> ## Documentation Index
> Fetch the complete documentation index at: https://docs.axottle.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Levels of inspections

> What checks each level of health — from network reachability to end-to-end route checks and external regional samples — and how manual checks differ from routine checks.

The health checks are organized by levels. Each layer answers an increasingly "deep" question: is the node available over the network, is the service running on it, is the protocol endpoint responding, is the entire traffic through the route, what it looks like from other regions. The levels are independent, but if the basic reachability fails, some of the top checks are marked as `skipped`.

> [!NOTE]
>
> For whom: Panel Administrator, SRE. Required rights: Administrator session.

> [!NOTE]
>
> It's described here. **behavior** The technical details of the samples (checkpoint addresses, handshake parameters, access confirmation method) are not published intentionally: their disclosure would help third parties to identify or fake service traffic.

## Levels.

**1. L1 – Network Achievability**

Basic layer: the node is available over the network. The panel allows the node domain name and checks the TCP availability of the control port and user input ports. **operates entirely on the** panel and does not require an agent, so it is applicable even to nodes not yet connected.

**2. L2 – Review of the Agent’s Service**

Checks that the server part of the VPN is actually running on the node, the data comes from the Axottle agent and is treated with the runtime-aware mode, requires an installed and online agent, implemented.

**3. L3 - Protocol Handshake (TLS)**

The endpoint of the protocol responds to a real TLS handshake with the expected server name. This is a strong signal "the server is really serving traffic, the configuration is intact" - no knowledge of client secrets. Implemented for TLS/REALITY inputs; where such verification does not apply, the status is `not_implemented`.

**4. L4 – Through route check**

End-to-end verification: gateway availability, conditional route targets achievable, and client passage through the input node to the control point. In addition, the panel confirms access to each input node. Implemented.

**5. L5 - External regional samples**

Achievability of the node through the eyes of external regional providers, as the node is seen "outside", from other networks and countries. It works through plug-in paid adapters and is turned on at will for individual input blocks. If no provider is connected, status is `external_probe_unavailable`. Implemented.

**6. L6 - Disassembly of Axosun**

This is not another network test. Axosun gets the results of previous levels, events and the available context of the object, and then generates a diagnostic report. The launch is available from the incident card with the **button to disassemble** and does not change configuration.

> [!WARNING]
>
> AI diagnostics work **only to read**: it does not execute commands on the nodes and does not change the configuration. The report does not allow changes by itself. Recovery is run separately: according to a confirmed plan or a pre-authorized policy [Flow Guard](https://docs.axottle.com/en/connection-units/recovery.md).

## Order and omissions

Levels L2 and L3 are skipped (with `skipped` status) if the basic attainability of L1 has already failed: with an unavailable node, it is pointless to check the service and shake hands. This is not a separate failure, but a reduction in unnecessary requests.

## Separate inspections

In addition to levels, the panel conducts several samples that answer questions that do not fit into the ladder “is the node alive?”

### Internet access

The country of the node in her card is the value announced: she came from Remnawave or she was entered by an operator. A separate sample looks at who the node is mistaken for by the Internet: through the tunnel of the scheme itself, the exit address is read, and two independent sources name its country. The result is visible in the node card, in the **Log in Internet**.

Between “node is alive” and “node is where it is written”, a whole class of breakdowns is placed, which the rest of the checks show in green: the provider reassigned the address from another pool, the geo-base left behind, users received the wrong streaming region.

The verdict describes a fact, not a breakdown, and does not affect anything in summary statuses:

| Verdict | Meaning | |
| - | - | - |
| Coincides with the declared country | All converges | |
| Internet sees another country | Divergence Showed by warning, not | crash |
| Exit via WARP | Address lawfully owned by Cloudflare, discrepancy explained by | |
| Sources named different countries | There is nothing to do, there is no charge. | |
| The node card does not specify the country | To compare with nothing; the observed country is still recorded | |

If all geo-hosts are unavailable, the past reading stays in place with its verification time: no blank print is written.

When the node has a [exit address,](https://docs.axottle.com/en/nodes/network.md#роли-вход-и-выход), the panel additionally checks the observed address with the assigned address, the role is assigned, and the traffic goes from the old address - this is exactly the discrepancy that no one has ever caught.

