fleet parameter reference

⚠️ Work in progress: not ready for real use. fleet is published early so the approach, and its comparison against malariasimulation, can be examined and argued with, not so that anyone can rely on its numbers: the API is unstable and nothing here has been peer reviewed. The known discrepancies against the IBM are open rather than resolved, the largest being all-age severe incidence, which runs about 4 to 6% below the IBM, alongside a roughly 9% excess on clinical and severe incidence across the 63-country site files that is not yet explained. The full statement is on the package front page; if you need results you can defend today, use malariasimulation.

Every user-facing malariasimulation parameterisation function, argument by argument, as of malariasimulation v3.0.0. Look a function up in the index below; each row says whether fleet reproduces that argument exactly, what it loses if not, and what to do about it.

Arguments named parameters are omitted throughout: they are the list being modified. For the system these arguments feed, see vignette("model"); for what to do differently because it is an ODE, vignette("using").

Legend

Mark	Meaning	What to do about it
✅ Exact	Modelled exactly, or algebraically equivalently for the mean field	Nothing. Differences from the IBM will sit inside its Monte-Carlo noise.
🟡 Approx.	Modelled, but the mean field loses something. The row says what.	Safe for comparing scenarios that share the approximation. Check against the IBM if the number you report is dominated by this mechanism.
➖ Inert	Accepted without error, but has no effect on the run — by design, not by omission	Nothing. Don’t tune it.
⚠ Ignored	Not modelled; the run continues (sometimes with a warning)	Decide whether you can live without it before trusting the run.
⛔ Rejected	Errors at input, or cannot reach the model at all	Remove it from the parameter list.

Where a row carries two marks, the second is a caveat on the first.

Index

Function	Status
get_parameters()	✅
set_species()	✅
set_equilibrium()	✅
set_demography()	🟡
set_drugs()	🟡
set_clinical_treatment()	🟡
set_antimalarial_resistance()	🟡
set_bednets()	🟡
set_spraying()	🟡
set_carrying_capacity()	🟡
set_mda(), set_smc()	🟡
set_pmc()	🟡
create_pev_profile()	✅
set_pev_epi()	🟡
set_mass_pev()	🟡
set_tbv()	🟡
set_epi_outputs()	✅
set_parameter_draw()	✅
Helpers and run functions	mixed

get_parameters(overrides, parasite)

Argument	malariasimulation meaning	fleet
overrides	Named list overriding any default parameter	✅ Passed through wholesale; see the group-by-group breakdown below
parasite	"falciparum" or "vivax"	✅ falciparum. ⛔ vivax: build_inputs() stops with an error

