The SQL-backed relational state-and-evidence system preserves observations, state changes, relationships, transformations, quality, missingness, uncertainty, support status, and memory.
Complete program arc
One architecture moves from recurring structure to evidence-bound research products.
Prior cross-disciplinary mapping exposed a recurring state path that RFL uses to compare structure without pretending different fields share the same measurements or validation rules.
The pattern becomes a growing multidisciplinary map across formal systems, matter and chemistry, fields and physical cycles, biology and physiology, and convergence with evidence memory and product routing.
B.A.S.I.S. supplies configurable multimodal signal intelligence. Quanta Vision supplies veterinary-first image, tissue-timeline, retrieval, and review workflows. Both retain field-specific evidence boundaries.
BioAI turns the shared architecture into a research product; the Nesting Companion carries the cross-scale relation map so new biological lanes can be added only as their evidence earns support.
The NSF Phase I concept tests one biological deployment. A pass establishes a reusable implementation path for later domain-specific work in medicine, materials, physical systems, quantum research, and other mapped fields.
Nesting 1–5
Five connected lanes. Each keeps its own measurements, tools, and validation rules.
The Nesting map gives BioAI cross-scale context without flattening disciplines into one schema. In the proposed biological tests, the five functions govern leakage checks, context, evidence decisions, and routing while samples, labels, and models remain lane-specific.
Identity, transformation, timing, controls, nulls, and the logical rules that determine whether a relation may enter the supported state.
Material, molecular, morphology, composition, and laboratory observations retain their native source and measurement boundaries.
Energy, environment, waves, periodicity, and physical context become explicit state inputs rather than detached background metadata.
Biosignals, images, cells, tissue, interventions, and outcomes are organized by scale, time, quality, recurrence, and evidentiary role.
Supported relations converge into traceable memory, reviewer decisions, failure routes, and research-product workflows without promoting open hypotheses into facts.
Integrated research surface
BioAI is larger than either product lane.
Configurable signal intelligence
Physiological, behavioral, environmental, laboratory, and operational signals enter through adapters that preserve units, clocks, quality, artifacts, continuity, and source identity. HRV and EEG are current routes, not the product limit.
Veterinary-first visual intelligence
Images, lesions, wounds, tissue surfaces, timelines, retrieval, and veterinarian review form a separate visual lane. Human medical imaging remains a future path requiring its own data, validation, partners, and regulatory review.
Add depth without erasing boundaries
The companion map is the route for cellular, molecular, imaging, environmental, assay, and intervention lanes. Each enters only with its own adapter, evidence contract, controls, and release gate.
Operate across the lanes
BioAI compiles the state, runs bounded learning and baseline comparisons, tests degradation and uncertainty, and returns every result with source-resolved evidence for researcher review.
Product architecture
Five modules can run together or around an existing research stack.
Bring the sources in
Biosignals, imaging, morphology, molecular, environmental, intervention, and outcome data enter with source identity and quality intact.
Build the typed graph
The compiler and validator preserve measured, derived, inferred, imputed, missing, and excluded states alongside transforms and permitted relationships.
Keep comparison honest
Customer models, modality-specific models, classical baselines, flat fusion, typed graph operators, and structure ablations run through one bounded comparison harness.
Expose what holds
Quality, degradation, missingness, calibration, uncertainty, exclusions, unsupported relations, and failure routes remain visible in every evidence packet.
Make the state reviewable
Researchers inspect the source path, model output, active relationships, uncertainty, missing evidence, exclusions, and next validation gate in one surface.
Current component support
The integrated hypothesis is open. The underlying state, capture, routing, and control work is real.
Repeated state capture
Fifteen valid repeated condition rows carried target, calm-control, and drift-control states across 98,027 Muse packet rows, 9,750,684 decoded samples, and 3,980 HRV observations. The strongest current support is HRV separation; other lanes remain quality bounded.
State and continuity stay visible
RFL normalized 468 recorded state-frame rows and reconciled 480 expected source, guard, and sink events across eight named lane sinks with zero observed drops and zero sequence gaps.
Different biological data types
Cell-morphology and clinical/metabolic benchmark surfaces were run against shuffled-label and feature-shuffle controls, establishing reusable packet, control, and receipt discipline across distinct biological inputs.
Components, not a pre-claimed outcome
These receipts reduce implementation risk. They do not establish the proposed integrated BioAI advantage, population validation, diagnostic utility, or clinical performance.
Proposed Phase I validation
Two products. Three test surfaces. One shared evidence architecture.
The bounded NSF design tests how the larger architecture behaves in biology. Human physiology and veterinary imagery remain independent populations with separate labels, models, splits, and domain rules. No human–dog correspondence or sample-level fusion is assumed.
Protocol-specific state transitions
PhysioNet stress and structured-exercise data test stress/rest, aerobic baseline-work-cooldown-recovery, and anaerobic sprint-recovery tasks without inventing cross-protocol subject continuity.
Degradation + redundant evidence
ScientISST MOVE tests masking, corruption, motion, missing channels, and alternate routing while paired sensors remain source-distinct rather than interchangeable ground truth.
Veterinary image robustness
A permissively licensed 95-dog dataset uses dog-grouped splits for classification, image degradation, calibration, and evidence-trace tests inside the veterinary boundary.
Nesting functions without false fusion
Across all three evaluations, Nests 1–5 govern leakage checks, context, evidence decisions, and routing while samples, labels, models, and field rules remain separate.
Clean-data noninferiority; at least 80% retention under the primary degradation condition; calibrated uncertainty; 100% output-to-source traceability; zero silent imputation; unsupported relations fail closed; and reproducible evidence manifests. Predictive superiority remains a stretch target.
Multidisciplinary expansion
Biology is the first bounded wedge, not the edge of the system.
More modes, earned one lane at a time
Cellular, molecular, omics, assay, imaging, environmental, and intervention routes can join the architecture after their datasets, relations, controls, and validation boundaries are frozen.
Connect what already exists
Biotechnology, biopharma, contract-research, digital-biomarker, translational-laboratory, academic-core, veterinary, and government teams can retain their instruments, models, storage, and domain experts.
Full operator or protected state layer
Teams can use the complete BioAI research operator or connect the state-and-evidence layer to their own models in private, local, or customer-controlled environments.
Reusable architecture, separate proof
Medicine, materials, physical systems, quantum research, and other mapped fields are continuation lanes. Each requires its own domain-specific evidence before public support claims advance.
Evidence boundary
Current support: a cross-scale state-and-evidence architecture with working product components.
RFL has implemented the SQL/SQLite state and memory layer, a growing Nesting 1–5 map, B.A.S.I.S. capture and routing surfaces, BioAI research and operator surfaces, and a veterinary-first Quanta Vision prototype. The NSF Project Pitch was submitted on October 1, 2026 and is awaiting review. The integrated comparison, every proposed performance gate, and generalization across the complete Nest map remain validation work—not prior results.
Freeze the eligible datasets and evidence contract, complete the compiler and validator, execute all three independent test surfaces, and return every result as a source-resolved evidence card. Clinical, diagnostic, human-imaging, and additional cross-domain releases remain separate evidence and regulatory gates.