01 · Architecture and governance
Established the separation between a proposed operation, required authority, execution, validation, and acceptance.
Implementation and validation
Physical results, with their test boundaries visible.
The reference implementation has progressed through synthesizable RTL, FPGA implementation, static timing closure, and identity-verified deployment on an AMD Kria KR260.
KR260Identity-verified FPGA deploymentPhysical reference implementation
22,071Accepted measured operationsO6 · approximately 30 minutes · queue depth 1
100 / 100Destinations validatedO7-A · functional I/O campaign
What the evidence establishes
Physical implementation, bounded sustained governed operation, verified functional I/O, and observed stopping and recovery behavior for the tested configuration.
Before optimization
Building the physical reference, step by step.
The optimization phases came after architecture, simulation, RTL, and FPGA implementation work. Each stage addressed a different engineering question for the reference configuration.
02 · Deterministic simulation
Exercised the reference behavior and validation program in a repeatable model before relying on physical hardware.
03 · Protocols and synthesizable RTL
Implemented the hardware-facing execution path and RTL for the tested reference configuration.
04 · FPGA implementation and timing closure
Progressed through synthesis, implementation, and routing to a timing-closed image for the KR260 configuration.
05 · Deployment and physical execution
Deployed the image, verified its identity, and established the physical operation path that the later optimization campaigns exercised.
O1–O5 then refined the host, control, completion, and evidence paths. O6 added the bounded sustained campaign, and O7-A demonstrated verified-I/O functional integration. Matched comparative qualification and evaluation-platform development remain the next engineering work.
Engineering progress through O1–O6
Control and host paths
The optimization program reduced control-state bridge round trips and repeated host work while preserving the required operation boundaries.
Durability and completion
The program refined durability handling and bounded native completion handling while retaining required ordering, correctness, and evidence semantics.
O1 through corrected O5 each completed an audited 2,100-operation governed physical campaign with zero retries. These engineering improvements are distinct from a qualified hardware-versus-software comparison.
O6 sustained governed operation
| Accepted measured operations | 22,071 |
|---|---|
| Measured duration | 1,800.029 seconds — approximately 30 minutes |
| Payload sizes | 15, 64, 256, 4,096, and 65,536 bytes |
| Queue depth | 1 |
| Publication windows | 22,071 |
| Durability boundaries | 88,284 |
| Health observations | 181 samples; maximum observed gap approximately 10.208 seconds |
| Recorded exceptions | Zero retries, validation errors, lost commands, duplicate executions, or unexpected faults |
What this supports: sustained governed data-movement operation over the observed population and duration.
Transaction counts are not application-level throughput. This campaign does not establish lifetime reliability, universal workload performance, certification, power advantage, or commercial superiority.
O7-A verified physical I/O
| Accepted operations | 100 — 20 at each of the five payload sizes |
|---|---|
| Validated destinations | 100 |
| Verified writes | 200 |
| Queue depth | 1 |
| Durability boundaries | 400 |
| Recorded exceptions | Zero retries, validation errors, lost commands, duplicate executions, or unexpected faults |
These results support functional integration and verified I/O in the tested configuration. This campaign was not a comparative performance benchmark.
Failure handling and controlled recovery
Recorded mismatch scenarios stopped without retry or silent acceptance. Separately authorized O6 and O7-A recovery procedures each performed one unload and one reload, preserved the verified package and bitstream identity, and returned the platform to a verified clean state.
These observed outcomes do not validate every failure mode or deployment environment.
Parity and acceleration remain open measurements.
Earlier same-host measurements did not reach parity. A later matched-comparison attempt completed its runtime population, but incomplete receipt coverage prevented qualification. The evidence was retained and the defect was diagnosed; the comparative result remains unqualified.
The measurement path must be corrected and qualified before parity or acceleration claims are made. Suitable measurement infrastructure and engineering capacity are also needed for power and energy characterization.
Review the implementation and its evidence.
The public Technical Status and Evaluation Brief provides further context. Detailed architecture and evidence materials are available for controlled technical diligence within an appropriate confidentiality and recipient scope.
Evidence basis: O6 sustained-campaign closure dated 7 October 2026 and O7-A functional-campaign closure dated 10 October 2026, with preserved manifests and audit records.
Prismaeon Inc. · Prismæon™
Governed execution. Verifiable outcomes. · nigel@prismaeon.com