Breakdown of overrides

Group	Parameters	fleet
Initial state proportions	s_proportion, d_proportion, a_proportion, u_proportion, t_proportion	⛔ Ignored. fleet seeds at the malariaEquilibrium fixed point for init_EIR (or, under a custom demography, for the EIR the IBM’s mosquito sizing supports, set_equilibrium()), not at user-supplied proportions
Initial immunity	init_ib, init_ica, init_iva, init_icm, init_ivm, init_id	⛔ Ignored, same reason
Population size	human_population	✅ Scales output counts only; the ODE is per-capita, so run time is independent of it
	human_population_timesteps	⛔ Ignored. Population is conserved; no dynamic population size
Baseline demography	average_age	✅ → $\eta = 1/\text{average_age}$: the constant death hazard and the equilibrium age structure when custom_demography = FALSE
	custom_demography	✅ Switches to the set_demography() path (set_demography())
Biting heterogeneity	a0, rho	✅ $\psi_i = 1 - \rho e^{-a_i/a_0}$
	sigma_squared	✅ Log-normal variance of $\zeta$
	n_heterogeneity_groups	✅ Number of Gauss–Hermite nodes $n_z$. Must be a whole number $\geq 3$: fewer nodes than that are not a quadrature on the log-normal at all, they just move every human off it
	enable_heterogeneity	✅ FALSE collapses to a single node with $\zeta = 1$
Aquatic mosquito	del, dl, dpl, me, ml, mup, gamma	✅ Used verbatim in the $E$/$L$/$P_L$ equations
Adult mosquito	mum, beta, total_M, blood_meal_rates, Q0, foraging_time, species, species_proportions	✅ Per-species; total_M is re-derived: from init_EIR under the default demography, as the IBM’s own value under a custom one (set_equilibrium())
	init_foim	⛔ Ignored: fleet recomputes FOIM from its own seeded human infectivity, which must be self-consistent with its seed
Seasonality	model_seasonality, g0, g, h, rainfall_floor	✅ Truncated Fourier rainfall drives $K_s(t) = K_{0,s}\,\text{scaler}(t)\,\text{rainfall}(t)/\bar{R}$, on a daily grid
Flexible carrying capacity	carrying_capacity, carrying_capacity_timesteps	✅ See set_carrying_capacity(). (carrying_capacity_scalers is not a get_parameters() default and cannot be passed in overrides; it is created only by set_carrying_capacity().)
	carrying_capacity_values	⛔ Not read (nor by malariasimulation; only the _scalers form is used)
Human disease & immunity constants	dd, dt, da, du	✅ Back-translated to $r_D, r_T, r_A, r_U$, then converted to the IBM’s per-day exit probability $1-e^{-1/d}$ so the realised dwell matches
	rb, rc, rva, rid, rm, rvm	✅ Immunity decay time constants $d_{I_B}, d_{I_{CA}}, d_{I_{VA}}, d_{I_D}$ and the maternal decay $d_m, d_{vm}$
	ub, uc, uv, ud	✅ Refractory periods, applied as $u_{\text{eff}} = \lceil u\rceil - 1$ (§B.5)
	pcm, pvm	✅ $P_M$, $P_{VM}$ maternal transfer fractions
	b0, b1, ib0, kb	✅ $b$ Hill function
	phi0, phi1, ic0, kc	✅ $\phi$ Hill function
	theta0, theta1, iv0, kv, fv0, av, gammav	✅ $\theta$ Hill function and its age modifier $f_v$
	d1, id0, kd, fd0, ad, gammad	✅ $q$ Hill function and its age modifier $f_d$
	cd, cu, ct, gamma1	✅ Infectivity by state; ct is superseded by the drug-linked $c_T(t)$ whenever clinical drugs are set
	de, delay_gam, dem	✅ $\tau_E$, $\tau_l$, $\tau_M$ (the three Erlang chains)
Vector-control shape	phi_bednets, phi_indoors, k0	✅ Per species, in the $a(t)$/$\mu(t)$ algebra
Output bands	age_group_rendering_*, incidence_rendering_*, clinical_incidence_rendering_*, severe_incidence_rendering_*, prevalence_rendering_*	✅ Each drives its own output family, exactly as in malariasimulation (§G)
	ib_/id_/ica_/iva_/idm_/icm_/ivm_rendering_*	⛔ fleet does not render mean immunity by band
Metapopulation test-and-treat	rdt_intercept, rdt_coeff	⛔ Ignored. They parameterise the PCR→RDT conversion used by run_metapop_simulation()’s mixing, which fleet does not support (Helpers and run functions)
Engine / solver	mosquito_limit, individual_mosquitoes, r_tol, a_tol, ode_max_steps, progress_bar	⛔ Ignored: fleet is always compartmental and exposes its own atol/rtol/step_size_max through ode_tuning()
Vivax-only	All hypnozoite parameters, drug_hypnozoite_*, n_with_hypnozoites_rendering_*	⛔ Not applicable

fleet-only extension fields

None of the three is a malariasimulation parameter; each may be added to the list, and each defaults to the validated choice.

