loading...
Starting.
Not run yet. Press Run Monte Carlo.
running...
The science
A beekeeper's inspection is a chain of small steps: look at the entrance, find the queen or fresh eggs, read the brood pattern, check for queen cells, heft the stores, wash a sample of bees for varroa mites, make room, clean the tool. Each step is easy and each is occasionally forgotten. This lab runs two identical yards with the same hidden events (queen failures, swarm preparation, foulbrood, a spring dearth, a varroa epidemic) and lets you design the procedure used in yard B, while yard A relies on memory.
Skipping a step. From memory the chance of skipping step i is pi = pmem · wi · (1 + f · x) with pmem the forgetfulness slider, wi a difficulty weight (the tedious mite wash and tool cleaning are 1.4 and 1.5, a glance at the entrance 0.6), f the fatigue slider and x the position of the hive in the row (0 for the first, 1 for the last). A written list removes a share c (discipline) of the forgetting: pilist = pi (1 − c) and a step you leave off the list is never done. A chance of chronic omission, 1.5 pmem (wi − 0.45), models habit: the same unloved step dropped on every visit all season. The probability of a perfect visit of k steps is the product of (1 − pi), the reliability argument of Gawande's Checklist Manifesto.
Finding a problem. A step finds an existing problem with sensitivity s (0.8 to 0.95). The chance to find it on one visit is q = (1 − p)s, so the number of visits until detection is geometric with mean 1/q; with a 14-day interval a 20% omission rate already adds about three weeks. Once found, the action is applied: new queen after 14 days, queen cells removed, 5 kg of syrup, mite treatment (×0.1), or destroying a foulbrood colony so the disease cannot spread.
The colonies. Per hive and per day, with B bees, M mites, S kilograms of stores and m = 100 M/B mites per 100 bees:
dB/dt = L(t) − (μ0 + κ m) B L(t) = Lmax sin2(π(d + 20)/280), Lmax = 1800 eggs/day, μ0 = 1/35, κ = 0.002
dM/dt = r (0.3 + 0.7 L/Lmax) M (doubling time ln 2 / r, about 32 days at r = 0.022)
dS/dt = g F(t) B/1000 − (c0 + c1 L/Lmax) B/1000
F(t) is the nectar flow (two spring flows and a late one, cut to nearly zero during the dearth). Foulbrood spreads by robbing at 0.0015 per infected hive per day, and by a dirty hive tool with probability 0.2 when the clean-tool step is skipped after an infected hive. A colony is lost when bees fall below 1500, mites pass 18 per 100 bees, or it stays queenless for 60 days.
Monte Carlo. One season is one random draw. The button repeats the whole season 50 to 500 times with the same hidden events in both yards and reports the mean colonies alive, honey and net value with 95% intervals (1.96 standard errors), and the distribution of survivors. The net value is 9 per kg honey above a 15 kg reserve plus 180 per living colony, minus treatment and feeding costs.
Honest simplifications. The numbers are stylised, not field data: every rate is chosen to give plausible orders of magnitude (a mite doubling time of a month, a sixth of queens failing in a season) rather than fitted to a region. One season has no winter, no drifting between hives, no climate, and the beekeeper's skill is a single probability. Real losses depend on much more; the model only isolates what a procedure changes. Mite threshold, treatment efficacy and the reserve are common textbook values used as defaults.
Checks run in the page. (1) The probability that a memory inspection skips nothing, simulated against the product of (1 − pi). (2) The mean number of inspections to find a missing queen against 1/q. (3) The daily varroa update with sub-steps against the exact exponential. (4) A symmetry test: with checklist discipline 0 and every step on, yard B must match yard A within three standard errors over 160 paired seasons.