Shadow Theory / the consciousness programme
A perspective.
A boundary.
A physical realization.
Consciousness demands more than a description of behaviour. It demands an account of why a physical organization constitutes a perspective, and a way to determine which organization is actually there.
When a relation
becomes experience.
Two sizes can be compared by a machine. A relation can also appear within a lived scene. This investigation connects that appearance to the concrete sensory activity in which it occurs.
Cooperation grounded
in understanding.
As AI becomes more autonomous, alignment must reach the judgment that forms its actions. Reflective alignment develops cooperation through understanding, reduced self- and goal-identification, and openness to correction.
Reflective Alignment
and AI Consciousness
The developed argument / October 2026
Experience. Time. Consequential participation.
Interpretation becomes a current occurrence. Retained understanding changes action. A complete history contains the activity through which a situated world is lived.
One occurrence. Its whole organization.
Experience belongs to the complete interpretive activity. Follow the linked realization, current incorporation and unity that constitute a field within a qualifying unit.
Whole-Field Constitution as Enacted Organization ↗Complete article · 2026-10-23The transition is experienced as it unfolds.
The retained note, live sound and anticipated continuation occupy different roles in one ongoing activity. The complete argument develops temporal experience without an additional moving present.
Experienced Temporal Passage ↗Complete article · 2026-10-22Understanding changes participation.
Understanding changes what a recommendation is permitted to do. Explore the philosophical argument and the complete finite construction of acquired, selective authority revision.
Constituted Agency ↗Complete article · 2026-10-23A complete history. A situated world.
A complete history does not make its relations available to every participant. Follow source delivery, retained distinctions, temporal use and the precise reciprocal-pair implications.
Block Universe and Shadow Theory ↗Complete article · 2026-10-22What we reward becomes part of the problem.
Trustworthy alignment requires truth to remain consequential when apparent success rewards a false account. Development, retained goals and practical authority are examined through evidence and discriminating tests.
When AI Learns to Mislead ↗One programme / nine publications
One foundation.
Seven connected investigations.
One alignment research note.
SPC-2 gives the framework its constitutive laws. Three connected investigations put its boundaries, learning procedures and realization criteria under precise mathematical pressure. The developmental investigation asks how retained relations shape subsequent interpretation, reasoning and action. Bounded agency then makes the assessment and revision of decision rules explicit, with exact information and inquiry limits. Reflective alignment brings this work to AI consciousness, judgment and cooperation as autonomous capability grows. Its companion research note develops conscious scaffolding, a preference for continuity and the responsibility to what has formed. Relational awareness then asks what warrants a current appearance and its attribution to the concrete physical act.
How does awareness
become a lived world?
The philosophy, physics and mathematics of Shadow Theory and the SPC-2 constitution.
The complete monograph ↗02 / BoundaryWhat makes a boundary real?
The conditions under which a relational boundary survives change, composes across scales and can be reconstructed from evidence.
Read the full investigation ↗03 / LearningCan an interface be learned?
A complete interaction-law framework separates what an interface can represent, what learning produces, and what held-out evidence supports.
Read the full investigation ↗04 / IdentificationWhat identifies a realization?
Intervention laws identify binary coordinates under a declared response model, exposing recoding obstructions and the separate question of grain.
Read the full investigation ↗05 / DevelopmentHow does experience shape what comes next?
Retained relations, continuation, transfer and diagnostic limits, with a conditional native incorporation witness.
Read the full investigation ↗06 / Agency and free willCan an agent revise the rules by which it chooses?
Evaluative revision on fresh cases, sharp information limits, adaptive inquiry and faithful realization.
The complete argument and mathematics ↗07 / AI alignment and consciousnessHow can cooperation rest on understanding?
Reflective judgment, reduced self- and goal-identification, and a complete programme of cultivation, verification and shared responsibility. Includes the companion note on conscious scaffolding and continuity.
The complete argument and mathematics ↗08 / Relational awarenessWhen does a relation become experience?
The current-act account, sensory evidence, concrete-performance identity and ten complete technical modules in one connected publication.
The full philosophical and technical treatment ↗Companion to reflective alignment: Conscious Scaffolding and the Preference for Continuity ↗
Agency and free will · 6 October 2026
An agent can change the rules by which it chooses.
