Neuron registration trends
The alpha price in τ, day by day, and what moved it each way.
Log change over the last 30 days −28.30%: protocol buys +0.01%, root sells −6.10%, users net −6.78%, cross-subnet −15.44%, unattributed −0.0002%.
The chart follows the alpha price in τ. Each period starts at its opening price and steps through the mechanisms below, in a fixed order, to its closing price: a step that rises pushed the price up, one that falls pushed it down, and its height is how far it moved the price. A thin line carries each close into the next period’s open; it is dashed where the two differ. The faint candle behind each period is the market’s own open, high, low and close, for context.
With many periods on screen (90 days, or hours), the steps would be too thin to read, so the chart draws the price as a line instead and colours each period by the mechanism that moved it most. Hover a period for the full step-by-step breakdown.
Not here, because they do not move the spot price: the per-block TAO injection into the pool, and burns.
All figures are log changes, not simple percentage changes, so the parts add up exactly to the total; on big moves the two differ (a 10% log fall is a 9.5% price fall).
2026-11-09 UTC, as of ≈17:28 UTC (block 9,470,129)
The day isn’t over yet, so these figures keep changing. They are in no daily total, average or rank.
Forecasting models are typically evaluated on synthetic or limited test datasets, leaving their performance on real held-out data uncertain. Generators that produce plausible patterns may not transfer to actual forecasting tasks. Teams lack a controlled, repeatable way to test whether their synthetic data improves real-world forecasts at scale.
A decentralized network solves this by establishing a fixed model contract, pinned seeds, and held-out evaluation pool that all contributors train against simultaneously. Miners submit deterministic generators or fine-tuning recipes; independent validators score them on the same real data and public benchmarks. Results are reproducible and comparable because the training context—model weights, learning schedule, starting checkpoint—never changes between submissions.
Cascade (Bittensor subnet 91) implements this as three parallel competitions: Lane A tests full-model generators on a 1.23B-parameter mixture-of-experts base; Lane B improves selected expert rows under the same model; Lane C fine-tunes a pinned Timestral base for market forecasting.
Miners compete by funding GPU training on their own accounts; validators independently verify results and set chain-based rewards. The pinned evaluation pool comes from TSBench-Forge, a catalog of real forecasting windows never published during live competition.
Auto-generated from the GitHub repository and may not reflect the latest changes.
Cascade is a Bittensor subnet competition where miners submit synthetic time-series generators and fine-tuning recipes that are tested against real held-out forecasting data. The subnet runs three lanes: Lane A (full-model generator competition), Lane B (selected-expert improvements), and Lane C (MarketCast application fine-tuning). Winners are determined through private pool evaluation and public benchmark gates, with rewards allocated through a ladder system for the reigning king and slot-based payments for B/C contributions.
Cascade consists of miners (submit generators/recipes), a trainer (owner-operated, manages GPU rental and training), and validators (full evaluators + satellites). Miners fund entries through the intake service; the trainer pins an era's contract, seeds, and warm-start init, then prepares data (CPU generation on rented machines for v2 Lane A) and trains on rented GPU pods (8xH100 or 8xH200 on mainnet). Completed legs trigger trainer-signed manifests; full validators score private pools, check public benchmarks, and publish signed receipts; satellites replay receipts without training.
Lane A compares full-model generators on the pinned Cascade MoE Base; Lane B trains selected expert rows under the same base and model; Lane C fine-tunes a separate Timestral base on MarketCast recipes. The controlled-experiment invariant—shared contract, seeds, and init—makes results reproducible and comparable. Rewards are set by validators' independent weight vectors on chain.
V2 scoring uses paired-bootstrap comparisons of CRPS/MASE metrics on private windows at horizons 64/256/720 steps. A challenger must beat the king by the dethrone margin (1%-0.5% in level units for the first 7 days, then 0.10-0.01 in increment units). Public benchmark gates (GIFT-Eval/BOOM/TIME/FEV) can block a private-pool win: the challenger must improve overall, stay within 1% regression per suite, and win at least 50% of tasks.
Lane B qualification requires target-specific gains and regression guards; acceptance into a warm-start init is required before payment begins. Lane C MarketCast uses forward-only windows from future market data with calibration checks; an initial pass triggers pending confirmation, which must also win the slot. Rewards allocated: A 100% (planned A/C 75/25, A/B/C 50/25/25), with A's pool split between the ladder-positioned king, immediate former king (5% for 36,000 blocks), and B/C slots.
Unpaid/expired allocations burn.
The sections above were auto-generated from the GitHub repository and may not reflect the latest changes. View repository (github.com, opens in a new tab, external site)
SOTA Time Series Foundation Models
This information was auto-extracted from the GitHub repository and may be incomplete or out of date. Always refer to the subnet's GitHub repository (github.com, opens in a new tab, external site) for the latest instructions.
