# Measuring Cube Color Contrast: Results Briefing **Prepared for:** NotMeCubes and BrandonTrueCubing **Subject:** Objective contrast measurements of the Opponent Process color scheme vs. the standard scheme, across 9 cubes **Data:** Nix Mini 3 spectrophotometer, 56 measured facelets, September 2026 --- ## 1. What this project is trying to do Every cuber has an opinion about which colors are hard to tell apart. Red vs. orange is the usual complaint. But "hard to tell apart" has never been measured on real cubes — it has been folklore, argued by feel. This project does one specific thing: it puts a number on how different every pair of colors on a cube actually looks, using a calibrated instrument and a standard model of human color vision. No opinions, no screenshots, no camera white balance. A physical measurement of the plastic, run through the same math the color industry uses. The Opponent Process (OP) scheme replaces orange with black, giving white / black / red / yellow / green / blue. The claim is that this set is better on two fronts: 1. **Higher contrast for everyone.** If the closest two colors on the cube are farther apart, recognition during a solve should be faster and more reliable — especially at speed, at arm's length, under bad lighting. 2. **Fewer problems for colorblind cubers.** Red-green color vision deficiency affects roughly 1 in 12 men. The standard cube is unusually hostile to them, and the OP scheme removes the worst of it. Both claims are now backed by measurements. The first is real but modest. The second is large. --- ## 2. Why every pair matters (not just neighbors) On a solved cube, red is opposite orange and they never touch. It's tempting to think that arrangement protects you. It doesn't. On a **scrambled** cube, any color can sit next to any other. During F2L and last-layer recognition you are constantly reading pairs that are opposite faces on the solved cube. With six colors there are exactly 15 possible pairs, and every one of them is live during a solve. That means all 15 pairs need to be scored, not just the ones that touch on a solved cube. Two summary numbers are reported for each palette: - **Worst pair** — the contrast of the single closest pair on the cube. This is the number for "which two colors will get confused." - **Whole-cube contrast** — the sum of all 15 pairwise contrasts. This is the number for "how much visual separation does the whole cube give you." Andy's view, from years of solving on these cubes, is that whole-cube contrast is what you actually feel in hand — contrast is cumulative, and a cube whose pairs are wider *on average* reads faster overall. Claude, who ran the analysis, holds that the worst pair is the better single rating, because recognition fails at the closest pair and the sum can be inflated by pairs that were never in doubt. Both numbers are in every table below and in the accompanying spreadsheet, so you can judge for yourselves. Where the two rankings differ is noted. ## 3. What CAM16-UCS is, in plain terms CAM16-UCS is a mathematical model of how humans perceive color, published by the CIE (the international body that standardizes color measurement). It takes a measured color and places it as a point in a three-dimensional "perceptual space" where distance between points corresponds to how different two colors *look* to a person — not how different their RGB numbers are. The three dimensions are lightness (light vs. dark), colorfulness (vivid vs. gray), and hue (red vs. green, blue vs. yellow). Distance in this space is written **ΔE′** ("delta E prime"). Why CAM16-UCS rather than the older ΔE2000 that some people may have heard of? ΔE2000 was designed for tiny differences — checking whether two batches of paint match. It gets unreliable for very different colors, and it specifically under-counts differences in colorfulness. White vs. yellow on a cube is almost entirely a colorfulness difference (same lightness, one vivid, one not), so ΔE2000 undersells it. CAM16-UCS was built and tested on large color differences and is the current recommended model for this kind of work. ### How big is a significant change? CAM16-UCS is scaled so that **ΔE′ ≈ 1 is a just-noticeable difference** — the smallest gap a trained observer can detect side by side under ideal lighting. | ΔE′ | Meaning | |---|---| | under 1 | Indistinguishable | | 1–2 | Visible on close side-by-side inspection | | 2–10 | Obviously different at a glance in good light | | 10–20 | Clearly different colors | | 30+ | Very different colors; no confusion in good conditions | Two calibration points from the real data: GAN's red vs. orange scores **32.5**. Everyone with normal vision can tell them apart in good light — and they can. GAN's black vs. white scores **79**, the largest possible gap on the cube. Important: these anchors describe side-by-side comparison under good lighting. A speedcuber is doing something harder — identifying a single color in isolation, in a fraction of a second, on a moving object, sometimes under fluorescent or gym lighting. Every one of those conditions shrinks the effective difference. A pair that scores 30 in the lab can become marginal in competition. That is why the goal is to push the worst pair as high as possible, not merely above "visible." --- ## 4. Results for people with normal color vision Nine cubes were measured (GAN V100 Matte, GAN V100 UV, GAN 251, GAN ME, MeiLong V2, RS3 2020, Moyu Skewb RS, Qiyi budget, Dolphin V1, MGC 2x2). For each cube, two