The money verb · demand response
Demand-response programs pay you to drop megawatts on a deadline. The only question is which machines — and answering it by rule of thumb under a fifteen-minute clock loses hashrate every event. We pre-compute the provably fewest machines to shed for each megawatt target you might be called for, must-runs honored, so the call becomes a lookup.
Send your last event The Playbook — $25K
Each tier is a certified answer: the fewest units whose shutdown meets the target, with the witness to prove no smaller set does. The full unit list ships to you, never to the site.
Where we sit in your stack
Your tools decide when and execute. Nobody computes — and signs — the assignment. That's us.
Keep Foreman. Keep Braiins. They're your hands. Your monitoring watches the fleet; your trigger tool fires when the price signal hits; your firmware tunes each chip. None of them answers the one question the payment turns on: of all the ways to drop the megawatts you were called for, which one loses the least hashrate — and can you prove it? That answer, with its proof, drops straight into the API your team already uses. Nothing of ours ever touches a machine.
Why rule-of-thumb leaves money on the table
≈ 2⁴¹⁰ ≈ 10¹²³ ways
A fleet of a few thousand machines has more distinct on/off combinations than there are atoms in the observable universe. Under a fifteen-minute deadline your team can only reach for a habit — shed the biggest room, shed the oldest models — and habits leave hashrate on the table on most events. We built an algorithm for this problem: it returns the provably best combination for each target, with a certificate you check in a spreadsheet. You don't take our word for which one is best — you verify it.
Season benchmark · computed, reproducible
Thirty curtailment events across one synthetic season, one synthetic fleet, three strategies. Hashrate lost is the quantity; less is better. The certified optimum is never beaten and the certificate tells you, per event, exactly how much more the others gave away.
| Strategy | Season hashrate lost | vs. certified optimum |
|---|---|---|
| Binary / block shed (whole rooms) | 28,149 PH·h | +168.7% more lost |
| Ranking greedy (oldest / worst-efficiency first) | 11,814 PH·h | +12.8% more lost |
| FarmBalance certified optimum | 10,476 PH·h | baseline — provably minimal |
Read the other way: the certified optimum lost 62.8% less than block shedding and 11.3% less than ranking greedy, over this season. Never less than your current tool — provably. The certificate tells you how much more, per event.
Definition, attached on purpose: quantity = PH·h of hashrate not produced during the curtailed slices; baseline = the same 30 events and the same fleet file scored under each strategy; fleet and season are synthetic and downloadable so you can reproduce every row. These figures measure the which-machines decision only — not when to curtail, not program economics — and are not comparable to any third-party percentage, which measures a different quantity on a different base.
The free hook · one signed number
Send what you shed on your most recent event — the units you powered down — plus the fleet file. Get back a single signed number: the gap between what your tool did and the provable minimum for that exact event. One number, never a list. If the gap is near zero, the certificate says so — your setup performed optimally, in writing, signed. If it isn't, you now know the size of the problem, proven, before you spend anything.
The Curtailment Playbook · $25K one-time
The Playbook is the full set of certified shed lists for your fleet — one per megawatt tier you designate — delivered before your next event, so the call is a lookup instead of a scramble. Your ops team executes; we never touch a machine. Lists ship in the dialect your tools speak:
Every certified line is covered by the Certified Number Guarantee: if a certified number is wrong, you don't pay. See exactly how you check it →
Farm·Balance