15% of channels in a published guide carry zero programmes

[quepasaquepasa]

Observation

In the guide published by one of the workers listed in GUIDES.md
(raw.githubusercontent.com/StrangeDrVN/epg/public/guide.xml, fetched today),
73 of 466 channels are declared in the file and carry zero programmes — 15%
of the guide. The XML is valid, the file is a healthy 40 MB, and nothing in CI is
failing.

Grouped by the site that declares those channel ids:

site	zero-programme channels / channels it declares in this guide
tataplay.com	60 / 403
airtelxstream.in	20 / 336
i.mjh.nz	9 / 16
zee5.com	5 / 56
tvtv.us	5 / 10

(A channel id can be declared by more than one site, so a row is "sites that
could have supplied this channel", not proof that a given grabber is at fault.)

Why this might be worth a look

The tvtv.us row lines up with iptv-org/epg#3187. Checking its endpoint directly today:

GET https://www.tvtv.us/api/v1/lineup/USA-NY71652-X/grid/2026-07-31.../2121
→ 404 Not found

So at least one cluster of empty channels corresponds to a source that is
genuinely broken and was already reported by hand. The rest have not been
verified — an empty channel can equally mean a station that went off air, or a
channel the worker simply does not grab.

What stands out is that this is measurable from the published output alone,
before anyone files a report. npm test passes for these sites because the
tests run against stored fixtures, so a source that changes or dies upstream
stays green indefinitely.

If it's useful

I put together a small tool that does this check: it reads a published
guide.xml, counts programmes per channel, stores a snapshot, and on later runs
reports which channels stopped carrying data, which degraded, and which
disappeared — with how many runs each has been failing.

https://github.com/quepasaquepasa/feedcanary

Deliberately, it never contacts the source sites — only the guide your
workers already publish. That avoids adding traffic to a few hundred third-party
sites, which I assume is why live checks aren't part of CI today. It ignores
channels that have always been empty (they're reported separately as idle), so
the output is confined to things that actually changed.

Something like this could run on a schedule against the guides in GUIDES.md,
or not at all — I'm not proposing a dependency, just sharing the measurement and
the tool behind it in case the numbers are of interest. Happy to open a PR
wiring it into a workflow if you'd want that, and equally happy to leave it
outside the repo.

I've been sending grabber fixes here recently (#3224, iptv-org/epg#3225, iptv-org/epg#3226), which is
how I ran into this: diagnosing each broken source by hand takes a while, and
the pattern seemed general enough to automate.