Per-provider health status (last search / last successful download / active gate) and a provider test that exercises the download path

[Svnow132]

Summary

Give the Providers page a live per-provider health rollup — last search, last successful download, and which gate (auth/breaker/budget/pool) is currently blocking it — plus a Test button that performs a real download, not just a search.

What problem does this feature solve?

Provider failures are invisible at the fleet level. On our install (see #NN), searches succeeded for 3 days while every download 401'd on an expired token, and ~20 providers were silently skipped by budget/pool gates — the UI showed nothing wrong anywhere. The per-item decision log is excellent for one item, but there is no rollup answering: "which providers are actually participating right now, and which are gated, and why?"

The search/download split matters specifically: a provider can pass every search while its download path is dead (that asymmetry is #NN Bug 1), so any health signal built only on searches shows green through the failure.

What do you propose?

On the Providers page, per provider:

• last search time and last successful download time as separate columns
• current gate state, live: ok / auth-expired / circuit-breaker open / budget-denied / no-pool-key / rate-limited / auto-disabled
• a Test button that exercises search and a real download of one result (content discarded afterwards, no history written) — a search-only test passes while the download path is dead

The data mostly exists already (provider_stats, circuit breakers, the decision-log skip reasons) — this is a surfacing feature more than a new subsystem.

Alternatives considered

• Per-item decision log ("why was nothing found?") — great forensics, but one item at a time, after the fact; it never shows the fleet.
• Log grepping — how we actually found everything in #NN; requires shell access and knowing which messages to look for (several gates log at DEBUG only).
• External monitoring — can probe the HTTP endpoint but cannot see internal gate states (budget, pool, breaker) at all.

[Abrechen2]

Implemented in v1.11.0, in three parts.

Search and download times are now recorded separately — they shared a single field before, which is precisely why "searches fine, downloads dead" was invisible. GET /providers/health reports both, plus a gate field naming the one thing currently keeping a provider out of searches, resolved in the order the search path applies its gates (auto-disable, breaker, rate limit, budget, pool). A gated provider now raises a warning on the Providers page instead of being counted as healthy — previously those two gates lived entirely inside the search loop, which is how providers came to be skipped with nothing in the UI saying so.

The Test button has a companion Test download that fetches one real subtitle and discards it. It is a separate action rather than always-on because a download spends one unit of the account's daily allowance.

One limitation, stated plainly: the download test probes with the first item from your wanted list, and for anime providers that query is too thin — AnimeTosho's matching depends on AniDB resolution that the real search pipeline performs and this probe does not. On our own instance the test found 0 results where the scheduler's own search found 18 for the same series. The result message says that a no-result outcome tells you nothing about the download path, but for that provider class the button does not yet deliver what it promises. Building the probe through the real search pipeline is the fix, and it is not done.