The Subnet Hierarchy

Our last report, The Validator Hierarchy, ranked the 25 largest Bittensor validators by what nominators actually earned on root. It found a 4.41-point spread between best and worst, and that the advertised take rate barely predicts realized return.

That report covered root staking only. This one asks the obvious follow-up: when you stake alpha on a specific subnet rather than root, does the same ranking hold? Or does each subnet have its own best validator, the way the common advice implies when it tells you to find specialists?

We took the same 25 validators and the same year of daily on-chain yield data, and measured how each validator performed on every one of the 128 subnets it validates. The answer is unusually clean.

Validator quality is not subnet-specific. It transfers. The validators that lead on root also lead almost everywhere else, and a handful of them win nearly the entire network.

The practical consequence for a staker is the opposite of the usual advice. You do not need to research a different validator for every subnet. The same small set of names that topped the root ranking also tops the subnet rankings, almost everywhere. One good validator choice covers your whole portfolio.

An important note on what these yields mean

Before the findings, a caveat that is central to reading this report honestly. The subnet yields here are alpha-denominated. A "58% APY" on a subnet means you earned 58% more of that subnet's alpha token, not 58% more value in TAO terms.

Our first report, The Yield Illusion, showed why that distinction matters: the median subnet alpha staker lost money in TAO last month despite headline APYs above 50%, because the alpha token's price fell faster than the yield accrued. Those high subnet APYs are mostly dilution, not profit.

So this report does not claim that staking alpha on a subnet at "58% APY" makes you money. It almost certainly does not, in TAO terms. What this report measures is something narrower and more reliable: among validators on the same subnet, who captured the most of whatever yield existed. Because every validator on a subnet faces the identical token price, the price decay cancels out of the comparison. The ranking of validators within a subnet is real even though the absolute yield number is an illusion.

Quality transfers: the root leaders win everywhere

For each validator we took its rank on every subnet it validates, then summarized how often it lands near the top. The pattern is a clean two-tier structure. A small group of validators places top-5 on the overwhelming majority of subnets. After that group, performance falls off a cliff.

Shaded rows are the five validators that place top-5 on more than a third of subnets. Root rank is from The Validator Hierarchy. Note how closely it tracks the subnet columns: the root-1 validator has a median subnet rank of 1; the root-2 validator a median of 2.

The correspondence with the root ranking is striking. 1T1B.AI, first on root, has a median subnet rank of 1 and places top-5 on 96 percent of subnets. tao.bot, second on root, has a median subnet rank of 2 and places top-5 on 98 percent. Datura, third on root, third on subnets. The validators that are good at validating are good at it everywhere. Quality is a property of the operator, not the subnet.

And the fall-off is sharp. After the top four (all above 80 percent top-5), the fifth validator drops to 39 percent and the sixth to 6 percent. There is no gentle gradient of slightly-worse validators. There is a small tier that is excellent across the board, and a large tier that is rarely near the top of anything.

Win concentration: two names, three-quarters of the network

The clearest way to see the concentration is to count outright wins: on how many subnets was each validator the single best performer?

1T1B.AI is the single best validator on 66 of 128 subnets, a clean 52 percent of the entire network. Add tao.bot and the top two cover 96 subnets, 75 percent. Add Ary van der Touw and K8STAKE.AI and the top four cover 91 percent. The remaining nine percent is split in ones and twos among everyone else.

This is the finding that overturns the common advice. If validator quality were subnet-specific, wins would be spread across many operators, each strong on the subnets they specialize in. Instead, a tiny number of validators win almost everywhere. The specialist-per-subnet strategy is chasing a pattern that is not in the data.

How much does the choice matter? The capture ratio

Knowing which validator is best on a subnet only matters if the gap to the others is meaningful. To measure it without the alpha-price distortion, we use the capture ratio: the best validator's yield divided by the median validator's yield on the same subnet. Because both face the identical token price, this ratio is a clean, price-neutral measure of how much validator choice is worth.

On the median subnet, the best validator captured 1.12 times what the median validator did, a 12 percent edge in yield for the same staked alpha, the same subnet, the same token price. The difference is purely the operator. On some subnets the edge is far larger: the best validator on subnet 65 captured 2.01 times the median.

