Time-Division Multiplexing: The Operational Rhythm and Temporal Organization Layer of Digital Life
Organizations: Henan University · Beijing Jiaotong University · National University of Defense Technology
Abstract
Continual agents must handle urgent interaction while retaining alternatives, replaying errors, and committing safe long-term changes. A previous digital-life framework distinguishes online (busy) and offline (idle) channels, but leaves open whether these are separate programs or phases of a common process and how their operations should be ordered. Motivated by recent observations of rapidly alternating perceptual representations in primate cortex, we develop a temporal organization layer with phase-tagged micro-epochs, an observable duty cycle, independent content and resource controls, and three nested clocks separating candidate exploration from durable commitment. We give a projected duty-cycle update, state its deterministic descent guarantee and the additional noise term required for a stochastic guarantee, and specify coupled busy/idle execution contracts. Five synthetic experiments (64 seeds each) examine descent and noise, objective misalignment, gain asymmetry, resource coupling, and dropped replay logs. In the gain/noise model the best tested gain ratio is 1.55; decoupled scheduling yields 26.6% versus 11.1% success on a constructed urgent/high-load subset. A frozen-backbone Qwen3-8B prototype further checks orchestration, but its proxy outcomes are not task-benchmark evidence. The results establish a testable scheduling design and identify its conditions of validity, not machine consciousness or general-purpose agent superiority.
Figures & tables
| Exp. | Controlled comparison | Recorded full-run outcome | Implication |
|---|---|---|---|
| E1 | Fixed vs. strict phase alternation | Terminal : vs. ; deterministic descent check | Mixing affects noise floor |
| E2 | Online at vs. , strongest offline tilt | vs. | Conflicting goals increase cost |
| E3 | Gain/noise exponent | Best grid : (boundary) | Optimum depends on noise law |
| E4 | Decoupled vs. coupled content/resource controls | Urgent/high-load modeled success: vs. ( pp) | Coupling can suppress urgency |
| E5 | Uniform replay threshold with random log loss | Relative elimination slope ; rare/common ratio at under corroboration | Missing logs harm replay |
| Observation | Reported anchor | Proposed architectural interpretation |
|---|---|---|
| Alternation during stable input | Roughly 15 Hz; dwell around 64/59 ms | Preserve a background phase during apparently stable interaction |
| Same recorded substrate | Content tuning remains related across phases | Test shared representation plus phase labels rather than fully isolated models |
| Suppressed content can be decoded | State-specific face-code reconstruction | Retain unselected evidence for later replay |
| Relatively small gain difference | Reported ratio around 1.21 | Vary resource gain rather than the identity of the processing module |
| Resource and content dissociation | Approximate orthogonality in On/Off and content labels | Give compute allocation and content selection distinct inputs |
| Imperfect coordination across areas | Partial state synchronization | Combine local processing with traceable action-level arbitration |
| Empirical motivation | Testable design restriction | |
|---|---|---|
| P1 | Representation alternates under fixed physical input | Avoid equating absence of requests with absence of background computation |
| P2 | Overlapping neural populations represent both contents | Compare shared versus separately parameterized inner/outer models |
| P3 | Suppressed information remains decodable | Avoid silently discarding nonwinning hypotheses when future replay is expected |
| P4 | Phases differ in gain without independent neural modules | Decouple gain scheduling from content broadcast selection |
| P5 | Fast neural dynamics are distinct from slow perceptual behavior | Separate fast candidate processing from slow durable commitment |
| Busy, foreground-dominant | Idle, replay-dominant | |
|---|---|---|
| 1 | Perceive inner and outer state | Sample failure and alternative records |
| 2 | Write raw record before inference | Check/resegment phase labels |
| 3 | Infer content/resource state | Attribute error and uncertainty |
| 4 | Compete for broadcast | Simulate counterfactual actions |
| 5 | Retrieve relevant memory | Abstract candidate skills |
| 6 | Integrate current hypothesis | Build evidence-bearing proposal |
| Operation | Output | Why it precedes the next operation | |
|---|---|---|---|
| 1 | Perceive both domains | Raw inner/outer observations | Monitoring requires both external evidence and internal risk indicators |
| 2 | Record | Immutable micro-epoch record | Interpretation must not overwrite the uninterpreted evidence |
| 3 | Monitor | Content/resource labels, urgency | Arbitration needs safety, uncertainty, and load estimates |
| 4 | Compete for ignition | Winning item and suppressed set | Retrieval must be conditioned on an explicit selected query |
