FLOWSNIPER/ RESEARCH
<< ON-CHAIN AUDIT · JUNE 2026 >>

The Yield Illusion

A 90-day audit of 128 Bittensor subnets testing the claim that staking subnet alpha at 50%+ APY beats root staking. The median staker lost 7.2% in TAO last month. Here's the on-chain proof.

If you spend any time in Bittensor circles, you've heard the pitch: why stake TAO on the root network for single-digit yield when subnet alpha tokens pay 50%, 100%, sometimes 200% APY?

It sounds like free money. We operate trading infrastructure on Bittensor full time, and we suspected the opposite, that alpha emissions inflate token supply faster than the yield accrues, and that the price decline this causes swamps the APY entirely. So we pulled 90 days of daily on-chain data for every one of the 128 active subnets and tested it.

The pitch does not survive contact with the data.

54%
Median marketed APY
(alpha-denominated)
-7.2%
Median realized 30-day
return (TAO-denominated)
73%
Subnets that lost money
in TAO over 30 days
24%
Subnets that beat
plain root staking
Figure 1. The path from the marketed number to the realized outcome for the median subnet over the most recent 30 days.

Terminology, because this is where the pitch hides

The entire illusion lives in a naming ambiguity, so we will be exact:

Alpha price (in TAO)The subnet token's AMM pool price, TAO per alpha. All "price" references in this report are this quantity.
Alpha supply growthThe percentage increase in the subnet's total issued alpha over the window. This is the dilution rate.
Staker yield (alpha)The growth of a nominator's staked alpha balance, compounded daily from measured validator APY, net of validator take.
Marketed APYThe annualized version of the staker yield. The number quoted in promotional material.
Realized return (TAO)Staker yield combined with the alpha price change. The number an investor actually experiences.
Root return (TAO)The return from staking TAO on the root network (netuid 0) with the same validator over the same window.
TAO / USDDeliberately excluded. It multiplies both strategies identically and cancels out of every comparison.

The relationship between these terms is one line of math:

realized return = (1 + staker yield) × (price_end / price_start) − 1

The pitch quotes the first term and stays silent about the second.

What 90 days of chain data shows

We measured every window from 7 to 90 days, ending June 10, 2026. Yields are actual nominator returns from tao.bot, the largest root validator, chosen because its 0% take removes fee distortion and it validates on all 129 subnets including root, which gives a perfectly controlled comparison: same validator, same operator, same infrastructure, only the asset differs.

WindowSubnetsMarketed APY Supply growthPrice change Realized return% beat rootRoot return
7d12753.8%+1.1%-4.7%-3.9%27%+0.15%
10d12752.6%+1.6%-5.5%-4.2%33%+0.22%
21d12753.2%+3.3%-8.1%-5.4%24%+0.44%
30d12754.0%+4.7%-10.1%-7.2%24%+0.65%
60d12654.4%+9.7%-9.5%-2.9%44%+1.38%
90d12656.0%+15.5%-17.0%-8.8%37%+2.09%
Figure 2. Share of subnets whose realized TAO-denominated return exceeded the measured root return, by window. At no horizon did even half of subnets win.

93 of 127 subnets lost money in TAO terms over the last 30 days. Only 31 beat root staking.

Where the staker yield actually comes from

This is the part of the analysis that explains why the pitch fails, mechanically, rather than just observing that it does.

The marketed APY is paid in newly emitted alpha. The yield is not income the subnet generates, it is a redistribution of dilution. Three medians, side by side:

Figure 3. Median alpha supply growth, staker yield, and alpha price change across 127 subnets by window. The yield never even keeps pace with dilution.

At 30 days, the median subnet's supply grew 4.7% while a staker's balance grew 3.6%. Your share of the network shrinks even while staked. (Holding unstaked alpha is strictly worse: full dilution, zero yield.) Meanwhile the alpha price fell 10.1%, nearly triple the yield, because the same emissions funding the APY create continuous sell pressure from miners, validators, and owners converting alpha to TAO.

