Four failures that get merged into one
| Symptom | What it establishes | What it does not | Next step |
|---|---|---|---|
| The public source returns 404 or the file is gone | This fetch could not retrieve a configuration | That every server in your stored profile is offline | Keep the URL, status, time and file hash; stop calling the old file “today's update” |
| The source returns 200 but the body is empty or the wrong format | Fetch succeeded; content or parse failed | Anything about node reachability | Check content type and the client's import result, without exposing the token |
| YAML parses, but a real HTTPS request times out | Format is fine; this request failed | Permanent death, or a packet-loss figure | Repeat on two different public exits and record the failure reason |
| Nine of ten requests succeed and a fixed download completes | At that moment, requests and transfer met the bar | Long-term stability, client compatibility, streaming capability | Check exit region against at least two IP sources and keep masked evidence |
Record the stage explicitly: source fetch → format parse → core config validation → HTTPS proxied request and fixed download → exit region. When an earlier stage fails, fix that one; do not keep adjusting TUN against a subscription that is no longer being served.
One failure in ten requests is a “10% request failure rate”. It is not “10% packet loss”. An HTTP or TLS failure is not the same measurement as a lost packet.
Step 1: is the source still updating
- Return to the original publishing page and check the update time, maintenance notes and format description. A mirror page claiming “working today” is not the source and does not prove your network can reach it.
- A 200 from the source only means the entry is reachable — not that any node carries traffic.
- Keep the retrieval time, the URL, the HTTP status and the file hash, so a later comparison is possible.
Step 2: fetch versus parse
- Keep the error text and mask the token.
- 401/403: check authorisation first. 404: check whether the address changed. 429: stop the rapid retries. Timeout: verify plain networking first.
- If the link opens, confirm the body is not a login page, a captcha or an HTML error page.
- Confirm the format matches the client: Clash YAML, plain URIs and sing-box JSON are not interchangeable.
- After an update, check node count and which profile is active — a downloaded profile that was never activated looks identical to a failed one.
Step 3: real connectivity versus a latency number
Client latency displays are produced by a specific probe, not by HTTPS requests through the proxy. Record at least: time, network type, number of real HTTPS proxied requests, failure reason, and the result of a fixed small download.
| Observation | Reasonable conclusion |
|---|---|
| Ten real requests, all failed | Not connected in this test — discard this node for now |
| No loss reported, but HTTPS latency very high | Connects, but page experience may be poor; observe only |
| Low RTT, pages still fail | Possibly only an entry or CDN responding; check TLS, DNS, route and the target |
| Results differ greatly between carriers | Line difference at this moment; not a general stability claim |
Limited comparison: one variable at a time
- Same subscription, a different node.
- Same node, Wi-Fi versus mobile data.
- Same network, an ordinary HTTPS page versus an exit-IP page.
- Another compatible client, only to test whether the config parses — never with two VPNs active.
Two or three comparisons are enough to establish a direction. After everything times out, do not refresh indefinitely or rotate accounts to bypass a rate limit.
Low latency but pages fail: check DNS and capture
Try a known IP and an ordinary domain. If the IP works and the domain fails, DNS is the first suspect. If both fail, check the node, the route or the capture. On desktop clients confirm the system proxy or TUN is actually active; on mobile confirm the VPN indicator appears.
Stop conditions
- The source asks you to install an unknown root certificate, run unfamiliar scripts, or grant unrelated administrator rights.
- The subscription requires uploading to a stranger's conversion site, or asking for help by publishing the full token.
- The node list changes constantly and no configuration survives a single day — the source is not providing a usable service, whatever its page claims.
When any of those hold, stop and restore direct connectivity rather than continuing to test.
What to record
- Source page, retrieval time, HTTP status, file hash.
- Client and version, format used, import result.
- Request count, failures and reasons, fixed download result.
- Two exit-IP observations, if region matters to you.
Mask tokens, node addresses and account identifiers in everything you keep or share.