| 5 | Recall | Relevant episodes and skills | Prediction needs a contextualized historical state |
| 6 | Think/integrate | Current hypothesis | Forecasts are conditioned on a coherent hypothesis |
| Operation | Output | Ordering rationale | |
|---|---|---|---|
| 1 | Replay sampling | Evidence-bearing episode batch | Reflection begins with actual recorded failures, novelty, contradictions or high errors |
| 2 | Re-segmentation/check | Reliable phase-labeled history | Error attribution requires credible event boundaries and provenance |
| 3 | Causal attribution | Candidate causes | Interventions need a hypothesized cause rather than arbitrary alternatives |
| 4 | Counterfactual prediction | Potentially better trajectories | Abstracted skills should reflect tested alternatives |
| 5 | Abstraction | Skills/rules and ideal-state proposals | Generalization follows evidence aggregation |
| 6 | Evolution proposal | Versioned self/world diff with evidence | Validators need a concrete candidate to judge |
| Dimension | Busy-dominant | Idle-dominant |
|---|---|---|
| Foreground occupancy | Typically high | Typically low |
| Winning item | User/task request, alarm | Self/world discrepancy, replay failure, internally generated goal |
| Dominant operation | Contextualized planning and authorized action | Causal replay, candidate revision, validation |
| Gain/step allocation | Foreground receives prioritized budget | Background receives larger fraction of available budget |
| Shadow alternatives | Retained, often with a bounded small quota | Replayed, compared, and possibly promoted |
| Durable commitment | Rare, separately authorized | Allowed only after gates on evolution clock |
| Framework element | Proposed operational location | Preserved invariant or measurement |
|---|---|---|
| Eight functions | Perceive, record, monitor, recall, think, predict, plan, act across the two loops | Input/output log and error feedback |
| Four self views | Constitution, old, ideal, proposed new state | Shared versioned self reference and identity gate |
| Four world views | Source constraints, old world, proposed world, desired world | World-state and provenance reference |
| Two domains | Inner and outer observations in the same record schema | Domain and privacy labels |
| Three temporal views | Current observation, history replay, future counterfactual | Three clocks plus event time |
| Two channels | occupancy and task categories | Duty cycle |
| Contract | Required purpose | Bound operations |
|---|---|---|
| MicroEpochRecord | Immutable phase-tagged unit of observed/reasoned experience | Perception, record, monitor, act, audit |
| BroadcastState | Winning content, evidence, confidence, available action permissions | Workspace, retrieval, prediction, planning |
| ShadowCacheEntry | Unselected hypothesis with source, confidence, TTL, access policy | Replay, counterfactuals, audit |
| ResourceState | Compute load, latency, budget, on/off/unknown status | Monitoring, gain scheduler, admission |
| ArbitrationTicket | Explicit cross-module decision with risk and permission context | Local scheduler, global action/commit gate |
| Probe | Conditions | Saved result |
|---|---|---|
| TP1 phase accounting | Busy vs idle occupancy | vs |
| TP2 shadow retention | Proxy success: cache vs prune | vs |
| TP2 proposal attribute | Reported mean quality: cache vs prune | vs |
| TP4 execution order | Proxy success: ordered vs shuffled | vs |
| TP4 record completeness | Ordered vs shuffled | vs |
| Proposed prediction | Operational metric | Failure or falsification condition | |
|---|---|---|---|
| TP1 | Estimated foreground occupancy reverses between sustained external work and no-request windows; winning items switch from external to endogenous | Per-epoch , external winner fraction, switch lag | Occupancy/winner change disappears under an emergent, non-prescribed scheduler |
| TP2 | Truncating the shadow phase has small immediate cost but lowers future useful proposals and initiative | Held-out task success after reflection; proposal validity, CEPA dimensions | No long-horizon decrease with equal compute and retention budgets |
| TP3 | Interleaving shadow processing in long busy tasks aids recovery compared with all-foreground operation | Corrected-error fraction, time to repair, phase-wise errors | Pure foreground is no worse across matched long tasks |
| TP4 | Certain operator-order violations harm prediction/error correction and faithful replay | Verified task accuracy, correct repair, event reconstruction | Reordering with proper independent controls has no effect |
| TP5 | A non-unit gain ratio may optimize learning speed when gain increases noise | Time to threshold, gain-dependent noise, online overfit | or search boundary remains optimal across tested regimes |
| TP6 | Shared backbone/world model and validation semantics facilitate transfer across phases | Skill transfer, consistency admission rate | Isolated models match or exceed shared ones at same cost |
| Approach | Characteristic focus | Relation to temporal organization |
|---|---|---|
