Network data.
Ready to read.

Two views.
Separate measures.

The feed URLs and schemas below are fixed. A request retrieves the configured public catalog.

Fetch once.
Read carefully.

The Python example uses the standard library. It checks snapshot and measurement timestamps against the client clock, then prints each public hostname, its ICMP status and current RTT when available.

A single request returns JSON. There are no date ranges, target selections or additional query parameters.

curl --max-time 15 --silent --show-error https://stealth.net/api/business-platforms

import json
import time
from urllib.error import HTTPError
from urllib.request import urlopen

url = "https://stealth.net/api/business-platforms"
try:
    response = urlopen(url, timeout=15)
except HTTPError as error:
    if error.code != 503:
        raise
    response = error  # 503 still contains a snapshot.
with response:
    data = json.load(response)
if data.get("schema_version") != 1:
    raise ValueError("Unsupported measurement schema")
if data.get("state") not in {"live", "empty"}:
    raise RuntimeError("Current measurements unavailable")
now = time.time()  # Keep the client clock accurate.
fresh = lambda stamp: type(stamp) in (int, float) and now - 120 <= stamp <= now + 5
if not all(fresh(data.get(key)) for key in ("server_time", "generated_at", "source_snapshot_at")):
    raise RuntimeError("Snapshot is old or client/server clocks differ")
for endpoint in data["endpoints"]:
    status = endpoint["probes"]["icmp"]
    metric = endpoint["metrics"]["icmp"]
    stamps = (metric["recorded_at"], status["success_at"], status["up_at"])
    current = metric["current_ms"] if status["state"] == "up" and all(map(fresh, stamps)) else None
    label = status["label"]
    if status["state"] in {"up", "down"} and not all(map(fresh, stamps[1:])):
        label = "Unconfirmed"
    print(endpoint["instance"], label, current)
# Request each feed at most once per minute. None means unavailable.

A Python None value means the current RTT is unavailable. It is never a zero-latency substitute.

• Authentication is not required. Make an HTTPS GET to one of the two fixed feed URLs. No query parameters, target selections, date ranges or arbitrary queries are accepted.
• Poll each feed at most once per minute. Snapshots are shared briefly on the server; faster polling does not create a new measurement. After a failure, wait at least a minute and back off rather than retrying rapidly.
• Measurement responses use Cache-Control: no-store. Keep source and observation timestamps with any separately stored result. Cross-origin browser access is not guaranteed; use a server-side or same-origin client.

Context travels
with the data.

These are network observations from Stealth NYC. They do not establish that an application workflow is functioning.

Measurements originate from Stealth NYC, 32 Avenue of the Americas. Hostnames can resolve to CDN or shared service edges. ICMP measures a round trip; TCP/TLS checks establish connection reachability, not login, messaging, inference or another application workflow.

Interpret each protocol independently. ICMP non-response alongside successful TCP/TLS checks is consistent with ICMP filtering, but does not establish its cause or an application outage. Not monitored is an intentional configuration state.

Use actual recorded_at, success_at and up_at timestamps. They are Unix seconds in UTC and can include fractional seconds. generated_at is the snapshot time and server_time is response presentation time; neither replaces a measurement timestamp. Current results require fresh measurements and healthy matching scrapes, normally within 120 seconds.

A null measurement, null chart point or null count is unavailable or unknown. Preserve gaps. Do not interpolate missing buckets, turn null into zero or infer uptime from successful-sample statistics. A genuine numeric zero remains a measurement.

The observation window spans 432,000 seconds with a nominal 300 second chart step. There is no older-history or export parameter. Treat the returned window as authoritative; cached stale behavior differs between feeds as described below.

Collection start and available history are explicit. collecting_history does not imply that every bucket has complete coverage. For AI, compare counts to aggregation.expected_samples. For Business, use the bucket duration and protocols[protocol].interval; a full 300 second bucket expects 300 divided by that interval (for example, ten at 30 seconds or five at 60 seconds). Partial or unknown counts need separate labeling.

Protocol states differ between feeds. AI uses up, down, unknown and stale. Business uses up, down, unconfirmed, unavailable and not_monitored. Use the accompanying labels and JSON Schema definitions for their fields.

Built for
careful reuse.

OpenAPI 3.1.1 describes the two GET operations. Each standalone schema uses JSON Schema 2020-12.

The response schema_version is currently 1. Read fields by name, accept additional fields and do not depend on object order, endpoint order, a fixed endpoint count or a hostname list. Check the version before interpreting a payload.

Top-level state is live, empty, stale or unavailable. HTTP 200 can carry stale or empty data; status 200 alone is not a health check. HTTP 503 carries an unavailable snapshot, while HTTP 400 carries an error object for a request with query parameters.

Only configured public measurement data is exposed. Business results are restricted to monitor_group=business_platforms, visibility=public and probe_host=millennium2; AI results remain in their separate super_intelligence group. Customer records, private tenants, router telemetry, detailed routes and RCA evidence are outside these feeds.

Keep the original source URL, observation window and UTC timestamps when citing or displaying results. Service commitments and remedies are defined in the applicable agreement, rather than inferred from a feed response.