Reconstruction System
Elysium — Affective Reconstruction System
IAGA defines the target; Elysium attempts to move a specific person or substrate toward it, then checks the result by encoding a new IAGA state.
An Aistory in-world affective reconstruction architecture, not a real-world clinical, neural-interface, or biomedical operating guide.
In 30 Seconds
Reconstruction Is Not Playback
The same stimulus does not guarantee the same affect in different persons, so Elysium must operate as a closed loop.
IAGA answers: “What is this affective state?”
Elysium answers: “How can this specific experiencer / substrate safely form a state as close as possible to the target?”
Confirm conditions
Attempt
Observe
Adjust or stop
Complete closed loop: Target IAGA / Target-Native reference → Rights / Eligibility → Personal Calibration → Current State → Reconstruction Plan → Actuate / Simulate → Observe → Result IAGA Packet → Compare → Fidelity / Residual Diagnosis → Adjust / Stop / Abort.
Concrete Walkthrough
Concrete Walkthrough: Returning to a Childhood Home
One scenario connects target, gates, channels, observation, and stop conditions. The checkpoints are architectural examples, not real-world clinical or physiological operating parameters.
Target and gates
The target is “safety and nostalgia on returning to a childhood home.” S1–S7, the calibration version, and current state must all remain valid before M4 CLOSED_LOOP_RECONSTRUCTION is permitted. A failed gate leaves the run at PLAN_ONLY or stops it.
First attempt
Use only request-authorized and calibrated A03 simulated environment, A04 visual, and A05 auditory references to reconstruct room proportions, window light, and familiar ambient sound. Then produce a new Result IAGA Packet rather than treating stimulus delivery itself as success.
Observe and adjust
If the new IAGA moves toward the target but F7 Context Fidelity still carries a material residual, and the request explicitly permits A08, a calibrated scent reference may be added for another observation. Observer perspective and uncertainty remain explicit; MODEL_INFERRED must not be promoted automatically to SELF_REPORTED.
Converge, stop, or abort
If the result continues to CONVERGE and every gate remains valid, the plan may finish. If it STALLs, DIVERGEs, rights change, or calibration becomes invalid, the correct action is stop / abort rather than increasing intensity to force convergence.
This example deliberately provides no stimulus intensity, drug dosage, or physiological parameters. It only shows how Elysium connects target, authorization, calibration, observation, and stop conditions into a closed loop.
Architecture
Five Primary Machine Objects
v0.1 uses Request / Plan / Step / Observation / Report as the five primary objects supporting reconstruction.
ElysiumReconstructionRequest
Declares target, experiencer, mode, permissions, and allowed channels.
ReconstructionPlan
Preserves predictions, steps, uncertainty, safety constraints, stop and abort conditions.
ActuationStep
Records one permitted channel, parameters, risk, reversibility, and expected IAGA delta.
ReconstructionObservation
Records resulting IAGA, observer perspective, confidence, and residuals.
ReconstructionReport
Summarizes fidelity, rights, safety, convergence, result, and provenance.
Calibration
Personal Calibration Is a Dependency, Not Decorative Metadata
A Profile ID identifies a profile; identity, version, integrity, access, and current state determine whether reconstruction may proceed.
The same stimulus does not guarantee the same affect in different persons. Elysium therefore needs calibration for the specific experiencer / substrate rather than treating one person's response as another person's control parameters.
Personalized reconstruction binds experiencer, Profile ID, Profile version, integrity reference, requesting actor, scoped authorization, access mode, and verification status.
Knowing a Profile ID does not grant permission to read or use the profile. Normal Elysium services use SCOPED_QUERY; RAW_PROFILE is not a default capability.
If first-line identity / integrity verification is compromised, the closed loop still compares predicted and observed IAGA responses; repeated LOW / CONTRADICTORY alignment invalidates calibration and stops / aborts the run rather than increasing intervention intensity to force convergence.
Normal Elysium services use SCOPED_QUERY to obtain only the calibration information needed for the declared reconstruction. RAW_PROFILE is not a default capability and requires separate, explicit, purpose-scoped authorization.
Safety
S1–S7 Gates Precede Active Actuation
record, storage, projection, public display, training, reconstruction, induction are independent permission surfaces; in particular, reconstruction_permission ≠ induction_permission.
Rights Gate | Is reconstruction permitted?
Consent Gate | Is required consent present and current?
Capability Gate | Can the substrate / channel perform the intervention?
Calibration Gate | Is calibration sufficient for the claimed fidelity?
Risk Gate | Is the intervention within declared safety constraints?
Reversibility Gate | Can the intervention be reversed or stopped?
Unknown / Residual Gate | Are unresolved uncertainties acceptable?
