Formalize precision-aware acceptance tolerances for HIL so results can route from Hardware Engineering → Observatory with no ambiguity.
Projector algebra residual:
(r_P := |P^2 - P|)
Kernel-annihilation residual:
(r_Q := |QP|)
Dynamic convergence residual:
(r_n := |x_n - P x_0|)
Norms used (fill in):
| Item | Value |
|---|---|
| Target platform | FPGA / ASIC / GPU / QPU-emulator |
| Arithmetic | float64 / float32 / fp16 / bfloat16 / fixed-point Qm.n |
| Unit roundoff (u) (or quantization step (Delta)) | |
| Rounding mode | nearest / stochastic / trunc / etc. |
| FMA | yes/no |
| Accumulator precision | same / wider (specify) |
| Saturation | yes/no (fixed-point) |
Describe the per-iteration kernel application path:
Provide an upper bound for the operation scale (N(n)):