We run Best IPTV Player, a catalog that tells people which IPTV player apps their device can run. Players only: no channels, no playlists, no subscriptions. Every "yes" in the catalog has to come from the publisher's own website, with a source URL and a check date. Anything else stays unknown.
On 20 November 2026 we rebuilt our SmartOne IPTV record. It had been all unknown since our first extract on 9 November. Fixing it was less about the facts and more about how our pipeline handles a publisher whose pages disagree, or do not load at all.
The four pages
SmartOne IPTV's site gave us four different "answers" depending on which URL we fetched:
| Page | What we got (20 Nov 2026) |
|---|---|
| Home page | "Samsung TVs (Orsay and Tizen), LG TVs (webOS and Netcast)", VIDAA, Foxxum, Vestel, Android TV, Google TV, Fire TV; Xtream Code API; m3u; "not offering any kind of channels" |
| Activation page | Lifetime €12.99 / 1 Year €2.50 "for each TV/device"; "15 days to test"; not refunded |
| LG page | "LG TV including webOS and Netcast series" |
| Samsung page | A bot-check interstitial ("Please wait while your request is being verified") |
On 9 November the activation page was the one behind the challenge, and our extractor recorded the activation URL as the player's "official" source. So the record pointed at a page we couldn't read, and every device stayed unknown. That was correct, but it didn't help anyone.
Rule 1: a challenge page is "no observation", not "no"
The worst bug you can have here is treating an interstitial as content. A bot-check page contains no platform names, so a naive "does the page mention Samsung?" check returns false, and false easily turns into "not supported". So the first step is to classify a fetch before reading anything from it:
def classify(html: str) -> str:
text = visible_text(html)
if len(text) < 300 and re.search(r"request is being verified|one moment, please", text, re.I):
return "challenged" # no observation; never emit a claim from this
return "ok"
A challenged fetch can't change a claim in either direction. The claim keeps its previous status and its previous check date.
Rule 2: claims are per (subject, property, source URL)
Don't store smartone.samsung = true. Store observations:
{"player": "smartone", "property": "device:samsung-tv",
"status": "verified_yes",
"quote": "Samsung TVs (Orsay and Tizen)",
"source": "https://smartone-iptv.com/", "checked": "2026-11-20"}
The published status for a property comes from resolving all observations for it:
- any
verified_yeswith a quote from a publisher-owned URL wins overunknown; - an explicit publisher "no" (for example "We are not offering any kind of channels") becomes
verified_no; - two publisher pages that really contradict each other (one says X, another says not-X) resolve to
partialand get flagged for a human. They don't get averaged.
In this round nothing actually contradicted anything. The pages were only differently complete. The home page has the device list, the activation page has the price, and the LG page adds detail about LG. Merging by property, not by page, gives the full record without any page "overriding" another.
Rule 3: keep the quote, not just the boolean
"Samsung TVs (Orsay and Tizen)" carries information a boolean throws away: Orsay is Samsung's pre-Tizen generation, and Samsung's developer specs list Tizen for model years 2015 to 2026. Storing the quote lets the page say something useful ("check your model year") without us inventing a model-year matrix the publisher never published.
Rule 4: a flip has to reach every derived view
The data change itself took one line. Finding everywhere the old value had been copied took much longer. In a statically exported site, one player's Samsung status had ended up in:
- the player page's fact cards,
- the compare table's client-side data chunk,
- a 12-character device bitstring (
data-d="uuuuuuuuuyuu") on each localized selector home page, plus the same string inside the framework's serialized payload, - the localized Samsung device pages (fr, ar, de, sv).
We missed #4 on the first pass a day earlier. The English Samsung page said "verified" while the four other languages still said "unknown". The lesson for us: after a flip, search every built output for the player's old status in every locale before calling it done. A better long-term fix is a check that rebuilds each view from the resolved claims and fails the build if any rendered status disagrees.
What stays unknown
Phones, tablets, computers, Apple TV and Roku are not named on the SmartOne pages we fetched, so they stay unknown on the SmartOne IPTV page. Unknown is not support. It just means the publisher hasn't said so, and a catalog that keeps that difference visible is more useful than one that guesses.
Top comments (0)