GOD'S HAND

(5 Oct, 16:55; data/_resol_patch.py on the log
176 stack). The scale follows the dot count instead (x0.97 a step under 95% of DOT_TARGET, x1.03 over it), the rescale
keeps every creature, and under the budget the dots holding most shares split in two (same genome, half the creatures
each), so creatures are never made or lost by the budget. Seed-4 microbe save, 12,000 ticks: seed 1 ran to the rail
with complex cells (1,509-1,546 dots) and the scale ran away to 1,264, 2 species; seeds 2 and 3 held under the rail but
bacteria took the sea (euk 34-231 dots) and fidelity fell 0.91 -> 0.68 and 0.69; 3-5 species against the page's 7-14.
The dot budget is PARKED here: six fixes measured (logs 166, 170, 176, 177), each right in kind, none holding a world.
What the page does (a creature-making scale) keeps worlds diverse; every honest version so far lets one group win.
Next step needs a different idea, decided with BLACK, before any more code. All patches stay in CODER's worktree data/.

(5 Oct, 15:05-15:12; BLACK's radius design and
two more from the measurements; patches in CODER's worktree data/: _radius_patch, _sdeath_patch, _perborn_patch, on
_nf + _sc + _expr + _sch). Seed-4 microbe save, 12,000 ticks; the pass mark is the fidelity dial holding (page: 0.91-0.92).
- A POPULATION COVERS GROUND (BLACK): a dot of share nf feeds and meets at a random point within
sqrt((nf - 1) x map area / DOT_TARGET / pi). Fidelity still fell: 0.91 -> 0.46, 0.70, 0.81 (3 seeds).
- EACH CREATURE ITS OWN FATE: disease, a waking virus and old age roll per share, taking dead shares out of the dot
(Scheffer et al. 1995); before, one roll killed every creature folded into a dot, so a line spread over many dots had
lower risk (free bet-hedging). Fidelity fell slower: 0.79, 0.88, 0.77.
- EVERY NEWBORN IS COPIED: at the budget a dividing dot copied at most BUD_MAX young, so a 20-share dot gave its 20
newborns 4 chances of a copy error. Now one copy per newborn share. Fidelity 0.85 and 0.84 (seed 2 dipped to 0.74).
- Underneath: in every arm built on _nf + _sc, complex cells fall in the first 2,000 ticks (61-153 dots against 429 on
the page) while their deaths are no higher (old age 1,071 against 1,383, disease 64 against 103): they are born less.
This is the base, not the expressed genome: log 166 measured eaten whole alone cutting complex cells 477 -> 300.
Conclusion: the three fixes are each right in kind and together slow the sloppy-copier drift about fourfold, but do not
stop it, and the base stack loses complex cells. Next: find what in EATEN WHOLE / the conserving rescale slows complex
cells' births (division refill at the cap, the rescale's step limits), then retest the full stack.
Addendum (15:12, births counted by group over 2,000 ticks, seed 1): eaten whole alone is not it (complex cells 444
against 429 on the page). The conserving rescale alone is: with no shares to follow, S.sc stays at 83.8, births halve
(complex cells 950 against 1,430, bacteria 566 against 1,206) and both groups shrink (107 and 83 dots by 2,000 ticks).
On the page S.sc falls 88 -> 51 as dots thin, so each dot stands for fewer creatures and grows back faster: that is
the lever that makes creatures from nothing (log 166). Without it the microbe save falls toward what its food really
holds. So the base is honest; the save's populations were partly made by the lever.

