Gonka PoC and inference: what the monitor shows and what it does not
The public API calls weight the epoch reward weight, and cap_weight the trust weight for governance, BLS, and PoC validation. cap_weight is present when cap_weight_applied is true on epochs formed by v0.2.16 or later. The monitor does not show cap_weight, poc_start_block_height, or inference_url. A model name is not proof that service is up.
| Item | What the official docs say | On the monitor |
|---|---|---|
| models | Models that Host supports | Listed after a match; a dash if none |
| inference_url | Inference service endpoint | Not shown |
| weight | Reward weight for this epoch | Shown after a match |
| cap_weight | Trust weight for governance, BLS, and PoC validation; only sometimes present | Not shown |
| poc_start_block_height | PoC start height on the participant group | Not shown |
After the Host matches, the Gonka monitor can tell you that the address is on the current list, what the epoch reward weight is, and which models are on the record. It cannot tell you whether PoC passed, or whether an inference request is succeeding now.
Fields the docs tie to these two questions
- models is the list of supported models. After a match, the page prints those names.
- inference_url is the inference service endpoint. The page does not show it, so you cannot open or probe your inference port from here.
- weight is the reward weight for this epoch. It is not the credited amount. See weight and GNK.
- The API docs say cap_weight is the trust weight for governance, BLS, and PoC validation. It is present when the epoch was formed by v0.2.16 or later and cap_weight_applied is true. On earlier epochs, voting power follows weight. The monitor does not show cap_weight, so you cannot compare the two weights here.
- poc_start_block_height sits on the current participant group. It is the PoC start block height. The page does not show that height, and it is not a “PoC passed” flag.
What you check yourself
- Model names only mean the list records those models. Confirm whether requests are served on the ML node and the official service.
- For join requirements, read the quickstart: a Host deploys a network node (chain node and API node) and at least one ML node. That is setup guidance, not a switch on this page.
- Whether the address itself is the right one is in which identifier to paste. An inference URL does not produce a models line.
What to do next
- If you only need to know whether it is in the epoch: use In this epoch, the epoch, and weight. See this epoch.
- If the question is PoC trust weight or the inference port: open the current-epoch response in the API docs, see whether cap_weight is present, and check the inference service on your own node. Do not treat the page weight as a PoC result.
- If the list fails to load, both PoC and inference stay unknown. Do not record them as failed or as zero.
Illustration
Illustration, not a live test: the document sample has one model name, an inference_url, and weight 10. That sample does not present cap_weight as a passed PoC. If your page shows only weight and model names, read only those two.
Limits
- Do not infer a PoC score from weight, or a live request count from how many model names appear.
- There is no promise of latency, success rate, or GNK. The site says developers pay GNK for inference through an OpenAI-compatible API, and providers host models and earn GNK. The request result is not in the monitor.
- This page does not operate containers. AI crash analysis is free and applies to IOTA Train at Home on a Mac. It cannot start or stop a Gonka node.
Common questions
If the model names appear, is the inference service up?
No. The names come from models on the current list. They mean those models are listed on the record. Whether a request was served successfully depends on your ML node and the official service, and it is not on this page.
Is the weight on the page a PoC pass?
No. The docs call weight the reward weight and cap_weight the trust weight used for PoC validation and related roles. The monitor shows only the first. Without cap_weight on the page, you cannot judge PoC trust weight here.
The quickstart requires at least one ML node. Does the monitor show that count?
No. The quickstart says you need at least one ML node to join, and it describes the network node and the ML node on one machine. After a match, the monitor shows the epoch, weight, and model names. It does not show an ML node count, and it does not check whether your containers are running.