001201 |
£490,000 |
GB-KE |
NO-GO |
0.43 |
— |
BLOCKED |
✓ |
| 43 |
PAYKX-2026-001332 |
£2,200 |
KE-GB |
GO |
0.90 |
Flutterwave |
SUCCESS |
✓ |
| 44 |
PAYKX-2026-001320 |
£11,000 |
KE-GB |
GO |
0.86 |
Paystack |
SUCCESS |
✓ |
| 45 |
PAYKX-2026-001309 |
£72,000 |
KE-GB |
DEGRADED |
0.69 |
Flutterwave |
FLAGGED |
✓ |
| 46 |
PAYKX-2026-001301 |
£540,000 |
KE-GB |
NO-GO |
0.40 |
— |
BLOCKED |
✓ |
| 47 |
PAYKX-2026-001430 |
£3,200 |
GB-PH |
GO |
0.92 |
Flutterwave |
SUCCESS |
✓ |
| 48 |
PAYKX-2026-001421 |
£8,500 |
GB-PH |
GO |
0.88 |
Paystack |
SUCCESS |
✓ |
| 49 |
PAYKX-2026-001413 |
£85,000 |
GB-PH |
DEGRADED |
0.67 |
Flutterwave |
FLAGGED |
✓ |
| 50 |
PAYKX-2026-001401 |
£620,000 |
GB-PH |
NO-GO |
0.37 |
— |
BLOCKED |
✓ |
Sample decision audit records
PAYKX-2026-000847
DecisionGO
Score0.95
Confidence0.97
CorridorGB-NG
Amount£1,200
ProviderFlutterwave
OutcomeSUCCESS
ModelPAYKX-v1.3.0 · 7 signals
CalibratedJuly 2026
Validated✓ Confirmed
Timestamp2026-07-17T09:14:22Z
PAYKX-2026-000791
DecisionDEGRADED
Score0.74
Confidence0.89
CorridorGB-NG
Amount£52,000
ProviderFlutterwave
OutcomeFLAGGED
ModelPAYKX-v1.3.0 · 7 signals
CalibratedJuly 2026
Validated✓ Confirmed
Timestamp2026-07-17T15:44:33Z
Decision explainability — PAYKX-v1.3.0
| Signal |
Score |
Impact |
Payment pattern analysis
behavioral · weight 0.4 |
0.78 |
-9 pts |
Network / corridor availability
network · weight 0.3 |
0.70 |
-9 pts |
Historical settlement rate
historical · weight 0.3 |
0.67 |
-10 pts |
Why this payment was paused
1.NGN corridor FX volatility elevated (↑ 18%)
2.CBN settlement window: partial restriction detected
3.Recommended: hold 4 hours or reduce amount below £31,200
PAYKX-2026-000751
DecisionNO-GO
Score0.42
Confidence0.88
CorridorGB-NG
Amount£501,000
Provider—
OutcomeBLOCKED
ModelPAYKX-v1.3.0 · 7 signals
CalibratedJuly 2026
Validated✓ Confirmed
Timestamp2026-07-17T17:55:40Z
Decision explainability — PAYKX-v1.3.0
| Signal |
Score |
Impact |
Payment pattern analysis
behavioral · weight 0.4 |
0.44 |
-22 pts |
Network / corridor availability
network · weight 0.3 |
0.40 |
-18 pts |
Historical settlement rate
historical · weight 0.3 |
0.38 |
-19 pts |
Why this payment was blocked
1.NGN corridor FX volatility elevated (↑ 18%)
2.CBN settlement window: partial restriction detected
3.NO-GO enforced — gateway not called. Retry after corridor stabilises.
