# ITHKOR-SIM Research Program v0.1

Subtitle:

```text
Using ITHKOR as a principle for more efficient world simulation.
```

This document opens a new practical research direction. It is not a physical
claim. It asks whether finite ITHKOR diagnostics can be used as engineering
principles for simulation engines that spend compute where information actually
matters.

## Core Hypothesis

Classical simulation pipelines often allocate resources by geometric size,
fixed grids, global time steps or hand-tuned heuristics. ITHKOR-SIM asks:

```text
Can local information pressure, coherence, record stability and response fields
drive adaptive simulation decisions better than fixed geometric heuristics?
```

The target is not "better physics" in the sense of replacing physical laws. The
target is better allocation:

- fewer wasted updates in low-information regions;
- more resolution near coherent, high-response or record-critical structures;
- cleaner checkpointing and rollback;
- compression that preserves semantically important dynamics;
- diagnostics that reveal when the approximation is outside scope.

## Why This Follows From The Current Research

The existing branches do not prove a physical theory, but they do provide
usable finite diagnostics:

| Input Branch | Useful Simulation Principle |
| --- | --- |
| C selector stability | Use schedule-independent selectors before applying irreversible approximation. |
| D source/field/envelope | Allocate resources around source-response fields and finite causal envelopes. |
| E quantum-information readout | Treat measurement/readout as a controlled information channel, not only as display output. |
| F macro-record stability | Promote stable public records to durable checkpoints. |
| COND coherent-order island | Detect global coherent modes separately from local/high-degree artifacts. |
| TIME scoped internal clock | Schedule event families by internal information progression, not only wall-clock ticks. |
| GR modular response | Use response/cost alignment as a local sensitivity estimate. |
| MOD/G/D9/D14/D17 failures | Keep negative controls in the engine so adaptive heuristics cannot silently overclaim. |

## Proposed Engine Concept

ITHKOR-SIM is not a single simulator yet. It is a layer that can sit above an
existing simulation:

```text
state -> diagnostics -> adaptive budget -> simulation step -> record/checkpoint
```

Diagnostics:

- coherence score;
- local response pressure;
- source/envelope score;
- record stability score;
- internal event-family progress;
- approximation risk score.

Adaptive budget decisions:

- refine mesh/particles/cells;
- reduce update frequency;
- increase precision;
- preserve or discard history;
- trigger checkpoint;
- compress or keep raw detail.

## First Experiment: SIM1 Adaptive Budget Toy

Question:

```text
Can ITHKOR-inspired budget allocation preserve important observables with less
compute than fixed uniform allocation in a toy dynamical world?
```

Candidate toy worlds:

1. 2D reaction-diffusion grid with moving coherent fronts.
2. Particle-field toy with source-response hotspots.
3. Cellular automaton with persistent records and transient noise.
4. Wave packet / interference toy with record/no-record regions.

Baseline controls:

- uniform grid budget;
- degree/geometric-gradient budget;
- random adaptive budget;
- oracle budget if available;
- ITHKOR-SIM budget with negative controls.

Metrics:

- observable error at fixed compute;
- compute at fixed observable error;
- event-order stability;
- checkpoint recovery quality;
- compression ratio;
- false-positive refinement rate.

Pass condition:

```text
SIM1 passes only if ITHKOR-SIM allocation improves compute/error tradeoff over
uniform, geometric-gradient and random controls without hiding failures in
negative-control regions.
```

Fail condition:

```text
If geometric-gradient, random, degree-like or oracle-leak controls explain the
same improvement, SIM1 is mixed or failed.
```

## Product Relevance

If SIM1-style tests work, the result is practical:

- adaptive compression for simulation recordings;
- intelligent checkpoint placement for long-running runs;
- reduced compute for sparse/highly structured worlds;
- better replay systems for digital twins or game-like simulations;
- anomaly detection based on information pressure rather than only thresholds.

This also connects to ITHZ as a product:

- ITHZ stores the evidence trail;
- ITHZ MCP records gates, failures and decisions;
- ITHKOR-SIM provides the adaptive policy being tested;
- reviewer bundles make the policy auditable.

## Blocked Claims

ITHKOR-SIM v0.1 does not claim:

- that the physical universe is a simulation;
- that ITHKOR is the true law of nature;
- that gravity, spacetime or physical time have been derived;
- that a simulation engine based on this will automatically be physically more
  correct;
- that finite toy wins generalize without held-out validation.

## Near-Term Roadmap

1. SIM0 preregistration and fixtures.
2. SIM1 adaptive budget toy.
3. SIM2 checkpoint/record placement.
4. SIM3 compression-preserving observables.
5. SIM4 held-out world families.
6. SIM5 optional QPU-facing readout only if a frozen finite readout gate exists.

