// ============================================================================ // XCHAN-AT29 01 — THE SEGMENTED MIXED SUPER-W TRIPLE CENSUS, AND @29 // (2026-08-19 — TODO item X's blind test of the ~3.8 joint-deficit candidate) // ============================================================================ // WHY. `research/history/staging/import-stein.md` §3.2 puts a zero-parameter // closed form on TODO item X's ~3.8 constant, 1 − J = 4·Σ_{x W} // + Σ_{r∈N} #{(q₁ ∈ D_a(r), q₂ W} // // with D_a(r) = {q ∈ qs : q | r}, D_b(r) = {q ∈ qs : q | r+2}. Every r // contributes through its own two divisor lists and nothing else, so the // whole census is LOCAL: it is a sum of a bounded function of the local // factorisation over a sieved interval, and the corpus's segmented-sieve // lever applies verbatim (`research/import-scanstat-04-score.js` streams // T₂₉ over the same 6.47e9 positions with a depth-64 ring buffer and never // materialises the gap word). |D_a| ≤ 6 and |D_b| ≤ 6 at @29 and at @31, // because 31·37·41·43·47·53 = 5.04e9 < W(29) and ·59 overflows it. // // (L2) SIEVE THE COMPRESSED SLOT LINE, NOT [0,W). Every r ∈ N has r ≡ 11 or 17 // (mod 30), so only W/15 of the line can matter. Indexing slots by // s = 2k + t with r = 30k + (t ? 17 : 11), a scour prime q marks the // a-side of class t at the arithmetic progression r ≡ CRT(t, 0) (mod 30q), // i.e. every 2q-th slot — so the marking cost falls by the full factor 15, // from 2W·Σ1/q to (2W/15)·Σ1/q. At @29 that is 9.9e8 marks instead of // 1.5e10. The CRT residues are computed once per (prime, class, side) and // re-based per segment with one division. // // EXACTNESS AT THE COMPARISON. The only inequality in the kernel is q₁q₂q₃ > W. // At @31, q₁q₂ < W = 2.006e11 and q₃ ≤ 447829, so the product reaches 9.0e16 and // is NOT exact in a double. It is never formed. Instead U[m] = ⌊W/q_m⌋ is // precomputed exactly (integer-corrected, not trusted from the division), and // the test is q₁q₂q₃ > W ⟺ q₁q₂ > U[m], valid because q_m ∤ W (every scour // prime exceeds x) so W/q_m is never an integer. q₁q₂ ≤ 2.006e11 is exact. // // MEMORY. Per segment: A (1 byte) + a (1) + b (1) + two divisor lists of // capacity 7 as Uint32 (56) = 59 bytes per slot, on 2^21 slots = 124 MB, at // every level. The level does not enter the memory bill at all; it enters only the // number of segments (206 at @29, 6376 at @31). // // --------------------------------------------------------------------------- // THE VALIDATION GATE (prereg §5). Every integer below is quoted from the // embedded OUTPUT block of `research/natal-cap-39-triple-census.js` and is // checked to the unit at @11, @13, @17, @19 and @23 before @29 is computed. If // any one fails the run reports the failure and computes no @29 value at all. // This is the gate that caught natal-cap-39's own Uint8Array wrap. // // --------------------------------------------------------------------------- // PRE-REGISTERED SCORING, quoted from `xchan-at29-prereg.md` §3 and §4: // TEST 1 (primary, the record's own yardstick) // z = ((1−J)meas − pred)/σ_J, σ_J = √(mixed super-W obs)/CRT // HIT |z| ≤ 3 | MARGINAL 3 < |z| ≤ 6 | MISS |z| > 6 // TEST 2 (secondary, scale accuracy) d = (1−J)meas/pred − 1 // TIGHT |d| ≤ 1.25% | CONSISTENT ≤ 2.5% | DRIFTING ≤ 5% | REFUTED > 5% // J-BAND CLAUSE: J(29) < 0.94 or > 1.00 ⇒ the object changed, not the constant // NON-SEPARATION, registered before the run: at @29 the candidate is 1.6σ_J // from a constant multiple fixed at @23 and 0.4σ_J from a 1/lnW law, so a // HIT confirms the scale and cannot single out 4S₂. At @31 the 1/lnW law is // ≈16σ_J away and does separate. // // THE @31 CLAUSE. `xchan-at29-prereg.md` §4 registers @31 as the level where the // candidate and the 1/lnW rival separate (≈16σ_J apart there against 0.4σ_J at // @29), and makes it conditional on affordability. It is affordable — the OUTPUT // block below carries its own price — so it is run: the default level list ends // at 31, and both blind levels are scored in one invocation by the same code. // // PRESENTATION NOTE, disclosed rather than left to be noticed. Each blind level // was computed once before the canonical embed run, with these predictions and // these bands already