Field	Default	Effect
bite_dedup	1	1 reproduces the IBM’s per-timestep bite deduplication (saturating hazard, §B.2). 0 uses the linear $b\varepsilon$ that malariaEquilibrium assumes, cutting the residual drift off the seed to well under 1% — but not to zero, since it gates only the hazard and not the boosting or sojourn forms. It is a switch: anything but 0 or 1 (TRUE/FALSE accepted) is an error
acquired_immunity_offset	0	$\delta$ in the $b$/$\phi$/$\theta$ Hill functions. 0.5 reproduces the IBM’s literal per-individual offset; 0 is the better mean-field match against the IBM ensemble mean. Must lie in $[0, 1]$
hold_init_EIR	FALSE	Only has any effect under set_demography() (set_demography()). By default fleet matches the IBM’s mosquito density rather than its nominal EIR, so the same list realises the same transmission in both models and the seed stays a fixed point; init_EIR is then the EIR the IBM was sized at, not the one fleet realises. TRUE seeds at init_EIR directly. Must be TRUE or FALSE

set_species(species, proportions)

Argument	malariasimulation meaning	fleet
species	List of species parameter lists (gamb_params, arab_params, fun_params, kol_params, steph_params, or custom)	✅ Each entry’s blood_meal_rates, foraging_time, Q0, phi_bednets, phi_indoors and mum become per-species vectors driving an independent mosquito sub-model, coupled only through the shared human population
proportions	Relative abundance, summing to 1	✅ Splits total_M and hence each species’ baseline carrying capacity. A species at proportion 0 stays inert; its $K_0$ is floored to a negligible positive value so the larval term does not evaluate $0/0$

set_equilibrium(init_EIR, eq_params, EIR_population_input)

Argument	malariasimulation meaning	fleet
init_EIR	Target EIR to seed from	✅ Stored as parameters$init_EIR, which is where run_simulation_ode() reads it from; it has no init_EIR argument of its own. It means the same thing in both models: under the default demography it is the EIR fleet realises; under set_demography() it sizes the mosquito population exactly as the IBM’s does and fleet seeds at the EIR that density supports (set_demography())
eq_params	Custom malariaEquilibrium parameter set	✅ Merged over fleet’s own back-translation and takes precedence. Note set_equilibrium() writes this field even when passed NULL (storing its own back-translation), so after any call to it, that stored set, not R/translate_params.R, supplies every shared constant (§F)
EIR_population_input	"adult" (default) or "total"	✅ Handled upstream: malariasimulation converts a total-population EIR to adult EIR before storing it, and fleet always interprets init_EIR as adult EIR (bites per adult per year)

set_equilibrium() also calls parameterise_mosquito_equilibrium() internally, which sizes the adult-mosquito population total_M from the human equilibrium under the IBM’s default exponential age structure (equilibrium_total_M(); custom mortality never enters). fleet never reads the stored parameters$total_M; it re-derives the same quantity itself, in one of two ways:

• Default demography. total_M is chosen so that $\sum_s a_s I_{M,s} = \text{init_EIR}/365$ holds exactly under fleet’s own seeded human infectivity, the IBM’s formula evaluated on fleet’s grid, so the two agree up to discretisation and init_EIR is the EIR realised.
• Custom demography (set_demography()). Holding init_EIR here would give the two models different mosquito populations, and hence different transmission, from the same parameter list: the IBM’s density supports less (or more) transmission under the custom age structure and it drifts there. So fleet reproduces the IBM’s total_M exactly (ibm_total_M(), the same human_equilibrium() call on the IBM’s 0–99.9 y grid and the same equilibrium_total_M() arithmetic) and root-finds the EIR at which its own equilibrium under the custom age structure has that density, seeding there. The seed is still a fixed point, so no burn-in is needed. parameters$hold_init_EIR = TRUE restores the previous behaviour (init_EIR as the realised EIR).

Calling parameterise_total_M() or parameterise_mosquito_equilibrium() directly is harmless but changes nothing.

set_demography(agegroups, timesteps, deathrates)

Argument	malariasimulation meaning	fleet
agegroups	Upper edges of the death-rate age groups, in days	✅ Model age groups are binned into them by midpoint using right-closed intervals, matching .bincode(age, c(0, agegroups))
timesteps	When each death-rate row takes effect	✅ Knots of a constant-interpolated $\mu_i(t)$, so custom demography is time-varying (a demographic transition is modelled, not frozen)
deathrates	[length(timesteps) × length(agegroups)] daily death rates	✅ Used directly as $\mu_i(t)$. The $t = 0$ row additionally sets the equilibrium age structure the human seed is rescaled onto; the open-ended top model group is looked up one day above its lower edge, so it takes the rate of the band it belongs to rather than the one below. 🟡 Model ages above the top agegroups edge take the top rate here, whereas the IBM removes them. A warning fires; raise default_age_lower(max_age =) or extend agegroups to match

