Technology

Architecture, Not Hype

The technical foundations of the NEPA platform are documented here for the purpose of procurement review, technical due diligence, and regulatory evaluation. Every architectural decision serves the platform's core requirement: deterministic, auditable, replay-verifiable inspection output.

01 — Deterministic Execution Model

Deterministic Execution Model

The NEPA engine is designed to produce identical output for identical input, independent of execution time, hardware state, or concurrent load. This property is enforced by architectural constraints, not by convention.

Determinism is not a feature that can be toggled. It is the foundational constraint from which the entire processing architecture is derived. Any deviation from deterministic behavior is treated as a critical fault.

  • 01No Floating-Point Non-Determinism — all arithmetic operations use fixed ordering and rounding modes enforced at compile time
  • 02Seeded Stochastic Processes — any process involving randomness uses deterministic seeding bound to the run ID
  • 03Locked Thread Scheduling — processing lane scheduling is fixed at run-start; no dynamic rebalancing during execution
  • 04Memory Isolation — each lane operates in isolated memory regions with no shared mutable state across lane boundaries
# Determinism Verification
run_a_hash: "c3f8a21d..."
run_b_hash: "c3f8a21d..."
match: true
fp_mode: IEEE754_STRICT
thread_sched: FIXED
shared_state: NONE
determinism_cert: PASS

Version Binding Mechanism

Every inspection run is cryptographically bound to a unique, immutable system state. The binding mechanism ensures that findings cannot be attributed to an ambiguous or unverifiable configuration.

Binary Fingerprinting

Every engine build is SHA-256 fingerprinted before deployment. The fingerprint is embedded in the audit record and cannot be altered post-hoc.

Configuration Hashing

The complete analytical parameter set — thresholds, lane weights, fusion coefficients — is hashed at run-start and recorded in the audit chain.

Sensor Certificate Binding

Camera intrinsics, LiDAR extrinsics, and thermal calibration records are linked to the run ID before the first frame is processed.

03 — Configuration Governance

Configuration Governance

Analytical configuration is treated as a versioned artifact subject to formal change control. No configuration change may enter a production run without a corresponding version record.

Configuration governance is not optional. Any inspection run that cannot be attributed to a known, versioned configuration state is flagged as non-compliant and excluded from the auditable evidence set.

Replay Validation Logic

The replay system allows any inspection finding to be independently re-derived from preserved sensor data. Replay is not a debugging tool — it is a formal validation mechanism with its own audit trail.

Pre-Replay State Verification

Before replay begins, the system verifies that the requested engine version, configuration hash, and sensor calibration records match the stored run manifest. Any discrepancy halts the replay and generates a compliance incident.

Byte-Level Deterministic Reconstruction

The replay engine reconstructs every intermediate processing state at the byte level. Output is compared to stored findings using a structured diff that flags any divergence as a critical validation failure.

Third-Party Auditor Interface

Auditors receive a signed attestation package containing the run manifest, replay output, diff report, and chain seal verification. No internal tooling is required on the auditor side.

Replay Audit Trail

Every replay attempt is itself logged in the audit chain, including who initiated it, when, the result, and whether any discrepancies were found.

05 — Change Control Framework

Change Control Framework

All changes to the NEPA engine, configuration system, and evidence framework follow a structured change control process. No change reaches production without a signed release record and regression validation.

Change TypeControl MechanismStatus
Engine Binary UpdatesCore processing logic modificationsGit-tagged signed release + automated regression suite + fingerprint updateActive
Configuration Contract ChangesAnalytical parameter modificationsVersion increment + structured diff + approval gate before deploymentActive
Audit Chain Schema ChangesEvidence storage format modificationsSchema versioning + migration validation + backward compatibility testActive
Sensor Calibration UpdatesHardware calibration record changesCertificate versioning + binding validation before next runActive
Deployment Architecture ChangesInfrastructure and topology modificationsArchitecture review board sign-off + integration test suiteIn Review

Technical Briefing Available

Detailed architecture documentation is available for procurement reviewers, technical due diligence teams, and regulatory evaluators upon request.

Request Briefing Governance Framework