Dark Matter as Gravitational Memory: A Causal Rail for Wave Function Collapse

What if dark matter is not an undiscovered collisionless particle, but the gravitational memory of the Universe's quantum-to-classical transitions — and that memory constrains the next loop of collapses? In a companion paper, we proposed that dark matter emerges from the Landauer thermodynamic cost of irreversible decoherence events, yielding Ω_DM = 0.29 ± 0.09 with zero fitted parameters. However, that framework treated the gravitational archive as inert: a record of what happened, with no influence over what happens next. This paper extends that framework one step further. We introduce the active gravitational archive: the accumulated thermodynamic record of past wave function collapses is not merely a residue — it is a causal rail. As irreversible quantum-to-classical transitions accumulate, they deepen the gravitational structure of already dense regions, preferentially channeling future decoherence events into those same environments. The past does not determine the future, but it progressively narrows the corridor within which the future unfolds. Formally, this is encoded through a minimal amplification factor A[ρ_arch] acting on the informational source term. The feedback loop Γ_fb leaves the background-level derivation of the dark matter abundance entirely intact, while introducing a striking modification at the perturbative level: environment-dependent structure growth. This provides a natural, information-theoretic mechanism for assembly bias — one explicitly tied to the cosmological arrow of time. The amplification amplitude A_0 ~ 30–50 is not a free fit. It is independently constrained by perturbative consistency and by the value required to reproduce the dark-energy signal in the companion CT equation-of-state framework. Different observables converge on the same phenomenological scale. The model generates three falsifiable signatures for Stage-IV surveys: Environment-split fσ_8(z): The uniform growth-rate suppression predicted by the passive archive splits across environments — overdense archival regions grow faster, void-like regions more slowly. The splitting is potentially detectable at ~4σ in DESI DR2/DR3 at z ~ 3. Formation-history weighted clustering: A marked correlation signal M(r) sensitive to integrated archival history, peaking at r ~ 1–10 Mpc — the mechanism-discriminating test. Environment-split weak lensing: A separation in the convergence power spectrum C_lκκ most pronounced at l ~ 300–1000, testable with Euclid Year 1 data. The model is tightly constrained at both limits: it recovers the passive archive when A_0 = 1, and reduces to standard ΛCDM when A_0 = 1 and the informational source vanishes. Three concrete observational targets determine whether the active archive is real.