hard-coded and the pre-registration already committed, so // neither figure was ever seen by an unclassified run. @29 was first computed on // a revision of this file differing only in two OUTPUT lines (the custody line, // and a corrected scour-mark work count — the count is printed, not used); @31 // was first computed on this revision, before the default level list gained 31 // and before this note and the @31 clause above were written. No band, no rule, // no threshold, no prediction and no computed quantity changed at any point, and // the embedded block reproduces both first runs to the digit. // // node --max-old-space-size=8192 research/xchan-at29-01-segmented.js // optional args: a list of levels, e.g. `-- 11 13 17` (default 11 13 17 19 23 29 31) // ============================================================================ 'use strict'; const T00 = Date.now(); const f = (v, d = 3) => Number.isFinite(v) ? v.toFixed(d) : String(v); const el = () => ((Date.now() - T00) / 1000).toFixed(1) + 's'; const say = (s) => process.stderr.write(s + '\n'); function primesUpTo(n){const s=new Uint8Array(n+1),o=[];for(let i=2;i<=n;i++){if(!s[i]){o.push(i);for(let j=i*i;j<=n;j+=i)s[j]=1;}}return o;} function inv(a,m){let r0=a%m;if(r0<0)r0+=m;let r1=m,s0=1,s1=0; while(r1!==0){const q=Math.floor(r0/r1),t=r0-q*r1,u=s0-q*s1;r0=r1;r1=t;s0=s1;s1=u;} let s=s0%m;if(s<0)s+=m;return s;} const C3=(n)=>n<3?0:n*(n-1)*(n-2)/6, C2=(n)=>n<2?0:n*(n-1)/2; // --------------------------------------------------------------------------- // REFERENCE TABLE — natal-cap-39-triple-census.js, embedded OUTPUT, verbatim. // [B3(3,0), B3(2,1), B3(1,2), B3(0,3)] ; sub = ∏QW // --------------------------------------------------------------------------- const REF = { 11:{W:2310, N:90, K:10, B3:[0,2,0,0], miss:0.006909, nSub:0, subAl:0, subMx:0, supMx:2, o21:2, o12:0, CRT:3.73, J:0.5361}, 13:{W:30030, N:990, K:34, B3:[12,89,70,11], miss:0.023518, nSub:71, subAl:23, subMx:16, supMx:143, o21:77, o12:66, CRT:139.70, J:1.0236}, 17:{W:510510, N:14850, K:120, B3:[848,2729,2713,847], miss:0.039784, nSub:5246, subAl:1695, subMx:2243, supMx:3199, o21:1622, o12:1577, CRT:3544.79, J:0.9025}, 19:{W:9699690, N:252450, K:435, B3:[27377,85713,85449,27391], miss:0.050943, nSub:162982, subAl:54768, subMx:97097, supMx:74065, o21:37139, o12:36926, CRT:77162.70, J:0.9599}, 23:{W:223092870, N:5301450, K:1739,B3:[879225,2779604,2779107,879385],miss:0.058834,nSub:4517592,subAl:1758610, subMx:3751046, supMx:1807665, o21:904409, o12:903256, CRT:1871421.20, J:0.9659}, }; // PRE-REGISTERED BLIND PREDICTIONS — import-stein.md §3.2, adopted verbatim. const PRED = {29:{oneMinusJ:0.028943, ratio:3.8399, F:0.007537, K:7863, y:80429}, 31:{oneMinusJ:0.024784, ratio:3.8612, F:0.006419, K:37534, y:447829}}; // the two rivals of prereg §4, also fixed before the run const RIVAL = {29:{n2:0.029145, n3:0.028990}, 31:{n2:0.024818, n3:0.025165}}; const T1=(z)=>Math.abs(z)<=3?'HIT':Math.abs(z)<=6?'MARGINAL':'MISS'; const T2=(d)=>Math.abs(d)<=0.0125?'TIGHT':Math.abs(d)<=0.025?'CONSISTENT':Math.abs(d)<=0.05?'DRIFTING':'REFUTED AS THE CONSTANT'; // --------------------------------------------------------------------------- // LEVEL DATA // --------------------------------------------------------------------------- function level(x){ const basePs=primesUpTo(x).filter(p=>p>=7); const W=30*basePs.reduce((a,b)=>a*b,1); const qs=primesUpTo(Math.floor(Math.sqrt(W))+1).filter(q=>q>x&&q*q<=W); const Nbar=2*basePs.reduce((a,p)=>a*(p-2),1); // U[m] = floor(W/qs[m]), integer-corrected: the division is not trusted const K=qs.length, U=new Float64Array(K); for(let m=0;mW)u--; while((u+1)*qs[m]<=W)u++; U[m]=u;} return {x,W,basePs,qs,K,U,Nbar,y:qs[K-1]}; } // --------------------------------------------------------------------------- // THE DENOMINATOR — Σ_{∏Q>W} 1/∏Q as a TAIL, never as e₃ minus the sub-W sum, // so no cancellation enters it. Monotone pointer over the third prime; the // comparison is the exact U-test. Also returns the sub-W triple count and // reciprocal sum, both of which the gate checks. // --------------------------------------------------------------------------- function tailMass(L){ const {qs,K,U}=L; const T=new Float64Array(K+1); for(let