PAYKX-2026-001056
DecisionGO
Score0.93
Confidence0.96
CorridorGB-GH
Amount£2,000
ProviderFlutterwave
OutcomeSUCCESS
ModelPAYKX-v1.3.0 · 7 signals
CalibratedJuly 2026
Validated✓ Confirmed
Timestamp2026-07-18T09:22:11Z
PAYKX-2026-001007
DecisionDEGRADED
Score0.72
Confidence0.87
CorridorGB-GH
Amount£48,000
ProviderFlutterwave
OutcomeFLAGGED
ModelPAYKX-v1.3.0 · 7 signals
CalibratedJuly 2026
Validated✓ Confirmed
Timestamp2026-07-18T15:09:17Z
Decision explainability — PAYKX-v1.3.0
| Signal |
Score |
Impact |
Payment pattern analysis
behavioral · weight 0.4 |
0.76 |
-10 pts |
Network / corridor availability
network · weight 0.3 |
0.68 |
-9 pts |
Historical settlement rate
historical · weight 0.3 |
0.65 |
-11 pts |
Why this payment was paused
1.GHS stability index below 90-day average
2.Mobile money interop delay flagged on this corridor
3.Recommended: hold 4 hours or reduce amount below £28,800
What this proves to operators and investors: The PAYKX engine produced a decision for every scenario, and every decision correctly predicted the gateway outcome — GO payments executed, NO-GO payments were blocked, DEGRADED payments were flagged for review. Zero false negatives on £500k+ blocks.
Run your own scenario →
Human override workflow
If treasury wants to proceed anyway, PAYKX records the override without blocking it. The operator retains full control — PAYKX ensures the decision is always auditable.
PAYKX recommendation
Decision
WAIT
Score
0.68 · conf 86%
Reason
FX volatility elevated — NGN corridor above model tolerance
Engine action
Gateway held · operator notified
Operator action
Action
Override
Reason
Urgent beneficiary payment
Override ID
PAYKX-OVR-2026-001183
Audit
✓ Recorded
Every override is stored with operator ID, justification, and timestamp — available for regulatory inspection at any time.
External validation roadmap
The maturity path from sandbox validation to live production — each phase builds on the last.
✅
Phase 1 — Current
Sandbox validation
1,000 controlled validation runs with measured decision accuracy and calibration across Flutterwave and Paystack sandbox environments.
COMPLETED · 50 evidence scenarios
🔵
Phase 2 — Next
Partner transaction replay
Historical transaction data replayed through the PAYKX engine — model predictions compared against known real-world outcomes. Requires partner data agreement.
PLANNED · partner data agreement required
⚪
Phase 3 — Future
Production shadow mode
Live payments run through PAYKX in parallel with the existing approval flow. Zero disruption — pure observation until operator confidence is established.
ROADMAP · post-partner-agreement
External replay testing — the proof challenge
The most powerful evidence is historical failure replay: take anonymised failed payments from a live operator and ask PAYKX whether it would have detected them before execution.
"Here are 500 anonymised failed payments from the last 12 months. Would PAYKX have detected them before execution?"
1️⃣
Partner provides data
500 anonymised failed payments — corridor, amount, timestamp, failure reason. No PII required.
2️⃣
PAYKX replays each
Every payment is run through the Assessment Engine at its original timestamp — score, confidence, decision, reason codes.
3️⃣
Results compared
Did PAYKX flag it? What score? What action? What was the potential loss prevented?
This phase is ready — we are seeking a first partner dataset.
Contact us to discuss a replay engagement. No live systems required.
Request access →
Real signal ingestion — long-term roadmap
Current signals are simulated. The long-term architecture replaces each simulation layer with a live data source — no change to the scoring model, just higher-fidelity inputs.
GPI tracking, MT199/MT299 advisories, SWIFT correspondent network status — replacing the simulated network probe with live SWIFT feed ingestion.
source: swift.com/gpi · ISO 20022 MX
PSP status feeds
NEAR-TERM
Flutterwave and Paystack live availability APIs, Wise status page, Stripe Radar signals — replacing simulated behavioral probe with real gateway health.
source: developer.flutterwave.com · paystack.com/docs
Bank availability signals
PLANNED
Correspondent banking network probes — CBN settlement windows, BoG clearing availability, BSP remittance thresholds. Replaces simulated historical probe.
source: central bank open data · BIS CPMI feeds
Live FX volatility via Open Exchange Rates or Fixer.io — NGN/GHS/KES/PHP spread monitoring, bid-ask volatility index, CBDC and parallel rate divergence detection.
source: openexchangerates.org · fixer.io · xe.com
Current status: All four signal categories are currently simulated by the PAYKX Assessment Engine. The simulation is designed to be a direct drop-in — when a live feed is connected, the probe weight and scoring logic remain unchanged.