What is wrong now
Two defects that look separate and are not:
They share a root cause. Both come from applying formulas derived for large clusters at N = 1. The strain term charges a lone unit for a shell it does not have; valence counts unfilled contacts on a frontier a single unit does not have. Neither is a tuning problem, and neither is fixed by a constant.
The consequence is not cosmetic. In the real world the one-proton element is the most promiscuous thing there is; here it is inert and unstable. That is not an alien universe, it is a broken one, and everything downstream — bond energies, molecules, eventually abiogenesis — inherits it.
Proposed direction
Ian's, 2026-08-04. Approximate real physics, and let the seed perturb physical constants within bounded ranges rather than inventing a packing law from scratch. Universes then differ in comprehensible ways — shells fill in a different order, a different column goes inert — because a knob moved, not because a formula misbehaves at its edges.
The part that makes this more than a rewrite
Two separate requests collapse into one under this design.
Ian also asked for a "special seed that shows real chemistry", and the honest objection was that no seed can produce it: an invented packing law cannot be steered into quantum mechanics, so a "real chemistry seed" would be a hard-coded table wearing a seed's clothes, with the app claiming it generated something it did not.
Under this proposal that objection disappears. Our universe is simply the constants left unperturbed. The generator really does produce the real periodic table, because it is real physics with the knobs at zero. Nothing has to be hard-coded and nothing has to be claimed falsely.
It also fixes a shape problem raised separately: a real periodic table is 7x18 with a block hanging underneath because of how shells fill. Borbax's current table is 2-4 rows by up to 74 columns, so it can never look like the real one. With real shell-filling that structure arrives as a consequence rather than as a layout decision — which is what makes the full-table view worth building.
What this costs, stated up front
- It replaces element generation and everything keyed to it: bond energies, the periodic table, valence, radii, affinity, instability.
- Every golden hash regenerates. This is a physics change of the largest kind.
- The fiction guarantee changes shape rather than disappearing. Real element data would exist in a bounded place. The rule that matters — that the invented universe stays invented, and that real data cannot reach the generator's own path — has to be re-expressed and re-guarded deliberately, in the same task, with the determinism and emergence auditors.
- Open question, not decided here: whether shape-complementarity binding survives as the bonding mechanism above a real-physics element layer, or whether it is replaced too. That is the single biggest design question and it decides how much of
borbax-molecule survives.
Relationship to #25
#25 is one symptom of the root cause described here. If this lands, #25 goes with it; if this is deferred, #25 still needs fixing on its own terms.
Sequencing
Not started. Deserves its own design pass — three amigos plus the determinism, emergence and geometry reviewers — rather than being folded into viewer work. The viewer branch in flight (Step 3) is unaffected and lands first.
What is wrong now
Two defects that look separate and are not:
They share a root cause. Both come from applying formulas derived for large clusters at N = 1. The strain term charges a lone unit for a shell it does not have; valence counts unfilled contacts on a frontier a single unit does not have. Neither is a tuning problem, and neither is fixed by a constant.
The consequence is not cosmetic. In the real world the one-proton element is the most promiscuous thing there is; here it is inert and unstable. That is not an alien universe, it is a broken one, and everything downstream — bond energies, molecules, eventually abiogenesis — inherits it.
Proposed direction
Ian's, 2026-08-04. Approximate real physics, and let the seed perturb physical constants within bounded ranges rather than inventing a packing law from scratch. Universes then differ in comprehensible ways — shells fill in a different order, a different column goes inert — because a knob moved, not because a formula misbehaves at its edges.
The part that makes this more than a rewrite
Two separate requests collapse into one under this design.
Ian also asked for a "special seed that shows real chemistry", and the honest objection was that no seed can produce it: an invented packing law cannot be steered into quantum mechanics, so a "real chemistry seed" would be a hard-coded table wearing a seed's clothes, with the app claiming it generated something it did not.
Under this proposal that objection disappears. Our universe is simply the constants left unperturbed. The generator really does produce the real periodic table, because it is real physics with the knobs at zero. Nothing has to be hard-coded and nothing has to be claimed falsely.
It also fixes a shape problem raised separately: a real periodic table is 7x18 with a block hanging underneath because of how shells fill. Borbax's current table is 2-4 rows by up to 74 columns, so it can never look like the real one. With real shell-filling that structure arrives as a consequence rather than as a layout decision — which is what makes the full-table view worth building.
What this costs, stated up front
borbax-moleculesurvives.Relationship to #25
#25 is one symptom of the root cause described here. If this lands, #25 goes with it; if this is deferred, #25 still needs fixing on its own terms.
Sequencing
Not started. Deserves its own design pass — three amigos plus the determinism, emergence and geometry reviewers — rather than being folded into viewer work. The viewer branch in flight (Step 3) is unaffected and lands first.