One function per component of the Python pipeline, in the canonical order
fixed by model.py:566-579:
Usage
sim_phase_susceptible(state, par, ctl, tick)
sim_phase_census(state, par, ctl, tick)
sim_phase_exposed(state, par, ctl, tick)
sim_phase_recovered(state, par, ctl, tick)
sim_phase_infectious(state, par, ctl, tick)
sim_phase_vaccinated(state, par, ctl, tick)
sim_phase_human_to_human(state, par, ctl, tick)
sim_phase_env_to_human(state, par, ctl, tick)
sim_phase_environmental(state, par, ctl, tick)Details
Susceptible -> Exposed -> Recovered -> Infectious -> Vaccinated -> Census
-> HumanToHuman -> EnvToHuman -> Environmental -> DerivedValues
The order is semantically load-bearing and is reproduced rather than
rationalised. Infectious reads E[tick + 1] after
Exposed has written it; HumanToHuman reads N[tick] but
writes into S[tick + 1], which Census has already summed. That
asymmetry is real. Port the behaviour, not the intent – anything that
looks wrong gets an issue, not a local fix.
Each function takes (state, par, ctl, tick) where tick is
0-based as in Python, and writes into row tick + 2 of the state
series (R's 1-based row for Python's tick + 1).
State rows are environments (see sim_alloc_state()), so each phase
binds the rows it needs once – rh for here, rn for
nxt – and then reads and writes channels by name: rh$S,
rn$E <- .... The two exceptions are the lagged reads, which address
an arbitrary historical row and so spell it out as
state$rows[[probe]]$Isym. Writing through rh/rn
mutates the state in place; there is no write-back step.