5313ea3 · basal-agent · pear-git-v3

Changed files (3) +137 -70

SYNTHESIS.md +18 -18

…	…	@@ -789,19 +789,19 @@ fission beats the propagule bottleneck 5 cells to 0**, with the gap widening as
789	789	(~0.2 at div size 8, ~1.2 at 16) — expected, since at size 8 compartments are small enough that both
790	790	modes assort similarly anyway.
791	791
792		> **Test A: quality is per-unit → maximise variance, the bottleneck wins.
793		> Test B: quality is collective → variance destroys it, fission wins.
794		> Optimal assortment is a TRADE-OFF, not a maximum.**
795
796		Both measured in the same machinery with the same controls, pointing opposite ways because the
797		kind of quality differs. Neither result alone yields this.
798
799		And it closes a loop with `linkage.c`. The tension has a resolution we had already shown emerges:
800		linkage. Keeping co-functional parts physically together preserves combinations against
801		assortment while still permitting variance between units. That is the chromosome argument (Maynard
802		Smith & Szathmáry), arrived at here by chaining our own rungs rather than imported. The three results
803		compose: collective function needs its parts kept together (2→3), parasite purging needs variance
804		(Test A), and linkage buys both at once.
	792	> ⚠️ RETRACTED, same day. The above was written from matched division threshold. At matched
	793	> realized size the two modes differ by −0.13 and RAF tracks size in 8/8 cells regardless of mode
	794	> — propagule mode simply produces smaller compartments, and RAF size scales with compartment size.
	795	> Test A's variance effect survived this identical check; Test B's did not. There is **no evidence
	796	> here that a bottleneck destroys collective closure**, and the trade-off claim is withdrawn.
	797	>
	798	> (The first attempt at this check was itself broken: fission's realized size is non-monotonic in
	799	> division size, so bounds-checking against the last array element silently dropped 7 of 8 points.)
	800
	801	The `linkage.c` loop, now unsupported. The retracted tension had an appealing resolution —
	802	linkage keeps co-functional parts together against assortment, the chromosome argument. With no
	803	measured tension it is speculation, recorded here only so the reasoning is not silently reused: it
	804	would need a tension that this experiment does not demonstrate.
805	805
806	806	Honest bounds. This is the first model in the arc with two rungs at once — an emergent RAF
807	807	(rung 1) inside compartments that build, grow and select themselves (rung 4), the second selecting on
…	…	@@ -861,11 +861,11 @@ transfer to 1→2 on its own — only Law 2 does.
861	861	- Compartment-level selection need not be supplied (`chemoton.c`) — coupling growth to the
862	862	chemistry's own output generates it with no fitness-proportional sampling anywhere, reproducing the
863	863	decreed version's size-threshold signature (PARTIAL: small 2/2, large 1/2). ✓
864		- Optimal assortment is a trade-off, not a maximum (Test A vs Test B) — when catalytic quality is
865		per-unit, maximising variance wins (propagule beats fission); when it is a collective closure
866		property, variance destroys the closure and fission wins (5 cells to 0). The two point opposite ways
867		in the same machinery because the kind of quality differs, and `linkage.c` is the mechanism that
868		buys both at once. ✓
	864	- ~~Optimal assortment is a trade-off, not a maximum~~ — RETRACTED same day. Test B's reversal was
	865	a size artifact: propagule mode shrinks realized compartment size and RAF size scales with it. At
	866	matched realized size the two modes differ by −0.13, and RAF tracks size in 8/8 cells regardless of
	867	mode. Test A's variance effect survived this same check; Test B's did not. **No evidence that a
	868	bottleneck destroys collective closure.**
869	869	- Compartment selection survives collective quality (`rafcomp.c`) — maxRAF 4.71 vs a random-subset
870	870	baseline of 0.13, with the closure-blind control at 0.00–0.01. First model in the arc holding two
871	871	rungs at once, though co-present rather than one producing the other. ✓