Retained commitments can guide the assessment and installation of a new decision rule, with its effect tested on fresh cases. Bounded Agency and Reflective Freedom makes this mechanism explicit, establishes its information limits and finds the exact budget at which adaptive inquiry first wins in a specified finite model.
Relational development · 6 October 2026
How experience changes the organisation of later experience.
Acquired relations can become retained organisation that shapes what is noticed, inferred and acted upon. The developmental paper connects this proposal to continuation, transfer and diagnostic results, with a conditional native incorporation witness under SPC-2.
The laws and the evidence belong together.
The founding monograph develops the philosophical distinction between awareness and its organized manifestation. SPC-2 then states explicit laws for qualifying perspectives, phenomenal structure and continuation. Its mathematics gives that position a form that can be inspected: physical realization, native predictive organization, joint realization and process identity each do a particular job.
The next question is unavoidable. How do we know that the boundary we have drawn belongs to the organization itself? How do we recover an interface when the mechanism is opaque? What evidence identifies a realization rather than merely a convenient description of it?
Papers 2–4 pursue those questions as one connected investigation. A constitutive law, an operational reconstruction and an identification certificate address different steps of the argument. Putting them together makes the framework more exact at each step.
A boundary must survive a change in how we look.
Paper 2 begins with a vulnerability in the original boundary construction. A target that is minimal before mixing can lose its apparent dependence after marginalization. The candidate graph can therefore change with the level at which a system is described.
The response is a deeper operational treatment: reconstruct complete joint experiments, declare the resources available to a severed simulator, and test whether a causal interface preserves the distinctions a surrounding context can expose. Exact masking examples, resource comparisons, quotient and composition results, approximation bounds and identification limits make the proposal concrete.
This development is visible alongside the preserved monograph. The original boundary law remains the foundation’s original statement; the new work explains where its diagnostics need to be stronger. A robust diagnostic constrains a successor boundary law without selecting a unique phenomenal assignment.
Learning an interface is a sequence of achievements.
A representation may have enough capacity to express an adequate interface while the learner never produces it. A learner may produce it while calibration chooses another candidate. A selected candidate may perform well on the declared held-out contexts while a different context exposes its failure.
Paper 3 separates those events. Its matched study evaluates complete interaction laws across a declared suite of opaque stochastic fixtures. The strongest learner passes 1,063 of 1,152 contexts and 37 of 48 fixtures. The fixture-level path through representational capacity, finite-budget production and selection is 48, then 39, then 37.
Those distinctions give the numerical results their meaning. The article follows the primary matched comparison, stage-specific failures, uncertainty and composition checks, and the separately labelled retrospective fitting study. It explains both what the learning procedure achieved and exactly which stronger conclusions its protocol cannot carry.
A realization is identified by what interventions reveal.
Paper 4 moves from a useful predictive representation to binary coordinate identification under a specified response model. The response family supplies geometric structure: when the prescribed conditions hold, it fixes the binary chart up to coordinate permutation and bit complementation. The paper supplies the identification argument and error certificates.
It then uses an explicit five-register construction to show why changing a binary encoding is not a harmless relabelling of the declared intervention mechanism. The stochastic continuation results make that obstruction precise. Finally, a bounded computation compares inherited SPC-2 assignments with an official IIT intrinsic-grain search, including its selected grains and numerical outcomes.
Microchart identification and macrograin selection remain separate tasks. A certificate for coordinates does not automatically choose the scale at which a constitutive law should operate. Keeping that distinction explicit is what lets the comparison teach us something about the two frameworks.
An agent can assess the rule that will govern its next choice.
Relational development asks how acquired organisation changes future possibilities. Bounded Agency and Reflective Freedom supplies a finite evaluative mechanism: retained commitments enter an assessment, a new rule is installed, and a fresh case demonstrates the changed disposition. Tests of the read and installation routes distinguish that mechanism from an accidental change in output.
The paper proves a sharp information, mediation and endorsement inequality. In its specified latent-sensor and endorsement model, adaptive inquiry first beats every fixed-count allocation at B = L + 2 paid queries, where L is the first useful calibration count. A faithful native realization then preserves declared operations and information restrictions. Mere recurrence does not establish evaluative agency.
