Has this validator actually been dependable?

A record that starts when Taosis started

The chain keeps no series of a validator's epochs: each epoch overwrites the dividend figures of the last. The public lite node prunes past state within minutes. An archive node keeps every block, so the history could in principle be rebuilt from it — one read per epoch, per subnet, per validator — but Taosis has not back-filled it.

So a reliability record only covers the epochs Taosis has sampled itself since it began, and every answer states how many days it covers rather than hiding that behind a confident-looking number.

Where to read it

The verdict is served by the public API, not shown on a page: /validators/reliability/{netuid}/{hotkey}, documented on the API page, returns the paying ratio, the median APR and APY, their spread, permit losses, the stake trend and the verdict for one validator on one subnet.

What is measured

The share of observations where the validator actually paid; the median yield rather than the mean, so one lucky epoch cannot carry the figure; the spread between best and worst; whether it lost its validator permit; and how its stake has trended.

A permit loss outranks everything else. A validator that stopped being a validator is not a good yield with an asterisk.

Why it refuses to answer

Under twelve observations or under two days, the verdict is "insufficient" and no score is given.

That is not a gap waiting to be filled by a better algorithm. It is the honest answer to a question the data cannot yet support.