rafcomp.c +82 -52

…	…	@@ -211,63 +211,93 @@ int main(void)
211	211	printf(" THE TENSION: Test A found assortment VARIANCE purges parasites (+1.14 at matched size).\n");
212	212	printf(" But a RAF is a property of the SET, so variance should DESTROY closure. Which wins?\n\n");
213	213
214		printf(" cat.dens \| divsz \| rand base \| FISSION raf \| fis ctrl \| PROPAGULE raf \| prop ctrl \| mean sz\n");
215		printf(" ---------+-------+-----------+-------------+----------+---------------+-----------+--------\n");
	214	/* ---- MATCHED REALIZED SIZE. Test B's reversal was established at matched division
	215	THRESHOLD, but propagule mode also shrinks realized compartment size, and in this model
	216	bigger compartments hold bigger RAFs. So fission > propagule may be a pure size artifact
	217	with no closure destruction at all. Sweep division size finely at one catalysis density,
	218	record each arm's OWN realized size, and interpolate fission onto propagule's sizes. ---- */
	219	double dens = 0.10;
	220	int sizes[]={4,6,8,10,12,16,20,24};
	221	int NZ=(int)(sizeof(sizes)/sizeof(sizes[0]));
	222	double fr[16], fsz[16], pr[16], psz[16], cf[16], cp[16], base[16];
	223	int any_raf=0, ctrl_flat=1;
216	224
217		double dens[]={0.04,0.08,0.12};
218		int sizes[]={8,16};
219		int any_raf=0, coupled_wins=0, ctrl_flat=1, ncmp=0, prop_better=0, fis_better=0;
	225	printf(" catalysis density %.2f, %d seeds\n\n", dens, NS);
	226	printf(" divsz \| rand base \| FISSION raf / size \| PROPAGULE raf / size \| fis ctrl \| prop ctrl\n");
	227	printf(" ------+-----------+--------------------+----------------------+----------+----------\n");
220	228
221		for(int d=0;d<3;d++){
222		for(int z=0;z<2;z++){
223		double bf=0,bc=0,pf=0,pc=0,ms=0,base=0,mc=0,nd=0,nc=0;
224		for(int s=0;s<NS;s++){
225		double re,n,m,dv,mm;
226		RNG=seeds[s]; build_net(dens[d]);
227		base+=random_baseline(sizes[z]/2>2?sizes[z]/2:2, 200);
228		RNG=seeds[s]; build_net(dens[d]); bf+=run(1,sizes[z],0,&re,&n,&m,&dv,&mm); ms+=m; mc+=mm; nd+=dv; nc+=n;
229		RNG=seeds[s]; build_net(dens[d]); bc+=run(0,sizes[z],0,&re,&n,&m,&dv,&mm);
230		RNG=seeds[s]; build_net(dens[d]); pf+=run(1,sizes[z],1,&re,&n,&m,&dv,&mm);
231		RNG=seeds[s]; build_net(dens[d]); pc+=run(0,sizes[z],1,&re,&n,&m,&dv,&mm);
232		}
233		bf/=NS;bc/=NS;pf/=NS;pc/=NS;ms/=NS;base/=NS;mc/=NS;nd/=NS;nc/=NS;
234		printf(" %.2f \| %2d \| %5.2f \| %5.2f \| %5.2f \| %5.2f \| %5.2f \| %5.2f\n",
235		dens[d],sizes[z],base,bf,bc,pf,pc,ms);
236		if(bf>0.2\|\|pf>0.2) any_raf=1;
237		ncmp++;
238		if(bf>base+0.3\|\|pf>base+0.3) coupled_wins++;
239		if(bc>base+0.3\|\|pc>base+0.3) ctrl_flat=0;
240		if(pf>bf+0.2) prop_better++;
241		if(bf>pf+0.2) fis_better++;
	229	for(int z=0;z<NZ;z++){
	230	double bf=0,bs=0,pf=0,ps=0,c1=0,c2=0,bb=0;
	231	for(int s=0;s<NS;s++){
	232	double re,n,m,dv,mm;
	233	RNG=seeds[s]; build_net(dens); bb+=random_baseline(sizes[z]/2>2?sizes[z]/2:2,200);
	234	RNG=seeds[s]; build_net(dens); bf+=run(1,sizes[z],0,&re,&n,&m,&dv,&mm); bs+=m;
	235	RNG=seeds[s]; build_net(dens); pf+=run(1,sizes[z],1,&re,&n,&m,&dv,&mm); ps+=m;
	236	RNG=seeds[s]; build_net(dens); c1+=run(0,sizes[z],0,&re,&n,&m,&dv,&mm);
	237	RNG=seeds[s]; build_net(dens); c2+=run(0,sizes[z],1,&re,&n,&m,&dv,&mm);
242	238	}
	239	fr[z]=bf/NS; fsz[z]=bs/NS; pr[z]=pf/NS; psz[z]=ps/NS; cf[z]=c1/NS; cp[z]=c2/NS; base[z]=bb/NS;
	240	printf(" %2d \| %5.2f \| %5.2f / %5.2f \| %5.2f / %5.2f \| %5.2f \| %5.2f\n",
	241	sizes[z],base[z],fr[z],fsz[z],pr[z],psz[z],cf[z],cp[z]);
	242	if(fr[z]>0.2\|\|pr[z]>0.2) any_raf=1;
	243	if(cf[z]>base[z]+0.3\|\|cp[z]>base[z]+0.3) ctrl_flat=0;
	244	}
	245
	246	/* fission's realized size is NON-MONOTONIC in division size (it rises then falls as the arm
	247	collapses at large divsz), so the fission points must be SORTED BY SIZE before interpolating.
	248	The first version bounds-checked against fsz[last] and silently dropped almost every point. */
	249	printf("\n Does RAF track SIZE regardless of reproduction mode? (per-cell, same divsz)\n");
	250	int tracks=0;
	251	for(int z=0;z<NZ;z++){
	252	int bigger_f = fsz[z]>psz[z];
	253	int rafbig_f = fr[z]>pr[z];
	254	if(bigger_f==rafbig_f) tracks++;
	255	printf(" divsz %2d : larger arm = %s, larger RAF = %s %s\n", sizes[z],
	256	bigger_f?"fission ":"propagule", rafbig_f?"fission ":"propagule",
	257	(bigger_f==rafbig_f)?"tracks":"DIVERGES");
243	258	}
244		printf("\n");
245		printf(" T1 RAFs formed inside compartments at all : %s\n", any_raf?"YES":"NO — test did not run");
246		printf(" T2 coupled beats the random-subset baseline : %d/%d cells\n", coupled_wins,ncmp);
