Skip to documentationOperations & observability
The read-only dashboard, service boundaries, monitoring and evidence used to investigate an incident.
On this page
Runtime boundaries
The app, read API and private operations dashboard are separate hosted applications sharing the deployment identity. Keeper, attester, prover and checks processes run as bounded services. The current release uses primary and standby AWS regions; an additional region is deferred.
Service signing uses the assigned KMS integration, while the randomness key remains enclave-protected. RPC access, indexing, authentication, publications and price sources are separate dependencies. An available homepage does not independently establish that each downstream service is healthy.
The private dashboard observes
The team dashboard shows launches, trading/fee activity, held assets, token/lens state, backing flags, current roles, prepared operations, routes, reward rounds, payments and credits. It uses the manifest, chain and indexer data and links to service-health tooling.
The dashboard is read-only: it holds no signing keys and submits no governance or financial transactions. Safe execution and keeper jobs occur through their own execution paths. Observing a prepared operation in the dashboard does not mean it has executed.
Signals and their interpretation
| Signal | Evidence to inspect |
|---|---|
| Coverage / floor change | Protected executable capacity and actual holder obligations, not nominal total asset value. |
| Unreconciled buckets or claims | Internal obligation categories and corresponding PoolManager/settlement state. |
| Stalled work | Due-job state, signer funding, simulation, submission, receipts and service progress. |
| Indexer lag | Head, processed block and finalized boundary, plus replay/restart behavior. |
| Issuer or yield-venue restrictions | Verified source events and affected markets/positions; do not guess event semantics from a token name. |
| HTTP errors / rate limits | Actual status, path, timeout and timing aggregates. Unknown/incomplete log status is not a confirmed HTTP failure. |
Investigate before changing behavior
A parent-block quote can be useful context but may be stale after intervening trades. Compare the executed path, applicable minimum, actual input/output, taxes and intervening state before attributing a conversion shortfall. Monitoring itself does not introduce a new price gate or confiscation power.
Application recovery can use an existing deployment rollback; service recovery can restart or restore the applicable provider path. On-chain changes still require the existing authority. Preserve old claims, credits and investment positions through recovery, and verify the original user’s outcome rather than only a new fixture.
Logs and metrics exclude keys, authentication tokens and credentialed endpoints. Availability, OAuth/profile behavior, chain correctness and completed end-to-end execution are reported as separate observations. Hosting spend notifications do not automatically pause the public app.