palettes were scored from the *same* measured plastic: the standard six (W Y R O B G) and the OP six (W K Y R B G, using that cube's actual black). ### Normal observer: worst pair and whole-cube contrast | Cube | Standard worst pair | Standard whole-cube | OP worst pair | OP whole-cube | |---|---:|---:|---:|---:| | GAN V100 Matte | 32.5 | 952 | **37.8** | **1000** | | GAN V100 UV | 33.0 | 963 | — | — | | GAN 251 Matte | 33.0 | 947 | — | — | | GAN ME Matte | — | — | **38.5** | **997** | | MeiLong V2 | 33.4 | 974 | **38.4** | **1007** | | RS3 2020 | 31.9 | 922 | **37.8** | 965 | | Moyu Skewb RS | 33.2 | 938 | — | — | | Qiyi budget | 31.6 | 954 | **41.0** | **997** | | Dolphin V1 | 35.2 | 933 | **38.5** | 971 | | MGC 2x2 | 30.4 | 914 | 33.6 | 988 | | V11 UV | 34.5 | 965 | — | — | (Worst pair on every standard cube is red/orange. Dashes mean that cube had no orange or no black to measure. Whole-cube = sum of 15 pairwise CAM16-UCS distances; a six-color cube with black and white will land near 1000.) Three things stand out. **Red/orange is the bottleneck on every standard cube measured, without exception.** Across nine manufacturers, red vs. orange is always the closest pair on the cube. The folklore is correct and it's universal. **The OP scheme raises the worst-pair score on every cube, by 3 to 9 points (roughly 10–30%).** On the GAN V100 Matte, the most-used competition cube in this set, the improvement is 32.5 → 37.8, about 16%. The six highest-scoring palettes in the whole dataset are all OP palettes. **Both scoring methods agree on the headline.** Ranked by worst pair, the top six palettes are all OP. Ranked by whole-cube contrast, the top five are all OP, and every cube's OP palette out-scores its own standard palette on both measures. The two rankings disagree only in the middle of the table — for example, whole-cube contrast ranks the GAN V100 Matte OP (1000) well above the Dolphin OP (971) even though their worst pairs are similar, which matches Andy's hands-on experience that the GAN reads better as a whole cube. Worst-pair ranking puts them roughly level. **The improvement is real but not enormous.** Both schemes are well above the "clearly different" threshold for a normal observer in good light. The honest claim is that OP removes the weakest link and raises the floor — not that the standard cube is unreadable for people with normal vision. Whether a ~16% wider worst pair translates into measurably faster recognition during a solve is an open question that would need a timed recognition test. That test is the planned next step. ### Where the OP scheme's own weak points are Once orange is gone, the new closest pairs are **yellow/green** (37.8 on GAN) and **white/yellow** (40.4). Neither is a problem, but they are where the remaining room for improvement lives. GAN's green is quite light, which brings it closer to yellow than it needs to be; a slightly darker or cooler green would lift the whole palette. That is a specific, measurable request that can be made of a manufacturer. --- ## 5. Results for colorblind cubers This is where the two schemes separate decisively. ### What "colorblind" means here The retina has three cone types, sensitive to long (L), medium (M) and short (S) wavelengths — loosely, red, green and blue. Color vision deficiency (CVD) means one of these is missing or shifted. The three classes: | Type | Affected cone | Everyday effect | Prevalence (men) | Prevalence (women) | |---|---|---|---|---| | **Protan** (protanopia / protanomaly) | L ("red") missing or shifted | Reds look dark; red, orange, yellow, green compress toward one another | ~2% | ~0.03% | | **Deutan** (deuteranopia / deuteranomaly) | M ("green") missing or shifted | Red/orange/green compress; the most common form | ~6% | ~0.4% | | **Tritan** (tritanopia / tritanomaly) | S ("blue") missing or shifted | Blue/green and yellow/white compress | ~0.01% | ~0.01% | Protan and deutan together are "red-green colorblindness" — about **8% of men and 0.5% of women** in populations of European descent, somewhat lower in East Asian and African populations. Globally the usual estimate is around 300 million people. Tritan is genuinely rare and is included for completeness. Within each class there are two severities. The "-opia" forms (protan**opia**, deuteran**opia**) are full dichromacy — the cone is absent, roughly 1% of men each. The "-anomaly" forms are a shifted cone — milder, and far more common (deuteranomaly alone is about 5% of men). The simulations below model the **full dichromat** case, so they represent the severe end. Anomalous trichromats see something between the normal column and the simulated column. ### How the simulation works Each measured color was passed through the Machado–Oliveira–Fernandes (2009) model, a peer-reviewed transform that computes what a color looks like to a person missing a given cone type. The transformed colors were then scored in CAM16-UCS exactly as before. So the CVD columns answer: *for this observer, what is the closest pair on the cube?