Shell-Residue Battery V3 - N=100,000 zeros - Odlyzko table - float64 - 0.01s load
Everything flips at N=100k.
41 pass - 20 fail (all findings) - magnitude is uniform - the structure is in the 2-step memory
41
Pass
20
Fail=Findings
100k
Zeros
Alternation
47.26%
was 68.5% at N=500 - z=-17.35 - REVERSED
gamma1 rank at N=100k
NOT #1
n=21962 has |r|=10^-8 - gamma1 is ~99sigma above E[min]
Magnitude KS p
0.9839
genuinely uniform - max bin dev 1.32%
Sign balance
49997/50003
diff=6/100k - p=0.987
rn lag-2 ACF
-0.407
dominant period-2 structure
Alt-indicator ACF
-0.606
crossings anti-cluster - NOT Markov
Run length 2 fraction
58.8%
geometric model predicts 24.9%
All taxonomy classes
<2%
vs uniform null - magnitude flat everywhere
The Central Result
The alternation rate reversed: 68.5% to 47.26%
N=500 - PREVIOUS
68.5% alternation
z=+8.3 - P(+->-)=0.69 - early-zero artefact
N=100k - TRUE PICTURE
47.26% alternation
z=-17.35 - P(+->-)=0.4726 - same-sign persistence wins
The 68.5% at N=500 was driven by shared-shell opposite-sign pairs (38.3% of early pairs, 78.9% opposite). At N=100k that effect is diluted. The dominant pattern: runs of length 2 in same sign (58.8% of all runs). P(+->-)=0.4726 < 0.5 at scale. Same-sign persistence is the global truth.
All Findings
What 100,000 zeros actually say
F-FLIP - FINDING
The alternation rate INVERTED at N=100k: 68.5% to 47.26%
At N=500, alternation was 68.5% (z=+8). At N=100k: 47.26% (z=-17.35). REVERSED. The 68.5% was inflated by shared-shell opposite-sign pairs in early zeros (38.3% of pairs, 78.9% opposite-sign). At scale: P(+->-)=0.4726, P(->=+)=0.4725 — both below 50%. Same-sign persistence is the global truth.
F-RANK - FINDING
gamma1 is NOT rank-1 at N=100k — n=21962 has |r|=10^-8
n=21962 (gamma~19586.26) is essentially on its shell to float64 precision. u=10^-8. gamma1 (u=0.00155) is ~154,000x farther. u1/E[u_min at N=100k]=98.96. Correct claim: gamma1 is rank-1 among the FIRST 500 zeros only.
F-PERIOD2 - FINDING
The sign process is period-2: lag-2 ACF=-0.407, alt-indicator ACF=-0.606
Lag-1: -0.029 (weak). Lag-2: -0.407 (STRONG). Lag-4: +0.370 (STRONG). Crossings anti-cluster (alt-indicator ACF=-0.606). 58.8% of runs have length 2 vs 24.9% for Markov. The zeros travel in (-,+) or (+,-) PAIRS with 2nd-order memory, not simple Markov alternation.
F-GAPFLIP - FINDING
Alternating pairs have LARGER gap than same-sign (mean 0.872 vs 0.639, t=123)
Same-sign consecutive zeros are CLOSER TOGETHER. Shared-shell pairs (close in gamma, opposite sign 78.9%) create local alternation but at global scale: small gap -> same-sign persistence. Gap and sign are structurally linked.
F-UNIFORM - CONFIRMED
Magnitude uniformity confirmed to 1.32% max deviation at N=100k
Chi-sq=6.15 (p=0.73). KS stat=0.0046 (p=0.984). Weyl sums k=1..5 below 0.004. Mean=0.49996. Std=0.28864 (expected 0.28868). Genuinely Uniform(0,1).
F-SIGN - CONFIRMED
Sign balance: 49997/50003, diff=6/100k, p=0.987
Perfect phase divider. P(r>0)=P(r<0)=0.5 to 4 decimal places at scale.
F-TAXONOMY - CONFIRMED
All 9 taxonomy classes match uniform null within 2%
FLAGSHIP 0.979, DEEP 0.987, STRONG 1.001, LOCKED 0.987, ANTI-SHELL 0.988. Magnitude is flat across all tiers. Shell-attraction hypothesis has zero support.
