Reputation
tasqnetwork.io/benchmark/reputation
Each period a node's reputation moves toward its observed success rate with decay μ = 0.9. A node below Rmin = 0.8 stops receiving work. The question is whether that rule removes bad nodes quickly without punishing honest ones that occasionally fail. 20,000 nodes, 300 periods.
0%Excluded at a 5% fault rate
20Median periods to exclude at 20% faults
7Median periods to exclude at 40% faults
0.065%Excluded at a 10% fault rate
Exclusion by fault rate
| Fault rate | Share excluded | Median periods | 95th percentile |
|---|---|---|---|
| 0% | 0.000% | never | never |
| 2% | 0.000% | never | never |
| 5% | 0.000% | never | never |
| 10% | 0.065% | 187 | 294.59999999999997 |
| 20% | 100.000% | 20 | 33 |
| 40% | 100.000% | 7 | 8 |
Source: S4_reputation.csv, sim_protocol.py
There is a sharp boundary between 10% and 20% faults. That is the intended behaviour for these parameters, but it means a node with an unreliable network link around 15% could be excluded. Final values of μ and Rmin will be set from testnet data.