This page preserves an earlier five-stage project tracker. “Reviewed,” “replicated,” and “published” below describe internal project stages—not academic peer review, independent replication, certification, or established science. Use the current evidence surface before quoting a claim.
Archived status: npm + PyPI v3.3.0 was published and the project recorded bijective round-trip checks for 6 tongues with 256 tokens each. This does not establish semantic quality or independent production validation.
An archived in-project report recorded 93% vs 88.7% on its stated comparison. The current public artifact must be checked for corpus, comparator mode, and false-positive scope. No military-grade certification is claimed.
Archived project tests recorded 93% detection and 0% false positives on their tested categories. Those results do not establish generalization or prove that adversarial inputs cannot mimic the signatures.
35 tests passing. Wired into runtime governance gate. Phi-scaled concentric shells in the Poincare ball create natural trust boundaries. Benchmarked for latency and accuracy.
The formula and golden vectors were exercised in project tests. Exponential score growth is a property of the chosen formula; it is not evidence that real attacks are computationally infeasible or military-certified.
46 tests passing. Live in governance gate. Trust scores accumulate on a Fibonacci schedule, making rapid trust manipulation impossible. Integrated with the 14-layer pipeline decision engine.
Tested on Gemini: 23.3% biblical probe alignment vs 33.3% control baseline and 0% noise. Suggests large language models retain structural residue from biblical training data visible in null-space projections.
32 tests passing. Balanced ternary encoding with phi-weighted bit positions creates a natural 3-state logic gate. Preliminary results show 40% fewer bit flips than binary for governance decisions.
Control probes vs biblical probes compared. Concept-aware scoring adds semantic category weights to the harmonic distance calculation, allowing the system to penalize domain-specific adversarial drift differently.
Theory: apply soap-film physics (Plateau’s laws of minimal surface tension) to learning localization. Each “bubble” is a knowledge domain; boundaries enforce natural separation of concerns in model weights.
Theory: use tangential projections in PHDM manifold space to derive operator coefficients that scale security enforcement. Code scaffolding exists but no formal test run yet.
Theory: modulate the Riemann zeta function with rock-paper-scissors ternary cycles to create a dual-ternary encoding. Maps governance states to critical-line zeros for anomaly detection.
Theory: embed semantic content at multiple resolution scales simultaneously (word, sentence, paragraph, document) using nested Poincare balls. Each scale inherits governance from the parent.
Theory: map covenantal agreement structures (promise, obligation, violation, restoration) to the 6 Sacred Tongues. Each tongue carries a natural covenant role in multi-agent trust negotiation.
Policy document drafted. Explores how AI systems can manage digital estates (data, models, credentials) with covenantal governance rules after principal incapacitation or death. Needs formal research.
Like a root beer sliding down the bar in a tapper game, each track moved through five internal project stages. The labels organize work; they do not replace external review or independent replication.