m=K-1;m>=0;m--)T[m]=T[m+1]+1/qs[m]; let miss=0,subRec=0,nSub=0,nSup=0; for(let i=0;i U[m]; non-increasing in j for(let j=i+1;j0&&t>U[p-1])p--; const st=p>j+1?p:j+1; const rt=1/t; miss+=rt*T[st]; subRec+=rt*(T[j+1]-T[st]); nSub+=st-(j+1);nSup+=K-st; } } return {miss,subRec,nSub,nSup}; } // elementary symmetric e₃(1/q), an independent reading of miss+subRec function e3of(qs){let e=[1,0,0,0];for(const q of qs){const p=1/q; e[3]+=e[2]*p;e[2]+=e[1]*p;e[1]+=e[0]*p;}return e[3];} // --------------------------------------------------------------------------- // THE SEGMENTED CENSUS. One pass over the compressed slot line s = 2k + t, // r = 30k + (t ? 17 : 11), in segments of SEGK values of k. // --------------------------------------------------------------------------- const LC=7; // divisor-list capacity per side function census(L,SEGK){ const {W,basePs,qs,K,U}=L; const kmax=W/30, nSeg=Math.ceil(kmax/SEGK), MS=2*SEGK; const A=new Uint8Array(MS),a=new Uint8Array(MS),b=new Uint8Array(MS); // la/lb store the INDEX into qs (read back as qv[ia[i]] and U[ia[k]]), not the // prime. These were Uint16Array, which silently wraps an index mod 65536. The // default ladder here tops out at @31 with K = 37534, 57% of the cap, and this // script takes its level list on the COMMAND LINE, so `-- 37` is one word away // from K = 198274 and a silent alias. That is the exact defect repaired in the // sibling xchan-at37-01-census.js on 2026-08-21, where it cost five hours of // compute and read as a clean refutation of a pre-registered law. It was masked // here only by the LC = 7 throw below — a guard on a DIFFERENT quantity — and // the at-37 history is precisely the story of that guard being raised and the // alias coming out from behind it. Widened to Uint32Array 2026-08-21 with the // throw below. Every internal identity in this file is a COUNT identity and // would have stayed PASS throughout, which is why only the value looks wrong. if(K>0xFFFFFFFF)throw new Error(`scour index ${K} exceeds the la/lb container width at @${L.x}`); const la=new Uint32Array(LC*MS),lb=new Uint32Array(LC*MS); // CRT residues mod 30·p, one per (prime, class, side), computed once const cls=[11,17]; const baseR=[]; // base primes: kill r ≡ 0 and ≡ −2 for(const p of basePs){const i30=inv(30%p,p),row=[]; for(let t=0;t<2;t++)for(const z of [0,(p-2)%p]){ let d=(z-cls[t])%p;if(d<0)d+=p;row.push(cls[t]+30*((d*i30)%p));} baseR.push({p,row});} const Ra=new Float64Array(2*K),Rb=new Float64Array(2*K); for(let m=0;m>1,R=row[c]; let v=R+m30*Math.ceil((V0-R)/m30); while(v=V0)v-=m30; for(let idx=2*((v-cls[t])/30-k0)+t;idx=V0)v-=m30; for(let idx=2*((v-cls[t])/30-k0)+t;idx=V0)w-=m30; for(let idx=2*((w-cls[t])/30-k0)+t;idxmaxA)maxA=av; if(bv>maxB)maxB=bv; B3[3]+=C3(av);B3[2]+=C2(av)*bv;B3[1]+=av*C2(bv);B3[0]+=C3(bv); if(av+bv<3)continue; if(av>=LC||bv>=LC)throw new Error(`divisor list capacity ${LC} exceeded: a=${av} b=${bv} at slot ${idx}`); const base=LC*idx; const mx0=sup[2]+sup[1]; // this slot's mixed super-W tally, before for(let i=0;iU[ia[k]])sup[3]++;else sub[3]++;} for(let k=0;kU[ib[k]])sup[2]++;else sub[2]++;}} for(let i=0;iU[ib[k]])sup[0]++;else sub[0]++;} for(let k=0;kU[ia[k]])sup[1]++;else sub[1]++;}} const xs=sup[2]+sup[1]-mx0;SX+=xs;SX2+=xs*xs; } if(nSeg>=16&&seg%Math.ceil(nSeg/16)===0)say(` seg ${seg+1}/${nSeg} [${el()}]`); } return {Nc,B3,sub,sup,nSeg,MS,maxA,maxB,SX,SX2, bytes:MS*(3+2*4*LC)}; } // --------------------------------------------------------------------------- // DRIVER // --------------------------------------------------------------------------- const ARG=process.argv.slice(2).map(Number).filter(v=>v>0); const LEVELS=ARG.length?ARG:[11,13,17,19,23,29,31]; const SEGK=1<<20; // 2^21 slots per segment, 124 MB console.log('XCHAN-AT29 — the segmented mixed super-W triple census'); console.log(`pre-registration: research/history/staging/xchan-at29-prereg.md levels: ${LEVELS.join(', ')} segment: 2^${Math.log2(2*SEGK)} slots\n`); let gateFail=0,gateRun=0; const ROWS=[]; let cSlot=null,cMark=null; // ns per slot / per mark, carried level to level for(const x of LEVELS){ const t0=Date.now(); const L=level(x); const {W,qs,K,Nbar,y}=L; // price before the run, from the previous level's two fitted constants let sMark=0,S2=0,lg=0; for(const q of qs){sMark+=2/q;S2+=1/(q*q);lg+=2*Math.log(1-1/q)-Math.log(1-2/q);} const forced=Math.exp(lg)-1; const nSlot=W/15,nMark=nSlot*sMark; const price=(cSlot!