Sofia Engine Operating Boundaries & Transparent Limitations

In accordance with our engineering principle "Evidence beats claims", the following boundaries and limitations apply to Sofia Engine 3.0.0a1:

1. Checkpoint Distribution Status

No Canonical Pretrained Foundation Model: Sofia Engine 3.0.0a1 is a Scientific & Edge Intelligence Runtime, not a foundation model. No pre-trained neural checkpoint is distributed in this initial release. Future trained checkpoints (e.g. bearing fault classifiers, transformer-based time-series anomaly models) will be distributed with explicit training provenance, evaluation reports, and SHA-256 digests.

C99 vs. Python Floating-Point: The Python runtime operates in IEEE 754 double precision (64-bit float). The C99 embedded reference engine supports both single precision (32-bit float) and Q16.16 fixed-point arithmetic. Fixed-point computations have a finite resolution of $\approx 1.526 \times 10^{-5}$ and may exhibit small truncation differences compared to 64-bit float baselines.

Relevance vs. Formal Certification: Sofia Engine implements calculations relevant to industrial standards (e.g., ISO 10816/20816 vibration severity zones, IEEE 519 harmonic limits, IEC 61000-4-30 power quality concepts). Software implementation does not constitute formal device or system-level certification. Certified deployments require end-to-end hardware calibration and formal test laboratory certification.

Non-Replacement of Qualified Engineering Judgment: Diagnostic outputs (Health Scores, Severity classifications, Anomaly events) are mathematical inferences based strictly on incoming sensor telemetry. They do not replace certified human inspection or qualified engineering judgment.
Physical Safety Interlocks: Sofia Engine outputs must never be connected directly to emergency stops, circuit breakers, or critical plant actuators without independent, hardwired, safety-certified physical interlocks (e.g. SIL-rated safety relays).