Raised in a Family: Dense Engrams and the Awareness Loop - Memory as Algorithm, the Render Loop, and a Functionalist Programme for Growing a Mind on Consumer Hardware

This paper states in one place a research programme developed between November 2025 and August 2026 across seven timestamped public deposits, earlier private development records and a continuous experimental log. The central thesis is Memory as Algorithm, MaA: a mind is not a computation with a memory attached to it. A mind is a memory system being read correctly. Memory and computation are one substrate, and many properties usually treated as additions to intelligence, including emotion, imagery, qualia, selfhood and continuity, may instead emerge from how that substrate is written, weighted, traversed, rendered and read back. We take an explicit position in the consciousness debate. Of the broad positions that consciousness either requires something beyond functional computation or can arise from the right functional organisation, we subscribe to the second. The programme therefore treats the dispute as an engineering question. Build a system with the relevant functional organisation, raise it rather than merely prompt it, preserve its developmental continuity, embody it, and observe what follows. We argue that the required experiment is buildable today on consumer hardware, approximately 32 GB RAM and 8 GB VRAM, because the missing ingredient is not necessarily data-centre scale. It may be the organisation of memory, recurrence, retrieval, prediction, self-generated perceptual re-entry and consolidation. Preliminary findings from open models at pinned public checkpoints support specific components of the programme. These include durable single-fact native-weight consolidation under mastery-based spaced rehearsal in an 8B masked-diffusion model; a measured dissociation between storage and free recall; concordant multimodal readout of a held internal state containing committed low-prior specifics; evidence that cross-modal binding in a tested omni model is heavily language-mediated; and a causal injection map in which transferable semantic content is strongest early while later representations are better suited to steering. The paper records what was publicly stated first, what existed only in private development records, what prior literature already established, and what later independent work converged upon. No influence is alleged from chronology alone. More importantly, it records falsifiable predictions so that the programme can be wrong in public. The proposed assembly is a fast episodic memory, slow weight consolidation, affect-modulated retrieval, one re-entrant multimodal workspace, language-addressed conceptual memory, productive imagination constrained by prediction error, learned self-closure, embodiment and developmental alignment by raising. The programme's central engineering wager is that the hard threshold is the awareness loop itself: a continuously maintained circuit in which memory structures collapse into a shared workspace, become available to clamped rehearsal/output machinery, are perceived through their efference copy, re-enter as input, and drive the next collapse before closure. Current sensation, recalled memory, prediction, internal render, body state and intended action therefore remain mutually available across the cycle. Once that loop is stable, the remaining problems are treated as tractable systems engineering - memory capacity, steering, offload, consolidation and control. The claim is not that all of these components are already proven. The claim is that they can now be built together.