Luminary Studies · environment 00 · where the criterion was derived
The assay
Before the tissues, this. Our selectivity criterion was not derived in a tumor — it was derived in a well containing three kinds of lymphocyte, a drug, and a clock. No vessels, no stroma, no pressure, no polarity, no barrier to cross. That simplicity is why the derivation is clean, and it is also the single largest limitation on everything built from it.
So this page does two things: it draws the administration and the predicted behavior in the environment the numbers actually come from, and it says plainly what this environment does not contain — which is the reason the other modules exist.
11 October 2026 · nothing here was measured by us · the six variants and all rate constants are published; the model and its predictions are ours
—
—
—
The administration
Exactly what is put in, and on what clock.
The vessel
100 µL well, swept 50–250 µL because the model's depletion term depends on it. Depletion comes to 2.4–3.3% of free concentration, so the free drug is effectively the dosed drug.
The cells
Lymphocytes treated as 10 µm spheres — 314 µm² of surface each. Swept 7–12 µm, and the diameter cancels out of the ratio.
Dose range
Set by each variant's own predicted EC₂₀: 0.026–0.036 nmol/L on the dual-positive population, 0.46–34.4 nmol/L on the single-positive one. The gap between those two columns is the selectivity.
The clock
6 hours, stated by the source. Swept fourfold, and the predicted slope moves by less than a quarter.
The anchoring arm
Held fixed across all six variants: k_on 2.0 × 10⁵ M⁻¹s⁻¹, k_off 4.9 × 10⁻³ s⁻¹, so K_D 24.5 nmol/L and a single-arm lifetime of 204 s.
The gating arm
The only thing varied: 0.9 → 70 nmol/L across the six published variants, a 78-fold span in affinity and 138-fold in off-rate.
What this environment does not contain
Each absence is a term that exists in a real tissue and has no counterpart here. This list is the argument for the other modules.
No vasculature. The drug is already at its target concentration everywhere. In tissue the surface sits 100–1,000× below plasma and penetration grows only as the square root of dose.
No interstitium or stroma. Nothing to diffuse through. In pancreatic tumors the pressure rivals arterial pressure and three quarters of vessels are collapsed.
No epithelial polarity. A cell in suspension has no apical and no basolateral face, so the accessibility question cannot arise — and it is the question that killed several programs.
No variation in the sugar coat. The glycocalyx stands 86–127 nm off a mucin-expressing membrane against the 14 nm this model's geometry assumes. Here it is uniform and ignorable; in tissue it is upstream of every term.
No immune context. No suppressive cells, no lactate, no exhaustion from continuous exposure, no stromal exclusion of T cells. The effectors are fresh and unopposed.
No tumor cell at all. The three populations are lymphocytes. Every tumor statement made from this series is an extrapolation, and the modules exist to test it.
The engineered step, drawn to scale. Once one arm is bound the other stops searching the well. It sweeps a 10 nm hemisphere of membrane, and inside that volume its partner antigen sits at roughly 22 µM against a dosed concentration of order a nanomolar. Apparent affinity is multiplied by 1 + C_eff/K — 150-fold here — and the complex’s lifetime by far more, which is the quantity the next section is about. The effective concentration is Rhoden’s construction evaluated at our assumed accessible density, not a measured local concentration.
The expected behavior, and what was measured
The model is driven entirely by published rate constants; the threshold is set once from the single-positive population and nothing is fitted to the dual-positive one. These are predictions, next to the source's own measurements.
variant
gating arm K_B
predicted EC₂₀ · single-positive
predicted EC₂₀ · dual-positive
predicted selectivity
measured
parental
0.9 nM
0.462 nM
0.0292 nM
16×
8×
VkY92A
10.0 nM
4.723 nM
0.0357 nM
132×
29×
VkY91A
17.0 nM
8.345 nM
0.0313 nM
267×
57×
VkR95A
42.0 nM
21.174 nM
0.0255 nM
831×
233×
VkR95A+VhD97A
63.0 nM
31.070 nM
0.0284 nM
1,093×
814×
VkY94A+VhY99A
70.0 nM
34.416 nM
0.0306 nM
1,126×
775×
Read the two right-hand columns as a trend, not a level. The prediction over-states absolute selectivity by up to 4.6-fold. What it gets right is the slope: selectivity rises in direct proportion to the gating arm's dissociation constant — +1.008 predicted against +1.068 measured, where the rival equilibrium account predicts zero. A level that is 2.4-fold off reads equally well as an accessible fraction of 0.41 or an effective exposure of 2.5 h, and one experiment separates those.
Why it works, and where it stops working
The same six variants, by how long each attachment persists against the 6-hour clock. The last two rows are the honest edge of the mechanism.
The same table as a picture. Held by one arm a complex lasts seconds to about an hour and turns over hundreds of times before the plate is read; bridged by two it lasts hours to weeks and never completes a single turnover. The read sits between them, which is the whole mechanism. The honest edge is also visible: the bridged band starts at 9.1 h, barely clear of the 6 h line, so the dual-positive population is intermediate rather than cleanly trapped — it is still settling when the plate is read.
variant
one arm bound
both arms bound
population settles in
against a 6 h read
parental
3,846 s
995.0 h
19.44 h
still in motion
VkY92A
526 s
115.5 h
16.64 h
still in motion
VkY91A
213 s
55.1 h
13.84 h
still in motion
VkR95A
43 s
14.4 h
7.27 h
still in motion
VkR95A+VhD97A
29 s
9.6 h
5.48 h
has largely settled
VkY94A+VhY99A
28 s
9.1 h
5.28 h
has largely settled
The dual-positive population is not irreversibly trapped, and the paper does not claim it is. It is intermediate — partway to equilibrium when the plate is read. That is enough to separate it from a single-positive cell that settles in 28 seconds to about an hour and is fully equilibrated, and it is not enough to call the bridged state irreversible. For the two weakest variants the dual-positive population settles before the read, which is exactly where the measured selectivity is weakest — the mechanism and the data agree about which end is the edge.
The one number the whole thing turns on
Two accounts of the same molecule, differing by a whole power rather than a margin. If selectivity comes from binding the target more tightly, weakening the gating arm cannot change it and the slope is zero. If selectivity comes from the two populations settling on opposite sides of the read, selectivity is proportional to the gating arm's dissociation constant and the slope is one. Nine points across two independent published series give 1.068. That is the experiment this study rests on, and it was run by somebody else before the model existed.
The criterion is clean here because this environment has almost nothing in it. Every module built since exists to find out which of the missing terms breaks it first.
Environment 00 of the Luminary Studies platform. The six variants, their rate constants and the measured selectivities are published by others; the three-species conserved kinetic model, the threshold, the predicted columns and the lifetime analysis are ours. No molecule described here has been synthesized and nothing was measured by us. The dual-positive population is described as intermediate rather than trapped throughout, because that is what the arithmetic supports.