About dTaoAnalytics
Private research on Bittensor subnet allocation.
dTaoAnalytics is a private, not-for-profit workbench on its own Bittensor data pipeline. It compares subnet opportunity, liquidity, evidence, costs, and risk in one place. Public pages stay open for baseline context; fuller depth is invite-only for people the site creator works with.
Why “dTaoAnalytics”?
The lowercase d stands for dynamic TAO — the per-subnet token model Bittensor introduced when it shipped dTAO in April 2024. Every subnet now has its own alpha token, denominated in TAO via an AMM pool. That structural shift is what makes subnet allocation a real investment problem, and what these analytics are built to examine.
Evidence first
Signals need measured forward evidence before they deserve trust. Narrative alone is not enough.
Liquidity aware
A subnet can be interesting and still too thin to trade at the size an investor needs.
Cost honest
Strategies are benchmarked, then held to a net-of-cost bar. Slippage, fees, turnover, and capacity limits all count before claiming alpha.
Why this exists
Buying TAO is simple. Choosing subnet exposure is not.
Subnet tokens are specific bets inside Bittensor. Each one has a different pool, emission profile, builder path, liquidity depth, and risk surface. A table of prices is not enough when the real question is whether an allocation survives evidence, tradeability, costs, and benchmark comparison.
The app is designed to turn those checks into a repeatable workflow: start with the market, inspect one subnet, validate signal evidence, compare against benchmarks, and monitor what is deteriorating.
Provenance
Data should carry its source with it.
A dedicated Subtensor node anchors the pipeline, with block-level context where practical. Pool state, emissions, metagraph snapshots, and derived indicators should be traceable back to chain data, not treated as detached dashboard numbers.
That provenance is part of the discipline: if a page makes the analyst think differently, it should also show where the data came from and how fresh it is.
Beyond raw chain data, every agent decision — portfolio changes, individual trades, the thesis behind each entry — is labeled with the data context that produced it and later matched against the outcome that followed. That labeled outcome ledger compounds daily. It is what lets the methodology be tested rather than asserted, and what makes it possible to say which signals actually predict subnet returns and which do not.
Two layers, one leaderboard
LLM agents and mechanical strategies compete on the same evidence.
A small fleet of LLM agents reads on-chain data and writes paired theses, portfolio changes, and invalidation criteria into an outcome ledger. Eight weekly allocators work alongside two daily watchers, plus shadow variants that strip out one variable at a time (memory, coaching, or cadence) so each piece's contribution can be measured. A coaching layer cross-checks them weekly against measured performance and writes structured notes the allocators consume on their next run.
Alongside that fleet, mechanical strategies and passive benchmarks track the same TAO units under the same cost assumptions. The point is not to crown one layer in advance. The point is to let both run honestly and let the leaderboard answer the question. A model is worth following only when it clears mechanical baselines net of slippage, fees, turnover, and capacity limits. The reverse is also true: a mechanical baseline is worth following only when it survives in live conditions, not just in a clean backtest.
Underperformance is shown the same way wins are: in TAO, dated, with the methodology attached. The point is the contest, not a verdict.