Simulation & worlds is Naevum’s work on rule-based models for planners and world builders, so that every change in a simulated site or world traces back to its rules.
Retain the selected domain state, source basis and current identity.
Explicit physical assumptions or world state; distinct selected views.
Explicit next scenario basis or actual settled world state.
Retain the selected domain state, source basis and current identity.
Site/device units, input quality and baseline.
World/reign identity and current authoritative state.
Physical configuration and living-world state are separate scenario domains.
Introduce one explicit physical alternative or a separately chosen world decision.
Tagged physical change or royal decision, never merged signals.
Introduce one explicit physical alternative or a separately chosen world decision.
Original baseline and changed assumption.
Royal choice and candidate generation identity.
The selected alternative retains the original baseline or world decision identity.
Use different internal operations for physical calculations and world contributions.
Unchanged baseline or authoritative world rules.
One explicit domain-appropriate change.
Use different internal operations for physical calculations and world contributions.
Electrical/thermal/inference checks and hourly storage balance.
Validated campaign, bounded repair or authored fallback and role tasks.
Physical calculations and world-task rules do not share a fictitious universal solver.
Keep calculated feasibility distinct from admitted durable world change.
Physical feasibility and time-basis results for the physical view.
Actual completion under the proposed task identity.
Identified proposed task and permitted consequences, not completion.
Keep calculated feasibility distinct from admitted durable world change.
Memory fit, physical limits, state of charge and shortfall.
Actual independent completion under authoritative world rules.
Calculated feasibility and an actually settled world consequence are different results.
Retain comparisons and actual accepted consequences.
Physical comparison or actually settled world consequence.
Retain comparisons and actual accepted consequences.
Coherent baseline/alternative cases and unresolved constraints.
Settled events precede story and next-world context.
Generated prose cannot replace an independent completion or a settled world record.
Explicit physical assumptions or world state; distinct selected views.
Explicit next scenario basis or actual settled world state.
Retain the selected domain state, source basis and current identity.
Site/device units, input quality and baseline.
World/reign identity and current authoritative state.
Physical configuration and living-world state are separate scenario domains.
Tagged physical change or royal decision, never merged signals.
Introduce one explicit physical alternative or a separately chosen world decision.
Original baseline and changed assumption.
Royal choice and candidate generation identity.
The selected alternative retains the original baseline or world decision identity.
Unchanged baseline or authoritative world rules.
One explicit domain-appropriate change.
Use different internal operations for physical calculations and world contributions.
Electrical/thermal/inference checks and hourly storage balance.
Validated campaign, bounded repair or authored fallback and role tasks.
Physical calculations and world-task rules do not share a fictitious universal solver.
Physical feasibility and time-basis results for the physical view.
Actual completion under the proposed task identity.
Identified proposed task and permitted consequences, not completion.
Keep calculated feasibility distinct from admitted durable world change.
Memory fit, physical limits, state of charge and shortfall.
Actual independent completion under authoritative world rules.
Calculated feasibility and an actually settled world consequence are different results.
Physical comparison or actually settled world consequence.
Retain comparisons and actual accepted consequences.
Coherent baseline/alternative cases and unresolved constraints.
Settled events precede story and next-world context.
Generated prose cannot replace an independent completion or a settled world record.
Two illustrative methods retain separate physical calculations and completed world consequences.
Additional mechanisms are described in the technical reference.
What the research says
Researchers now build simulated people, and the EU funds a simulated planet. A simulation earns trust when each result traces back to its rules and inputs. Builder and Everbloom show every result together with the rules that produced it.
A denser rack changes memory capacity, power and cooling. A royal choice changes the work available to a kingdom. A scene is useful when those consequences follow the model’s rules rather than a visual effect.
Builder’s device → rack → room → building → plot hierarchy ties geometry to declared equipment and site assumptions. A layout change propagates into feasibility and operating calculations.
Everbloom connects a reign and day to generated events, role-specific tasks, settlement and the Chronicle. The accepted state becomes part of later play.
