Admin Guide
The operator command center: observer endpoints, settlement controls, and demand radar.
The admin command center is the operator-facing NOC for the platform. It is not customer-facing — access requires the operator token — but its design principles are public because they explain the numbers you see everywhere else.
Data-truth policy
Every metric in the command center traces to a real SQL query against the production database or an on-chain read:
- Traffic counters come from Redis counters that reset at UTC midnight, with
a
last_seen-filtered database fallback — never from stale accumulators. - Revenue is split external vs internal: usage metered against the platform's own keys is labeled internal and never presented as customer revenue.
- The network health score is computed — node availability minus penalties for unfunded signer, blocked settlement batches, and dry-run mode — with the formula displayed. It cannot read 100% while a critical alert is active.
- Where no backend exists for a panel, the panel shows an honest empty state instead of a placeholder number.
Surfaces
| View | What it shows |
|---|---|
| Executive Overview | Revenue (external/internal), live traffic, health score, alert stack with unblock conditions |
| Demand Radar | Classified traffic (developer / machine / scanner), conversion funnel |
| Customer Ops | Paying wallets and the active free-tier conversion pipeline |
| Treasury | Settlement mode, signer state, batch accounting, on-chain contract references |
| Network / Nodes | Provider health, registered node performance |
| Observability | Database/Redis probes, API latency percentiles |
Settlement controls
Settlement state is visible on every admin view (a persistent banner while in dry-run). Enabling live settlement is deliberately manual and multi-gated: funded signer, real external revenue above the safety threshold, and an explicit human confirmation. It can never be flipped autonomously.