The headline APY and the price decline are not independent quantities. They are the same emission stream measured from two different angles.

"But what about the subnets that pumped?"

Fair question. Some subnets crushed it, and honesty requires showing them.

Best performers (30 days)

SNNameMarketed APYPrice changeRealized returnvs Root
SN116 156.8%+237.7%+264.9%+264.3%
SN122 86.5%+232.3%+249.8%+249.1%
SN92 192.2%+189.2%+215.9%+215.2%
SN118Ditto49.1%+141.5%+149.5%+148.9%
SN110Green Compute68.8%+88.8%+97.1%+96.4%

Worst performers (30 days)

SNNameMarketed APYPrice changeRealized returnvs Root
SN82Compelle184.0%-78.1%-76.2%-76.8%
SN84 136.7%-68.9%-66.6%-67.2%
SN102ConnitoAI167.0%-57.0%-53.4%-54.0%
SN76Byzantium152.8%-44.0%-39.5%-40.2%
SN29Coldint43.8%-40.7%-38.9%-39.5%

Look at the decomposition. The five winners returned +97% to +265% in TAO, of which the staker yield contributed three to nine percentage points. The rest was alpha price appreciation. The losers carried similar yields against price declines of 40 to 78 percent.

In both directions, the conclusion is the same: alpha staking is a price bet with a small yield attached, not a yield product. If you have a thesis that a specific subnet's token will appreciate, that can be a fine trade. But the APY is not a reason to make it, and it was never the thing driving your outcome.

Figure 4. Marketed APY against realized 30-day return across all 127 subnets. If the pitch were sound, this cloud would slope upward. Correlation: effectively zero.

By market cap quartile (30 days)

QuartileSubnetsMedian realized return% beating root
Q1 (smallest)33-15.9%18%
Q231-7.5%6%
Q332-2.4%44%
Q4 (largest)31-6.5%29%

Losses are not confined to illiquid small caps. Every quartile posted a negative median realized return, including the largest subnets by market capitalization.

Robustness across validators

Headline figures use tao.bot. Repeating the analysis with two other major validators, at 9% and 18% take read directly from chain state, does not change the conclusion:

Validator (take)Subnets coveredMedian 30d return% positive
tao.bot (0%)128-7.8%28%
Kraken (18%)121-8.2%26%
Yuma (9%)127-8.0%28%

Take rates make a losing proposition slightly more losing. They do not change direction.

The breakeven, in one sentence

▌ THE STRUCTURAL FINDING

At a 54% marketed APY (≈3.6% over 30 days) and the measured root return of +0.65%, the alpha price needed to fall less than about 2.9% over the month for alpha staking to stay ahead of root. The median subnet's alpha price fell 10.1%. That gap is the entire story.

Figure 5. Every subnet's actual 30-day outcome. The yield lifts each dot a few points; the price axis decides everything.

Methodology

DATA SOURCES
All historical series pulled from the Taostats API: daily pool snapshots (alpha price in TAO, total alpha supply, alpha staked) from /api/dtao/pool/history/v1 and daily validator nominator APY from /api/dtao/validator/yield/history/v1, covering the 95 days ending June 10, 2026. Subnet identities from /api/subnet/identity/v1.
VERIFICATION
A self-operated Bittensor mainnet node (production build, finney) was queried for current chain state. Across 128 subnets, the median difference between the Taostats end-of-day snapshot and live chain pool price was 1.43%, fully attributable to the 0-24 hour gap between snapshot time and query time. Validator take rates were read directly from on-chain Delegates storage rather than assumed.
STAKER YIELD
Daily nominator APY values are net of validator take. Converted to daily compounded rates via (1 + APY)^(1/365) − 1 and compounded across each window. Windows require at least 80% of expected yield days; partial coverage is geometrically scaled. The headline series uses tao.bot exclusively (0% take, full subnet coverage).
ROOT HURDLE
The primary hurdle is the measured return of the same validator staking netuid 0 over the identical window, controlling for take, operator, and infrastructure. A fixed 12% annualized assumption serves as a secondary check.
SUPPLY GROWTH
Computed from the change in total issued alpha between the snapshots nearest each window's endpoints.
COVERAGE
Of 129 active subnets, 128 are non-root. 127 of 128 are included in the 7-30 day analysis; 126 of 128 in 60-90 day windows. Subnet 104 ("for sale, burn to uid1") is abandoned with insufficient validator data (10/30 valid days at 30 days). Subnet 96 (Verathos) joined validator coverage on April 25, 2026, clearing 30d but not the longer windows.
OUTLIER FLAG
Subnet 96 (Verathos) recorded +235% alpha supply growth over the 30-day window, approximately 50 times the median. This reflects a one-time stake migration rather than ordinary emission. Included in all aggregates; medians are robust to it.