With a custom demography the mosquito population is sized as the IBM sizes it (set_equilibrium()): fleet takes the IBM’s total_M and seeds at the EIR its own equilibrium under this age structure supports, which is generally not init_EIR (an older population sustains less transmission from the same mosquitoes; the comparison article’s scenario seeds at EIR 13.9 for init_EIR = 20, matching the IBM’s realised 13.8). parameters$hold_init_EIR = TRUE makes init_EIR the realised EIR instead.

set_drugs(drugs)

Argument	malariasimulation meaning	fleet
drugs	List of 4-element vectors c(efficacy, rel_c, prophylaxis_shape, prophylaxis_scale) (e.g. AL_params, DHA_PQP_params, SP_AQ_params)	🟡 Each element is handled as below
drug_efficacy	Probability treatment clears the infection	✅ Multiplies coverage into $f_t^{\text{eff}}$; also multiplies chemoprevention pulse fractions
drug_rel_c	Infectivity of a treated case relative to cd	✅ $c_T = c_d \times \text{rel_c}$, coverage-share-weighted across active drugs and time-varying
drug_prophylaxis_shape, drug_prophylaxis_scale	Weibull prophylaxis survival	✅ Both moments used. The chain mean is $\bar{d}_W - \int e^{-r_T t}W\,dt$, the protection left after the $T$ sojourn (§B.4), and its length $k_P$ matches the variance of the whole $T + P$ sojourn: 16 stages for SP-AQ, 20 for DHA-PQP, 1 for AL (the chemoprevention chain, with no $T$ stage in front, uses $1/\text{CV}_W^2$: 14 / 15). Capped at 20. 🟡 An Erlang chain is not a Weibull, and the exponential $T$ stage adds variability the IBM’s parallel clock does not have; for AL the protection curve is reproduced in its integral, not its shape. ode_tuning(n_ph =) overrides the count
drug_hypnozoite_*	P. vivax only	⛔ Not applicable

set_clinical_treatment(drug, timesteps, coverages)

Argument	malariasimulation meaning	fleet
drug	Index into the drug table	✅ Identifies which drug’s efficacy / rel_c / prophylaxis enter the blend
timesteps	When coverage changes	✅ Knots of the step-interpolated $f_t(t)$, and of the drug-mix series
coverages	Fraction of clinical cases treated with this drug	✅ $f_t(t) = \min\left(1, \sum_{\text{drugs}} \text{cov}_d(t)\right)$. set_clinical_treatment() errors if the summed coverage exceeds 1 at any timestep, so the min(1, ·) fleet applies is a defensive floor that never binds on a valid parameter set. 🟡 The drug-linked quantities $\text{eff}(t)$, $c_T(t)$, $r_P(t)$ are blended by each drug’s instantaneous coverage share, so a first-line switch is modelled; a genuinely mixed first line is represented by its blend rather than as parallel sub-populations

set_antimalarial_resistance(...)

Argument	malariasimulation meaning	fleet
drug	Which clinical drug carries resistance	✅ Matched to the clinical-treatment drug index; also matched against the MDA/SMC/PMC drug so a resistant chemoprevention drug clears fewer infections
timesteps	When resistance levels change	✅ Knots of $\text{ETF}(t)$ and $\text{SPC}(t)$, merged with the treatment-coverage change times
artemisinin_resistance_proportion	Fraction of infections that are artemisinin-resistant	✅ Multiplies both the ETF and SPC probabilities
partner_drug_resistance_proportion	Partner-drug resistance	⛔ malariasimulation requires 0; cannot reach the model
slow_parasite_clearance_probability	Probability a treated resistant case clears slowly	✅ $\text{SPC}(t)$, the split of the treated inflow between $T$ and $T_s$. Coverage-share-weighted across drugs
early_treatment_failure_probability	Probability treatment fails early	✅ $\text{ETF}(t)$: reduces $f_t^{\text{eff}}$, diverting those cases to $D$
late_clinical_failure_probability	Late clinical failure	⛔ malariasimulation requires 0
late_parasitological_failure_probability	Late parasitological failure	⛔ malariasimulation requires 0
reinfection_during_prophylaxis_probability	Reinfection while prophylactic	⛔ malariasimulation requires 0
slow_parasite_clearance_time	Mean duration of slow clearance	🟡 $r_T^{\text{slow}} = 1 - e^{-1/\text{dt_slow}}$, a single scalar carrying the same whole-day conversion as the other sojourns (§B.4). With one resistant drug the blend is exact; with several it is the coverage-weighted blend evaluated at peak resistance (peak, not final, so a rise-then-fall schedule is not collapsed to zero)

