 }})
Physical Combinatorics
An innovation discovery engine that explores the combinatorial space of physical systems — generating, constraining, scoring, and ranking entity combinations against real-world physics.
An innovation discovery engine that explores the combinatorial space of physical systems — generating, constraining, scoring, and ranking entity combinations against real-world physics.
PhysCom takes entities from different dimensions, forms their Cartesian product, then filters and scores every combination through a 5-pass pipeline.
Each entity declares physical dependencies — what it requires, provides, or excludes. Incompatible combinations are blocked before any estimation begins.
vacuum_or_thin atmosphere —
Human Pedalling requires none fuel infrastructure
Surviving combinations get raw metric estimates — speed, cost, safety, range — via heuristic stubs or an LLM provider that reasons about the physical properties of each pairing.
Raw estimates are log-normalized against domain-specific bounds, then combined into a single composite score via weighted geometric mean. Combinations are ranked within their domain.
urban_commuting weights: speed 25%, cost 25%,
safety 25%, availability 15%, range 10%
Top-scoring combinations are sent to a language model for plausibility and novelty assessment — catching physically valid but practically absurd pairings.
The final pass surfaces results for expert evaluation. Reviewers can approve, flag, or annotate each combination with domain knowledge that no model captures.
The building blocks. Each entity belongs to a dimension (e.g. platform, actuator, energy_storage) and carries typed dependencies that define its physical properties and constraints.
The evaluation lens. A domain defines which metrics matter and their normalization bounds — the same combination scores differently under "urban commuting" vs "interplanetary travel."
The physics layer. Typed key-value constraints
(requires, provides, excludes,
range_min, range_max) that gate which
combinations are physically possible.
Quantitative axes like speed, cost, safety, and range. Each metric has a domain-specific weight and normalization range. Some are inverted — lower cost is better.