Limitations

Survivorship. Subnets that deregistered during the 90-day window are absent from the dataset. Their inclusion would make every aggregate worse. The results here are a conservative upper bound on the aggregate alpha staking experience, which strengthens rather than weakens the conclusion.

Costs not modeled. AMM slippage on entry and exit, trading fees, and unstaking mechanics are excluded. All are costs borne by the alpha staker and not by the root staker; including them would further widen the gap.

Window dependence. All windows end June 10, 2026. A period of broad alpha price appreciation would produce friendlier numbers; the five best performers show what that looks like. The structural finding, that realized returns are dominated by alpha price action and that the yield does not outpace dilution, is window-independent.

What this means in practice

  1. If someone quotes you a subnet APY, ask for the realized TAO-denominated return over the same period. They are different numbers, and the difference is usually the whole trade.
  2. Staking alpha you already hold is rational, it claws back most of the dilution. Buying alpha for the yield is the error.
  3. Root staking is the honest benchmark. Boring, single-digit, and it beat three quarters of subnets last month.
  4. Subnet selection is everything. The winners were price stories, identifiable (if at all) by fundamentals and flows, not by APY, which carried zero predictive signal.
APPENDIX, COMPLETE PER-SUBNET RESULTS, 30 DAYS (sorted best to worst)

128 subnets analyzed, 127 with sufficient data at the 30-day window. Scroll inside the table to see all rows. Click any column header to view the report PDF for further analysis.

