Author: Lucas Luhur Last updated: 2026-06-26
This is the theory write-up for build-order steps 3–4 (consensus → broadcast network → adversaries → measures of anonymity). The two are combined because, a privacy property = an attack + a measure.
When a node wins a slot it broadcasts a block proposal; the global passive adversary (GPA) watches the whole network but cannot see inside an honest node or break crypto. The same observed trace supports two distinct anonymity questions on two timescales, each a property + attack + measure:
| Property | Question | Attack | Measure |
|---|---|---|---|
| Stake privacy | which node holds high stake? | stake inference | confidence, time-to-link, Jaccard |
| Sender unlinkability | who proposed this block? | message attribution | anonymity entropy, FIFO, correlation / intersection |
Everything is a timing/rate attack on the observable traffic pattern, under ideal crypto, and the model is layer-agnostic — it consumes whatever trace the (future) anonymisation layer produces.
| Parameter | Symbol | Meaning | Value |
|---|---|---|---|
| Observation fraction | $q$ | fraction of slots the GPA observes ($q=1$ = global view) | 1.0 |
| Accuracy | $γ$ | half-width of the target stake interval $[α(1±γ)]$ | 0.1 |
| Confidence threshold | $δ / θ$ | target confidence for time-to-link | 0.5 |
| Compromised fraction | $β$ | fraction of corrupted (observe-only) nodes — not yet coded | 0 (pure GPA) |
| Active-slots coefficient | $f$ | per-slot win rate → block time (inherited) | 1/30 |
| Epoch length | $T$ | slots per epoch = stake-inference window (inherited) | 388,800 |
| Pareto shape | $k$ | stake inequality (inherited) | 2 |
$γ, δ/\theta$ have defaults set (Resilience and Anonymity, pp.10–12).
Everything the adversary reasons about is derived from one object: a trace, a timestamped record of network activity.
The system hides exactly two things:
A broadcast as an event set: A single broadcast from source $s$ at slot $t$ is an infection process: node $v$ receives it at (Modelling of Network Delays):
$$ \tau_{s \to v} = t + \Delta_{sv}, $$
where $t$ is the winner slot and $Δ_{sv}$ the weighted shortest-path delay on the quenched random regular graph. At most the adversary records the full set of timestamped link activations — which edge fired when.