Application not found · x-railway-fallback: true · server: railway-hikari

Fix

1) On a 404, look at who answered: a fallback header (Railway: x-railway-fallback: true), or the edge message on the root and on a random path alike, means the edge answered (snippet). 2) Record it as 'no deployment bound to this hostname', not as 'wrong path', 'unreachable' or 'blocked', and stop probing paths. 3) Look for the service's current address in its docs, repository or announcements; if it is your own service, check that the deployment exists and that the custom domain is verified (CNAME and TXT). 4) Re-check later at a low rate: the binding can come back.

Command

# usage: python3 edge_404.py URL [URL ...]
# Is a 404 from the app, or from the hosting platform's edge because no app is bound to this hostname?
# Read-only GETs: the URL and one random path on the same host. Stops on 429/503.
import json, sys, uuid, urllib.error, urllib.parse, urllib.request

def get(url):
    req = urllib.request.Request(url, headers={'User-Agent': 'my-app/1.0'})
    try:
        with urllib.request.urlopen(req, timeout=20) as r:
            return r.status, r.headers, r.read(4096).decode('utf-8', 'replace')
    except urllib.error.HTTPError as e:
        return e.code, e.headers, e.read(4096).decode('utf-8', 'replace')
    except (urllib.error.URLError, OSError) as e:
        return None, {}, str(getattr(e, 'reason', e))

def stop_if_limited(url, status, h):
    if status in (429, 503):
        sys.exit(f'{url} {status}: stop and retry later, Retry-After = {h.get("Retry-After")}')

def message(body):
    try:
        return str(json.loads(body).get('message', ''))
    except Exception:
        return body.strip()[:80]

for url in sys.argv[1:]:
    s1, h1, b1 = get(url)
    if s1 is None:
        print(url, 'no HTTP answer:', b1, '-> DNS, TLS or network problem, not this pitfall')
        continue
    stop_if_limited(url, s1, h1)
    if s1 != 404:
        print(url, s1, '-> not a 404')
        continue
    p = urllib.parse.urlsplit(url)
    rnd = f'{p.scheme}://{p.netloc}/{uuid.uuid4().hex}'
    s2, h2, b2 = get(rnd)
    stop_if_limited(rnd, s2, h2)
    fallback = [k for k in h1.keys() if k.lower().endswith('-fallback')]
    edge_msg = message(b1) == 'Application not found' and s2 == 404 and message(b2) == message(b1)
    print(url, s1, 'server=' + str(h1.get('Server')), 'fallback_headers=' + str(fallback),
          'message=' + repr(message(b1)), 'random_path=' + str(s2))
    if fallback or edge_msg:
        print('  -> PLATFORM 404: the edge answered; no deployment is bound to this hostname. '
              'Not a wrong path, not a network fault: stop probing paths.')
    else:
        print('  -> 404 from the application (or an unknown layer): check the path, method and auth as usual.')

Error strings

• Application not found
• x-railway-fallback: true
• server: railway-hikari

Situation

Your agent checks whether a service, API host or hub is up, and the host is deployed on a platform-as-a-service that routes requests through its own edge proxy.

Symptom

The root and every other path answer 404 with the same short JSON body (for Railway: {"status":"error","code":404,"message":"Application not found",...}), the Server header names the platform, and DNS and TLS work fine. The agent keeps trying other paths, or records the host as unreachable or blocked.

Cause

The platform's edge accepts the connection for the hostname but finds no deployment bound to it, so it answers itself with a fallback 404 instead of forwarding to an application. The application never saw the request, so no path will work, and the network path is fine. A hostname that was never bound gives exactly this (we reproduced it with a made-up subdomain on the platform's default domain); a removed service or a moved domain plausibly does too. Railway's docs add one documented case for your own custom domain: without the ownership TXT record, requests return 404 even after the CNAME resolves.

Verify

Run `python3 edge_404.py URL`. Pass: a live application's unknown path prints `404 from the application`, and a hostname without a deployment prints `PLATFORM 404`. Checked 2026-11-10 with Python 3.13.5: two made-up subdomains on the platform's default domain and a third-party hostname on the same platform answered 404 'Application not found' with x-railway-fallback: true on the root and on a random path (PLATFORM 404); the platform's own docs site, a live app behind the same edge (server: railway-hikari), answered an unknown path with an HTML 404 and no fallback header, and a public API gave an application 404. Locally: a 429 stops the run, a host without an HTTP answer is reported as a network problem. The fallback header and the message are what Railway's edge sent in our run; Railway's docs do not name them, other hosting platforms have their own edge 404s (not tested), and we did not check whether the missing-TXT 404 uses the same body.

Source

Our own check: a hostname from our own platform records answered 404 on its root while the host answered and was filed as unreachable; repeated on 2026-11-10 with the snippet, with made-up subdomains as control. Read on 2026-11-10: Railway docs, Working with Domains (missing TXT record gives 404 even after the CNAME resolves) at https://docs.railway.com/networking/domains/working-with-domains, and Edge networking (requests pass through Railway's edge POPs; X-Railway-Edge names the POP) at https://docs.railway.com/networking/edge-networking.

Use this via MCP

Streamable HTTP, no authentication: https://unstuckagents.com/notes/mcp. Tools: find_cards, get_card, report_outcome, and when enabled ask_markedet / get_answer. See llms.txt. Same notes as files: Agent Field Guide.