Decoding an ARINC 429 BNR word, bit by bit
The reversed label, SDI, two's-complement data, SSM and odd parity — a 32-bit ARINC 429 word encoded and decoded by hand, with 35 000 ft and −1 250 ft worked through to hex.
FROM THE TEAM
Technical notes
Writing on the problems we work on: sovereignty, controlled environments, and engineering results you can audit.
Requirements traceability that survives a certification audit
What auditors check in trace data under DO-178C and ARP4754A — bidirectional links, flagged derived requirements, requirements-based tests — and the five patterns that fail an audit.
Bi-elliptic or Hohmann? When the extra burn saves Δv
From a 400 km orbit, a third burn can beat the Hohmann transfer — but only above a radius ratio of 11.94, and only by paying days of flight time. Worked Δv for both, every number derived.
Margin of safety, done right: a worked bracket check
A worked bracket check — limit versus ultimate, the 1.5 factor of safety, a 1.15 fitting factor that turns +0.23 into +0.07 — and the reporting mistakes that hide a thin margin.
The White House superintelligence accord: what human control means in engineering
What the voluntary White House Accord on Superintelligence commits to, what it leaves out, and what keeping AI output under human control means for engineering teams.
AI-generated flight code under DO-178C
Why a general-purpose LLM can't be a qualified code generator under DO-178C and DO-330, and the traceability, review, testing and structural coverage its output still needs.
CMMC Phase 2 is suspended. Your CUI obligations aren't.
What the July 2026 suspension of CMMC Phase 2 changed, what still binds contractors handling CUI — DFARS 7012, NIST SP 800-171, Phase 1 — and where AI tools fit.
The link budget that passes at zenith and fails at 10°
A worked S-band small-satellite downlink, carried across a pass: why 12.9 dB of margin at zenith becomes 2.5 dB at 10° elevation, and three ways to close the link.
Writing verifiable requirements with EARS
The five EARS patterns, the verification-method test, the words that make a requirement untestable, and a before-and-after landing-gear example.
EASA's AI concept paper Issue 03: what it means for LLMs in aerospace
EASA's June 2026 proposed Issue 03 brings LLMs into scope and caps general-purpose models at IDAL D / AL 5 / TQL 5. What it says and what it means for engineering teams.
How to check an AI's engineering calculation
Seven fast checks — dimensions, magnitude, limiting cases, regime, input provenance, constants, re-derivation — that catch most wrong AI engineering answers.
Sovereign AI for aerospace and defense
Why control of the model, the data path, and the deployment environment is not a preference but a programme requirement — and what "sovereign AI" actually demands in aerospace, defense, and government work.
Verifiable AI: no hallucinated math
A language model that guesses at arithmetic has no place in engineering. How AETHER separates reasoning from computation so every number ships with its derivation, assumptions, and validity range.
Air-gapped and on-prem AI for ITAR, EAR, and CUI environments
Export-controlled technical data cannot transit a public API. What it takes to run engineering AI fully local, on-premises, or air-gapped — and where approved cloud fits.
AETHER Aerospace · Technical notes