set_bednets()

set_bednets(parameters, timesteps, coverages, dn0, rn, rnm, gamman, retention, logistic_half_life, logistic_k)

Module status: 🟡. Net efficacy is averaged over the net-using fraction into per-species $a_s(t)$ and $\mu_s(t)$, so protection is not correlated within individuals across bites. Within that, every argument below is handled exactly.

Argument	malariasimulation meaning	fleet
timesteps	Distribution dates	✅ Every past distribution contributes: at each grid time the population is the mixture over all rounds, each weighted by most-recent-receipt probability $\text{cov}_d\prod_{j>d}(1-\text{cov}_j)$
coverages	Fraction receiving a net at each distribution	✅ As above; repeated distributions accumulate correctly rather than only the latest applying
dn0	[n_timesteps × n_species] probability a net kills a mosquito	✅ Decayed by net age: $d_{n}(s_n) = d_{n0}e^{-s_n/\gamma_n}$
rn	[n_timesteps × n_species] initial repelling probability	✅ Decayed toward rnm: $r_n(s_n) = (r_n - r_{nm})e^{-s_n/\gamma_n} + r_{nm}$
rnm	[n_timesteps × n_species] minimum (asymptotic) repelling probability	✅ As above
gamman	Insecticide decay time constant per distribution	✅ The $\gamma_n$ above; each distribution decays on its own clock
retention	Mean net-retention time (log-uniform model)	✅ Exponential usage survival $e^{-s_n/\text{retention}}$
logistic_half_life, logistic_k	Alternative logistic retention	✅ Exact survival of the IBM’s logistic retention time: $S(s_n) = \exp\!\left(-k\frac{r^2}{1-r^2}\right)$ for $r = s_n/l < 1$ (else 0), with $l = \text{half_life}/\sqrt{1 - k/(k-\log 0.5)}$. Verified $S(\text{half_life}) = 0.5$ exactly

set_spraying()

set_spraying(parameters, timesteps, coverages, ls_theta, ls_gamma, ks_theta, ks_gamma, ms_theta, ms_gamma)

Module status: 🟡. As for nets: population-averaged, exact within. Sprayed protection never expires in the IBM, so the mixture over past rounds carries no retention factor.

Argument	malariasimulation meaning	fleet
timesteps	Spray dates	✅ Mixture over all past rounds, weighted by most-recent-spray probability
coverages	Fraction of houses sprayed	✅ As above
ls_theta, ls_gamma	[n_timesteps × n_species] mortality logistic parameters	✅ $l_s(s_s) = \text{logit}^{-1}(\theta + \gamma s_s)$
ks_theta, ks_gamma	Feeding-success logistic parameters	✅ $k_s = k_0\,\text{logit}^{-1}(\theta + \gamma s_s)$
ms_theta, ms_gamma	Deterrence logistic parameters	✅ $m_s = \text{logit}^{-1}(\theta + \gamma s_s)$

The repellency $r_s$ and survival $s_s$ derived from these are the same spray outcome per individual, so fleet forms the joint mean $\overline{(1-r_s)s_s}$ rather than multiplying two independent averages. Because the IBM recomputes these every timestep, $a_s(t)$/$\mu_s(t)$ are linearly interpolated so within-round logistic decay is a ramp, not a step.

