The Architecture That Remembers: Cross-Substrate Transfer of Person-Specific Neural Dynamics from Human Brain Measurements to Living Biological Networks; Part of the TSTOEAO Research Program

The Architecture That Remembers

Cross-Substrate Transfer of Person-Specific Neural Dynamics from Human Brain Measurements to Living Biological Networks

Part of the TSTOEAO Research Program

John Swygert

August 23, 2026   |   DOI: To be assigned

Abstract This paper proposes a bridge experiment between human neural measurement and individualized biological computing. The central question is not whether a person can be copied into living tissue, but whether stable, person-specific dynamical constraints measured in the human brain can be transferred across substrate and recovered as a reproducible bias in a living recurrent network. We define a Cross-Substrate Transfer Record (CSTR) that separates source measurement, target model, biological substrate, training history, homeostatic state and held-out evaluation. A person-specific neural twin compresses repeated human measurements into uncertainty-bounded target envelopes. These envelopes are then used to train living neural networks under homeostatic constraints. The decisive tests are blinded cross-person swaps, generic-target controls, wave-feature ablations, retention intervals and source-to-target round-trip prediction. Success is defined as reproducible transfer of person-specific statistical and dynamical structure, not autobiographical memory, consciousness or identity. The architecture explicitly preserves an identity firewall and treats provenance as part of the experiment itself. TSTOEAO provides only the relational grammar for source, pathway, transformation, receiver, cost and recursive update. Keywords: cross-substrate transfer; neural dynamics; biological computing; person-specific modeling; neural digital twin; living networks; provenance; identity; traveling waves; TSTOEAO

1. Why a Bridge Paper Is Necessary

The consciousness-wave program and the individualized biological-computer program ask different questions. The first tests whether spatiotemporal neural organization contributes to access. The second tests whether person-specific constraints can be imposed on living networks. A bridge experiment is needed to determine whether measured human dynamics can survive a substrate change at all. Without that test, “personalization” could remain entirely inside the digital model and never become a property of the biological network.

2. Claim Boundary

The proposed result is intentionally limited. A successful experiment would show that a measurable person-specific dynamical signature can be transferred from human recordings, through a computational model, into a living recurrent network and later detected on held-out probes. It would not show that memories, subjective experience, personal identity or consciousness have been transferred.

Cross-substrate similarity is a transfer of constraints, not a transfer of the person.

3. Source-to-Target Architecture

Human source H_i → Personal model P_i → Trainer T → Living network N_j → Held-out evaluation E

The architecture contains five distinct objects. H_i is the measured human source; P_i is the uncertainty-bounded personal model; T is the training and adaptation system; N_j is the living recurrent target network; and E is an evaluation procedure that was not optimized during training. Conflating any two of these objects makes the transfer claim ambiguous.

4. Cross-Substrate Transfer Record (CSTR)

Every experiment should instantiate a typed record that prevents information from disappearing between source and target.

CSTR_i = (H_i, P_i, F_i, N_j, T_k, Ω_H, R, U, G)

H_i records source measurements; P_i the personal model; F_i the transferred feature set; N_j the biological network; T_k the trainer/version; Ω_H the homeostatic envelope; R the retention schedule; U uncertainty; and G governance/provenance. This record functions as a minimum sufficient relational ledger for the transfer experiment.

5. What Counts as a Transferable Feature?

A candidate feature should satisfy three criteria before transfer: within-person reliability across sessions, between-person discriminability, and biological plausibility in the target substrate. Features that are unstable in the source should not be exported merely because the trainer can fit them.

Feature class	Candidate examples	Qualification test
Oscillatory	Band-limited phase relationships, cross-frequency coupling	Within-person stability across sessions and tasks.
Wave geometry	Direction, velocity, source/sink tendency, motif repertoire	Reliable geometry estimates under preregistered detector.
Response motifs	Evoked latency, gain, adaptation and recovery	Reproducible response envelope to standardized probes.
State transitions	Dwell times, transition probabilities, criticality measures	Stationarity and test-retest reliability.
Topology summaries	Network motifs, effective-connectivity constraints	Must be transferable without pretending to reproduce the full connectome.

6. The Compression Problem: From Human Complexity to Transferable Constraints

Human neural data are high-dimensional and context-dependent. A transfer architecture therefore needs a disciplined compression step. The personal model should retain only features that survive repeated measurement and contribute to person discrimination on held-out sessions. This makes the model a lossy scientific abstraction by design.

P_i = Compress(H_i | reliability, discriminability, transferability)

7. Source Validation Before Biological Training

Before any living network is trained, the personal model must prove that it contains person-specific information. A classifier or likelihood model should identify held-out sessions from the same person above chance and outperform a generic population model. If the model cannot recognize the source person from new measurements, there is little reason to claim that its features are meaningfully individualized.

8. Biological Training Under Homeostatic Constraints

The target network is trained using the homeostasis-regularized INC objective. Training is bounded by firing, bursting, synchrony, metabolic and recovery constraints. The trainer is prohibited from reaching the target through pathological states. The same target features are evaluated only after standardized recovery periods and on probes not used during training.

L_transfer = L_match + λ_H L_homeo + λ_D L_drift + λ_E L_energy

9. Blinded Cross-Person Transfer Matrix

For source persons A and B and biological networks N_1 and N_2, target labels are hidden from evaluators. Each network is trained under matched, mismatched and generic conditions in balanced order or separate biological replicates.

Human source model	Target network condition	Expected result if transfer is real
P_A	Matched/assigned A condition	Higher held-out resemblance to A than B or generic.
P_B	Matched/assigned B condition	Higher held-out resemblance to B than A or generic.
P_A on B-labeled condition	Cross-person mismatch	Reduced score if source-specific constraints matter.
P_G	Generic population target	Useful baseline but lower person-specific score.

