Toy-scale replica of the model in Min-Kyung Chae, Woo-Sik Son and Sang Hoon Lee, Scientific Reports (2026), doi:10.1038/s41598-026-66867-6
A few hundred synthetic residents live in households, go to classrooms and workplaces, meet friends and bump into strangers across four districts. An emerging infection spreads through those contacts while investigators trace confirmed cases. Make the tracing leaky in two different ways and watch how the outbreak and its transmission tree change.
Repeat the outbreak many times at each omission rate on this town, with every other setting as it is above. Infector omission sweeps 0–80% with contact omission off; contact omission sweeps 0–90% with infector omission off. Bands show the middle 90% of runs.
Kept from the paper. Five contact layers (household daily, classroom and workplace on weekdays, friends and local community each with a 1-in-7 daily chance), age-homophilic Barabási–Albert friendships (h = 0.9), and around 28% of workers and 9% of students commuting to another district. Disease follows an extended SEIR model: gamma-distributed latent period (shape 1.926, scale 1.775), infectiousness starting two days before symptoms and following a viral-shedding curve (mean 3.067, sd 2.109), 20% asymptomatic, recovery after 8 days, and per-contact transmission 1 − exp(−β·t·φ) with contact durations drawn from the Korean survey distributions.
Tracing protocol. Half of symptomatic people self-quarantine and get tested. A positive result means seven days of isolation and a trace of everyone met in the previous ten days; those contacts quarantine, are tested on release, and positives start a new round. Untraced cases are isolated but never investigated; missed contacts simply carry on.
Simplified. The paper simulates 9.5 million people in Seoul and 3.3 million in Busan; here a town of a few hundred stands in, so absolute thresholds (about 4% IO for Seoul, 10% for Busan) do not transfer. Classrooms are grouped in three-year age bands, friend gatherings are pairwise, the relative-infectiousness scale is absorbed into the transmissibility slider, and a test only detects an infection at least two days old.
What to look for. Raising infector omission lengthens transmission chains and grows the outbreak faster than an equal rate of missed contacts, because one untraced case hides a whole downstream branch while one missed contact hides a single link. Try a one- or two-day tracing delay to see timeliness overwhelm completeness, as in the paper's supplementary analysis.