SNNameMarketed APY30d yieldPrice changeRealized returnBeats root
SN116 156.8%8.06%+237.7%+264.9%YES
SN122 86.5%5.26%+232.3%+249.8%YES
SN92 192.2%9.21%+189.2%+215.9%YES
SN118Ditto49.1%3.34%+141.5%+149.5%YES
SN110Green Compute68.8%4.40%+88.8%+97.1%YES
SN23Trishool37.3%2.64%+86.5%+91.4%YES
SN38colosseum119.2%6.66%+67.9%+79.1%YES
SN111oneoneone39.8%2.79%+69.6%+74.3%YES
SN77Liquidity39.1%2.75%+66.7%+71.3%YES
SN95Actual75.8%4.75%+62.6%+70.3%YES
SN28gm35.5%2.53%+60.7%+64.8%YES
SN18Zeus33.7%2.41%+55.6%+59.4%YES
SN9iota39.6%2.78%+48.2%+52.4%YES
SN14Cacheon76.1%4.76%+38.6%+45.2%YES
SN114SOMA113.5%6.43%+29.8%+38.2%YES
SN107Minos122.9%6.81%+23.9%+32.4%YES
SN83CliqueAI61.6%4.02%+26.1%+31.2%YES
SN105Beam76.3%4.77%+15.1%+20.6%YES
SN46Zipcode32.8%2.36%+10.0%+12.6%YES
SN91Bitstarter #1227.3%10.24%+1.1%+11.4%YES
SN59Babelbit32.3%2.33%+8.1%+10.6%YES
SN33ReadyAI57.6%3.81%+5.1%+9.1%YES
SN49Nepher Robotics85.3%5.20%+3.6%+9.0%YES
SN79MVTRX30.9%2.24%+3.8%+6.1%YES
SN41Almanac49.5%3.36%+2.5%+6.0%YES
SN101 144.0%7.61%-2.1%+5.4%YES
SN10Swap49.7%3.37%+1.9%+5.3%YES
SN68NOVA38.2%2.70%+2.0%+4.8%YES
SN39deprecated76.2%4.77%-0.1%+4.6%YES
SN30Endure Network38.1%2.69%+1.2%+4.0%YES
SN48Quantum Compute55.5%3.69%-1.3%+2.3%YES
SN63Enigma47.8%3.26%-2.6%+0.5%NO
SN126Poker44149.2%7.79%-6.8%+0.4%NO
SN11TrajectoryRL52.9%3.55%-3.2%+0.3%NO
SN112minotaur78.2%4.86%-4.8%-0.2%NO
SN34BitMind53.8%3.60%-4.0%-0.5%NO
SN128ByteLeap39.7%2.79%-3.4%-0.7%NO
SN45Talisman AI45.0%3.10%-4.0%-1.0%NO
SN43Graphite44.2%3.06%-4.4%-1.4%NO
SN81deprecated81.5%5.02%-6.4%-1.7%NO
SN106Nodexo61.9%4.04%-6.7%-2.9%NO
SN52Dojo56.2%3.73%-6.5%-3.0%NO
SN60Bitsec.ai38.9%2.74%-5.9%-3.3%NO
SN53EfficientFrontier43.5%3.01%-6.2%-3.3%NO
SN123MANTIS54.3%3.63%-6.8%-3.4%NO
SN8Vanta46.7%3.20%-6.7%-3.8%NO
SN2DSperse41.8%2.91%-6.8%-4.1%NO
SN124Swarm50.3%3.41%-7.3%-4.2%NO
SN20GroundLayer52.7%3.54%-7.6%-4.3%NO
SN25Mainframe62.5%4.07%-8.2%-4.5%NO
SN1258 Ball111.3%6.34%-10.2%-4.5%NO
SN1Apex60.4%3.96%-8.2%-4.5%NO
SN89InfiniteHash49.2%3.34%-7.7%-4.6%NO
SN71Leadpoet32.5%2.34%-7.0%-4.8%NO
SN121sundae_bar64.5%4.17%-8.7%-4.8%NO
SN13Data Universe53.9%3.61%-8.3%-5.0%NO
SN80dogelayer95.7%5.67%-10.2%-5.1%NO
SN40Chunking43.8%3.03%-8.4%-5.7%NO
SN72StreetVision by NATIX41.7%2.91%-9.1%-6.5%NO
SN4Targon38.3%2.70%-9.0%-6.5%NO
SN109Academia118.0%6.61%-12.4%-6.6%NO
SN21AdTAO45.5%3.13%-9.5%-6.7%NO
SN37Aurelius45.1%3.11%-9.7%-6.9%NO
SN64Chutes44.5%3.07%-10.0%-7.2%NO
SN56Gradients52.1%3.50%-10.4%-7.2%NO
SN54Yanez MIID32.4%2.34%-9.5%-7.4%NO
SN98ForeverMoney62.5%4.07%-11.1%-7.5%NO
SN19blockmachine57.8%3.82%-11.0%-7.6%NO
SN127Astrid76.8%4.79%-11.8%-7.6%NO