set_carrying_capacity(timesteps, carrying_capacity_scalers)

Argument	malariasimulation meaning	fleet
timesteps	When each scaler row takes effect	✅ Step change in $K_s(t)$, composed multiplicatively with seasonality. $K_s$ is interpolated linearly, so the grid carries a knot the day before each change and the step is confined to a single day (§H)
carrying_capacity_scalers	[n_timesteps × n_species] multipliers on baseline $K_0$	✅ Multiplies baseline $K_0$ per species. 🟡 A scaler of exactly 0 is floored at $K_0 \times 10^{-4}$ to keep the larval term finite, so complete vector elimination is approached but not reproduced exactly

set_mda() and set_smc()

set_mda(parameters, drug, timesteps, coverages, min_ages, max_ages). set_smc() has the identical signature and identical handling.

Module status: 🟡. A mass campaign is applied as a pulse rather than as per-individual events, and the brief treated-infectious phase of cleared cases is omitted (negligible for the fast-clearing drugs used).

Argument	malariasimulation meaning	fleet
drug	Drug administered	✅ Its drug_efficacy scales the cleared fraction; its Weibull mean sets $r_{P_c}$ and its Weibull shape the chain length $k_{P_c}$. Antimalarial-resistance ETF on this drug reduces the cleared fraction
timesteps	Round dates	✅ Integration is split at each; the pulse takes effect the day after the scheduled timestep
coverages	Fraction of the target band reached	✅ Cleared fraction $= \text{coverage}\times\text{efficacy}\times(1-\text{ETF})\times o_i$. A zero-coverage round is a no-op
min_ages, max_ages	Per-round vectors (one entry per timesteps entry) giving the target band in days, inclusive at both ends	✅ Per-round bands are honoured. Mapped onto the model age grid by fractional overlap $o_i$, so narrow bands are not dropped and coarse groups are not over-treated. The absorbing top group is treated as fully covered iff the band reaches its lower edge. 🟡 fleet treats the band as half-open $[\ell, u)$, one day narrower than the IBM — negligible except for very narrow (PMC-style) bands

Co-deployed chemoprevention types (SMC + MDA + PMC) share the single $P_c$ compartment. Its decay rate is the mean of their drugs’ protection durations, weighted by each type’s total scheduled coverage (the sum over all of its rounds), so a type with more rounds carries proportionally more weight regardless of its per-round coverage.

set_pmc(drug, timesteps, coverages, ages)

Argument	malariasimulation meaning	fleet
drug	Drug administered	✅ As for MDA/SMC
timesteps	Coverage-change schedule (not dose dates; PMC delivery is age-triggered)	✅ Read as a coverage schedule
coverages	Fraction of infants receiving each dose	✅ The coverage in force at each pulse time is used
ages	Ages (days) at which doses are given	🟡 The IBM triggers on an individual reaching each dose age. fleet approximates this as pulses at a ~30-day cadence over a 30-day band starting at each dose age, with per-group overlap weighting, keeping the effective dose at approximately coverage × efficacy

create_pev_profile(vmax, alpha, beta, cs, rho, ds, dl)

Argument	malariasimulation meaning	fleet
vmax	Maximum efficacy	✅
alpha, beta	Hill shape and scale mapping antibody titre to efficacy	✅
cs	c(mu, sigma) of $\log$ peak antibody titre	✅ Including the sigma
rho	c(mu, sigma) of the logit short-lived antibody fraction	✅ Including the sigma
ds, dl	c(mu, sigma) of $\log$ short/long antibody half-lives	✅ Including the sigmas

malariasimulation re-draws all four antibody parameters from their profile distributions at every timestep, for every vaccinated individual; nothing is stored per person. Population efficacy is therefore exactly $\mathbb{E}\left[\text{Hill}(\text{Ab})\right]$, not $\text{Hill}(\text{Ab at the median})$: the sigmas are large and the Hill function is nonlinear, so the two differ materially (evaluating at the median overstated R21 efficacy by up to ~4.7 percentage points). fleet integrates over the 4-D distribution with a tensor Gauss–Hermite rule (options(fleet.pev_gq =), default 7 nodes per axis), validated against a Monte Carlo of the IBM’s own sampler to under $5\times10^{-4}$. Because the IBM’s draws are independent between days and between people, there is no within-person correlation for the mean field to lose here: the quadrature is exact rather than approximate. The resulting mean-efficacy curve is cached per profile on a daily grid.