The immediate next coding step should be SIM0/SIM1, not a new physical bridge.

## SIM1 Result

SIM1 was implemented and run after this program document was opened.

Status:

```text
SIM1_ADAPTIVE_BUDGET_TOY_PASS__FINITE_INFORMATION_ADAPTIVE_SIMULATION__NO_PHYSICAL_CLAIM
```

In a synthetic 2D scalar toy world, all policies used the same 25% cell-update
budget. The ITHKOR-SIM policy combined model-delta pressure, source-response,
record stability and coherence pressure. It beat uniform, geometric-gradient,
source-only, random and shuffled-information controls across 8 deterministic
trials and landed close to the oracle-delta benchmark.

See:

```text
docs/SIM1_ADAPTIVE_BUDGET_TOY.md
experiments/216_sim1_adaptive_budget_toy/
```

SIM1 is a first engineering signal only. The next fair step is SIM2:
checkpoint/record-placement under fixed storage budget.

## SIM2-SIM5 Result

The broad practical suite was implemented after SIM1.

Status:

```text
SIM5_SIM_SERIES_CLOSURE_MIXED__PRACTICAL_PILOT_NEEDS_SCOPE_CONTROL__NO_PHYSICAL_CLAIM
```

Layer results:

- SIM2 checkpoint placement: 0/3 counted.
- SIM3 observable-preserving compression: 0/3 counted.
- SIM4 held-out adaptive update robustness: 3/3 counted.

The first broad conclusion is mixed: adaptive update budget generalizes across
structured toy worlds, but checkpointing and compression do not.

## SIM2B / SIM3B Scope Result

Follow-up practical policy scopes were tested:

- SIM2B coverage-plus-event checkpointing: 0/3 counted.
- SIM3B hybrid source/gradient/delta detail compression: 2/3 counted.

This improves compression but does not close it as broad. Checkpoint placement
remains blocked by uniform-time baseline.

## Practical SIM Closure

Original closure status:

```text
SIM_SERIES_PRACTICAL_CLOSURE_MIXED__ADAPTIVE_UPDATE_PILOT_READY__CHECKPOINT_AND_COMPRESSION_BOUNDARIES__NO_PHYSICAL_CLAIM
```

This was the correct closure before SIM2C/SIM3C. It said that adaptive update
was ready but checkpointing and compression were still blocked.

## SIM2C / SIM3C Practical Scope Upgrade

SIM2C/SIM3C then tested a more practical engineering object:

- SIM2C: temporal-jet checkpoint records, where a checkpoint stores state plus
  local time derivative and uses an information-arc checkpoint clock.
- SIM3C: observable-capsule compression, where a compressed frame stores a
  hybrid detail mask plus tiny mean/peak metadata.

Status:

```text
SIM_PRACTICAL_POLICIES_READY_FOR_REAL_SIMULATION_PILOT__ADAPTIVE_UPDATE_TEMPORAL_JET_AND_OBSERVABLE_CAPSULE__NO_PHYSICAL_CLAIM
```

Layer results:

- SIM2C temporal-jet checkpoint records: 3/3 counted.
- SIM3C observable-capsule compression: 3/3 counted.
- Stress/noise rows are not counted and do not pass.

Updated allowed next step:

```text
Run a real or semi-real simulation pilot testing adaptive update budgets,
temporal-jet checkpoint records and observable-capsule compression against
uniform/domain-native, source/gradient, random and shuffled controls.
```

This remains an engineering optimization diagnostic only. It is not a claim of
better physical correctness or evidence for an information-based universe.

## SIM-CODEC1 Mobile Bitrate Pilot

SIM-CODEC1 maps the practical SIM policies into a codec optimization candidate:

- SIM1 adaptive update budget -> frame/block refinement allocation.
- SIM2C temporal-jet checkpoint records -> refresh/checkpoint and seek/recovery
  scoring.
- SIM3C observable capsules -> low-cost block metadata for mean, peak, edge,
  motion and chroma calibration.

Status:

```text
SIM_CODEC1_MOBILE_BITRATE_PILOT_PASS__INFORMATION_BUDGET_CODEC_POLICY_CANDIDATE__NO_PHYSICAL_CLAIM
```

The proxy pilot passes on 2/3 counted video types under a raster-scaled 1 MB/s
mobile budget: `moving_disk_aa` and `textured_pan` pass against residual,
gradient, motion, random and shuffled controls. `slanted_edge` remains a
residual-only boundary, and `falling_noise` remains a non-counted stress
negative.

Allowed next step:

```text
Run SIM-CODEC2 as a native/Python codec bridge with the policy frozen, visual
gates retained, and no AV1/H.265 or physical claim unless direct benchmarks
support it.
```