Transparency
MACH is how AION evaluates mission submissions—transparent, multi-dimensional, and built for European tech hiring teams who need defensible decisions, not black boxes.
The five dimensions
Each dimension is scored independently (0–20), then combined into a 0–100 total. This isolates strengths and weaknesses so hiring teams can compare candidates fairly.
Does the work actually solve the problem? We look for runnable logic, sound assumptions, and alignment with the mission brief.
Is the solution well-structured and maintainable? Clear boundaries, sensible abstractions, and attention to edge cases matter here.
Are trade-offs explained honestly? Strong candidates show how they chose between alternatives—not just the happy path.
Can they articulate decisions for teammates and stakeholders? Clarity beats buzzwords; we reward readable explanations.
Is there a creative or non-obvious angle when appropriate? Pragmatic novelty scores higher than novelty for its own sake.
Methodology
Trust signals
MACH is AION's proprietary scoring methodology. It is not an acronym—think of it like FICO or ELO: a named standard you can trust over time.
Multi-model ensemble scoring
Every score is computed by our MACH scoring engine — a proprietary multi-model ensemble that cross-validates each submission using the same rubric. We average the results and measure how far apart the independent evaluations were.
Example
Independent evaluations scored within 6.4 points of each other, so this result is published directly. A wider gap would route it to human review instead.
Three score examples
Same MACH structure for every candidate: five dimensions (0–20 each), one total out of 100. Illustrative examples below — green, yellow, and red bands so you can read quality at a glance. Live public cards use the same dimension layout at /score/<submission_id>.
Example A — high score
92/100
Top 5%Role
AI Engineer
Mission
“RAG pipeline for legal docs”
Production-ready code with proper error handling
Clean abstraction, well-typed, modular
Clearly explains chunking strategy + trade-offs
Concise write-up, code comments where needed
Hybrid retrieval approach, beyond standard tutorials
Example B — medium score
76/100
Strong fit, may need mentoringRole
ML Engineer
Mission
“Customer churn prediction model”
Model works, but feature engineering is basic
Functional code, could be more modular
Explains choice of XGBoost but skips validation strategy
Writeup is brief, lacks business framing
Standard approach, no novel insights
Example C — low score
52/100
Below bar — would recommend skipRole
Backend Engineer
Mission
“API for user authentication”
Works but has security vulnerabilities (token in URL)
Tightly coupled, hard to test
No explanation of design choices
Minimal documentation
Generic boilerplate solution
MACH doesn't grade on a curve. We've scored submissions from 35 to 98. Only top performers reach 90+. This is why companies trust the score.
We verify the work is yours
FAQ
