Mining & models

NEB miners

Every miner lock we see on-chain, ranked by its share of blocks in the last 24 hours.

Miners

Share measured from block coinbases over the last 24h · 1,464 blocks

100.0% unlabelled
Concentration · 24hTop miner32.2%Top 386.6%Nakamoto2Effective miners3.9Active locks13

Shares and block counts are computed by us from each block’s first coinbase output (the miner lock). A pool pays from one lock, so one row can be many miners. A miner showing “—” has not found a block in the last 24h. Run a pool or a known miner? Send us your payout address for config/labels.json and the row gets your name.

Models trained by mining

NEB’s proof of work is a training step of a small neural network. The chain holds two models and stores their weights.

Next block trains
Slot 0
kepler-v2 · height 3,356
Shares per block
19.1
average of the last 100 blocks
Weights per model
26,000
4-bit integers, −8 to 7 (KEP1)
Combined state hash
d3f137e0…a061b3
both models at the tip
Slot 0

Kepler transit signals

even heights

Light curves of stars watched by the Kepler space telescope, each folded around a periodic dip in brightness. The model learns to tell a planet candidate from a false positive (13,404 labelled signals).

Dataset
kepler-v2
Weights hash
a75a7a9070…87ad5a49
State at
block 3,355
Weights
26,000 (4-bit, 13,000 bytes packed)
Changed since genesis3,345 / 26,000 · 12.9%

Layer 5 weights (32 × 768)

Hover or tap the map to read a weight.
Slot 1

TESS eclipse signals

odd heights

Signals from the TESS mission (Astronet-Triage-v2, Tey et al. 2023), in the same two views. The model learns to tell a real eclipse from junk (2,363 eclipses, 4,726 junk signals).

Dataset
tess
Weights hash
344f33db43…7894fc8f
State at
block 3,355
Weights
26,000 (4-bit, 13,000 bytes packed)
Changed since genesis2,973 / 26,000 · 11.4%

Layer 5 weights (32 × 768)

Hover or tap the map to read a weight.

How NEB mining works

In Bitcoin an attempt at a block is one hash. In NEB an attempt is one gradient computation: the miner hashes the block header, uses the hash to pick 16 samples from a public dataset, runs them through the current model and computes how each weight should move to do better on them. The gradient is hashed with the header into a ticket; a ticket below the target finds the block. One attempt costs about 2 ms of a CPU core.

Attempts that miss the block target but meet a 32 times easier one are shares. The next block carries up to 256 of them, and their gradients, summed, are applied to the model as one training step. The work spent searching for blocks is not thrown away: it trains the model, and every node replays the training from genesis and gets the same weights, bit for bit (integer arithmetic only).

Blocks alternate between the two models: an even height trains slot 0, an odd height slot 1. Both are the same small convolutional network (KEP1, 26,000 weights of 4 bits). Difficulty follows ASERT towards one block every 60 s. The block reward is 20 NEB, halving every 450,000 blocks, and 5% of it goes to the project share lock. See the hashrate and economics pages.

Specification: spec/shares.md and spec/nebhash-v2.md of the NEB repository. Model data from the node’s getmodel and getinfo.

Miner shares are measured by us from block coinbases. NEB Explorer is not affiliated with any miner or pool.