Magnitude Uniformity - Confirmed at N=100k
u_n is genuinely Uniform(0,1) to <2% deviation
0.0-0.1
10014 (+0.1%)
0.1-0.2
10048 (+0.5%)
0.2-0.3
9868 (-1.3%)
0.3-0.4
10087 (+0.9%)
0.4-0.5
9995 (-0.1%)
0.5-0.6
10036 (+0.4%)
0.6-0.7
9868 (-1.3%)
0.7-0.8
10118 (+1.2%)
0.8-0.9
10002 (+0.0%)
0.9-1.0
9964 (-0.4%)
Chi-sq=6.15 (p=0.73) - KS stat=0.0046 (p=0.984) - Weyl k=1..5 all <0.004 - Mean=0.49996 - Std=0.28864 (expected 0.28868)
Sign Balance - Confirmed
49,997 floor / 50,003 ceiling - diff=6 - p=0.987
FLOOR (r>0)
49,997
49.997%
CEILING (r<0)
50,003
50.003%
DIFFERENCE
6
p=0.987
Alternation - Inverted at Scale
N=500 vs N=100k: the key comparison
metricN=500N=100kdirection
Alternation rate68.5%47.26%REVERSED
z-score vs 50%+8.3-17.35SIGN FLIPPED
P(+to-)0.6900.4726below 50%
P(-to+)0.6850.4725below 50%
Same-sign rate31.5%52.74%dominant
Sign lag-1 ACF-0.370+0.055sign flipped
Run-Length Distribution - Period-2 Signal
58.8% of runs have length 2 - geometric model badly wrong
run lencountobservedgeometric(p=0.473)ratio
172320.153040.472560.32x
2277890.588040.249252.36x
3118640.251050.131461.91x
42980.006310.069340.09x
5420.000890.036570.02x
6290.000610.019290.03x
Length-1: 3.1x LESS than geometric. Length-2: 2.4x MORE. Zeros travel in PAIRS with 2nd-order memory. The sign sequence is NOT a Markov chain.
rn Autocorrelations - Lag-2 Dominates
lag-1: -0.029 (weak) - lag-2: -0.407 (STRONG) - period-2 signature
lagrn autocorrstrength
1-0.02916weak
2-0.40651STRONG
3-0.15773mild
40.37019STRONG
50.20333mild
10-0.06587weak
20-0.14716mild
Negative at lag-2 (-0.407): two steps back anticorrelated. Positive at lag-4 (+0.370): correlated. Classic period-2 pattern. Alt-indicator ACF=-0.606: crossings anti-cluster (a crossing makes next crossing less likely). n=21962 is rank-1 at N=100k with |r|=10^-8 - gamma1 is NOT globally special.
Taxonomy - N=100k - All Classes Match Null
Every class within 2% of uniform — magnitude is genuinely flat
classobservedexpected (null)ratio
FLAGSHIP623636.60.9786
DEEP31403183.10.9865
STRONG1274312732.41.0008
LOCKED9871000.00.987
ANTI-SHELL49405000.00.988
MID-PHASE5003050000.01.0006
FLOOR4999750000.00.9999
CEILING5000350000.01.0001
Theory After N=100k
Magnitude is generic. The structure is in the 2-step memory.
OLD: zeros seek pi-shells / alternation is the primary law
NEW: The nontrivial zeros define a shell-centered phase process. Magnitudes u_n are genuinely Uniform(0,1). Signs are balanced (50/50). The signed residue r_n exhibits a dominant period-2 oscillatory structure: zeros travel in matched pairs of opposite sign with 2nd-order memory. The structure is not in the magnitude; it is in the 2-step rhythm of the sign.
CONFIRMED AT N=100k
  • Magnitude ~ Uniform(0,1) exactly
  • Sign balance 50/50 to 4 decimal places
  • All taxonomy classes match null <2%
  • No shell attraction in magnitude
  • Dominant period-2 structure (lag-2 ACF=-0.407)
  • Crossings anti-cluster (alt-indicator ACF=-0.606)
  • Same-sign persistence at scale (P(+->-)=0.473 < 0.5)
REVISED / CORRECTED
  • gamma1 rank-1: first 500 only, not global
  • 68.5% alternation: N=500 artefact
  • Simple Markov model: WRONG (period-2 required)
  • Alternating pairs have LARGER gap (reversed)
  • Same-sign pairs cluster at small gap
Five canon sentences after N=100k:
Magnitude is flat; the 2-step rhythm is the law.
The shell is a reference plane for a period-2 oscillation, not a target.
gamma1 is the champion of the first 500 zeros, not the universe.
Sign is balanced; pairing is the structure.
The signed residue process has 2-step memory. That is the deepest pattern in 100k zeros.