This is a small, standalone demonstration of five software mechanisms extracted from an earlier Aura stack. It lets you inspect how attention is chosen, recent events are carried forward, predictions are compared with outcomes, and internal control values change.
The repository keeps its original name, aura-consciousness-proof-kernel.
Its output is a component demonstration. It does not run Aura's current ISC
battery or establish subjective experience.
| You want to… | Read… |
|---|---|
| Get the idea without knowing the project | The demonstration explained |
| Understand each variable, with examples and metaphors | What the numbers mean |
| See where the current Aura research stands | Current ISC research |
| See why these mechanisms matter and how they relate to other work | Demos and related work |
| Trace the scientific ideas and prepare citations | Scientific foundations and their limits |
| Inspect the implementation and repeat it | Technical guide |
| Find every page and distinguish old output from current information | Document map |
Imagine a small control panel with several competing concerns: investigate something, protect operating condition, or attend to a social cue. Each cycle, the demo scores those concerns and chooses one. It records the event, updates its prediction of what will happen next, and changes its internal control values. A separate small numerical network supplies perturbation measurements.
The concerns and input signals are generated by the example program. They are not observations of the current Aura desktop or reports from its language model. The control-panel comparison explains the moving parts; it does not imply a person inside the demonstration.
| Mechanism | What it does here | Practical example |
|---|---|---|
| Workspace | Chooses a scored candidate for attention | A focus boost lets one concern beat a higher base-priority competitor |
| Temporal binding | Keeps a recent event window and a short narrative | Earlier selected events remain available on later cycles |
| Self-prediction | Compares predicted state with the state that follows | A changed focus or valence increases prediction error |
| Homeostasis | Changes internal control values in response to supplied signals | A simulated fault reduces the integrity value |
| Structural opacity | Compares internal and projected-output changes after small perturbations | A hidden-state change can differ in size from its visible projection |
The six tests check those component behaviors and report generation. The main Aura research asks broader questions about how many parts work together and whether a proposed core survives controlled interventions.
Python 3.10 or newer is required. The dependencies are NumPy and, for the tests, pytest. No language model, model download or Aura desktop process is needed.
python -m venv .venv
source .venv/bin/activate
python -m pip install -e '.[dev]'
python -m pytest -q
mkdir -p demo-output
aura-proof-report --cycles 20 \
--json demo-output/report.json \
--markdown demo-output/report.mdRead the technical guide for a source-checkout route that does not install the package. Write fresh output to a separate directory. The committed report.md and report.json are the original 12-cycle sample, retained unchanged. Some perturbations and timestamps can vary between runs, so the sample's exact numbers are not a reproduction bar.
As checked on 25 October 2026, the main repository's completed seed-23 experiment recorded 22 of 24 checks passing under ISC-v6, with positive content and withheld-report outputs. Later repairs to shared histories and restored method ownership require rechecking those paired results. The least-domain closure-margin rule was withdrawn. The later night addendum now names K*, a numeric margin about 0.0196 and an additional interventional closure test; fresh confirmation is still required. The carrier and split-system control questions remain unsettled. These are results from the main Aura experiment, not scores produced by this demonstration.
Read the plain-English ISC introduction and the current status page. The original April README is preserved for history. Package names, public APIs and sample reports retain their original names so existing usage continues to work.
The main documentation content audit records the later reconciliation of capability claims, mathematical arguments and operator instructions. A corrected explanation is not a new experimental result; this small demonstration still measures only its own components.