#include "chemistry.h" #include #include #include namespace sleela { namespace chemistry { namespace { double clamp01(double x) { if (!std::isfinite(x)) return 0.0; return std::max(0.0, std::min(1.0, x)); } } double ratio(double numerator, double denominator) { if (!std::isfinite(numerator) || !std::isfinite(denominator) || denominator == 0.0) return 0.0; return numerator / denominator; } double similarity(double a, double b) { if (!std::isfinite(a) || !std::isfinite(b)) return 0.0; const double scale = std::max({std::fabs(a), std::fabs(b), 1.0}); return clamp01(1.0 - std::fabs(a - b) / scale); } double stochastic_factor(std::uint64_t seed) { // SplitMix64-style deterministic mixing. The output is deliberately // narrow: stochastic ranking may perturb a hypothesis, but cannot create // evidence or overturn a strong contradictory observation. std::uint64_t x = seed + 0x9e3779b97f4a7c15ULL; x = (x ^ (x >> 30)) * 0xbf58476d1ce4e5b9ULL; x = (x ^ (x >> 27)) * 0x94d049bb133111ebULL; x ^= x >> 31; const double u = static_cast(x & 0xFFFFFFFFULL) / 4294967295.0; return 0.98 + 0.04 * u; } InferenceScore evaluate(double observation, double bond, double valence, double similarity_score, double signal, double symmetry, std::uint64_t seed) { const double evidence[6] = { clamp01(observation), clamp01(bond), clamp01(valence), clamp01(similarity_score), clamp01(signal), clamp01(symmetry) }; // Observation and valence receive the strongest weights. Similarity and // symmetry are supporting evidence, not substitutes for observation. const double weights[6] = {0.24, 0.20, 0.22, 0.13, 0.12, 0.09}; double weighted = 0.0; for (int i = 0; i < 6; ++i) weighted += evidence[i] * weights[i]; const double stochastic = stochastic_factor(seed); const double support = clamp01(weighted * stochastic); int level = static_cast(std::floor(support * 25.0)); if (level > 24) level = 24; // Uncertainty remains explicit and cannot fall to zero merely because a // model receives a high score. const double spread = std::max({evidence[0], evidence[1], evidence[2], evidence[3], evidence[4], evidence[5]}) - std::min({evidence[0], evidence[1], evidence[2], evidence[3], evidence[4], evidence[5]}); const double uncertainty = clamp01(0.55 * (1.0 - support) + 0.45 * spread); return {level, support, uncertainty}; } SubjectNorm presume(const std::string& subject, double observation, double bond, double valence, double similarity_score, double signal, double symmetry, std::uint64_t seed) { SubjectNorm result; result.subject = subject; result.observations.push_back("normalized observation score"); result.bonds.push_back("bond compatibility"); result.signals.push_back("available experimental signal"); result.conferrers.push_back("periodic/structural analogy"); result.conferrers.push_back("valence consistency"); result.conferrers.push_back("convergent signal support"); result.symmetry.push_back("causal evidence ordering"); result.symmetry.push_back("compound-family similarity"); result.ratios = { {"observation", clamp01(observation), 0.24}, {"bond", clamp01(bond), 0.20}, {"valence", clamp01(valence), 0.22}, {"similarity", clamp01(similarity_score), 0.13}, {"signal", clamp01(signal), 0.12}, {"symmetry", clamp01(symmetry), 0.09} }; result.score = evaluate(observation, bond, valence, similarity_score, signal, symmetry, seed); return result; } const std::vector& causal_order() { static const std::vector order = { "observation", "measurement", "structure", "bond", "valence", "geometry", "electronic_state", "reaction_relationship", "compound_family_similarity", "thermodynamic_consequence", "astronomical_context" }; return order; } } // namespace chemistry } // namespace sleela