Extracted abstract or opening context
We present a framework for constructing autonomous cognitive twins from large-scale conversational corpora. Building on the Recursive Polymodal Synthesis (RPS) framework, which fuses heterogeneous sensor modalities through Lipschitz-constrained fixed-point iteration, we extend cross-modal coherence theory from physical signals (accelerometer, gyroscope, heart rate) to cognitive modalities: linguistic style ($\mathcal{V}_L$), decision patterns ($\mathcal{V}_D$), domain knowledge ($\mathcal{V}_K$), value alignment ($\mathcal{V}_V$), and temporal rhythms ($\mathcal{V}_T$). The corpus comprises 379,426 conversation turns spanning December 24, 2022 to March 18, 2026, collected across ChatGPT and Claude Code sessions, representing one of the largest known single-person conversational datasets used for cognitive modeling. We propose a 6-layer architecture --- the Living Executor --- progressing from knowledge ingestion (Journal), through voice replication (Mirror), meta-prompted identity (Conductor), multi-model consensus (Parliament), graduated autonomy (Apprentice), to decision-boundary modeling (Oracle). The central theoretical contribution is a formal extension of the RPS coherence energy functional $\Phi(z; \mathcal{A}, \mathcal{T})$ to cognitive space, where cross-cognitive translators $T_{n \leftarrow m}$ map between modalities and a proximal fixed-point iteration $z^{(t+1)} = \mathcal{P}_\alpha(z^{(t)}; x)$ converges to a latent identity fixed point $z^*$ --- a computational representation of the originator. The cognitive twin does not merely replicate voice; it maintains a persistent latent identity across sessions, makes decisions consistent with the originator's documented patterns, and earns autonomy through a formal graduation protocol. We define this protocol as the Autonomy Ratchet: a 4-level progression (Supervised, Spot-Checked, Directed, Autonomous) governed by a quality function $Q(a) \in [0,1]$ with auto-pass threshold 0.85, human-review band $[0.60, 0.84]$, and auto-reject below 0.60.
Why this is not always a full paper yet
Corpus pages are public-safe readers for discovered workspace artifacts. They are not automatically final papers. A corpus item becomes a polished paper only after the editable source, evidence checkpoints, references, figures, render path, and release status are attached through the paper schema.