set_pev_epi()

set_pev_epi(parameters, profile, coverages, timesteps, age, min_wait, booster_spacing, booster_coverage, booster_profile, seasonal_boosters)

Module status: 🟡. The routine schedule and full booster sequence are modelled, but seasonal boosters are approximated (warned).

Argument	malariasimulation meaning	fleet
profile	Primary-series PEV profile	✅
coverages	Coverage of the routine programme over time	✅ Time-varying: each cohort is protected at the coverage in force on its own first-dose date, not at a single coverage
timesteps	When each coverage takes effect	✅ Also gates eligibility: a cohort is protected only once its first dose falls inside the programme. Gating follows malariasimulation, which keys on pev_epi_coverages/pev_epi_timesteps and never reads parameters$pev
age	Age (days) at the first dose	✅ Efficacy starts at age + max(pev_doses), i.e. after the final primary dose
min_wait	Minimum time since the last vaccination	➖ For a routine EPI programme this guards re-vaccination and seasonal-booster timing, not the primary-plus-booster schedule fleet models, so it has no effect on the result. (For mass campaigns it does bite; see set_mass_pev().)
booster_spacing	Days from the final primary dose to each booster	✅ The full booster sequence is modelled: the vaccinated are partitioned by the most recent booster each has reached, and each stratum carries its own decayed efficacy
booster_coverage	Matrix of conditional booster coverages	✅ Each booster’s coverage is read at its own administration date (matching coverage[match_timestep(timesteps, admin_time), booster]), reducing to the single-row lookup in the usual case
booster_profile	Per-booster profiles	✅ Each booster uses its own profile, integrated over the antibody distribution as in create_pev_profile()
seasonal_boosters	Time the first booster relative to the start of the calendar year: booster_spacing[1] becomes a day-of-year. It makes no reference to the fitted seasonality; you choose the day yourself, e.g. via peak_season_offset()	🟡 Approximated as a fixed days-since-primary schedule. fleet warns when this is set

set_mass_pev()

set_mass_pev(parameters, profile, timesteps, coverages, min_ages, max_ages, min_wait, booster_spacing, booster_coverage, booster_profile)

Module status: 🟡. Repeated campaigns are the weak point: min_wait re-vaccination exclusion is not applied, campaigns combine as independent protections rather than most-recent-receipt, and the vaccinated cohort does not age out of its band. A single campaign is modelled faithfully.

Argument	malariasimulation meaning	fleet
profile	Primary-series profile	✅
timesteps	Campaign dates	🟡 Every campaign contributes, combined as $1-\prod_k(1-e_k)$. The IBM instead keeps only the most recent vaccination per person, so its population protection is a most-recent-receipt mixture $\sum_k \text{cov}_k \prod_{j>k}(1-\text{cov}_j)\,e_k$, the same structure fleet already uses for bed nets (set_bednets()). The two agree for a single campaign and diverge when campaigns overlap
coverages	Coverage per campaign (recycled if length 1)	✅
min_ages, max_ages	Target age bands	✅ Every band is applied at every campaign, each mapped onto the model age grid by fractional overlap as for MDA/SMC (set_mda() and set_smc()), so a band narrower than the groups it falls between still vaccinates its share instead of nobody. The bands of one campaign are one campaign: their weights add into a single covered fraction per group (capped at 1) before coverage is applied, and only campaigns combine as independent protections. A band overlapping no group warns
min_wait	Minimum time since the last vaccination	⚠ Ignored, and it matters here. In malariasimulation each campaign excludes anyone vaccinated within min_wait of it, so with repeated campaigns this is the primary control on who gets re-vaccinated. fleet treats every campaign as reaching its whole target band independently, so a min_wait longer than the campaign spacing will over-vaccinate
booster_spacing	Days from the final primary dose to each booster	✅ Full sequence, as in set_pev_epi()
booster_coverage	Conditional booster coverage matrix	✅ Read at each booster’s administration date
booster_profile	Per-booster profiles	✅

