Why AI Might Solve Bouldering’s Biggest Controversy
1. The Myth of the Objective Grade: A Universal Climbing Truth
Walk into any bouldering gym on a busy Tuesday night, and you are guaranteed to witness the exact same phenomenon. At the coordination-heavy slab, an elite climber who flashes 7B/V8 outdoors is hopelessly stuck on a technical 6B/V4, muttering about “sandbagged sets.” Meanwhile, three lanes over, a tall first-timer skips three crucial handholds on a steep overhang, easily muscling their way to the top of a 6C/V5, loudly proclaiming that the gym’s grading scale must be “way too soft.”
This is the chaotic, beautiful, and deeply frustrating reality of the climbing grading system. From the professional circuit down to absolute beginners, we discuss, dissect, and bicker over grades as if they represent a static mathematical law.
But the reality? There is no objective truth in this subjective climbing grading system.
The classic V-scale or font scales are inherently flawed because it attempts to apply a rigid numerical value to an elastic physical experience. A grade is nothing more than a community consensus—a collective average of opinions. Your perception of a boulder problem’s difficulty doesn’t just depend on your raw power; it is heavily dictated by:
- Your unique body proportions (height, arm span, and weight).
- Your specific technical background (slab precision vs. dynamic power).
- Your friction on any given day (ambient gym temperature, skin health, and shoe rubber compound).
Because humans can never look past their own physiological biases, the climbing grading system remains an imperfect, highly debatable ecosystem.
2. Enter the Machine: How AI is Breaking Down the Board
If humans are too biased to grade a climb objectively, can artificial intelligence do it for us?
Over the last few years, data scientists and climbing enthusiasts have started pointing machine learning algorithms directly at standardized climbing training walls. Because systems like the MoonBoard and Kilter Board feature perfectly identical hold layouts worldwide, they provide a massive, standardized digital playground for data-driven analysis.
Recent academic research has taken huge strides in this exact domain:
- The Kilter Board Matrix: A groundbreaking 2026 thesis by Lauri Heijari at Aalto University processed over 66,000 real-world Kilter Board routes by converting the grid positions into visual color-coded pixels (mapping start, hand, and foot roles explicitly). By running this through a complex Convolutional Neural Network (CNN), the AI predicted exact route grades with an astonishing 54% exact accuracy and an 84% accuracy within a single grade threshold—actually outperforming the visual grading accuracy of human experts.
- The MoonBoard Pathfinders: Across the aisle, data models built on the MoonBoard database (like the Petashvili & Rodda cross-board frameworks) have successfully trained neural networks to detect intricate spatial hold groupings, uncovering exactly how specific sequences of micro-crimps dictate rapid spikes in mechanical difficulty.
The AI Generation Experiment
We are even seeing early experiments where neural networks are asked to work in reverse: instead of grading an existing climb, the user inputs a targeted difficulty—say, V5/6C—and asks a generative algorithm to select a balanced sequence of hold coordinates.
The result? We are getting closer to the truth, but we aren’t quite there yet.
While an AI can reliably recognize that placing a shallow pocket two meters away from a slick footholder implies a high difficulty tier, it still misses the organic nuances of body tension, momentum, and the micro-adjustments of a dynamic deadpoint. Generative routes often produce awkward, biomechanically jarring movements that feel highly unnatural to a human climber. AI can crunch the coordinates, but it doesn’t know what it feels like to cut feet.
3. The Future of Grading: Scanners, Algorithms, and Hyper-Personalization
So, where does this leave us five or ten years down the line? The trajectory of camera hardware and machine learning suggests a massive paradigm shift in how we evaluate our training.
Imagine topping out on a brutal, sandbagged project at your local gym, pulling out your phone, and using a built-in LiDAR app to instantly scan the three-dimensional wall structure. Within seconds, a cloud-based algorithm cross-references the wall’s precise volume depth, hold orientations, and angles against a global library of millions of indexed climbs. The app spits out an unyielding, completely unbiased, mathematical calculation of the route’s baseline physical stress. No ego, no sandbagging—just pure data.
The Ultimate Shift: The Personalized Scale
But perhaps an absolute “objective” grade isn’t even the ultimate solution. What if the future of climbing grading system is hyper-individualized?
Instead of a generic plastic tag on the wall that reads “V6,” a smart training app could evaluate the climb specifically for you. By integrating your individual athlete profile into its calculations, the algorithm would run a personalized diagnostic query:
“Based on your height of 180cm, your current finger-strength-to-weight ratio, your preference for static compression, and your slightly shorter arm span, this specific physical layout will feel like a stiff V6/7a. However, for your training partner who is 10cm taller, the reach allows an easier sequence, dropping their personalized grade down to a soft V5/6C.”
This technology would fundamentally transform how we track our athletic progression. Rather than chasing arbitrary numbers stamped by a route setter who might be a foot taller than you, you would be training against a dynamic, fluid metric tailored directly to your own mechanical bottlenecks and biomechanical strengths.
AI might not completely kill off our favorite post-session gym arguments anytime soon, but it is paving the way for a training environment where data finally meets the unique realities of our bodies.
Sources
Addressing grading bias in rock climbing: machine and deep learning approaches
Machine learning-based prediction of kilter board climbing routes
Board-to-Board: Evaluating Moonboard Grade Prediction Generalization
Recurrent Neural Network for MoonBoard Climbing Route Classification and Generation