A program with several kinds of evidence
Joseph's project joins two sustained questions: how to build a distributed, adaptive software system, and how to describe or measure the transitions he proposes as relevant to consciousness. The recovered work gives those questions different forms. Some documents define state updates. Some organize named agents and devices. Some contain recorded counters or experiments. Later manuscripts become more precise about the difference between a computational measurement and the meaning assigned to it.
The names MEGAMIND, Pantheon, SEFERIM, G16, DNA, Epona, Rhiannon and the other builds belong to that history. They should remain searchable even where their relationships change between versions. A name is not a guarantee that two repositories implement the same system, and a revised branch is not automatically the next chronological generation. Each build page therefore needs its own source revision, architecture, equations, artifacts, and evidence dates.
The Fibonacci and resonance manuscript
The earlier JAIR manuscript proposes a unity score, Fibonacci modulation, a weighted aggregate, and a federated architecture. It describes measurements from named devices and labels their changes as consciousness emergence. Its exact equations and original claims remain part of the record.
The recovered audit identifies concrete problems with that interpretation: the written inverse-variance equation cannot produce the reported baseline, the convergence argument chooses a desired input without proving convergence of the full dynamics, and several reported statistics disagree with the attached exports. These findings concern this manuscript and its evidence package. The website presents the original formulation alongside the audit so the evolution is visible.
Declaration and persistent state change
The later JAIC manuscript, Declaration and Persistent State Change in Artificial Systems, explicitly treats consciousness relevance as a hypothesis. It separates parameters, persistent memory, runtime context, and observations. It also gives declaration and subsequent change independent detectors, allowing either temporal order.
Its completed empirical component analyzes one public LoRA training trajectory. It reports parameter and effective-operator changes; it does not claim that this dataset measured a declaration or subjective experience. The proposed feedback intervention remains a prospective test. This is a substantive change in method and claim scope, not a renaming of the earlier resonance experiment.
The exact normalized-partition benchmark
The October 12 Quantum Flywheel packet contains a separate completed numerical experiment. Sixty generated linear stochastic networks have fully enumerated bipartition tables. Queyranne, simulated annealing, two simulated QAOA depths, and uniform sampling are compared against those tables.
The experiment answers a specific search question for a specific normalized stochastic-interaction objective. Its network models are synthetic, its QAOA execution is classical simulation, and its exact reference costs are precomputed exhaustively. Those boundaries give the results a clear interpretation and distinguish them from both Pantheon runtime telemetry and the LoRA study.
How to follow a generation
Read a build as a linked record: what problem it addressed; what input and state it used; which update rules it implemented; how it persisted or communicated state; what output was measured; and which later build reused or changed that work. Preserve an explicit link when a source states a relationship. Keep a relationship tentative when it is only suggested by similar names or formulas.
The same equation can legitimately appear in more than one build. Each occurrence should point to its original context, while a shared mathematical page can explain the common operation. Equations with changed coefficients, signs, state definitions, or index conventions must remain visible as variants.