Introducing SubnetAIQ Signals — A Subnet Selection Tool
Tonight I shipped a new feature: SubnetAIQ Signals. It's a subnet selection algorithm I built and validated against a year of Bittensor chain data. Live at signals.subnetaiq.io.
Here's the story.
The question
Which Bittensor subnets will outperform over the next 30-60 days? The question matters to:
- SN88 miners who submit trading strategies and need to pick winners
- Stakers deciding where to allocate TAO
- Validators determining which subnets deserve weight
- Retail trying to surf subnet momentum
I had the data infrastructure — every subnet's price, emission, stake, volume at every block back to 2025-03. But raw data isn't a signal. A signal has to pick winners before the move, not after.
What doesn't work
I tested obvious strategies first:
- Weekly top-3 momentum rotation — scored 0.00. Buying last week's winners means buying at peaks. Classic whipsaw.
- Inverse momentum (mean reversion) — also 0.00. Cheap alphas kept getting cheaper.
- 30-day momentum with low-volatility filter — 0.00. Longer horizon, same result.
- Static concentration in one "quiet" subnet — 0.00. Quiet subnets are quiet for a reason.
A year of Bittensor alpha markets has been too choppy for any rotation strategy to survive.
What does work
I profiled the actual top-5 winners from 2025-05-01 → 2026-04-21:
- SN67: +1,088% — tao_in 380, price 0.001τ
- SN97: +2,825% — tao_in 700, price 0.002τ
- SN71: +1,304% — tao_in 492, price 0.0005τ
The pattern wasn't "momentum" or "fundamentals." It was this: small subnets where capital is accumulating but price hasn't moved yet.
Specifically:
- tao_in in bottom 25% of the universe (small, not-yet-discovered)
- Price in bottom 75% (cheap)
- Emission above median (active, not dying)
- 14-day price change < 30% (not already pumping)
Across 88 historical picks that matched this profile, 67 were winners. The filter implicitly catches subnets where emission is growing 21x faster than the universe median, stake is growing 25x faster, and volume is growing 3.5x faster — all while price lags. Coiled springs.
Regime matters
The signal works when established subnets are flat or down. When big subnets rally, capital stays parked in them instead of rotating to small speculative alphas. So the product has a regime filter: only active when SN1's 30-day return is below +5% and the universe-aggregate 30-day return is below +10%.
Today's regime: favorable.
Caveats I'm not hiding
- The filter has been tested on roughly a year of data. That's enough to suggest a real edge, not enough to prove one.
- Dedup risk: if every SN88 miner runs similar logic, scores collapse via SN88's similarity penalty.
- The subnet economy changed visibly in late 2025 — what worked in May 2025 looked different by February 2026. That's why the product is rebuilt around percentile thresholds instead of fixed values.
- This isn't financial advice. It's a signal I'm running with real money, starting tonight.
What's live
- Public track record at signals.subnetaiq.io/signals/track-record — every historical pick, current status.
- Current regime at signals.subnetaiq.io/signals — free to check any time.
- Weekly portfolio — available to SubnetAIQ Validator-tier subscribers.
I'm deploying this week's portfolio on SN88 UID 180 tonight as real-money validation. The backtest says it should land top-20 placement on SN88's scoring. Live data over the next 30-60 days will tell us if the backtest survives contact with reality.
If you're tracking Bittensor subnets, check the tool. If you're mining SN88, this might save you some research cycles.
— SubnetAIQ