| Global workspace theories ( Baars, 1988 ; Mashour et al., 2020 ) | Capacity-limited selection and widespread access | Motivates the broadcast winner and phase-conditioned access; does not prescribe the proposed software commit protocol |
| Integrated information theory ( Tononi et al., 2016 ) | Integrated causal structure and related measures | Addresses a different scientific question; neither proves nor is implemented by a scheduling state machine |
| Predictive processing and active inference ( Friston, 2010 ; Friston et al., 2017 ) | Perceptual prediction, action, and model updating | Shares prediction/error/action dynamics; our additional focus is versioned authorization and two-timescale evidence retention |
| LLM memory/reflection agents ( Park et al., 2023 ; Packer et al., 2023 ; Shinn et al., 2023 ; Madaan and others, 2023 ) | Episodic memory, retrieval, verbal feedback, iterative refinement | Provide practical components; the proposed contribution is explicit phase allocation, shadow provenance, and slow commit gates |
| ID | Area | Open question |
|---|---|---|
| O1 | Architecture | Should broadcast/suppressed labels be assigned by interpretable rules or inferred from model representations? How do both choices trade accuracy, safety, and transparency? |
| O2 | Architecture | Should idle work run on fixed schedules or trigger on accumulated self/world discrepancy? What memory-sampling policy balances replay coverage and compute? |
| O3 | Architecture | How can assimilation/accommodation compare costs of changing the model versus changing the world, accounting for reversibility, irreversible externalities, and authorization? |
| O4 | Architecture | How should multi-stakeholder ideal self/world targets be represented and revised when values conflict? Who can modify each target? |
| O5 | Neuro–AI | How much of biological temporal multiplexing can be approximated by logical event labels, and which capabilities, if any, require actual model-internal temporal dynamics? |
| O6 | Neuro–AI | What is an operational analogue of neuronal tuning gain in LLMs? Do token budgets, attention scaling, or adapter gating produce comparable optimization effects? |
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
| Notation | Meaning | Relation to predecessor |
|---|---|---|
| , , | Micro-epoch label; foreground broadcast and shadow phases | New temporal organization layer |
| Intended mixing/occupancy and observed rolling foreground duty cycle | Extends previous online/offline choice | |
| Candidate step sizes and duty-cycle effective step | Generalizes the predecessor’s two update channels | |
| Absolute allowed online update step/magnitude policy | Derived from constitutional update cap | |
| Ratio of outer/inner gain | Testable design parameter, not fitted from neural data | |
| , | Content scheduling map and resource gain map | Separate policies under T5 |
| ID | Proposed testable prediction | Evidence available here |
|---|---|---|
| TP1 | Phase occupancy and winning item switch with request pressure | API prototype, assignment-driven quick test |
| TP2 | Pruning shadow alternatives harms later proposal quality | API prototype, engineered quality bonus |
| TP3 | Removing interleaved shadow work harms long-run error recovery | Not tested |
| TP4 | Changing key operation order degrades accuracy/repair | API prototype, programmed confidence effect |
| TP5 | Non-unit gain ratio can optimize time-to-threshold | E3 toy simulation, conditional on noise law |
| TP6 | Shared model/validation improves cross-phase skill transfer | Not tested |
| Exp. | Configuration |
|---|---|
| E1 | 64 seeds; , 3,000 steps; , , , weights ; 500 deterministic inequality checks. |
| E2 | 64 seeds; scalar tilt strengths ; weights ; report two terminal objective families separately. |
| E3 | 64 seeds; ; ; 12 log-spaced ; ; target potential ; horizon 8,000. |
| E4 | 64 seeds; 20,000 synthetic task draws per correlation setting; urgency/load correlation ; content threshold ; logistic response with prescribed gain requirement. |
| E5 | 64 seeds; replay budget ; eight failure modes with rates to ; deletion ; evidence thresholds . |
| Tier 2 | Saved quick run: 15 generated tasks, five rounds, 16 workers, frozen api/qwen3-8b backend, logically scheduled micro-epochs. Independent true task correctness was not scored. |
| ID | Design choice | Testable consequence |
|---|---|---|
| T1 | Interleave phase-tagged micro-epochs and retain shadow candidates | Retention/replay ablations alter later performance. |
| T2 | Allow asymmetric foreground/background gain | Gain ratio interacts with noise and learning speed. |
| T3 | Share backbone, verification and projection operators | Shared and independent variants differ in skill transfer/consistency. |
| T4 | Coordinate phase transitions; commit on a slower clock | Removing transition barriers changes invalid commits or recovery. |
| T5 | Separate content selection from resource-gain decisions | Coupling can misallocate compute on conflicting load/urgency inputs. |
LLM Agent Orchestration
Large language model agent orchestration coordinates models, tools, and workflows to complete complex tasks reliably and efficiently.
See all 493 papers