A merchant defends against a double-spend by shipping only after the paying transaction has confirmations (Definition 2.3). This section fixes one convention of the paper’s attack timeline and computes, in the paper’s first-seen model, the exact probability that the attacker overtakes after confirmations.
The paper assumes the attacker holds one block before the race is
scored: “we assume one block was pre-mined by the attacker before
commencing the attack” (§4). The attacker forks from the public leaf
at the moment the merchant’s transaction is broadcast, and times the
broadcast to coincide with a private block he has found but not yet
announced. At the moment the honest chain reaches
confirmations, the attacker who has found further blocks
therefore holds
The convention is a modelling choice: an attacker with no pre-mined
block starts at
By Proposition 3.5, the number of attacker blocks found while the honest network finds its is negative binomial. Averaging the catch-up probability of Theorem 4.4 over gives the success probability under the first-seen tie rule of the specification and of Rosenfeld, which Zebra also applies from version 6.4.0 (Section 8.1). The model does not give an exact rollback probability for the deployed network.
Conditioning on and applying Theorem 4.4 at
valid for
which by Lemma 3.6 is the probability that the
attacker’s -th block arrives while the honest network holds ;
summed over
Under the one-pre-mined-block convention, the catch-up half —
the honest chain reaches first and the attacker later overtakes —
has exactly the same probability as the already-ahead half, in
which the attacker reaches first. A first-to- winner may have
fallen behind and recovered along the way; neither half asserts
otherwise. The double-spend probability is therefore twice the
probability of winning a first-to- race. The identity also covers the
boundary case: at
Take
| product | |||
| catch-up half (sum of products) | |||
| already-ahead half, | |||
The two halves are equal exactly, by Remark 5.3, and the closed form (2) evaluates to the same . For a merchant who ships after six confirmations against a twenty-percent attacker, the model gives a probability of a strict overtake.
The whitepaper’s §11 computes the same quantity with one
approximation and two offsetting conventions: the attacker’s progress
is approximated as Poisson with mean
| for every | ||||||||||