(5 Oct,
14:20-14:45; CODER with BLACK's design from log 166; patches in CODER's worktree data/_expr_patch.py and _sch_patch.py,
on top of _nf_patch and _sc_patch). Seed-4 microbe save, 3 seeds, up to 20,000 ticks.
- The idea holds in numbers: of 2,140 cell births on the page, 12.7% had the parent's exact DNA but 85.4% the same
EXPRESSED genome (same steps working, every dial sum T+T2 equal, same mind genes and virus); the silent errors sat in
step genes not yet working (the step strand copies 4x faster) or in the instinct strand a cell never reads. Two-parent
cell births were 17 of ~2,150. So under the budget only ~15% of young need a dot of their own, not 85%.
- Built: young with the parent's expressed genome join its dot; same-square dots of one expressed genome merge; the dot's
DNA becomes a joiner's with the joiner's share (a neutral variant fixes with probability equal to its frequency:
Kimura 1962, Genetics 47:713); o.nv counts neutral variants; a dot's age is its members' share-weighted mean age;
saves keep shares above one. MEETINGS BY CREATURES: swallowing, filter feeding, cell poison and gene transfer happen
as often as the dot's creatures would meet (rate x share) and partners are picked by creatures, not dots (the
super-individual method: Scheffer, Baveco, DeAngelis, Rose & van Nes 1995, Ecol. Model. 80:161).
- Two bugs of mine found on the way: the young-stay replacement ended its line with a // comment that swallowed the
page's own line taking the young's share out of the parent, so below the target every division made creatures from
nothing (this was also in log 166's young-stay test); and EATEN WHOLE reset a leftover share above one to exactly one.
After both: a division ledger (data/_split_ledger.py) is exact, 0 of 5,231 splits off (10-12 before).
- Result with creatures conserved: no stable world. Each run ends with one group holding the sea (bacteria 630-930 dots,
or complex cells 1,400-1,490 dots at the rail), species 2-49 (page 13-33), and the runs at the rail have the most
mutant young (59-65%). The cause, measured: COPYING GETS SLOPPY. The fidelity dial (fid, mean over cell creatures)
falls 0.91 -> 0.47 and 0.91 -> 0.79 in 12,000 ticks on two seeds, while the page holds 0.91-0.92. A clone folds into
its parent's dot at the parent's spot; a mutant gets a dot of its own and goes its own way, so a sloppy copier's line
spreads over more squares of light and nutrient than a faithful one's. That is a selection for mutation the budget
makes, not nature (in real asexual populations mutators rise only by riding along with a good mutation and are then
selected down: Sniegowski, Gerrish & Lenski 1997, Nature 387:703).
Conclusion: folding clones needs the folded dot to stand for creatures spread over space (its feeding, light and nutrient
spread over the squares its members would cover), or the budget must be held some other way. ROADMAP: THE DOT BUDGET.

(5 Oct, 12:15-14:10; CODER with
BLACK, audit items (a) step 2 and the cell death flux). Seed-4 microbe save (1,400 Ma), 8,000-20,000 ticks, 3 seeds; deaths
counted in creatures. Patches kept in CODER's worktree (C:/repos/GENESIS-coder/data/_*_patch.py) for the next attempt.
Found (each is a real fault in the current page, measured):
- Old age is 81-85% of cell deaths; on Earth viruses and grazers each take about half and old age none (Fuhrman 1999,
Nature 399:541; Suttle 2007). Taking it off (a cell dot is a population turning over within a tick) exposed competitive
exclusion that old age had been hiding by thinning every line evenly: the lines lost were out-shaded sun-eaters and
complex cells "of old age, too few young".
- A swallowing cell could never out-earn its upkeep: it fed one tick in twenty (PHAGO_P 0.05), best intake ~0.0014 a tick
against upkeep 0.0018 and a sun-eater's ~0.04. Grazers cannot hold bacteria down on today's page.
- A cell's body is free both ways (a division pays only a share of the store, a swallowed cell gives only its store), so
starving bacteria are worthless food. A one-ledger version (BODY_C 0.4 paid at division, yielded when eaten or dead) was
built and checked exact (data/_cell_ledger.py: 8 divisions and 8 swallows, before = after).
- A sleeper dies at 1/4,000 a tick while a tick holds 2,000 years: a protist cyst sleeps for millions of years. With cell old
age off, starving cells went to sleep and their cysts filled the lab (1,471 of 1,474 complex-cell dots asleep).
- Changing S.sc multiplies every dot's creatures at once (creatures made or destroyed by the budget). A conserving rescale
(each dot's share nf moves the other way; data/_sc_check.py: 12 rescales, error <= 3e-14) leaves the budget no lever on
the dot count.
- Merging dots to hold the budget is hidden selection unless it is fair in every rule: across a 64-patch it cut food (a dot
feeds from its own map square; merging only bacteria let complex cells win 1,569 to 22 dots, merging only complex cells
let bacteria win 1,537 to 29); merging on the same square is too rare to hold the budget; folding young into the parent
throws away mutations (species fell to 8-9); keeping every mutant its own dot fails because 85% of young carry a copy
error (2,064 of 2,433 by 2,000 ticks).
Ablation (seed 1, complex-cell dots at 2,000 ticks, patches added in turn): main 477; + eaten whole (nf) 300; + strain
disease 1,098; + light colours 419; + no cell old age 1,348 (at the cap); + conserving rescale 1,313; + sleepers die by
years 1,275; + merging 48; + cell body 44.
Single pieces on main (20,000 ticks, species at the end; main 15/14/16, though main swings 9-16 within a run): prey found
by creatures (pickW) 10/10/10, both groups kept; swallowers feed every tick 9/10/24 (bacteria grazed to ~20 dots on 2 of 3
seeds); sleepers die by years 5/8/18; the whole ledger set with old age kept: complex cells crash to ~20, species 10/11/5.
Conclusion: none of this ships as is. The super-individual needs a design decision first (ROADMAP: THE DOT BUDGET), then the
cell pieces go back in one at a time.