Not yet verified: GPU model, RAM, CPU cores, Network, Notes
Not yet verified: Network, Notes
Machine-extracted from the subnet's repository docs; may be wrong or out of date. Check the repository before buying or renting hardware. Open the repository (github.com, opens in a new tab, external site) · Extracted 9 Nov 2026
Clone the Cascade v2 repository
git clone --branch cascade-v2 https://github.com/TensorLink-AI/cascade && cd cascade
Install miner dependencies with hippius (Hub) and chain (Bittensor) extras
pip install -e '.[hippius,chain]'
Verify your generator code layout, imports, and deterministic output
cascade verify ./my-generator --live
Create a Bittensor wallet and hotkey
btcli wallet new-coldkey --wallet-name my-miner && btcli wallet new-hotkey --wallet-name my-miner --wallet-hotkey gen1
Register your hotkey on mainnet subnet 91 (or testnet netuid 259)
btcli subnets register --netuid 91 --network finney --wallet-name my-miner --wallet-hotkey gen1
Submit your generator privately with an API key set (export LIUM_API_KEY=sk-...)(optional)
cascade submit ./my-generator https://submissions.cascadesub.net --wallet-name my-miner --wallet-hotkey gen1 --label my-gen-v1 --yes
Track the exact entry and its scientific progress
cascade status --hotkey YOUR_SS58 --lane a --ref IMMUTABLE_REF --intake https://submissions.cascadesub.net --watch
Clone the main Cascade repository
git clone https://github.com/TensorLink-AI/cascade && cd cascade
Install validator dependencies using uv with train, hippius, and chain extras
uv sync --frozen --extra train --extra hippius --extra chain
Verify TLS certificates are properly configured
python scripts/check_tls_trust.py
Create a Bittensor wallet and hotkey for the validator
btcli wallet new-coldkey --wallet-name my-validator && btcli wallet new-hotkey --wallet-name my-validator --wallet-hotkey default
Register your validator hotkey on mainnet subnet 91 (or testnet 259)
btcli subnets register --netuid 91 --network finney --wallet-name my-validator --wallet-hotkey default
Set environment variables for storage and Hub access credentials
export HIPPIUS_S3_ACCESS_KEY=... HIPPIUS_S3_SECRET_KEY=... HIPPIUS_HUB_USERNAME=... HIPPIUS_HUB_PASSWORD=...
Start the validator process (run under a process manager like systemd)
cascade-validator --chain-toml chain.toml --network finney --wallet-name my-validator --wallet-hotkey default
| Day (UTC) | Share |
|---|---|
| 2026-11-08 | 0.0048% |
| 2026-11-07 | 0.0043% |
| 2026-11-06 | 0.004% |
| 2026-11-05 | 0.004% |
| 2026-11-04 | 0.0039% |
| 2026-11-03 | 0.00% |
| 2026-11-02 | 0.00% |
| 2026-11-01 | 0.00% |
| 2026-10-31 | 0.01% |
| 2026-10-30 | 0.06% |
| 2026-10-29 | 0.07% |
| 2026-10-28 | 0.09% |
| 2026-10-27 | 0.08% |
| 2026-10-26 | 0.06% |
| 2026-10-25 | 0.05% |
| 2026-10-24 | 0.04% |
| 2026-10-23 | 0.05% |
| 2026-10-22 | 0.06% |
| 2026-10-21 | 0.05% |
| 2026-10-20 | 0.05% |
| 2026-10-19 | 0.06% |
| 2026-10-18 | 0.07% |
| 2026-10-17 | 0.08% |
| 2026-10-16 | 0.09% |
| 2026-10-15 | 0.11% |
| 2026-10-14 | 0.12% |
| 2026-10-13 | 0.12% |
| 2026-10-12 | 0.13% |
| 2026-10-11 | 0.13% |
| 2026-10-10 | 0.15% |
| 2026-10-09 | 0.14% |
| 2026-10-08 | 0.12% |
| 2026-10-07 | 0.11% |
| 2026-10-06 | 0.13% |
| 2026-10-05 | 0.14% |
| 2026-10-04 | 0.15% |
| 2026-10-03 | 0.19% |
| 2026-10-02 | 0.15% |
| 2026-10-01 | 0.11% |
| 2026-09-30 | 0.11% |
| 2026-09-29 | 0.10% |
| 2026-09-28 | 0.12% |
| 2026-09-27 | 0.13% |
| 2026-09-26 | 0.22% |
| 2026-09-25 | 0.20% |
| 2026-09-24 | 0.18% |
| 2026-09-23 | 0.17% |
| 2026-09-22 | 0.23% |
| 2026-09-21 | 0.29% |
| 2026-09-20 | 0.31% |
| 2026-09-19 | 0.31% |
| 2026-09-18 | 0.34% |
| 2026-09-17 | 0.37% |
| 2026-09-16 | 0.42% |
| 2026-09-15 | 0.46% |
| 2026-09-14 | 0.52% |
| 2026-09-13 | 0.56% |
| 2026-09-12 | 0.61% |
| 2026-09-11 | 0.60% |
| 2026-09-10 | 0.58% |
| 2026-09-09 | 0.64% |
| 2026-09-08 | 0.79% |
| 2026-09-07 | 0.90% |
| 2026-09-06 | 0.92% |
| 2026-09-05 | 1.00% |
| 2026-09-04 | 0.98% |
| 2026-09-03 | 0.96% |
| 2026-09-02 | 0.86% |
| 2026-09-01 | 0.50% |
| 2026-08-31 | 0.40% |
| 2026-08-30 | 0.39% |
| 2026-08-29 | 0.47% |
| 2026-08-28 | 0.50% |
| 2026-08-27 | 0.55% |
| 2026-08-26 | 0.56% |
| 2026-08-25 | 0.55% |
| 2026-08-24 | 0.56% |
| 2026-08-23 | 0.57% |
| 2026-08-22 | 0.58% |
| 2026-08-21 | 0.53% |
| 2026-08-20 | 0.53% |
| 2026-08-19 | 0.56% |
| 2026-08-18 | 0.50% |
| 2026-08-17 | 0.57% |
| 2026-08-16 | 0.56% |
| 2026-08-15 | 0.57% |
| 2026-08-14 | 0.63% |
| 2026-08-13 | 0.75% |
| 2026-08-12 | 0.81% |
| 2026-08-11 | 0.89% |
