Route Readiness
How B3IQ explains why a machine or model is eligible or blocked for routed inference.
Route readiness explains a machine's connection, runtime, and model blockers. For owner inference, first list models with your application's gateway key using the OpenAI SDK quickstart. The retained diagnostic API below helps investigate machine state; it is not a test of that key's access.
Machine diagnosis
For your machine's connection, runtime, and model state, sign in to the machine dashboard and open its Models or Health tab. A separate public-safe readiness endpoint exposes the routing diagnosis:
textGET /v1/public/nodes/:node_id/route-readiness?model=<model>
This retained lookup does not receive your gateway key. Public explorer
pages are retired; use your signed-in machine dashboard first. It includes policy
diagnostics for outside work, including the owner's participation election,
and can return a separate promotion_blockers list. Those findings are not
a substitute for authenticated discovery on your own account.
Model readiness
explains the promotion gates that owner inference skips. Machine health,
installed-model matching, ownership, key restrictions, and incident controls
still apply.
The response can include route blockers, model blockers, next actions, and
operator prompts. For consumer_edge profiles, model_profiles[].consumer_edge
reports model_class, a conservative concurrency_cap, benchmark age and
freshness, the uptime gates (online, tunnel ready, service persistent), recent
completions and errors with an error_rate_pct, the blockers, the next actions,
and an earning_status of earning_ready, paused, or blocked. It must not
include tunnel IDs, local URLs, runtime IDs, prompts, outputs, request IDs, API
keys, customer IDs, raw tokens, route secrets, or private tunnel details.
Gate order
Blockers and actions
| Blocker | Meaning | Operator action |
|---|---|---|
| Enrollment | Node is not fully enrolled | Complete managed enrollment |
| Heartbeat | Heartbeat is stale, missing, or offline | Bring host agent online and sync a fresh heartbeat |
| Runtime | Inference is not installed, or its engine is not ready or reachable | Install inference from the machine's page in your dashboard, add an engine on its Models tab, or restart a stopped engine from Node control |
| Service persistence | The host agent is not installed as a persistent service, or does not report its persistence state | Install the host agent as a systemd service (or enable systemd linger) and let it publish a fresh heartbeat |
| Public route | The tunnel or the public inference route is missing, unhealthy, or not advertised to the router | Rotate or repair the broker-managed tunnel from the host menu, then re-publish the route in heartbeat metadata |
| Claim | Node is not linked to an owner account | Redeem the claim code from the node's setup screen in your dashboard |
| Route tier | A promotion policy rejects the tier, discovery filters it, or an incident control pauses the class | Check which policy reports the blocker. Owner chat skips economic tier gates; discovery tier filters are not an owner-chat access boundary |
| Benchmark proof | Observed benchmark evidence is missing or stale under a policy that requires it | Missing benchmark evidence alone does not block owner use. Inspect current discovery and runtime health before taking action |
| Queue health | The runtime reports a blocked or unhealthy queue; the profile gate may be skipped for owner use | Inspect runtime health and actual request results; address overload or queue failures |
| Customer policy | The key's machine/model scope rejects the request, discovery filters the model, or the profile does not support its modality, task, or capabilities | Check the intended key scope and request. Change policy only if the application should have that access |
A machine switched to Earn mode is excluded from every b3iq routing query at
the SQL loader. It never appears as a candidate, so route readiness reports it
as route_missing until you switch it back to Bare metal from the machine's
dashboard page.
The short answer
The operator-facing answer to "why is nothing routing to my machine?" should be direct:
Owner inference needs an enrolled, online, claimed machine with a healthy runtime and route, an installed matching model, and account/key policy that permits the request. Marketplace promotion flags and benchmark evidence are not prerequisites for owner use. Verify discovery and a small request with the same key your application will use.
A public readiness result does not establish earnings or access for a particular key. Owner inference has no per-request B3IQ charge. See Claiming and earnings for what a machine does earn from.
Stale benchmark prompts
benchmark.smoke is not an operator-runnable command. Owner inference does not
require a benchmark. If a separate promotion policy reports stale observed
benchmark evidence, keep the node online
with a healthy runtime after any runtime, driver, model artifact,
quantization, context, or queue-policy change. B3IQ re-benchmarks eligible
profiles that lack fresh observed evidence on its own schedule.