🟡 The vaccinated cohort does not age out of its target band: efficacy decays in place with time since vaccination, but the protected fraction stays attached to the age band rather than moving up the age grid with the cohort.

set_tbv(timesteps, coverages, ages)

Argument	malariasimulation meaning	fleet
timesteps	Vaccination dates	🟡 Antibody titre decays from each campaign date; campaigns combine as $1-\prod_k(1-e_k)$. As for mass PEV, the IBM overwrites each person’s vaccination date, giving a most-recent-receipt mixture instead. Identical for a single campaign; divergent when campaigns overlap
coverages	Fraction vaccinated	✅

It has to be fractional, because above age 15 the default grid puts its edges at 15, 16.67, 18.33 and 20, so a whole-year target lines up with no group: 18:20 spans parts of three of them.

Transmission-reducing activity is mapped to state-specific transmission-blocking activity via the IBM’s own transform $\text{TBA}(m_x, k, \text{TRA})$ with per-state $m_x \in \{\texttt{tbv_mu}, \texttt{tbv_ma}, \texttt{tbv_md}, \texttt{tbv_mt}\}$, giving four independent per-age multipliers on $U$, $A$, $D$ and $T$ infectivity. All the tbv_* shape constants (tbv_tau, tbv_rho, tbv_ds, tbv_dl, tbv_tra_mu, tbv_gamma1, tbv_gamma2, tbv_k) are used verbatim.

set_epi_outputs(...)

Argument	malariasimulation meaning	fleet
age_group	Bands for population counts	✅ → n_age_* (also emitted over the union of all other families’ bands, as malariasimulation does)
incidence	Bands for all-infection incidence	✅ → n_inc_*
clinical_incidence	Bands for clinical incidence	✅ → n_inc_clinical_*
severe_incidence	Bands for severe incidence	✅ → n_inc_severe_*
prevalence	Bands for prevalence	✅ → n_detect_lm_*, p_detect_lm_*, n_detect_pcr_*
ica, icm, iva, ivm, id, ib	Bands for mean immunity	⛔ Not rendered by fleet
n_with_hypnozoites, hypnozoites, iaa, iam	P. vivax only	⛔ Not applicable

When a family’s band list is empty, fleet falls back to the convenience bands 2–10 y and all-ages for that family only. A bare get_parameters() leaves four of the five families empty; it defaults the prevalence bands to 2–10 y.

set_parameter_draw(draw)

Argument	malariasimulation meaning	fleet
draw	Index 1–1000 into the fitted posterior draws	✅ malariasimulation overwrites the immunity/disease constants in the parameter list in place; they then flow through fleet’s back-translation like any other parameter. Must be called before set_equilibrium(), as in the IBM

Helpers and run functions

Function / argument	malariasimulation meaning	fleet
peak_season_offset(parameters)	Day of peak seasonal transmission	✅ Pure helper on g0/g/h, which fleet reads; useful for timing SMC/booster schedules
get_correlation_parameters(parameters)	Correlation between intervention recipients	⛔ Individual-level; no mean-field analogue
run_simulation(timesteps, parameters, correlations)	Run the IBM	✅ run_simulation_ode(timesteps, parameters, correlations, tuning) mirrors the first three arguments exactly; tuning carries the ODE-only settings, so nothing epidemiological sits outside the parameter list. A non-NULL correlations is warned and ignored: the ODE assumes independence between interventions
run_simulation_with_repetitions(...)	Repeat stochastic runs	⛔ Not applicable. fleet is deterministic, so one run is the ensemble mean
run_metapop_simulation(timesteps, parameters, ..., export_mixing, import_mixing)	Coupled patches	⛔ Not supported. fleet is a single patch; there is no guard, so do not pass a metapopulation setup expecting it to be honoured
run_resumable_simulation(...)	Resume from saved state	⛔ Not implemented. fleet runs are cheap enough to re-run from the seed