SN44Score32.2%2.32%-10.1%-8.0%NO
SN7Allways33.1%2.38%-10.2%-8.1%NO
SN32ItsAI31.5%2.28%-10.1%-8.1%NO
SN73Parked54.0%3.61%-11.6%-8.4%NO
SN65TAO Private Network41.0%2.86%-11.1%-8.5%NO
SN93Bitcast46.8%3.21%-12.0%-9.2%NO
SN12Compute Horde42.8%2.97%-12.2%-9.6%NO
SN96Verathos215.5%9.90%-18.0%-9.9%NO
SN115HashiChain118.3%6.63%-15.6%-10.0%NO
SN42 43.7%3.02%-13.0%-10.3%NO
SN35OxMarkets33.5%2.40%-13.1%-11.0%NO
SN120Affine46.3%3.18%-14.4%-11.7%NO
SN117 66.3%4.27%-15.3%-11.7%NO
SN103Djinn46.4%3.18%-14.8%-12.1%NO
SN5Hone111.6%6.36%-17.3%-12.1%NO
SN22Desearch61.0%3.99%-17.7%-14.5%NO
SN67Harnyx167.0%8.41%-21.3%-14.7%NO
SN88Investing34.8%2.49%-17.4%-15.4%NO
SN6Numinous41.6%2.90%-17.8%-15.4%NO
SN31Recall109.5%6.27%-20.6%-15.7%NO
SN119Satori160.0%8.17%-22.1%-15.7%NO
SN87unknown113.4%6.43%-21.0%-15.9%NO
SN51lium.io58.8%3.87%-19.3%-16.1%NO
SN55NIOME35.5%2.53%-18.3%-16.2%NO
SN100Plaτform100.1%5.87%-21.1%-16.4%NO
SN75Hippius37.2%2.63%-18.6%-16.5%NO
SN3deprecated59.1%3.89%-19.8%-16.7%NO
SN50Synth36.2%2.57%-20.8%-18.8%NO
SN17404 GEN42.2%2.93%-21.4%-19.0%NO
SN74Gittensor39.3%2.76%-22.3%-20.2%NO
SN15ORO106.8%6.15%-24.8%-20.2%NO
SN47EvolAI126.6%6.95%-25.8%-20.6%NO
SN58Pending48.1%3.28%-23.8%-21.3%NO
SN61RedTeam36.1%2.57%-24.2%-22.3%NO
SN27Nodexo41.4%2.89%-24.7%-22.5%NO
SN85Vidaio43.8%3.03%-26.1%-23.8%NO
SN24Quasar34.9%2.49%-26.6%-24.7%NO
SN57 158.1%8.10%-31.1%-25.5%NO
SN16BitAds47.3%3.24%-29.5%-27.2%NO
SN36Eirel188.0%9.08%-33.7%-27.6%NO
SN113TensorUSD108.8%6.24%-32.0%-27.7%NO
SN108TalkHead100.5%5.88%-31.9%-27.8%NO
SN90 133.4%7.22%-33.0%-28.2%NO
SN62Ridges38.9%2.74%-30.7%-28.8%NO
SN99Leoma97.0%5.73%-32.8%-28.9%NO
SN97Albedo112.7%6.40%-33.5%-29.2%NO
SN8698.6%5.80%-33.1%-29.2%NO
SN66ninja30.3%2.20%-31.7%-30.2%NO
SN69 95.9%5.68%-35.4%-31.7%NO
SN94Bitsota114.8%6.48%-36.0%-31.9%NO
SN78Vocence151.0%7.86%-39.2%-34.4%NO
SN70NexisGen165.6%8.36%-39.7%-34.6%NO
SN26Perturb116.4%6.55%-42.1%-38.3%NO
SN29Coldint43.8%3.03%-40.7%-38.9%NO
SN76Byzantium152.8%7.92%-44.0%-39.5%NO
SN102ConnitoAI167.0%8.41%-57.0%-53.4%NO
SN84 136.7%7.34%-68.9%-66.6%NO
SN82Compelle184.0%8.96%-78.1%-76.2%NO
This document is research prepared by FlowSniper Research from public on-chain data. It is not financial advice and no representation is made regarding future performance of any token, subnet, or staking strategy. Bittensor dTAO is a high-volatility ecosystem; all figures are historical and window-dependent. The underlying dataset, SQL tables, and analysis code are retained and available for verification. Contact: research@flowsniper.ai
DOWNLOAD PDF MORE REPORTS SHARE ON X