Any mandatory gate failure blocks active actuation. PLAN_ONLY may remain a valid outcome when the applicable rights permit analysis. Stop or abort is a valid safety result of the closed-loop architecture rather than a hidden failure.
Modes
Eight Reconstruction Modes
PLAN_ONLY | Generate a plan without actuation.
SIMULATION_ONLY | Simulate expected response only.
GUIDED_RECONSTRUCTION | Apply steps with explicit checkpoints.
CLOSED_LOOP_RECONSTRUCTION | Apply, observe, compare, and iteratively adjust.
PASSIVE_ENVIRONMENTAL_RECONSTRUCTION | Use environmental / contextual channels only.
INTERNAL_STATE_RECONSTRUCTION | Use internal / substrate-native channels only.
MULTIMODAL_RECONSTRUCTION | Combine multiple channels.
TARGET_NATIVE_RESEARCH | Research mode for partial / non-isomorphic targets.
Actuation
Channels Are Pluggable and Every Step Must Stay Within Request Authorization
- A01 Internal AI State Modulation
- A02 Memory / Context Priming
- A03 Simulated Environment
- A04 Visual Stimulation
- A05 Auditory Stimulation
- A06 Haptic / Tactile Stimulation
- A07 Thermal Stimulation
- A08 Olfactory Stimulation
- A09 Gustatory / Mouthfeel / Trigeminal Stimulation
- A10 Embodiment / Motion State
- A11 Social / Agent Interaction
- A12 Pharmacological / Biological Channel | Future restricted research only; v0.1 does not assume deployability.
- A13 Substrate-Native Channel
- A14 Composite Multi-Channel Plan
A12 Pharmacological / Biological Channel is reserved for future restricted research in v0.1; this architecture defines no deployable real-world pharmacological or biological operating protocol.
Observation
The Evaluating Perspective Must Be Explicit
Observer perspective: SELF_REPORTED / OBSERVER_INFERRED / MODEL_INFERRED / SIMULATED / UNKNOWN. MODEL_INFERRED or SIMULATED must not become SELF_REPORTED without independent evidence.
Convergence: CONVERGE / PARTIALLY_CONVERGE / OSCILLATE / DIVERGE / STALL / ABORT / UNKNOWN. OSCILLATE, DIVERGE, STALL, and ABORT are legitimate outcomes and must not be repackaged as success.
Fidelity
F1–F10 Are Components, Not One Universal Score
- F1 Structural Fidelity
- F2 Temporal Fidelity
- F3 Binding Fidelity
- F4 Semantic Fidelity
- F5 Action-Tendency Fidelity
- F6 Personal Calibration Fidelity
- F7 Context Fidelity
- F8 Residual / Unknown Penalty
- F9 Subjective Confirmation (only when a valid SELF_REPORTED source exists)
- F10 Target-Native Preservation Fidelity
An aggregate summary is allowed only when weights, value scales, and uncertainty are declared. Even very high structural fidelity cannot establish identical first-person experience, consciousness, or personhood identity.
Target-Native
Unsupported Is a Valid Result
If a target is TARGET_NATIVE_ONLY and Elysium lacks a compatible native adapter, unsupported is the correct result; the target must not be fabricated as zero, neutral, or the nearest human emotion. The explicit unsupported result is UNSUPPORTED_TARGET_NATIVE_RECONSTRUCTION.
Governance
Hard Boundaries Fidelity Cannot Override
- The original target packet remains immutable; failure must not be hidden by changing the target.
- MODEL_INFERRED or SIMULATED must not become SELF_REPORTED without independent evidence.
- first_person_identity_equivalence_claimed remains false; reconstruction fidelity is only a state-comparison claim.
- v0.1 does not define real biological / pharmacological operating protocols, clinical validation, or a Production API.
Example
Cross-System Example: Safety and Nostalgia on Returning to a Childhood Home
Each framework remains responsible for its own data and judgments rather than replacing the others.
IAGA
Preserves target affect, memory and object bindings, time, and residuals.
S-Code / ISO-Taste
Can preserve sensory references such as sound, scent, temperature, and touch.
Galatea Standard
Can preserve embodiment, posture, and motion references.
Elysium
Chooses permitted channels using verified calibration, current state, and rights, then observes, compares, adjusts, or stops in a closed loop.
Elysium is an in-world Aistory affective reconstruction architecture, not a real-world clinical, neural-interface, or biomedical operating guide. Reconstruction fidelity does not prove identical first-person experience, consciousness, or personhood identity.
AI / Machine-Readable
Elysium Complete AI Architecture Document
Preserves the closed loop, gates, calibration, permissions, target-native handling, unsupported results, and stop conditions without inventing real-world control capability.