Can Computing Power Become a New Form of Money?

Chapter 1 — What is being measured?

Separate four things that marketing merges: compute resources (machines, energy), computational work (operations performed), useful outputs (results somebody needed), and verifiable outputs (results a stranger can check cheaply). Contribution accounting turns verified outputs into reward units; credit is an internal ledger concept, money is a social institution — a protocol can mint the former and merely apply for the latter. A CPU cycle has no universal price: its value depends on the task, the deadline and the cost of checking the answer.

Chapter 2 — Existing approaches, honestly compared

Chapter 3 — A useful-compute protocol (hypothetical)

State transitions: REGISTERED → ASSIGNED → SUBMITTED → VERIFIED → SETTLED, with DISPUTED reachable from SUBMITTED. Trust assumptions are printed on the box: task specs must be deterministic or statistically checkable, result formats canonical (byte-identical duplicates detectable), and every transition logged for audit. RESEARCH HYPOTHESIS: such a protocol can reward useful work. Everything after this chapter is the attempt to break that hypothesis before reality does.

Chapter 4 — The verification problem

Verification strategies form a cost ladder: recompute everything (perfect, pointless), replicate k-of-n with majority vote (collusion-sensitive), spot-check probabilistically (statistical guarantees), cryptographic commitments + reveal (binds, doesn't validate), zero-knowledge proofs of correct execution (powerful, heavy prover cost, statement-specific), trusted hardware (shifts trust to a vendor). ESTABLISHED constraint: for some tasks no cheap verification is known at all — and there the protocol's economics die, however elegant the token design.

Chapter 5 — Threat model

Chapter 6 — Economic model

# Toy viability model — illustrative numbers only, no market claims.
def net_value(task_value, compute_cost, verify_cost):
    return task_value - compute_cost - verify_cost

print(net_value(100, 30, 10))   # 60: viable in the toy (executed)
print(net_value(100, 30, 90))   # -20: verification ate the protocol (executed)

A contribution is viable when task value > compute cost + verification cost + network cost, with the reward sized so honest expected value beats fraud expected value after detection probability and penalties. Drag the sliders in the lab: the moment verification cost crosses the line, no tokenomics vocabulary can resurrect the protocol. That inequality — not the consensus algorithm — is the research contribution to check first.

Chapter 7 — Python simulation

import hashlib, random
random.seed(42)  # deterministic, reproducible

def result_hash(task_id, answer):
    return hashlib.sha256(f'{task_id}:{answer}'.encode()).hexdigest()

seen = {}
def submit(worker, task_id, answer):
    h = result_hash(task_id, answer)
    if h in seen:
        return f'rejected: duplicate of {seen[h]}'
    seen[h] = worker
    return 'accepted'

print(submit('honest-1', 't1', 42))
print(submit('copycat', 't1', 42))  # rejected: duplicate (executed pattern)

Chapter 8 — Cryptographic and protocol limitations

ESTABLISHED, and constantly forgotten: a hash proves data matches data; a signature proves a key signed; a commitment proves you cannot change your answer later; a zero-knowledge proof proves one precisely stated claim under stated assumptions. None of them, alone, proves a useful computation was performed correctly. Protocols fail exactly where these get conflated.

Chapter 9 — Deployment and governance

Open governance questions dominate any real deployment: who verifies the verifiers, how disputes resolve, how upgrades ship without breaking audit history, what privacy task data gets, which regulations a reward unit triggers, and how abuse gets excluded without centralizing the gate. Conclusion: before useful computation can function like money, someone must demonstrate cheap verification for a valuable task class, Sybil resistance that survives wealthy adversaries, and an economic loop where value enters from outside. Until then: a beautiful research problem, modelled honestly below.