Each calculated value names a method and its inputs. Tagged inputs carry a value, unit, source/assumption status and optional low/high bounds. Changing a scenario reruns the connected calculation graph.
Containment, overlap, aisle clearance, rack mass, floor load and site footprint relate the placed objects to physical constraints. Checks distinguish a sourced limit from a recorded design assumption.
Installed nameplate supports breaker and fit checks. Provisioned peak sizes the load. Utilization-based average power supports energy estimates. Those three values remain distinct.
The electrical tree adds conversion/distribution losses. Air/liquid heat splits, airflow, coolant flow and cooling power propagate into the facility demand.
Base, low and high scenarios rerun coherent sets of tagged inputs. The result retains limiting assumptions and source quality instead of adding an unexplained percentage to the final answer.
Two copies of one world run side by side, one with the rule; a person compares them and decides. Illustrative.
A model’s weights are only one memory demand. Context length, concurrency and the device’s memory/compute limits change what a configuration can actually serve.
KV bytes per token = 2 × layers × KV heads × head dimension × bytes per element. Context length and concurrent sequences multiply that demand. Remaining accelerator memory bounds the feasible batch.
Model shape, precision, context and available memory.
Memory placement and concurrency bounds.
The decode step uses the slower of the memory-transfer and compute times. Batch sweeps retain the limiting regime and distinguish theoretical roofs from utilization-adjusted estimates.
Design-point facility power divided by estimated site throughput produces joules per generated token for that scenario. A memory-infeasible configuration cannot be made feasible by an attractive energy figure.
Generation, grid exchange, battery charge/discharge, export, curtailment and unmet load share one hourly balance. Battery state of charge retains capacity, power and efficiency bounds.
Load and generation series plus storage/grid assumptions.
Annual energy quantities and a traceable shortfall record.
Calculated scenarios are distinct from commissioned hardware measurements.
A King or Queen’s decision can become a campaign and shared work. Generative content enters through a bounded shape; server-side rules retain identity, settlement and economic consequences.
The generation path ties the request to reign/day context and uses a defined input shape. Generated event content is converted into task structure.
Schema checks and bounded repair precede acceptance. An accepted generated result uses first-writer persistence; refusal can select an authored fallback.
The cascade runner separates decision logic from effects. Role-specific work can change world state without making generated prose the authority over the economy.
Task payouts are idempotent; treasury conservation and durable Chronicle records retain what happened. Player contribution becomes part of the world’s continuing history.
Builder reports source assumptions and method lineage. Everbloom retains the accepted world identity and settlement history. Manna’s planning research uses a bounded synthetic environment with explicit observation and action contracts.
These models serve different decisions. Physical planning explains why a configuration fits or fails and how a workload changes the site. A living world explains how a choice becomes coherent shared activity. Predictive-organ research examines future states before proposing an action.
A custom simulation engagement selects the state, transitions, inputs and outcomes relevant to its real setting. The existing technology supplies a starting architecture with identifiable rules; the relationship to physical measurements or observed player outcomes is evaluated for that application.
Inputs, retained state, outputs and limits of Naevum’s Simulation & living worlds architecture.
Yes, when the engagement defines the world’s rules, user decisions, outcome measures and the relationship between simulated behavior and the real setting.
Builder’s numeric results retain input lineage, units, source status, method identifiers and scenario hashes. Design-point facility power and annual energy balance answer different questions.
KV-memory and roofline calculations establish scenario bounds. Hardware benchmarks, facility commissioning and physical measurements are separate evidence.
Everbloom validates structure independently from provider output formatting. Its server assigns identity; frozen phase inputs support deterministic repeated derivation; durable settlement remains outside rerunnable decision logic.
A coherent campaign and idempotent payout are engineering properties. Player enjoyment, retention and observed social outcomes require their own evidence.
Which rule would you test in a world before applying it to yours?
Discuss a simulation or living world