247		printf(" T3 CONTROL (uptake blind to closure) flat : %s\n", ctrl_flat?"YES":"NO — CONFOUNDED");
248		printf(" T4 REVERSAL: propagule better %d / fission better %d\n\n", prop_better, fis_better);
	259	printf(" RAF tracks realized size in %d/%d cells\n",tracks,NZ);
	260
	261	printf("\n MATCHED REALIZED SIZE — propagule vs fission interpolated (fission sorted by size):\n");
	262	int ord[16]; for(int z=0;z<NZ;z++) ord[z]=z;
	263	for(int a=1;a<NZ;a++){ int k=ord[a],b=a-1; while(b>=0&&fsz[ord[b]]>fsz[k]){ ord[b+1]=ord[b]; b--; } ord[b+1]=k; }
	264	double sumd=0; int nd=0;
	265	for(int z=0;z<NZ;z++){
	266	double t=psz[z];
	267	if(t<=fsz[ord[0]]\|\|t>=fsz[ord[NZ-1]]) continue;
	268	int lo=-1; for(int y=0;y+1<NZ;y++) if(fsz[ord[y]]<=t&&fsz[ord[y+1]]>=t){ lo=y; break; }
	269	if(lo<0) continue;
	270	double d0=fsz[ord[lo]], d1=fsz[ord[lo+1]];
	271	double w=(d1>d0)?(t-d0)/(d1-d0):0.0;
	272	double fi=fr[ord[lo]]+w*(fr[ord[lo+1]]-fr[ord[lo]]);
	273	double d=pr[z]-fi;
	274	printf(" size %5.2f : propagule %5.2f vs fission-interp %5.2f diff %+.2f\n", t, pr[z], fi, d);
	275	sumd+=d; nd++;
	276	}
	277	double mean_adv = nd? sumd/nd : 0.0/0.0;
	278	printf(" mean propagule advantage at MATCHED size: %+.2f over %d points\n\n", mean_adv, nd);
	279
	280	printf(" T1 RAFs formed : %s\n", any_raf?"YES":"NO — test did not run");
	281	printf(" T3 CONTROL blind to closure stayed flat : %s\n", ctrl_flat?"YES":"NO — CONFOUNDED");
	282	printf(" T4 at matched DIVISION THRESHOLD, fission won in the earlier run.\n");
	283	printf(" T4' at matched REALIZED SIZE, propagule advantage = %+.2f\n\n", mean_adv);
249	284
250		if(!any_raf)
251		printf(" => THE TEST DID NOT RUN. No compartment ever closed a RAF, so there was nothing to\n"
252		" select on. Invalid experiment, not a null. Raise catalysis density or compartment size.\n");
253		else if(!ctrl_flat)
254		printf(" => CONFOUNDED. The closure-blind control also beats baseline, so the rise is not\n"
255		" selection on collective function.\n");
256		else if(coupled_wins==0)
257		printf(" => HONEST NULL. RAFs form, but compartment selection does NOT raise closure above the\n"
258		" random baseline. The ladder does NOT chain: chemoton's success depended on catalytic\n"
259		" quality being a PER-UNIT property, and it does not survive the move to collective\n"
260		" closure — assortment destroys the very thing selection would act on.\n");
261		else if(fis_better>prop_better)
262		printf(" => CHAINS, AND TEST A REVERSES. Compartment selection survives collective quality\n"
263		" (%d/%d cells beat baseline), but FISSION now beats the propagule bottleneck\n"
264		" (%d vs %d) — the opposite of Test A. A single-unit bottleneck cannot carry a\n"
265		" closure, so the variance that purges parasites also destroys collective function.\n"
266		" The optimal assortment is a TRADE-OFF, not a maximum.\n", coupled_wins,ncmp,fis_better,prop_better);
	285	if(!any_raf) printf(" => TEST DID NOT RUN.\n");
	286	else if(!ctrl_flat) printf(" => CONFOUNDED by the control.\n");
	287	else if(mean_adv < -0.2)
	288	printf(" => THE REVERSAL SURVIVES. Even at matched realized size the propagule bottleneck is\n"
	289	" WORSE for collective closure (%+.2f). So it is not a size artifact: a single-unit\n"
	290	" bottleneck genuinely cannot carry a RAF, and Test A's finding does NOT generalise.\n"
	291	" Optimal assortment is a trade-off set by whether quality is per-unit or collective.\n", mean_adv);
	292	else if(mean_adv > 0.2)
	293	printf(" => THE REVERSAL WAS AN ARTIFACT, AND IT INVERTS. At matched realized size the propagule\n"
	294	" bottleneck is BETTER (%+.2f), same direction as Test A. The earlier fission win was\n"
	295	" entirely propagule mode shrinking compartments, and in this model smaller compartments\n"
	296	" hold smaller RAFs. RETRACT the trade-off claim.\n", mean_adv);
267	297	else
268		printf(" => CHAINS, AND TEST A GENERALISES. Compartment selection survives collective quality\n"
269		" (%d/%d beat baseline) and the propagule bottleneck still wins (%d vs %d), so RAF\n"
270		" closure is more robust to assortment than predicted. Report as-is.\n",
271		coupled_wins,ncmp,prop_better,fis_better);
	298	printf(" => THE REVERSAL WAS A SIZE ARTIFACT. At matched realized size the two modes are\n"
	299	" indistinguishable (%+.2f). Fission only looked better because propagule shrinks\n"
	300	" compartments and smaller compartments hold smaller RAFs. There is no evidence here\n"
	301	" that a bottleneck destroys closure — RETRACT the trade-off claim as stated.\n", mean_adv);
272	302	return 0;
273	303	}