==null)?`${f((nSlot*cSlot+nMark*cMark)/1e9,1)} s predicted`:'no calibration yet (first level)'; console.log(`===== @${x}: W=${W} N̄(formula)=${Nbar} K=${K} (${qs[0]}..${y})`); console.log(` WORK COUNTS (the code's own loop counts): ${nSlot} slots, ${f(nMark,0)} scour marks, ${Math.ceil((W/30)/SEGK)} segments`); console.log(` CUSTODY on the record's own closed form: 4S₂ = 4·Σ_{xt1){cSlot=(t2-t1)*1e6/nSlot*0.5;cMark=(t2-t1)*1e6/nMark*0.5;} const N=C.Nc; const supMx=C.sup[2]+C.sup[1],subMx=C.sub[2]+C.sub[1]; const supAl=C.sup[3]+C.sup[0],subAl=C.sub[3]+C.sub[0]; const CRT=6*N*tm.miss, CRTal=2*N*tm.miss; const J=supMx/CRT, sJ=Math.sqrt(Math.max(supMx,0))/CRT; // The registered σ stands as registered (prereg §3); the corrected one is // reported beside it. Var of a sum over N̄ independent slots is // Σx² − (Σx)²/N̄, and Σx is supMx by construction, which is the check below. if(Math.abs(C.SX-supMx)>1e-6)throw new Error(`per-slot subtotals do not sum to supMx: ${C.SX} vs ${supMx}`); const varSlot=Math.max(C.SX2-C.SX*C.SX/N,0); const sJslot=Math.sqrt(varSlot)/CRT, infl=sJ>0?sJslot/sJ:NaN; const e3=e3of(qs); console.log(` MEMORY: ${f(C.bytes/1048576,1)} MB of arrays, independent of the level [census ${f((t2-t1)/1000,2)}s]`); console.log(` N̄ counted = ${N} (formula ${Nbar}: ${N===Nbar?'PASS':'FAIL'}) max |D_a| = ${C.maxA}, max |D_b| = ${C.maxB}`); console.log(` B_3 FROM THE (a,b) COUNTS = ${f(C.B3[3]+C.B3[2]+C.B3[1]+C.B3[0],0)} split (3,0)=${f(C.B3[3],0)} (2,1)=${f(C.B3[2],0)} (1,2)=${f(C.B3[1],0)} (0,3)=${f(C.B3[0],0)}`); let ident=0; for(let c=0;c<4;c++)if(Math.abs(C.sub[c]+C.sup[c]-C.B3[c])>1e-9)ident++; console.log(` INTERNAL IDENTITY sub+sup = B_3 class by class: ${ident===0?'PASS':'FAIL ('+ident+' classes)'}`); console.log(` SUB-W (∏QW}1/∏Q = ${f(tm.miss,6)} (e₃ = miss + sub = ${f(tm.miss+tm.subRec,8)} vs esym ${f(e3,8)}; C(K,3) = ${tm.nSub+tm.nSup} vs ${K*(K-1)*(K-2)/6})`); console.log(` SUPER-W (∏Q>W):`); console.log(` ALIGNED obs = ${f(supAl,0)} [P1 predicts exactly 0: ${supAl===0?'PASS':'FAIL'}] CRT=${f(CRTal,2)}`); console.log(` MIXED obs = ${f(supMx,0)} CRT=${f(CRT,2)} by orientation (2,1) = ${f(C.sup[2],0)} and (1,2) = ${f(C.sup[1],0)} against ${f(CRT/2,2)} each`); console.log(` >>> JOINT DEFICIT J = ${f(J,6)} 1−J = ${f(1-J,6)} σ_J = ${f(sJ,6)} [as REGISTERED: Poisson on the triple count]`); console.log(` σ_J SLOT-CLUSTERED (the sample unit is the slot) = ${f(sJslot,6)} inflation ×${f(infl,3)}`); console.log(` This is still a FLOOR. It prices within-slot correlation exactly and prices`); console.log(` between-slot clustering not at all, so a disjoint-block estimate is larger`); console.log(` again where it is affordable. An understated σ makes |z| too LARGE, so every`); console.log(` z below is a ceiling on its own significance, not a floor.`); // ---- the gate const R=REF[x]; if(R){ gateRun++; const chk=[['W',W,R.W],['N̄',N,R.N],['K',K,R.K], ['B3(3,0)',C.B3[3],R.B3[0]],['B3(2,1)',C.B3[2],R.B3[1]],['B3(1,2)',C.B3[1],R.B3[2]],['B3(0,3)',C.B3[0],R.B3[3]], ['sub-W triples',tm.nSub,R.nSub],['sub-W aligned obs',subAl,R.subAl],['sub-W mixed obs',subMx,R.subMx], ['super-W aligned obs',supAl,0],['super-W mixed obs',supMx,R.supMx],['(2,1)',C.sup[2],R.o21],['(1,2)',C.sup[1],R.o12]]; let bad=0,line=''; for(const [nm,got,want] of chk){if(got!==want){bad++;line+=`\n MISMATCH ${nm}: got ${got}, natal-cap-39 has ${want}`;}} // the two rounded reals from the record const rr=[['Σ_{∏Q>W}1/∏Q',Number(f(tm.miss,6)),R.miss],['CRT',Number(f(CRT,2)),R.CRT],['J',Number(f(J,4)),R.J]]; for(const [nm,got,want] of rr){if(got!