How it works
Methodology
Freshness is a feature. This page explains what our status labels mean, how they are produced, and what we refuse to infer.
What we list
The core directory covers AI-native live media: continuous AI streams, audience-directed shows, interactive generative video, interactive generative worlds and AI-native live channels. Adjacent projects (live transformation, world-model research previews, creator tooling, real-time inference infrastructure) can be listed but are always tagged as such. Ordinary one-shot video generators and human live streams with AI captions are out of scope.
Every published entry needs a unique name, a reachable official URL, a one-line description, a type, a current status with a last-verified time, an original editorial summary and at least one public source. Fields we cannot confirm are shown as “Not publicly disclosed”, never guessed.
Two separate questions: is the site up, and is it broadcasting?
Website reachability (Online, Blocked, Authentication Required, Unknown) and stream status (Live, Scheduled, On Demand, Paused, Offline, Discontinued, Unknown) are recorded as separate fields. A page that returns HTTP 200 tells us the site is online. It tells us nothing about whether anything is on air, so it can never produce a LIVE label.
How LIVE is decided
Detectors run against public endpoints only, in this order of trust:
- Stream manifest — a public HLS/DASH playlist whose media sequence advances between two reads. High confidence.
- Status endpoint — the operator’s own public status feed reporting live, with a fresh timestamp and recent generation activity. Medium confidence.
- Page signal — a stable marker in the page. Medium at best; not used on its own for LIVE in this phase.
- Editor verified — a human watched the stream. Always carries an expiry.
Red LIVE appears only when a detector with High or Medium confidence confirmed a broadcast and the result has not expired. Low-confidence live signals are shown as Unverified.
Expiry and degradation
| Situation | Shown as |
|---|---|
| Live confirmed within the last 60 minutes | LIVE (red) |
| Any check older than its validity window | STALE — treated as unverified until the next check |
| Detector timed out, errored or returned unusable data | UNVERIFIED — never Offline |
| First negative read after a live result | UNVERIFIED — a second consecutive negative is required before Offline |
| No public detector exists for the stream | UNVERIFIED, with the reason stated on the platform page |
This site is published as static files from a build-time status snapshot. The pages re-check expiry in the browser, so a snapshot that has aged past its window degrades to STALE on screen and drops out of the Live sections even before the next build. We do not claim a real-time ten-minute refresh in this phase; the snapshot time is printed on every page.
What we deliberately do not record
- Concurrent viewer counts or “today” counters from operator pages — they are dynamic UI values, not facts.
- Launch dates, prices or models that appear only in third-party summaries.
- Marketing copy — every summary on this site is written by us.
Sources and claim keys
Each fact field on a platform or model page maps to a claim key (for example platform.infinite-slop.launch_date) that must be covered by at least one public source record with an observation timestamp. The build fails if a fact has no source. Sources are labelled Officially stated, Observed by AI Stream Radar, Reported by third party, Not publicly disclosed or Unable to verify.
Editorial, trending and sponsorship are kept apart
Live is a fact. Editor’s Picks are our judgement. Trending would be an algorithm and is switched off until the site has enough traffic to compute it honestly. Sponsored placements, when they exist, will be labelled as such and will never be mixed into Editor’s Picks.
Corrections
Every platform page links to a report-an-issue entry. Corrections that change a fact field are recorded with a new source.