Give each species parameter default a single home

[gustavdelius]

Summary

Every species parameter default now has exactly one home, and which home is decided by what consumes the parameter. Two commits, same principle pulling in opposite directions:

• a and b (0b5b661) are consumed by the length-weight conversion itself — l2w()/w2l() are called from check_and_convert_species_params() inside species_params.data.frame(). Their defaults move into species_params rather than being applied ad-hoc inside l2w()/w2l().
• p, z_ext, d, E_ext, D_ext, interaction_resource (c610222) are each read by exactly one rate setter. Their defaults move out of the central table to that setter — where they already were, duplicated.

Which parameters stay central

A parameter is defaulted by validSpeciesParams() only when no single rate setter owns it, which happens for five reasons:

Reason	Parameters
Several setters read it	n (six of them: setExtDiffusion, setExtEncounter, setExtMort, setMaxIntakeRate, setReproduction, setSearchVolume); w_mat, w_inf
Only the projection reads it; no setter touches it	alpha
The size grid is built from it, before any setter runs	w_min, w_max
The table's own length-weight conversion needs it	a, b
Reporting only	is_background

Everything else belongs to its setter: p/k to setMetabolicRate(), z0/z_ext/d to setExtMort(), E_ext to setExtEncounter(), D_ext to setExtDiffusion(), interaction_resource to setInteraction(), q/gamma to setSearchVolume(), erepro/m/w_mat25/R_max to setReproduction(), beta/sigma to setPredKernel().

The setter-local defaults are not redundancy: the rate setters are public API that can be called directly, without setParams() and hence without validSpeciesParams(), so each setter has to supply what it needs on its own. Tests already relied on this (e.g. test-setExtMort.R NULLs z_ext/d and requires setExtMort() to restore them).

The p default

Three places set p, with three different values:

1. newMultispeciesParams()'s own p argument, default 0.7, injected into the species parameter table at newMultispeciesParams.R#L179-L180 before validation;
2. species_params.data.frame()'s p = n;
3. setMetabolicRate()'s p = 3/4.

(1) preempts both others, so it is p = 0.7 — not p = n — that models built in the usual way have been getting. (2) applied only when validSpeciesParams() was called directly on a bare data frame, and it shadowed (3).

This PR removes (2). That promotes (3) from shadowed to live, so it is corrected from 3/4 to n for consistency with the exponent it tracks. (1) is untouched and still wins for newMultispeciesParams(), which is why no model changes.

The p argument of setMetabolicRate() (deprecated here)

There is a fourth would-be home: setMetabolicRate()'s own p argument. It never had any effect on a MizerParams object, because it is applied with set_species_param_default() — which only fills a missing or NA column and never overwrites — and (1) guarantees the column is always already there. The condition it was documented for, "if the exponent is not specified in a p column", is unreachable. It was silently ignored.

It is now deprecated in favour of species_params(params)$p <- value, which triggers setParams() and does what was meant. This also makes the family consistent: setSearchVolume() has no q argument and setMaxIntakeRate() has no n argument. What the argument does is unchanged while deprecated; it just warns now.

newMultispeciesParams()'s p is a different argument, is not forwarded to setMetabolicRate(), and still works. It is now documented explicitly there so it does not inherit the deprecation badge, and is excluded from the arguments setParams() advertises in its dots.

Closes #459.

Behaviour

Built models are unchanged — setParams() calls every rate setter, so each deleted default is immediately re-supplied by its owner. Verified with a golden-model diff of NS_params across the change: no species_params column lost, and no numeric change to search_vol, intake_max, metab, mu_b, psi, maturity, ext_encounter, ext_diffusion, initial_n or interaction.

User-visible change: validSpeciesParams() applied to a bare species parameter data frame now returns fewer columns; the setter-owned ones appear once the model is built. Its roxygen previously documented returning n, p, z_ext, d, E_ext, D_ext and interaction_resource, and has been rewritten.

Also

• Dropped the interaction_resource note in setInteraction(). It was dead while the column was set centrally and would otherwise have fired on every model build; "all species feed on the resource" is the unremarkable default.
• Removed the unreachable a = 0.004 / b = 3 defaults in plot_growth_curves(), which conflicted with the central a = 0.01.

Tests

• test-validSpeciesParams.R: asserts the setter-owned columns are absent from a bare validSpeciesParams() result but present on a built model; guards the no-maximum-size case.
• test-setMetabolicRate.R: pins p == n rather than a hardcoded 3/4, and that the construction and standalone-setter paths agree.
• test-setExtEncounter.R / test-setExtDiffusion.R: new standalone-setter tests for E_ext / D_ext restoration.
• test-setMetabolicRate.R: the p argument warns; setting the species parameter works; newMultispeciesParams(sp, p =) still works and does not warn. The existing test that pinned the argument's fill-only semantics is kept (they are unchanged) and renamed.

Full suite green apart from test-plots.R:501, which fails identically on master and is annotated in-source as machine-dependent.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QZop1EdWQNm9SXGCqwWg8k