The Application's run outcomes, latency, tool errors and quality loop over an optional window.
Runs, deliveries, latency, tool calls and judge verdicts are the Application's own production runs, selected on the run's Application: cost-only bookkeeping (the relevance pre-gate, judges, reviews) is not a run, and pre-flight rejections are counted apart from the runs that started. Latency is over completed runs, the newest 20,000 when there are more (`sampled`). Tools are the eight most called plus any other with at least five errors at 10% or more. The review loop — reviews, findings, improvement cards awaiting a decision, the last finished evaluation run — is counted over the Application's runnable member agents. `from`/`to` are Unix ms (half-open), defaulting to the last 30 days; all time is `from=0`.
Authorization
Bearer JWT Authorization header using the Bearer scheme. Enter 'Bearer' [space] and then your token.
In: header
Path Parameters
The Application (Network) id.
uuidThe tenant identifier
Query Parameters
Window start, Unix ms inclusive. Defaults to 30 days before to; 0 for all time.
int64Window end, Unix ms exclusive. Defaults to now.
int64Response Body
curl -X GET "https://api.fruxon.com/v1/tenants/string/networks/497f6eca-6276-4993-bfeb-53cbbbba6f08/reliability"{
"networkId": "00000000-0000-0000-0000-000000000000",
"window": {
"fromMs": 0,
"toMs": 0
},
"countedFromMs": 0,
"runs": {
"started": 0,
"completed": 0,
"failed": 0,
"cancelled": 0,
"cancelledSuperseded": 0,
"inFlight": 0,
"rejected": 0
},
"delivery": {
"failed": 0,
"pending": 0
},
"latency": {
"p50Ms": 0,
"p95Ms": 0,
"timedRuns": 0,
"sampled": false
},
"tools": [
{
"toolIntegrationId": "string",
"toolId": "string",
"toolType": "API",
"displayName": "string",
"invocations": 0,
"errors": 0,
"cost": 0,
"totalDurationMs": 0,
"toolKey": "string.string",
"errorRate": null,
"avgDurationMs": null
}
],
"review": {
"reviewedAgents": 0,
"unitsReviewed": 0,
"findings": 0,
"recommendationsAwaiting": 0,
"recommendationsAwaitingByAgent": [
{
"agentId": "string",
"displayName": "string",
"count": 0
}
]
},
"judge": {
"passed": 0,
"failed": 0,
"scored": 0,
"errors": 0,
"passRate": 0
},
"lastEvaluationRun": {
"runId": "00000000-0000-0000-0000-000000000000",
"agentId": "string",
"agentDisplayName": "string",
"status": "PENDING",
"score": 0,
"completedAt": 0
}
}{
"type": "string",
"title": "string",
"status": 0,
"detail": "string",
"instance": "string",
"property1": null,
"property2": null
}{
"type": "string",
"title": "string",
"status": 0,
"detail": "string",
"instance": "string",
"property1": null,
"property2": null
}Creates or replaces the Application's automatic-enrollment policy.
Full replace. The fallback must be `REQUEST_ACCESS` or `REJECT`, the environment rule must be specified, and the hourly cap at least 1 — anything else is a 400. Applies immediately: senders admitted under `PROVISION` are matched against this policy from the next message on. Save this before flipping an admission select to `PROVISION`; the Network PATCH refuses the enum with no policy in place.
Projects the Network's roster roles with holders and referencing agent contacts.
Inverts the People surface: instead of listing participants and the roles each one carries, this lists every roster role in the Network and, for each, who holds it and which agent contacts defer to it. A holder is reported with every tier that grants the role — `EXPERTISE` for a tenant-level expertise tag that applies everywhere, `EDGE_ROLE` for a tag on a member edge in this Network — and an edge-granted holder names the agents whose edge grants it, since that is the only place the tag can be removed. `consultable` reports whether a consult-allowed member edge exists: a holder without one is still reachable for delivery through a role-bound contact but is refused for `consult_for_role`. A role with references and no holders is the actionable gap this surface exists to expose — deliveries and consults to it refuse at runtime. Roles group case-insensitively and are reported using the tenant capability catalog's casing; `inCatalog` distinguishes a catalog role from one that exists only as an ad-hoc tag. Read-only and fully derived from existing expertise, member-edge, and contact-slot state — it adds no storage of its own, so nothing here can drift from the People surface. The whole role vocabulary is projected in memory before paging — it is tens of names, not thousands (participants are paged on the People surface, not here) — so a page is a slice of the complete, fully ordered result and omitting `pageSize` returns everything.