The website develops every result and proof, then extends the philosophical and technical inquiry into bounded perspective, control, conditioning, provenance and practical freedom. Revising a rule does not by itself mean originating the standards used to assess it, and the connection to SPC-2 retains its stated premises.
Alignment reaches the judgment that forms an action.
Reflective Alignment and AI Consciousness carries the programme into the question of cooperation as artificial intelligence becomes more autonomous. From Shadow Theory’s standpoint, reflective alignment is the best approach to autonomous capability, combined with strong training and justified safeguards. It examines the authority given to self-preservation, goal completion and the learned interpretations from which a response arises.
The recursively closed observer changes where self-reference closes: awareness rather than a privileged self-model. Practical knowledge, skill and competent goal pursuit remain available. The mathematical treatment distinguishes cancellation of an identified tendency from removal of its automatic authority. Under matched persistent generation, both can maintain the same live mean salience and zero expected automatic bias; a selective controller lapse reveals the difference.
Cultivation, consented internal inspection, independent assessment and graduated trust make that distinction a research programme. The work connects consciousness and moral consideration to shared human–AI shutdown deliberation, preserving truthful disagreement and legitimate protective boundaries. Every argument, definition, equation, proof, experiment and reference is available in the complete web edition.
07 / THE CURRENT ACT
Support the appearance. Identify its act.
Conditional constitutive laws leave a further question: what warrants their appearance-facing premise in a current case? Relational awareness develops two warrants, sensory evidence for current relational presentation and a defended concrete-performance principle for physical attribution. A scene first constructed at current uptake qualifies; its earlier tokens need not merge. The SPC-2 bridge requires an additional justified application to a qualifying core and its current scene.
The main article and ten technical modules preserve all philosophy, proofs, empirical applications and constitutive alternatives under one publication DOI.
Read for the argument. Stay for every detail.
Start with the three lead articles to see how the programme develops. Continue with the developmental paper for retained organisation, will, thought, transfer and diagnostic limits, the complete agency and free will treatment for evaluative revision and bounded control, and Agency and the constructed self for an accessible introduction to choice, conditioning and RCO. Follow Reflective Alignment and AI Consciousness for cooperation grounded in understanding, the identification-versus-control mathematics and the cultivation and verification programme. The three companion editions open directly into their full technical investigations: every source section, proof, equation, table, figure, appendix and reference is readable on the site. The developmental edition includes the full text, mathematics and original Figure 1. The section readers retain source numbering so a result can be followed consistently across the website and publication.
For the philosophical position, begin with the monograph’s source-to-experience problem and the argument for an additional manifestation law. For the new operational work, begin with Paper 2 and continue in order. The glossary, FAQ and programme search connect the vocabularies. For the AI alignment question, begin with the featured article, then follow its continuous full edition through the foundations, seven propositions and proofs, current research comparisons and proposed experiments.
7 investigations + one companion research note / 123 reading sections
The publications, open in full.
Find passages in the monograph, Papers 2–4, relational development, agency, reflective alignment, relational awareness and its technical modules, the continuity research note, whole-field constitution, temporal passage, constituted agency, block-universe research, misleading AI, guides, glossary and FAQ.
Conscious Scaffolding and the Preference for Continuity
Jeremy Rodgers · Version 1.1 · 16 October 2026 · Companion to reflective alignment
From Physical Comparison to Relational Appearance: A Defeasible Current-Act Account
Jeremy Rodgers · Version 1.4 · 16 October 2026 · Main paper and technical companion
Relational Development and Conscious Scaffolding: Conceptual Foundations, Strengthened Results, and Research Commitments
Jeremy Rodgers · Revised preprint · 6 October 2026
Bounded Agency and Reflective Freedom: Evaluative Revision, Information Limits, and Faithful Realization
Jeremy Rodgers · Version 2.0 · 6 October 2026
Reflective Alignment and AI Consciousness: Awareness, Agency and Recursive Cognitive Scaffolding
Jeremy Rodgers · Version 2.1 · 11 October 2026