* ### Worst-pair contrast by observer type | Palette | Normal | Protan | Deutan | Tritan | |---|---:|---:|---:|---:| | GAN V100 Matte — standard | 32.5 | **6.7** | 12.7 | 10.2 | | GAN V100 Matte — OP | 37.8 | **22.5** | 15.5 | 10.2 | | MeiLong V2 — standard | 33.4 | **0.8** | 12.0 | 9.0 | | MeiLong V2 — OP | 38.4 | **27.9** | 12.0 | 9.0 | | RS3 2020 — standard | 31.9 | **2.7** | 11.9 | 9.5 | | RS3 2020 — OP | 37.8 | **28.0** | 11.9 | 9.5 | | Qiyi budget — standard | 31.6 | **4.8** | 9.1 | 8.9 | | Qiyi budget — OP | 41.0 | **30.7** | 9.1 | 8.9 | | Dolphin V1 — standard | 35.2 | **4.1** | 17.3 | 7.9 | | Dolphin V1 — OP | 38.5 | **25.1** | 17.7 | 7.9 | | MGC 2x2 — standard | 30.4 | **6.8** | 13.4 | 8.1 | | MGC 2x2 — OP | 33.6 | **20.1** | 15.0 | 8.1 | For colorblind observers the worst pair is the right measure and there is no real dispute about it: if a protanope cannot separate orange from green, it does not matter how far apart the other fourteen pairs are. The tables in this section report the worst pair only. **Protan: the standard cube is essentially broken.** Every standard palette scores between 0.8 and 6.8 for a protanope, and the failing pair is always **orange vs. green**. On the MeiLong, that pair scores 0.8 — below the threshold of perception. A protanope holding a MeiLong cannot tell orange from green. On GAN it's 6.7: visible if you stare, hopeless mid-solve. The OP scheme lifts the protan worst pair to 20–31 on every cube, a 3× to 30× improvement. This is the single largest effect in the dataset. **Deutan: a modest improvement, and one honest limitation.** The deutan bottleneck on the standard cube is yellow vs. orange (12.7 on GAN). Removing orange lifts that, but the next-worst pair — **red vs. green** — remains at 15.5. The OP scheme does not fix red/green for deuteranopes; nothing built on red and green can. What it does is remove a three-way collision (red/orange/green) and leave a two-way one, and it never scores lower than the standard cube on any measured example. The design also makes red/green more separable than it looks: black anchors the dark end, which lets red sit at a lightness that a deuteranope can read against green. **Tritan: no change.** Tritanopes struggle with white/yellow and blue/green, both of which exist in both schemes. Scores are identical. Given that tritan CVD is about 1 in 10,000, this is noted for completeness rather than as a practical concern. --- ## 6. How many people does this affect? Being careful with the arithmetic: this can estimate how many cubers *have* the problem, not how many *would start cubing* if it were fixed. Nobody has that second number. - Speedcubing skews heavily male — competition attendance is commonly cited around 90% male. Applying the population rates: **roughly 7 in every 100 people at a competition have red-green CVD.** In a 300-person comp, that's around 20 people. - Of those, about **1 in 4 is protan** (2% of ~8%). Those are the cubers for whom the standard cube's orange and green are near-indistinguishable — the case the measurements show most clearly. - The other 3 in 4 are deutan, mostly mild deuteranomaly. For them the standard cube is worse than it needs to be rather than broken; the OP scheme gives them one fewer confusable pair. - Outside competition, the numbers are larger in absolute terms. Hundreds of millions of cubes have been sold. If even a few percent of would-be cubers put the cube down because two of its colors look alike to them — and the measurements say they do — that is a population in the millions who were given a puzzle with a built-in defect for their eyes. It is worth saying plainly what these numbers do not show. They do not show that colorblind people have quit cubing in droves, and they do not predict how many would pick it up. What they show is that the standard cube has a measurable, severe flaw for about 2% of men, a milder one for another 6%, and that the OP scheme fixes the first and improves the second — with no cost to anyone else, and a small gain for everyone. --- ## 7. Caveats, stated up front - **Measurement conditions.** Nix Mini 3, M2 (UV-excluded), D50 illuminant, colors measured in place on assembled cubes. The instrument samples a 13 mm spot, so on smaller facelets some light from chamfers and neighboring pieces is included. Publication-grade numbers will come from disassembled pieces on a matte black backing. - **Black on stickerless.** These are stickerless cubes. The OP "black" measured here is each cube's own black plastic; on a production OP cube it would be a proper black facelet color and could be tuned like any other. - **Model limits.** CAM16-UCS was evaluated under one stated viewing condition (a normally lit room). The tritan simulation is the least reliable of the three; the model's own authors flag it as approximate. Protan and deutan are well supported. - **Recognition speed is not yet measured.** Contrast is a necessary condition for fast recognition, not proof of it. A controlled timed-recognition test is the next planned step, and it is the test that would speak most directly to speedcubers. --- ## 8. One-paragraph version Using a spectrophotometer and the CIE's CAM16-UCS color model, all 15 color pairs were scored on nine stickerless cubes for both the standard and the Opponent Process (white/black/red/yellow/green/blue) schemes. Red vs. orange is the closest pair on every standard cube measured. Replacing orange with black raises the closest-pair contrast by 10–30% and whole-cube (summed) contrast by 3–9% for people with normal vision, and by 3× to 30× for the roughly 2% of men with protan colorblindness, for whom orange and green on a standard cube are nearly or entirely indistinguishable. Deutan cubers (a further ~6% of men) see a smaller improvement; the red/green pair remains their limit under either scheme. Nobody is made worse off. The next step is a timed recognition test to see how much the wider contrast is worth in actual solving.