Frogue Living Worlds That Explain Themselves

A native causal simulation engine where residents perceive, remember, act, explain their choices, and recalculate alternate worlds.

Overview

Frogue is a native C++ causal simulation and authoring engine. The Build Week release demonstrates four materially different residents interpreting the same novelty and damage events according to their own pressures, identity, embodiment, and memory.

The submission includes:

• a guided native Mirror Room demo,
• four persistent causal residents,
• a harmless resonant-prism micro-simulation,
• a deterministic ceramic-mug/mirror incident,
• shared-tick causal inspection,
• structural memory for the architectural resident,
• a reinforced-mirror counterfactual,
• guarded GPT-5.6 authoring,
• an offline local fallback,
• a portable Windows release,
• an end-to-end automated self-test.

Fastest test portable release

Supported platform:

Windows 10/11 x64

1. Download the latest Build Week release ZIP from:

   https://github.com/LexCapp/FROGUE-Mirror-Room/releases/tag/build-week-2026
   

2. Extract it to a normal writable folder.

3. Run:

   Run_FROGUE_BuildWeek.bat
   

4. Choose Begin Guided Demo.

No Unity, Visual Studio, CMake, source build, OpenAI account, or API key is required.

Automated test

Run:

Run_FROGUE_SelfTest.bat

Expected ending:

FROGUE BUILD WEEK SELF TEST: PASS

The report is written to:

logs/frogue_buildweek_self_test.txt

Guided demonstration route

1. Select the four residents and compare Active Pressures.
2. Introduce the resonant prism.
3. Compare how the same harmless object acquires different relevance.
4. Trigger the ceramic-mug incident.
5. Inspect all four reactions at one shared simulation tick.
6. Compare the reinforced-mirror counterfactual.
7. Open Author Change.
8. Apply a reviewable behavioural proposal to a reset copy.
9. Replay the same incident and inspect the changed explanation.

GPT-5.6 authoring

When OPENAI_API_KEY is present, Frogue can use GPT-5.6 to interpret a request into a constrained allowed route.

Example:

Make Huginn curious enough to inspect the impact.

The model does not mutate world state. Frogue constructs and validates the actual proposal, displays the change, requires explicit approval, and reruns the causal simulation.

Without a key, Frogue automatically uses its guarded local interpreter.

Optional GPT launch

From Command Prompt:

set OPENAI_API_KEY=YOUR_KEY_HERE
Run_FROGUE_BuildWeek.bat

Never store a key in the repository or launcher.

Development build

Requirements:

• Windows 10/11 x64
• Visual Studio 2022 Build Tools with MSVC
• CMake
• OpenGL-capable graphics driver

From the project root:

Remove-Item ".\build_hardened" -Recurse -Force -ErrorAction SilentlyContinue
.\Run_FROGUE_Hardened.bat

Important controls

Mouse        Click resident tabs and buttons
F9           Presentation Mode
F10          Engineering view / bounds
R            Reset presentation camera
0            Deterministic scenario reset
O            Introduce or stir prism
I            Trigger incident
Space        Run / pause
N            Step
V            Snapshot / live view
X            Observed / reinforced world
T            Author Change

Architecture

See:

• docs/TECHNICAL_OVERVIEW.md
• docs/CODEX_AND_GPT56_USE.md
• docs/JUDGING_EVIDENCE_MAP.md

How Codex and GPT-5.6 were used

Codex accelerated the codebase audit, hardening, persistent biosim integration, native human interface, hero-asset residency, deterministic fixtures, self-test, and portable packaging.

GPT-5.6 is used inside the final product as a constrained natural-language interpreter. Frogue remains authoritative over proposal validation, world mutation, simulation, memory, counterfactuals, and explanation.

Repository and asset notes

The final submission must include all licenses and attribution required by third-party source, texture, model, and library dependencies. No API credentials or machine-specific secrets belong in source control.

Status

The portable gold-master route passed the complete automated Build Week self-test and exited cleanly.