An exact thermodynamic execution sandbox.
Reduces continuous physical domains to deterministic integer routing via a strict C ABI.
As a C++20 execution engine, Tensorless operates on bounded, discrete routing mechanics across a periodic three-dimensional integer grid.
Core = mechanics. Adapter = meaning.
- State Evolution: The core library advances state, routing, scheduling, transport, coarse-graining, conservation ledgers, and diagnostic measurements.
- Strict Agnosticism: The core never decides what the state means.
- Meaning Assignment: Domain adapters assign meaning to the bounded execution rules, translating them into diffusion, markets, quantum circuits, thermal routing, network queues, pathfinding, or other domains.
- Unidirectional Dependency: The dependency direction is strictly one-way: adapters link to the public ABI, and the core implementation remains entirely domain-blind.
- Zero External Baggage: Tensorless does not depend on an external tensor framework, numerical runtime, plugin system, or domain-specific core branch.
- Purity Enforcement: The core does not link to adapters, call adapter callbacks, load plugins, inspect domain names, or branch on domain-specific concepts.
- Mandatory Erasure: All state transformations require discrete erasure.
-
Physical Bounds: The engine enforces the
$kT \ln 2$ lower bound on bit erasure directly at the routing layer. - Mathematical Starvation: If a routing adapter attempts to execute an operation without sufficient subunits to pay the required entropic toll, the operation mathematically starves and the trace faults.
- Subunit Exactness: Both energy and action use exact one-third-micro-action subunits, and successful ticks verify energy and momentum conservation identities.
- Ledger Segregation: Accepted and rejected momentum are accounted separately, while external exhaust and starvation have distinct ledgers.
Tensorless strictly distinguishes implemented mechanics from external interpretation. A passing test suite establishes agreement with the implemented contracts, such as enforcing capacity ceilings and energy conservation identities.
It does not establish:
- Empirical validity of Finite Possibility Mechanics.
- Correctness of an adapter’s external model.
- Universal physical equivalence between unrelated domains.
⚠️ Note: A zero conservation residual validates the represented ledger identity. It does not validate an external domain interpretation.
To prevent "hidden tuning" attacks, the TensorlessParamLedger records every run parameter with its classification (FIXED, DERIVED, FREE, or FITTED), source, and value. The ledger computes a non-cryptographic FNV-1a fingerprint over the declared entries for drift detection, ensuring that any changes to core constants or adapter variables leave a traceable record.
This public repository serves as the API reference and validation harness. The private core implementation is strictly black-boxed and excluded. You cannot modify the core thermodynamic ledger; you can only route through it.
To validate the thermodynamic bounds locally, you must download the pre-compiled FPM binaries:
1. Download and extract the latest release
Download the .zip for your platform (Linux, macOS, or Windows) from the Releases page. Extract the archive and open a terminal in the extracted folder.
2. Generate the build files
# Targets the 'source' folder bundled in the zip release
cmake -S source -B build -DCMAKE_BUILD_TYPE=Release3. Compile the adapters
cmake --build build --parallel4. Run the exact accounting experiments
ctest --test-dir build --output-on-failure5. Run specific validators directly
# On Windows, this will be build/Release/tensorless_landauer_erasure_validator.exe
./build/tensorless_landauer_erasure_validator