Twelve percent may sound modest next to the headline subnet APYs, but remember those APYs are largely illusory in TAO terms. The capture ratio is the part that is real. A consistent 12 percent edge in actual yield capture, compounded across a year and across every subnet position you hold, is the difference between a good validator choice and a mediocre one, and it costs nothing to choose well.

The per-subnet lookup

The complete table: for each of the 128 subnets, the validator that delivered the best realized nominator yield over the trailing 90 days, the capture ratio versus the median validator, and the spread between the best and worst validator on that subnet. Use the search box to jump to a subnet by number.

Capture ratio and spread are the reliable, price-neutral measures. Eight subnets are flagged outlier where a short-lived emission spike produced an implausible annualized figure (marked with *); their ranking is shown but the magnitude should be ignored. Spread is in percentage points of alpha-denominated APY and is comparable only within a subnet.

What this means if you stake alpha on subnets

You do not need a different validator for each subnet. The single most useful takeaway: the validators that are best on root are best almost everywhere. Picking one of the top handful covers your entire subnet portfolio. The effort of researching a specialist for each position buys you nothing, because the specialists mostly do not exist.

The same names keep winning. 1T1B.AI and tao.bot together are the best validator on three-quarters of all subnets. If you stake alpha and have not thought about your validator, defaulting to one of the consistent top performers is the safe, simple choice, the same conclusion the root report reached, now confirmed across the whole network.

Choosing well is worth a real, if modest, edge. The best validator captures about 12 percent more yield than the median on the typical subnet, more on some. That edge is price-neutral and recurring. It will not rescue a losing alpha position, but among validators it is free money left on the table if you pick poorly.

But the validator is the smallest of your subnet decisions. The honest hierarchy of what determines your outcome when you stake alpha: first, whether the subnet's token holds its value (mostly it does not, see the Yield Illusion); second, your timing; a distant third, which validator you pick. This report optimizes the third. Do not let it distract from the first.

Methodology

This report uses the same dataset as The Validator Hierarchy: 860,642 daily nominator-yield snapshots pulled from the Taostats API endpoint /api/dtao/validator/yield/history/v1, covering 25 validators across all 129 subnets including root, for 365 days ending June 12, 2026.

Universe. The 25 validators are the largest commercial root validators by global nominator count, active on at least 50 subnets. Subnet-primary operators that validate only their own subnet are excluded, as in the prior report.

Per-subnet ranking. For each subnet we average each validator's daily one-day nominator APY over the trailing 90 days, then rank validators within that subnet. The validator with the highest average realized yield is the subnet's best. Validators with fewer than 70 percent of the window's days, or fewer than 25 days, are excluded from that subnet's ranking.

Capture ratio. The price-neutral quality measure: the best validator's average APY divided by the median validator's average APY on the same subnet. Because all validators on a subnet earn the same alpha token and face the same token price, dividing one by another cancels the price effect, leaving a clean comparison of validator performance.

Outlier handling. Eight subnets where the best validator's annualized APY exceeded 200 percent are flagged as outliers. These reflect short-lived emission spikes or low-liquidity dislocations that produce implausible annualized figures, the most extreme being a single subnet reading above 1,400 percent. We show their validator ranking but the magnitudes should be disregarded; they are excluded from the summary statistics.

Transfer analysis. For each validator we compute its rank on every subnet it validates, then report the median subnet rank and the share of subnets where it places top-5, top-3, and first. Comparing these to the root ranking shows whether quality transfers across subnets. It does.

What this is not. The subnet APY figures are alpha-denominated and are not TAO returns. This report ranks validators within subnets; it does not endorse subnet staking as profitable. See The Yield Illusion for the TAO-denominated picture.

Limitations

Realized yield is backward-looking and reflects the trailing 90 days; validator performance can change. The subnet APY figures are alpha-denominated and should not be read as TAO returns. The capture ratio and within-subnet rankings are reliable; the absolute yield magnitudes are not, and on the eight flagged outlier subnets even the magnitude of the spread should be disregarded. Ary van der Touw appears strong where present but validates fewer subnets (108 of 128) with a shorter history, so its standing carries more uncertainty than the fully-covered validators. This is a single 90-day snapshot within the broader year of data; we intend to refresh it alongside the root ranking quarterly.