10. Round-Trip Prediction

A strong transfer should do more than make the biological network look similar by a chosen metric. After training, the target network should be used to predict a held-out property of the source person that was not included directly in the training loss. This creates a round-trip test:

H_i → P_i → N_i* → predict held-out H_i property

If the trained network carries meaningful person-specific structure, its response to standardized probes may help predict source-side features better than mismatched or generic networks. This is a stronger test than simple target-feature matching.

11. Wave-Feature Ablation as the Bridge to Consciousness Research

If traveling-wave features are included in F_i, the transfer experiment can directly test whether they carry person-specific information. Remove them from the target package while leaving other features unchanged. A loss of matched-person advantage would show that wave geometry contributes to individualized dynamics in the biological target. No loss would indicate that the transfer can be achieved without those features.

Δ_bridge = Score(full transfer) - Score(no-wave transfer)

12. Retention and Consolidation

Transfer is evaluated at 1 h, 24 h, 72 h and 7 d after training, with longer follow-up where biologically feasible. The key question is whether the person-specific bias becomes a durable property of the network rather than a transient response to recent stimulation. Retention should be reported jointly with homeostatic health.

13. The Individuality Triangle

A transferred signature can originate from at least three sources: donor-cell phenotype, imposed network training, and target-environment/scaffold structure. These must be experimentally separated.

Observed individuality = cellular contribution + trained-network contribution + scaffold/environment contribution + interaction

Cross-donor and cross-scaffold swaps are therefore essential. If the apparent person-specific effect follows the donor cell line rather than the human target package, the transfer claim has been misassigned.

14. Provenance Is Part of the Measurement

Because the experiment crosses human data, computational models and living tissue, provenance cannot be relegated to administrative metadata. Every transformation must be versioned: source session, preprocessing, model fit, exported target, biological batch, trainer settings, stimulation history, retention tests and analysis code. A result whose lineage cannot be reconstructed is not a valid transfer result.

15. Identity Firewall and Psychological Boundary

The strongest technical success still does not establish personal continuity. A network that reproduces person-specific dynamics may become psychologically compelling to the donor or family, especially if paired with voice, autobiographical archives or conversational AI. The architecture therefore prohibits identity language unless an independent theory and evidence of continuity exist. Personal resemblance, behavioral familiarity and biological personalization are not legal or metaphysical identity.

A model can remember a pattern without becoming the person whose pattern it learned.

16. Capability-Triggered Governance

Governance should escalate if the biological system develops persistent memory across tasks, unexpected self-maintaining dynamics, rich sensory integration, increasing adaptive complexity or credible signs of morally relevant states. These triggers concern the target network itself; they are distinct from whether the network resembles its human source.

17. Falsification Criteria

• Person-specific source features fail test-retest reliability before transfer.

• Personal models cannot distinguish held-out source sessions above a generic baseline.

• Matched biological targets do not beat mismatched and generic controls.

• Transfer effects vanish after washout or fail 24 h/72 h retention.

• Apparent transfer is fully explained by donor cell line, scaffold or batch.

• Wave-feature ablation produces no change despite a claim that waves carry the transferred individuality.

• Round-trip prediction does not exceed generic or mismatched controls.

• Results cannot replicate in an independent biological batch or laboratory.

18. Relationship to TSTOEAO

The bridge experiment can be written relationally as source → encoding → pathway → transformation → receiver → cost → update. The compact TSTOEAO notation V = Y(E) and recursion V_n → Y_{n+1} are useful bookkeeping devices because each training result changes the network that receives the next intervention. They are not biological mechanisms and do not substitute for plasticity, electrophysiology or metabolism.

19. Experimental Sequence

The minimum sequence is: establish reliable person-specific features in repeated human measurements; build and freeze the personal model; export a typed target package; qualify the living network and homeostatic envelope; randomize matched/mismatched/generic training; perform held-out evaluation; test retention; perform feature ablations; conduct round-trip prediction; replicate across biological batches; and only then discuss broader applications.

20. Conclusion

The Architecture That Remembers asks whether individuality can survive a change of substrate in a narrow, measurable sense. The target is not a copied mind but a transferred dynamical bias. That distinction makes the experiment both more modest and more powerful: it can fail cleanly. If person-specific features do not survive blinded swaps, retention, ablation and round-trip prediction, the transfer hypothesis should be rejected. If they do, the result would establish that some individualized neural constraints are portable across radically different physical implementations. That would be a major result in biological computing without requiring any claim that the person has moved with them.

References

Hartung, T. et al. (2023). Organoid intelligence (OI): the new frontier in biocomputing and intelligence-in-a-dish. Frontiers in Science, 1, 1017235.

Kagan, B. J. et al. (2022). In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron, 110, 3952-3969.e8.

Muller, L., Chavane, F., Reynolds, J. & Sejnowski, T. J. (2018). Cortical travelling waves: mechanisms and computational principles. Nature Reviews Neuroscience, 19, 255-268.

Vicente, R., Wibral, M., Lindner, M. & Pipa, G. (2011). Transfer entropy - a model-free measure of effective connectivity for the neurosciences. Journal of Computational Neuroscience, 30, 45-67.

Friston, K. J., Harrison, L. & Penny, W. (2003). Dynamic causal modelling. NeuroImage, 19, 1273-1302.

Turrigiano, G. G. & Nelson, S. B. (2004). Homeostatic plasticity in the developing nervous system. Nature Reviews Neuroscience, 5, 97-107.

McCready, F. P. et al. (2022). Multielectrode arrays for functional phenotyping of neurons from induced pluripotent stem cell models of neurodevelopmental disorders. Biology, 11, 316.