Bonus
One shard's weight doubles. Nodes there earn roughly 1.8× for the hour, paid for by everyone else.
$SHARD
Run nodes on NEAR. Each one sits on a shard, wears out a little every hour, and earns $SHARD by the second while you keep it alive.
- Shards
- 8
- Halving
- 30 d
- Supply cap
- 1,000,000,000
- Wear
- −2% / d
- Clock
- block_timestamp_ms
A night on shard 3
$SHARD is a fork of the idea behind $GPU, moved to NEAR and made to feel like a small hosting business. You buy hardware, choose where it lives, and decide when it's worth paying to keep it running.
- S3 congested again. Moved nothing, waited it out. Cheaper than migrating.
- Reseated N-117. Uptime was down to 81%. Paid the upkeep to bring it back.
- Someone parked four racks on S5 overnight. My share there halved. Noted.
- Bonus rolled on S2. Not mine. Of course.
Hardware
Every node is an NFT (NEP-171). Bigger hardware earns more, but less than you'd think: rewards follow the square root of compute, so one rack is worth eight boards, not sixty-four.
NodeComputeReward weightMintMove shard
PISingle-board. A board on a shelf. Cheap to run, cheap to fix.1√1 = 15 NEAR3 days of rewards, min 500
1URack server. Two fans, one job.4√4 = 218 NEAR3 days of rewards, min 1,000
GPU-4UGPU chassis. Six cards and a noise complaint.16√16 = 460 NEAR3 days of rewards, min 2,000
RACK-42UFull cabinet. Liquid-cooled. Costs the most to keep at 100%.64√64 = 8200 NEAR3 days of rewards, min 4,000Mint is priced in NEAR. The contract also accepts an Ondo tokenized NVDA if one ever becomes holdable by a NEAR contract; today it isn't, so NEAR it is. Payments go to the treasury.
Earnings
Emission is split evenly across the eight shards, then shared inside each shard by reward weight.
Events
Once an hour, anyone can roll the network. The roll picks one of three things, or nothing, and it lasts until the next roll.
Congestion
One shard's weight halves. Nothing breaks. It just pays less until the next roll.
Reshard
Every shard draws a new weight from a fixed uneven set, from 1.6× down to 0.5×. Your quiet shard can end up the busiest one.
Wear and upkeep
Uptime drops 2 points a day from the last time you touched the node. Rewards scale with the average uptime since then. What a worn node doesn't earn is never minted.
- Maintain
- Back to 100%. Each missing point costs 0.15 days of that node's rewards (at least 2 × weight $SHARD), burned. From 0%, that's about 15 days of earnings.
- Upgrade
- Move up one tier. About 20 days of what the extra weight earns (at least 1,000 × new compute $SHARD), burned.
- Migrate
- Move to another shard. Three days of the node's rewards (at least 500 × weight $SHARD), burned.
- Neglect
- At 0% a node keeps its seat on the shard and still dilutes everyone there. It just earns nothing.
Emission halves every 30 days and stops at the 1B cap. Every way to earn more (upkeep, bigger hardware, a better shard) costs $SHARD that is destroyed, so there's no loop that farms forever.
Where the dice come from
NEAR gives contracts env::random_seed(). Validators can't choose it, but the one producing a block can decide not to produce it, and every call in the same block sees the same seed. So a roll takes two blocks: commit_event locks the first seed, reveal_event mixes in a later one. Both are open to anyone and pay 1 $SHARD to whoever calls them (from a pool that takes 0.1% of emission), so nobody can sit on a roll they don't like.
The full analysis, and what it still doesn't protect against, is in docs/PHASE0.md.