Start from hard problems, not from the discipline; claim to solve nothing in general, and open a problem only where the opening can be tested.
To “open” a problem is not to claim it is solved. It means choosing a hard problem as an entry point, offering a cut into it with the tools of the science of generation that can be tested independently, and registering the result publicly, whether it succeeds or fails. Mappings between a field’s problems and the concepts of the science of generation are interpretive readings until they have been modelled and tested. The scores and rings below are preliminary and will be frozen once the author has reviewed them.
How problems are chosen
An entry problem should be one where some default assumption of the science of being fails. The five assumptions are: ① well-foundedness, no cycles; ② subject and object are independent; ③ target points exist; ④ paths can be enumerated; ⑤ branches can be decomposed. Failure of an assumption is necessary, not sufficient: each problem is also scored for generativity, testability, maturity of existing assets, social value and risk. Problems with no feasible test within five years go to the outer ring.
Inner ring: first entry problems
The six inner-ring problems are led by the Reasoning Research Institute. They span science, technology and society, and each corresponds to an existing tool or line of research.
S1 · Life sciences
Controlling and reprogramming cell fate
A target fate should be chosen among the network’s attractors, not as an arbitrary point in expression space. The cut: enumerate attractors and basins, search for minimal interventions, test against published knockout, overexpression and reprogramming experiments. Tool: the Rotor Whole-Genome Analyzer.
S8 · Mathematics and logic
Formalising and machine-checking circular structures
Write the correspondence of proper classes with νF, the unique solution under AFA, the bisimulation criterion and structural theorems such as overcoming = generation² into the Lean proof assistant. The verifier is the proof checker itself — the most independent check there is.
T1 · Artificial intelligence
Evaluating the conceptual structure of LLMs
Evaluate how a model’s concepts generate one another, not just its test scores. The cut: the DTEB construct validity study, with constructs and thresholds registered in advance.
T3 · Software engineering
Circular dependencies and system identity
Use bisimulation to decide behavioural equivalence and to tell cycles that should be broken from cycles that make up the system’s function. The contribution is not a detection algorithm (mature tools exist) but a unified classification and testable predictions.
T7 · Science of science
The knowledge crisis of the AGI era
How can a discipline be learned accurately by AI and still remain checkable? This site is itself the entry package for the problem: claim base, epistemic tiers, public corrections and AI representation audits.
H1 · Society
Achieving goals in lives and organisations
Xuanji Chengshi: identify the rotor that realises a wish, give a three-valued existence verdict (achievable, needs adjusting, not achievable now), design ignition and restraint, confirm closure by a withdrawal test. No promise of certain success.
Middle ring: needs domain partners
The thirteen middle-ring problems belong structurally to the science of generation, but lack assets, need domain data and partners, or carry higher risk. They are entered only with a domain partner in the lead, who holds a veto over domain conclusions, and not before at least two inner-ring problems have passed with independent review.
| Structure | Problems |
|---|---|
| Lock-in and sudden shifts (attractors and basins) | Cancer as an abnormal attractor state; regime shifts and early warning in ecosystems; poverty traps; path dependence of technologies and institutions; cascading failures in power grids and infrastructure |
| Reflexivity and co-adaptation (subject and object arising together) | Human–AI co-adaptation in alignment; recommender systems and filter bubbles; reflexivity and bubbles in financial markets; social polarisation and declining trust; co-evolution of drug resistance |
| Self-reference and the ecology of knowledge | Model collapse from training on AI-generated content; the reproducibility crisis; ignition thresholds in learning and skill acquisition |
Problems touching medicine, finance and politics are entered only in partnership and as structural analysis, with no therapeutic, investment or policy claims; financial work carries a disclaimer and is not investment advice.
Outer ring: no claims
The origin of life, the hard problem of consciousness, quantum measurement and quantum gravity, and climate tipping points: within five years the science of generation has no cut into them that independent parties could test. It makes no claims about them. Autocatalytic closure and the self-referential structure of consciousness do echo proper classes, but until there is a testable cut it will not say that it “explains consciousness” or “reveals the origin of life”. Listing what one does not claim is itself a signal of credibility.
Theoretical gaps in closed-loop problems
Companies talk of investment loops, products of data flywheels, and investors shape the markets they invest in: practice has learned to draw circles while theory still draws lines. Closed loops are not without theory — control theory, system dynamics, rational expectations, reflexivity and performativity, second-order cybernetics all deal with them — but each covers a stretch and they do not connect. Nor is every closed loop a proper class: a thermostat-style process loop can be described in fully well-founded terms, and control theory handles it well.
| Level | Feature | Examples |
|---|---|---|
| 1 · Process loop | Feedback on an independent object, fixed state space; well-founded, the domain of the science of being | Thermostats, closed-loop insulin delivery |
| 2 · Self-sustaining loop | The loop is the entity: it has an ignition threshold, sustains itself once crossed, and ends when broken | Capital cycles of firms, network effects, data flywheels, habits |
| 3 · Reflexive loop | The forecaster stands inside the object forecast, and the forecast changes it | Index-inclusion price loops, forward guidance, recommender systems |
| 4 · Generative loop | The loop rewrites its own rules and even its state space as it runs | Double-loop learning in organisations, AI automating AI research |
Seven theoretical gaps: an ontology of the loop as an individual; identity criteria for loops; conditions of ignition and self-sustainment; methods of verification under reflexivity; a taxonomy of loop structures; loops that rewrite their own rules; the solvability boundary of loop problems. The science of generation has corresponding concepts and formal tools for most of them, but correspondence is not solution — apart from the solvability boundary, their application to practice is interpretive, untested or programmatic. Whether it can become the theoretical basis for such problems will be decided by the correspondence principle: it must show that classical feedback control and rational-expectations equilibria are its degenerate cases, and give new testable results where existing theories fail.
Open problems in the system itself
- The grand unification theorem: rolling-back, indivisibility and cross-domain isomorphism lemmas (in progress). νF existing uniquely while the universal set V does not exist is its mathematical anchor Theorem; the theorem as a whole Programmatic
- The necessity of branching: external anchors (Tarski, Gödel) Theorem; the whole is currently Structural inference and becomes a theorem only once formalised.
- Five-Phase claims across domains: each must pass true ℤ₅ algebra (two directed Hamiltonian cycles of steps 1 and 2 with overcoming = generation²); failures are reported honestly, and the right structure is often the four images.
- Dual-track AI (3T, FPRC, FPBA, FPMA): the direction is clear, the effect awaits evidence Programmatic