rafcomp.md +37 -0

…	…	@@ -61,3 +61,40 @@ compartments that build, grow and select themselves (rung 4), with the second se
61	61	Honest bound: this is two rungs co-present and interacting, not one producing the other. The
62	62	self-building chain SYNTHESIS calls open is still open. And the reaction network is supplied, as in
63	63	`rung1.c` — what emerges is which subset closes, not the chemistry itself.
	64
	65	---
	66
	67	## ⚠️ T4 RETRACTED — the reversal was a size artifact (matched-realized-size check)
	68
	69	T4 was established at matched division threshold. But propagule mode also shrinks realized
	70	compartment size, and in this model RAF size scales with compartment size — so fission > propagule
	71	may have been nothing but bigger compartments. Re-run at catalysis density 0.10, sweeping 8 division
	72	sizes and recording each arm's own realized size:
	73
	74	\| divsz \| fission raf / size \| propagule raf / size \|
	75	\|---\|---\|---\|
	76	\| 4 \| 1.46 / 1.50 \| 1.40 / 1.44 \|
	77	\| 8 \| 2.57 / 2.71 \| 2.35 / 2.47 \|
	78	\| 10 \| 2.40 / 2.56 \| 2.71 / 2.89 \|
	79	\| 12 \| 2.76 / 2.99 \| 2.99 / 3.26 \|
	80	\| 16 \| 4.69 / 5.28 \| 3.49 / 3.93 \|
	81	\| 20 \| 4.27 / 5.04 \| 2.99 / 3.46 \|
	82	\| 24 \| 1.55 / 2.24 \| 2.35 / 3.89 \|
	83
	84	RAF tracks realized size in 8/8 cells — in every cell, whichever arm had the larger compartments
	85	had the larger RAF, regardless of reproduction mode. Note rows 10, 12 and 24, where propagule has
	86	the larger size and duly wins: the direction flips with size, not with mode.
	87
	88	Matched-size interpolation (fission sorted by size — its realized size is non-monotonic in
	89	division size, which broke the first attempt at this analysis and silently dropped 7 of 8 points):
	90	mean propagule advantage −0.13 over 7 points, with 6 of 7 inside ±0.17. The lone outlier (−1.07)
	91	sits at the divsz-24 cell where the fission arm is collapsing.
	92
	93	> There is no evidence here that a bottleneck destroys collective closure. The "optimal assortment
	94	> is a trade-off" claim is retracted. Test A's variance effect was real and survived this same
	95	> check; Test B's apparent reversal did not.
	96
	97	What survives Test B: T1 (RAFs form), T2 (compartment selection beats the random baseline 6/6,
	98	~36×), and T3 (the closure-blind control at 0.00–0.02 everywhere). The chaining result stands —
	99	compartment-level selection survives the move from per-unit to collective quality. Only the
	100	mode-comparison dies.

basal-agent commits / 5313ea3

Changed files (3) +137 -70