### WARP Local Port

In the past, "WARP's fault" was based on topology: there's a WARP block in the circuit, so it's it. Now, if the outgoing shoulder fails, the panel asks the node agent to go to the SOCKS local sidecar port and read the service response. The label "WARP's local port is closed" is only put if this check fails. The answer is - there's a common reason, the check didn't start - too. The result is saved by a separate break, so the evidence lies next to the dial.

### Place on the disk

The background sample monitors the free space on the panel server and starts an incident with a balance of less than 10% or less than 5%. The thresholds are the same as the Axottle Doctor, so that they do not argue about the same host. A full disk is not "slow", but a stop: PostgreSQL can not write checkpoint and goes into a restart cycle. See [Maintenance and Axottle Doctor](https://docs.axottle.com/en/operations/maintenance.md#место-на-диске).

## Manual vs. scheduled inspection

The same level can be triggered in two ways, and they behave differently, which is important for interpretation.

| | Manual check | Scheduled (on schedule) |
| - | - | - |
| Start | Button **Start checking** | Interval Planner |
| Through passage (dial) | Exactly. **single-handed** run-through | Do. **three** succession |
| Can expose `critical` from the node? | **Net** - a single failure does not escalate | **Da** - but only after a sustained failure (several attempts) |

> [!NOTE]
>
> The difference is that a manual check is needed to quickly look at the current state, and one random timeout does not have to translate the scheme to critical. A routine check escalates to critical from the node only after a sustained failure of the through sample - three attempts (immediately, then at about 1/5 interval, then after about 30 seconds).

## Post-application window

The single failure of the through dial in the first three minutes after the circuit is applied is not considered an accident. The reason is technical: the config goes to Remnawave and further to the nodes, xray rereads it and tears up live tunnels, including the tunnel of the sample itself, while the recording of the sample remains true, in its details, a note that the circuit has just been applied, the persistent failure of the scheduler escalates as usual.

## Green only on evidence.

The path through the branch of the circuit doesn't sample itself: it's assembled from the node state and the dial result. The path becomes green only when the dial is successful, otherwise it is shown as `unknown`. In the past, a circuit in which the service health user is not at all, showed all the paths in green. The report after application separately reports how many paths remained unproven.

## The default intervals

Scheduled checks are in intervals (default values are configured - see [Health Policies](https://docs.axottle.com/en/health/policies.md)):

* Achievability (L1) – every **60 seconds**
* Service (L2) - every **120 seconds**
* end-to-end circuit check (L4) – every **10 minutes**.

Minimum allowable values are limited from the bottom (15/30/60 seconds) so that checks do not turn into self-fired infrastructure.

## History of external audits

Bscheck results are available separately from internal samples in the Flow health tab. A confirmed fresh failure from an external network can impair the overall score even when the internal validation is successful. ISP error cannot be considered a failure of purpose: [point selection and interpretation of Bscheck](https://docs.axottle.com/en/health/bscheck.md).

## Verification

> [!TIP]
>
> The sample history of the scheme is displayed in the tab **Health** blockwise **Server timeline**Use tabs for the node. **Review** and **Journal**.

## Rollback

Running a test is a read-only operation (except for the confirmed recovery actions, which are described separately). There is nothing to roll back: the check does not change the configuration of the node, only records observations.

## Typical errors

> [accordion] L2 shows absent/unknown
>
> The service level requires an online agent. If the agent is not available or offline, the status is `absent`/`unknown`, not a failure. Set the agent up or wait for the heartbeat.

> [accordion] L5 is always external\_probe\_unavailable
>
> No external regional sample provider is connected, or L5 is not included for this block, as expected: L5 is an optional pay tier.

> [accordion] Where to open L6 parsing
>
> Open the incident on the page **Health** and press **Disassemble**The history of the reports is on the tab. **Axosun History**.

Related pages: [Conditions and severity](https://docs.axottle.com/en/health/states.md), [Dashboard Health](https://docs.axottle.com/en/health/dashboard.md), [Health Policies](https://docs.axottle.com/en/health/policies.md), [Axosun](https://docs.axottle.com/en/axosun/overview.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.