Skip to main content

// Section 7.5 · Protocol

How Detection Scales with the Network

3 min7.5Protocol
// 5 of 5 · threat model

Detection cost shapes the security review focus.

// 7.5 · how each layer scales as the network grows

The launch detection mechanisms work at small network size, where nodes are few and individually observable. As the permissionless active node set grows, each detection layer scales differently. The two that scale gracefully defend the most mechanical attacks. The one that does not is backstopped by the strongest economic disincentive.

Five detection mechanisms

// cost per scale dimension · verdict per layer

// 7.5.1 · mechanism

Hash-Commitment Verification

O(1)SCALES GRACEFULLY

One SHA-256 reconstruction per request at the coordinator. Total cost scales linearly with request throughput. At a target of 10,000 requests per second, the coordinator performs 10,000 SHA-256 verifications per second; a modern server handles this in under one core of CPU. The mechanism is essentially free to scale.

// per request1 SHA-256 reconstruction
// at 10K/s<1 CPU core
// scaling factorLinear with throughput
// 7.5.2 · mechanism

Sampled Redundant Re-dispatch

O(R·t)SCALES WITH THROUGHPUT

A fraction (the sampling rate) of requests runs on a second node. Total cost scales as throughput × sampling_rate. The mechanism trades sampling rate against throughput: at a 10% rate the network does 10% extra work but catches dishonest nodes 10 times faster. The planned calibration target is 5% rate against the permissionless active node set; the rate adapts upward when validation disagreements exceed a threshold.

// target rate5% baseline
// extra cost5% of throughput
// adaptationRate up if disagreements spike
// 7.5.3 · mechanism

Statistical Agreement (Collusion Detection)

O(n²)DOES NOT SCALE

Pairwise comparisons of result patterns across the active node set. This is the detection layer that does not scale well: a 10,000-node network would require ~50 million pairwise comparisons per epoch. A planned clustering-based approximation reduces the cost to O(n log n) at the price of slightly weaker detection on small colluding groups. The approximation is acceptable because the economic boundary already makes small-group collusion unprofitable: a flagged node is dropped from the active set and earns nothing while its stake sits idle.

// at 10K nodes~50M pairs/epoch
// plannedO(n log n) clustering
// backstopForfeited earnings (stake idle)
// 7.5.4 · mechanism

Behavioural Fingerprinting (Sybil Detection)

O(r)SCALES GRACEFULLY

A constant-time hardware/IP/network-fingerprint check at registration plus a periodic re-check during operation. Scales linearly with the registration rate, not with the active node set. A 10% monthly churn at a 10,000-node steady state means 1,000 registrations per month, well within the budget of a fingerprinting service running as a background coordinator process.

// per registrationConstant time
// at 10K nodes~1K regs/month
// scaling factorLinear with registration rate
// 7.5.5 · mechanism

Reputation Tracking

O(n)SCALES GRACEFULLY

Incremental update of a per-node uptime and validation-success counter per request. Memory cost scales linearly with network size; CPU cost is negligible. The reputation snapshot publishes on-chain at epoch boundaries (planned design) which adds a fixed per-epoch L1 gas cost independent of throughput.

// memoryLinear with network
// CPUNegligible per request
// on-chain costFixed per-epoch L1 gas

The synthesis

// the network is balanced for its target operating range

Two layers that scale defend the mechanical attacks. The one that does not is backstopped by the strongest economic disincentive.

The two detection mechanisms that scale gracefully (hash verification, fingerprinting) are the same two that defend against the most mechanical attacks. The one mechanism that does not scale gracefully (collusion-detection via pairwise agreement analysis) is the one backstopped by the strongest economic disincentive: a flagged node is dropped from the active set and earns nothing while its capital stays locked.

Security review should focus on the calibration of the sampling rate (mechanism 2) and the collusion-detection approximation (mechanism 3). These are the two parameters that move as the network grows.