==want){bad++;line+=`\n MISMATCH ${nm}: got ${got}, natal-cap-39 has ${want}`;}} gateFail+=bad; console.log(` GATE vs natal-cap-39 embedded OUTPUT (${chk.length} integers + ${rr.length} rounded reals): ${bad===0?'ALL PASS':bad+' FAIL'}${line}`); } // ---- the blind score const P=PRED[x]; if(P){ if(gateFail>0){console.log(' SCORE WITHHELD: the validation gate failed, prereg §5 forbids reporting this level.');} else{ const om=1-J,z=(om-P.oneMinusJ)/sJ,d=om/P.oneMinusJ-1; const V=RIVAL[x]; console.log(` ---- BLIND SCORE against import-stein.md §3.2, adopted verbatim in the prereg`); console.log(` predicted 1−J = ${f(P.oneMinusJ,6)} measured 1−J = ${f(om,6)} σ_J = ${f(sJ,6)}`); console.log(` TEST 1 z = ${f(z,2)} -> ${T1(z)} [the REGISTERED verdict, on the registered σ]`); const zs=(om-P.oneMinusJ)/sJslot; console.log(` TEST 1 on the slot-clustered σ: z = ${f(zs,2)} -> ${T1(zs)} [reported beside the registered verdict, never in place of it]`); console.log(` TEST 2 d = ${f(100*d,2)}% -> ${T2(d)}`); console.log(` J-BAND CLAUSE (J < 0.94 or J > 1.00 ⇒ the object changed): J = ${f(J,6)} -> ${(J<0.94||J>1.00)?'FIRES':'does not fire'}`); console.log(` RIVALS, fixed before the run: N2 constant·F ${f(V.n2,6)} (z=${f((om-V.n2)/sJ,2)}) | N3 ∝1/lnW ${f(V.n3,6)} (z=${f((om-V.n3)/sJ,2)})`); console.log(` the same rivals on the slot-clustered σ: N2 z=${f((om-V.n2)/sJslot,2)} | N3 z=${f((om-V.n3)/sJslot,2)}`); console.log(` ratio (1−J)/F with F from the record (${f(100*P.F,4)}%): measured ${f(om/P.F,4)} vs predicted ${f(P.ratio,4)}`); console.log(` custody: 4S₂ recomputed here = ${f(4*S2,6)} vs the record's ${f(P.oneMinusJ,6)} (${Math.abs(4*S2-P.oneMinusJ)<5e-7?'agrees to the printed digit':'DISAGREES — the record is wrong'}); ratio on this file's own F (${f(100*forced,4)}%) = ${f(om/forced,4)}`); } } ROWS.push({x,lnW:Math.log(W),N,supMx,CRT,J,sJ,sJslot,infl,miss:tm.miss,nSub:tm.nSub,pred:P?P.oneMinusJ:null}); console.log(` [level time ${f((Date.now()-t0)/1000,2)}s]\n`); } console.log('===== SUMMARY — THE JOINT DEFICIT ACROSS LEVELS ====='); console.log(' x | lnW | mixed super-W obs | CRT | J | 1−J | σ_J(reg) | σ_J(slot) | infl | predicted 1−J | z(reg) | z(slot) | d'); for(const r of ROWS){ const om=1-r.J; console.log(` ${String(r.x).padStart(2)} | ${f(r.lnW,3)} | ${f(r.supMx,0).padStart(17)} | ${f(r.CRT,2).padStart(13)} | ${f(r.J,6)} | ${f(om,6)} | ${f(r.sJ,6)} | ${f(r.sJslot,6).padStart(9)} | ${f(r.infl,2).padStart(4)} | ${(r.pred?f(r.pred,6):'—').padStart(13)} | ${(r.pred?f((om-r.pred)/r.sJ,2):'—').padStart(7)} | ${(r.pred?f((om-r.pred)/r.sJslot,2):'—').padStart(7)} | ${r.pred?f(100*(om/r.pred-1),2)+'%':'—'}`); } console.log(' σ_J(reg) is the pre-registered Poisson floor on the TRIPLE count, kept as registered.'); console.log(' σ_J(slot) prices the sample unit the census actually walks, the SLOT, exactly.'); console.log(' Both are floors: neither prices clustering BETWEEN slots. z(reg) is therefore an'); console.log(' upper bound on significance, and TEST 2\'s relative d has no σ in it at all.'); console.log(`\nGATE: ${gateRun} reference levels checked, ${gateFail} mismatches.`); console.log(`[total ${f((Date.now()-T00)/1000,1)}s]`); // ============================================================================ // OUTPUT — EMBEDDED, do not hand-edit. Regenerate: // node research/qc/embed.js research/xchan-at29-01-segmented.js // invocation: node research/xchan-at29-01-segmented.js // code-sha256: 0977b0423e156d8694a3efd58c799dc9340c78efe315258e8d32aa4861a826f1 // out-sha256: d2e5a1b6f94a5c1968faebea08948a18515d9a91637256131f0472d7402e6465 // body-lines: 191 // streams: stdout // node: v22.21.0 // embedded: 2026-08-21 // elapsed: 450.1 s // ============================================================================ // XCHAN-AT29 — the segmented mixed super-W triple census // pre-registration: research/history/staging/xchan-at29-prereg.md levels: 11, 13, 17, 19, 23, 29, 31 segment: 2^21 slots // // ===== @11: W=2310 N̄(formula)=90 K=10 (13..47) // WORK COUNTS (the code's own loop counts): 154 slots, 122 scour marks, 1 segments // CUSTODY on the record's own closed form: 4S₂ = 4·Σ_{xW}1/∏Q = 0.006909 (e₃ = miss + sub = 0.00690899 vs esym 0.00690899; C(K,3) = 120 vs 120) // SUPER-W (∏Q>W): // ALIGNED obs = 0 [P1 predicts exactly 0: PASS] CRT=1.24 // MIXED obs = 2 CRT=3.73 by orientation (2,1) = 2 and (1,2) = 0 against 1.87 each // >>> JOINT DEFICIT J = 0.536070 1−J = 0.463930 σ_J = 0.379059 [as REGISTERED: Poisson on the triple count] // σ_J SLOT-CLUSTERED (the sample unit is the slot) = 0.374823 inflation ×0.989 // This is still a FLOOR. It prices within-slot correlation exactly and prices // between-slot clustering not at all, so a disjoint-block estimate is larger // again where it is affordable. An understated σ makes |z| too LARGE, so every // z below is a ceiling on its own significance, not a floor. // GATE vs natal-cap-39 embedded OUTPUT (14 integers + 3 rounded reals): ALL PASS // [level time 0.00s] // // ===== @13: W=30030 N̄(formula)=990 K=34 (17..173) // WORK COUNTS (the code's own loop counts): 2002 slots, 2296 scour marks, 1 segments // CUSTODY on the record's own closed form: 4S₂ = 4·Σ_{xW}1/∏Q = 0.023518 (e₃ = miss + sub = 0.02713696 vs esym 0.02713696; C(K,3) = 5984 vs 5984) // SUPER-W (∏Q>W): // ALIGNED obs = 0 [P1 predicts exactly 0: PASS] CRT=46.57 // MIXED obs = 143 CRT=139.70 by orientation (2,1) = 77 and (1,2) = 66 against 69.85 each // >>> JOINT DEFICIT J = 1.023632 1−J = -0.023632 σ_J = 0.085600 [as REGISTERED: Poisson on the triple count] // σ_J SLOT-CLUSTERED (the sample unit is the slot) = 0.154583 inflation ×1.806 // This is still a FLOOR. It prices within-slot correlation exactly and prices // between-slot clustering not at all, so a disjoint-block estimate is larger // again where it is affordable. An understated σ makes |z| too LARGE, so every // z below is a ceiling on its own significance, not a floor. // GATE vs natal-cap-39 embedded OUTPUT (14 integers + 3 rounded reals): ALL PASS // [level time 0.01s] // // ===== @17: W=510510 N̄(formula)=14850 K=120 (19..709) // WORK COUNTS (the code's own loop counts): 34034 slots, 50840 scour marks, 1 segments // CUSTODY on the record's own closed form: 4S₂ = 4·Σ_{xW}1/∏Q = 0.039784 (e₃ = miss + sub = 0.06475786 vs esym 0.06475786; C(K,3) = 280840 vs 280840) // SUPER-W (∏Q>W): // ALIGNED obs = 0 [P1 predicts exactly 0: PASS] CRT=1181.60 // MIXED obs = 3199 CRT=3544.79 by orientation (2,1) = 1622 and (1,2) = 1577 against 1772.39 each // >>> JOINT DEFICIT J = 0.902452 1−J = 0.097548 σ_J = 0.015956 [as REGISTERED: Poisson on the triple count] // σ_J SLOT-CLUSTERED (the sample unit is the slot) = 0.029937 inflation ×1.876 // This is still a FLOOR. It prices within-slot correlation exactly and prices // between-slot clustering not at all, so a disjoint-block estimate is larger // again where it is affordable. An understated σ makes |z| too LARGE, so every // z below is a ceiling on its own significance, not a floor. // GATE vs natal-cap-39 embedded OUTPUT (14 integers + 3 rounded reals): ALL PASS // [level time 0.02s] // // ===== @19: W=9699690 N̄(formula)=252450 K=435 (23..3109) // WORK COUNTS (the code's own loop counts): 646646 slots, 1154693 scour marks, 1 segments // CUSTODY on the record's own closed form: 4S₂ = 4·Σ_{xW}1/∏Q = 0.050943 (e₃ = miss + sub = 0.11411552 vs esym 0.11411552; C(K,3) = 13624345 vs 13624345) // SUPER-W (∏Q>W): // ALIGNED obs = 0 [P1 predicts exactly 0: PASS] CRT=25720.90 // MIXED obs = 74065 CRT=77162.70 by orientation (2,1) = 37139 and (1,2) = 36926 against 38581.35 each // >>> JOINT DEFICIT J = 0.959855 1−J = 0.040145 σ_J = 0.003527 [as REGISTERED: Poisson on the triple count] // σ_J SLOT-CLUSTERED (the sample unit is the slot) = 0.007107 inflation ×2.015 // This is still a FLOOR. It prices within-slot correlation exactly and prices // between-slot clustering not at all, so a disjoint-block estimate is larger // again where it is affordable. An understated σ makes |z| too LARGE, so every // z below is a ceiling on its own significance, not a floor. // GATE vs natal-cap-39 embedded OUTPUT (14 integers + 3 rounded reals): ALL PASS // [level time 0.06s] // // ===== @23: W=223092870 N̄(formula)=5301450 K=1739 (29..14929) // WORK COUNTS (the code's own loop counts): 14872858 slots, 30527907 scour marks, 8 segments // CUSTODY on the record's own closed form: 4S₂ = 4·Σ_{xW}1/∏Q = 0.058834 (e₃ = miss + sub = 0.17590031 vs esym 0.17590031; C(K,3) = 874979589 vs 874979589) // SUPER-W (∏Q>W): // ALIGNED obs = 0 [P1 predicts exactly 0: PASS] CRT=623807.07 // MIXED obs = 1807665 CRT=1871421.20 by orientation (2,1) = 904409 and (1,2) = 903256 against 935710.60 each // >>> JOINT DEFICIT J = 0.965932 1−J = 0.034068 σ_J = 0.000718 [as REGISTERED: Poisson on the triple count] // σ_J SLOT-CLUSTERED (the sample unit is the slot) = 0.001493 inflation ×2.078 // This is still a FLOOR. It prices within-slot correlation exactly and prices // between-slot