The whole of §6 in one place: what couples to what (the map), the unit basis and the orderings each cluster needs (the skeleton), and the review amendments that reconcile §1–§5. No fitted numbers — those come from tuning against the running sim.
Eight balance-clusters; an edge means they share a constant that must be tuned for both at once. Edge colour = the shared hub.
Three distinct things kept separate: dimensions, the units that measure them, and the non-dimensionalizations that pin the free scales.
C and N are two species of M, so C:N is a pure ratio. Derived: light = E·L⁻²·T⁻¹, concentration = M·L⁻² (2-D), velocity = L·T⁻¹, rate = T⁻¹.
M and E magnitudes are free — the dynamics are invariant under rescaling all matter (or all energy) consistently.
Plus surface light intensity ≡ 1. Parameter-absorptions, not unit definitions.
Size ≡ structural mass S (in m). Radius r = √(S ⁄ πρ) with structural density ρ constant; surface (perimeter) ∝ r; area ∝ r² ∝ S. Every layer reads size this way: surface-machinery cap ∝ r, internal cap ∝ S, engulf compares S, gradient reliability rises with r.
Necessary orderings in the locked units (basal maintenance = 1, reference-cell structure = 1, node-spacing = 1, surface light = 1). A value sheet must land inside these.
Reserves are matter; the night bill is energy. The bridge is respiration yield (energy per m of carbon). A cell of structure S pays S each night-tick, so banked reserve carbon, converted, must clear it. [amended #3 — was "reserve_cap (energy-equiv)"]
Attenuation steep enough for a dark basin; the compensation depth falls inside the column.
Patches sized between a cell and the world; idle standing-machinery upkeep must bite, so a generalist loses to a specialist in a mono-resource patch. [re-read per #1/#2: upkeep is machinery mass × maintenance]
Predators big enough to engulf; an achievable clump crosses the refuge threshold; bonding cheaper than the predation it escapes; predation pays for itself.
Enzyme edge beats passive by more than it costs; DOM reaches absorbers; POM lingers to be eaten. The last term places decomposition in the column vs the floor.
Gradients span several nodes but are smaller than the world; halving node-spacing or tick mustn't change outcomes.
Match within a factor so either element can become limiting and switch; excretion sheds the surplus.
Marker range a local cue — bigger than cell spacing, smaller than the world; emission costs something.
Honest limit: these are necessary, not sufficient. Walk the calibration order against the running sim; a value lands when it sits inside every ordering and the canary holds.
Five fixes from the full review. These are authoritative — where they differ from the original layer docs, these win. (Back-portable into §1/§3/§4/§5 on request.)
GRN outputs are three kinds: standing machinery (uptake, fixation, respiration, active excretion, enzyme) — output sets a target allocation, machinery relaxes toward it (retool lag = the plasticity cost); per-tick efforts (swim, buoyancy, emit-marker, transfer) — paid per use; discrete events (divide, engulf, adhere/break-bond) — threshold-triggered, paid per event. Standing machinery is built mass, hard-capped in two geometric pools: surface-bound ≤ surface ∝ r, internal ≤ structure ∝ r². Allocation is the split within each pool. This unifies §3.3 with §1's surface-vs-volume scaling and is why the size cap exists.
Three distinct costs, no double-count. Standing maintenance = a basal term on inert structural mass (the §6 anchor ≡ 1) plus an activity surcharge on machinery mass (rate > 1) — active transporters/enzymes burn upkeep beyond their mass (ion gradients, protein turnover), and idle machinery still pays the surcharge. Opex — per-use cost of running an effector. GRN cost — a separate weak per-node term. Machinery is built mass, so it is part of structure; the surcharge is what makes it cost more than inert mass. [kept biologically real: surcharge, not mass-only]
Light powers the cell directly: the light reactions give immediate usable energy (efficiency η_fix), spent the same tick on maintenance and work, while fixation also builds reserve carbon. Reserve carbon is the buffer — respired only to cover shortfalls (night, shade, or demand above the light supply), at efficiency η_resp. So per tick: energy ≈ (light captured · η_fix) + (reserve C respired · resp_yield · η_resp); at night the first term is zero. The η_fix·η_resp round-trip loss falls only on the buffered fraction, not on direct daytime use. Energy is still one-way and never stored as energy — only as carbon; the night ordering (B2·1) is the buffered regime. [revised: real phototrophy — ATP is fleeting, carbon is the store]
Size ≡ structural mass S; radius and surface are derived (2-D). Settles the loose use of "size" across engulf ratios, the surface-vs-volume cap, and gradient reliability.
A cell that dies mid-act still sits in the tick-start cluster snapshot, so colony-effective size can count a dead member for the rest of that tick — accept the one-tick lag, or drop dead cells from clusters on death. And buoyancy carries a (small) cost — no free vertical migration.
§6 consolidated: coupling map + dimensional skeleton + five review amendments. Unit basis locked; orderings stated; §1–§5 reconciled. v2: amendment #3 revised to the realistic two-path energy model (light direct, carbon as buffer) after the biology check. No fitted numbers — values come from tuning against code, in the calibration order above. Supersedes evolab-constants-v0.1, evolab-constants-skeleton-v0.1, and evolab-section6-v1.