Many Lakes, One Watchtower Coupling in the Reading Layer, Common-Mode Synchrony, and the Takeover of the Cognitive Layer

Abstract The companion paper writes a person as a barrier lake with named components. This paper asks what happens when many lakes are coupled through the reading layer. The unit of a lake should be read as the range that shares one watchtower: many lakes merge in the reading layer, while bodies, gates and circumstances remain separate. The paper gives four results and one conditional inference. First, when the errors of the watchtowers of n lakes are correlated at ρ, the effective number of independent watchtowers is n/(1 + (n − 1)ρ), never more than 1/ρ. Ten thousand lakes at ρ = 0.3 have only 3.33, and a hundred lakes 3.26. The order-two effective number of a correlation matrix is a different quantity, 11.1 under the same conditions, and any citation must state which is meant. Second, common mode puts a floor under the probability of synchronous failure. The fluctuation brought by a shared reading does not shrink with the size of a lake, so the effective number of units in the mean activity of a population has a cap that depends neither on the size nor on the number of lakes. Without direct coupling between lakes, the lakes are conditionally independent given the path of the shared reading; the fraction failing in a window converges to the conditional failure probability, and the probability of synchronous failure converges to a positive plateau. In a reproducible simulation the mean failing fraction is about 2% per window in every group. With independent noise, windows in which at least a tenth of the lakes fail together occur zero times among 100 and among 1000 lakes, over 5000 windows per group. When the common mode carries 0.4 of the noise variance, the probability is 0.045, 0.044 and 0.039 for 100, 1000 and 10000 lakes and no longer falls with the number of lakes; the size at which the plateau appears is far larger than 1/ρ. Finite-size jumps are still not phase transitions by criterion T12, but the large-number reason for setting them aside as a risk no longer holds in a population that shares a reading. Third, same-source monitors fail to see precisely when something goes wrong. At a shared correlation of 0.6 the miss rate given a violation is 0.65, against an unconditional miss rate of 0.10. Fourth, lakes have become fewer over history. Every jump in transport capacity raised the numerator of the screening length; how much of the rise of the screening length came from the numerator and how much from the denominator is tested by a registered analysis protocol, and this paper does not answer it in advance. The boundaries between the three tiers are operational. Between the first and second, the correlation length of the network is compared with the physically screenable distance; between the second and third, one asks whether the screening length is finite and whether the turning point is independent of one's own action. On tier three the screening length has no definition: absorption remains, but there is no distance over which it can act. None of these four results depends on artificial intelligence. Section 6 treats artificial intelligence as a conditional inference. What is taken over is the reading, the gating and part of the set point; the takeover proceeds as a ratchet; and when hunger, a shared support, is withdrawn, several readings fall together. The paper also notes that “one watchtower” is the corner point of concentration in the interpretive layer, while reading is an additive function with no structural reason to be concentrated. If the share of an interpretive source has a self-reinforcing gain whose elasticity exceeds one somewhere, it locks in by a fold condition free of parameterisation, and after lock-in lowering connectivity need not restore many watchtowers. The paper gives an audit, four registered predictions and four directions for building cold zones, and prints its full reproduction program at the end.