clustering not at all, so a disjoint-block estimate is larger // again where it is affordable. An understated σ makes |z| too LARGE, so every // z below is a ceiling on its own significance, not a floor. // GATE vs natal-cap-39 embedded OUTPUT (14 integers + 3 rounded reals): ALL PASS // [level time 0.42s] // // ===== @29: W=6469693230 N̄(formula)=143139150 K=7863 (31..80429) // WORK COUNTS (the code's own loop counts): 431312882 slots, 994345351 scour marks, 206 segments // CUSTODY on the record's own closed form: 4S₂ = 4·Σ_{xW}1/∏Q = 0.064335 (e₃ = miss + sub = 0.25113952 vs esym 0.25113952; C(K,3) = 80993070011 vs 80993070011) // SUPER-W (∏Q>W): // ALIGNED obs = 0 [P1 predicts exactly 0: PASS] CRT=18417582.39 // MIXED obs = 53660192 CRT=55252747.16 by orientation (2,1) = 26827582 and (1,2) = 26832610 against 27626373.58 each // >>> JOINT DEFICIT J = 0.971177 1−J = 0.028823 σ_J = 0.000133 [as REGISTERED: Poisson on the triple count] // σ_J SLOT-CLUSTERED (the sample unit is the slot) = 0.000279 inflation ×2.106 // This is still a FLOOR. It prices within-slot correlation exactly and prices // between-slot clustering not at all, so a disjoint-block estimate is larger // again where it is affordable. An understated σ makes |z| too LARGE, so every // z below is a ceiling on its own significance, not a floor. // ---- BLIND SCORE against import-stein.md §3.2, adopted verbatim in the prereg // predicted 1−J = 0.028943 measured 1−J = 0.028823 σ_J = 0.000133 // TEST 1 z = -0.90 -> HIT [the REGISTERED verdict, on the registered σ] // TEST 1 on the slot-clustered σ: z = -0.43 -> HIT [reported beside the registered verdict, never in place of it] // TEST 2 d = -0.41% -> TIGHT // J-BAND CLAUSE (J < 0.94 or J > 1.00 ⇒ the object changed): J = 0.971177 -> does not fire // RIVALS, fixed before the run: N2 constant·F 0.029145 (z=-2.43) | N3 ∝1/lnW 0.028990 (z=-1.26) // the same rivals on the slot-clustered σ: N2 z=-1.15 | N3 z=-0.60 // ratio (1−J)/F with F from the record (0.7537%): measured 3.8242 vs predicted 3.8399 // custody: 4S₂ recomputed here = 0.028943 vs the record's 0.028943 (agrees to the printed digit); ratio on this file's own F (0.7537%) = 3.8240 // [level time 13.17s] // // ===== @31: W=200560490130 N̄(formula)=4151035350 K=37534 (37..447829) // WORK COUNTS (the code's own loop counts): 13370699342 slots, 33742377278 scour marks, 6376 segments // CUSTODY on the record's own closed form: 4S₂ = 4·Σ_{xW}1/∏Q = 0.068057 (e₃ = miss + sub = 0.33095269 vs esym 0.33095269; C(K,3) = 8812286043484 vs 8812286043484) // SUPER-W (∏Q>W): // ALIGNED obs = 0 [P1 predicts exactly 0: PASS] CRT=565017131.17 // MIXED obs = 1653241687 CRT=1695051393.52 by orientation (2,1) = 826645240 and (1,2) = 826596447 against 847525696.76 each // >>> JOINT DEFICIT J = 0.975334 1−J = 0.024666 σ_J = 0.000024 [as REGISTERED: Poisson on the triple count] // σ_J SLOT-CLUSTERED (the sample unit is the slot) = 0.000051 inflation ×2.124 // This is still a FLOOR. It prices within-slot correlation exactly and prices // between-slot clustering not at all, so a disjoint-block estimate is larger // again where it is affordable. An understated σ makes |z| too LARGE, so every // z below is a ceiling on its own significance, not a floor. // ---- BLIND SCORE against import-stein.md §3.2, adopted verbatim in the prereg // predicted 1−J = 0.024784 measured 1−J = 0.024666 σ_J = 0.000024 // TEST 1 z = -4.93 -> MARGINAL [the REGISTERED verdict, on the registered σ] // TEST 1 on the slot-clustered σ: z = -2.32 -> HIT [reported beside the registered verdict, never in place of it] // TEST 2 d = -0.48% -> TIGHT // J-BAND CLAUSE (J < 0.94 or J > 1.00 ⇒ the object changed): J = 0.975334 -> does not fire // RIVALS, fixed before the run: N2 constant·F 0.024818 (z=-6.35) | N3 ∝1/lnW 0.025165 (z=-20.81) // the same rivals on the slot-clustered σ: N2 z=-2.99 | N3 z=-9.80 // ratio (1−J)/F with F from the record (0.6419%): measured 3.8426 vs predicted 3.8612 // custody: 4S₂ recomputed here = 0.024784 vs the record's 0.024784 (agrees to the printed digit); ratio on this file's own F (0.6419%) = 3.8427 // [level time 436.29s] // // ===== SUMMARY — THE JOINT DEFICIT ACROSS LEVELS ===== // x | lnW | mixed super-W obs | CRT | J | 1−J | σ_J(reg) | σ_J(slot) | infl | predicted 1−J | z(reg) | z(slot) | d // 11 | 7.745 | 2 | 3.73 | 0.536070 | 0.463930 | 0.379059 | 0.374823 | 0.99 | — | — | — | — // 13 | 10.310 | 143 | 139.70 | 1.023632 | -0.023632 | 0.085600 | 0.154583 | 1.81 | — | — | — | — // 17 | 13.143 | 3199 | 3544.79 | 0.902452 | 0.097548 | 0.015956 | 0.029937 | 1.88 | — | — | — | — // 19 | 16.088 | 74065 | 77162.70 | 0.959855 | 0.040145 | 0.003527 | 0.007107 | 2.01 | — | — | — | — // 23 | 19.223 | 1807665 | 1871421.20 | 0.965932 | 0.034068 | 0.000718 | 0.001493 | 2.08 | — | — | — | — // 29 | 22.590 | 53660192 | 55252747.16 | 0.971177 | 0.028823 | 0.000133 | 0.000279 | 2.11 | 0.028943 | -0.90 | -0.43 | -0.41% // 31 | 26.024 | 1653241687 | 1695051393.52 | 0.975334 | 0.024666 | 0.000024 | 0.000051 | 2.12 | 0.024784 | -4.93 | -2.32 | -0.48% // σ_J(reg) is the pre-registered Poisson floor on the TRIPLE count, kept as registered. // σ_J(slot) prices the sample unit the census actually walks, the SLOT, exactly. // Both are floors: neither prices clustering BETWEEN slots. z(reg) is therefore an // upper bound on significance, and TEST 2's relative d has no σ in it at all. // // GATE: 5 reference levels checked, 0 mismatches. // [total 450.0s] // ============================================================================ // READINGS // // R1. THE GATE HOLDS AT ALL FIVE REFERENCE LEVELS. "GATE: 5 reference levels // checked, 0 mismatches" — fourteen integers and three rounded reals per // level against natal-cap-39's embedded OUTPUT, including 3199, 74065 and // 1807665 mixed super-W and J = 0.902452, 0.959855, 0.965932. The two // instruments share no data structure. @23 costs 0.56s here against that // file's 169.89 s and 1.203 GB, on 118.0 MB (printed as 62.0 until the // 2026-08-21 bytes-formula repair; the arrays were 118.0 MB all along). // // R2. @29 IS A HIT AND @31 IS NOT. z = -0.90 (HIT, TIGHT) against the record's // blind 0.028943; z = -4.93 (MARGINAL, TIGHT) against its blind 0.024784. // The closed form is right in scale and wrong in the fourth digit. // // R3. THE OFFSET IS STABLE AND IS ONE DETECTION. d = -0.41% and -0.48%; // 0.028943 - 0.028823 = 0.000120 and 0.024784 - 0.024666 = 0.000118. But // sigma_J is 0.000133 at @29, so only @31 detects it. No closed form is // proposed for it here. // // R4. BOTH RIVALS DIE AT @31, AS THE PREREG SAID THEY WOULD. N3 (1/lnW) at // z = -20.81 and N2 (constant multiple fixed at @23) at z = -6.35, against // -1.26 and -2.43 at @29 where the prereg registered them inseparable. // // R5. P1 IS ENUMERATED DIRECTLY AT @29 AND @31. "ALIGNED obs = 0 ... PASS" at // both, from the divisor lists rather than from a bridge subtraction. // // R6. J STAYS MONOTONE AND STAYS LARGE. 0.902452, 0.959855, 0.965932, // 0.971177, 0.975334; at @31 the shortfall 1695051393.52 - 1653241687 is // 41,809,706.52 coincidences against sigma_J = 0.000024. // --------------------------------------------------------------------------- // FIGURE PROVENANCE (added 2026-08-20, readings-traceability pass). What every // figure in the readings above that the embedded run of this file does not // contain verbatim actually is. No number above was changed. // // DERIVED IN THIS READING by arithmetic over printed values: // 0.000120 and 0.000118 in R3 are the two predicted-minus-measured offsets. // Both operands of each are printed: "predicted 1-J = 0.028943 measured // 1-J = 0.028823" at @29 and "predicted 1-J = 0.024784 measured 1-J = // 0.024666" at @31. // 41,809,706.52 in R6 is the @31 shortfall, printed as its two operands in // "MIXED obs = 1653241687 CRT=1695051393.52" and as the same pair in the @31 // summary row. The subtraction is stated in the reading itself. // // BORROWED, and the producer flags it as off-block: // 1.203 GB in R1 is natal-cap-39-triple-census.js's peak memory. Its run time // 169.89 s is in that file's embedded output, but the memory figure is not: // that file records it as "R11 [MEASURED, off-block] PEAK MEMORY 1.203 GB, // measured with /usr/bin/time -l on a separate run of identical code". So the // comparison in R1 pairs one embedded figure with one declared off-block one. // ---------------------------------------------------------------------------