97 comments

[ 527 ms ] story [ 2396 ms ] thread
https://imgur.com/a/3j3DJ5n

Perfect score, but someone forgot to validate their inputs :P

Same happened for me but I got a score of 2.
I got 2 as well, but I also got frustrated when moving tiles didn't seem to register or something. I tried again, and it definitely seems like there are tiles out of place and I try to change them but they snap back sometimes.
In some cases moving one tile over another didn't work but it did work in the opposite direction. Definitely made the test a bit more difficult because it wasn't always obvious when a change had or hadn't worked.
I start doubting my perceptions when I can't tell if my actions are having an effect.
Me as well. Funny thing is that with a non-calibrated monitor, I scored better than everyone in my art department including the art director... and I'm just a developer who knows a fringe level of Photoshop / Illustrator.
I scored 0 on a phone in quick time, and many other people here are reporting perfect scores as well. The fact that the scores of your art dpt are noticeably lower is thought-provoking.
Not necessarily. Maybe the best student of my course (Architecture) discovered that he was color blind some years later.
Same for me

Best Score for your Gender -2147483648

Worst Score for your Gender 2147483647

Alternatively: how good is your monitor's colour accuracy?

I find it a little odd that they didn't mention that as being one of the main factors that can affect the outcome of the test.

Or the settings of the monitor. I did it in "reader" mode and got a score of 19 struggling as a few boxes looked the same. Switched to "photo" mode and got a score of 0 easily.
Wide gamut ips. 0 on first attempt.
Anecdata: I too got a score of 0 on my first try (having never seen a test like this before) using the builtin LCD on my old ThinkPad X61.

I may have leaned around a little to take advantage of the slight vertical color shifting in this particular LCD but that was about it.

There were a couple of cases where two cells seemed wrong, no matter the order, so I also suspected the monitor is the culprit.

Also, what's the scale? They say lower is better but how better?

This is probably also extra difficult if you have f.lux or something similar running.
Farnsworth Munsell is based on a subset of sRGB, so unless you're using a cathode ray tube, your eyes, not your screen, will discern the hue.
Cheap (and even not-so-cheap) LCDs can be pretty inaccurate:

https://en.wikipedia.org/wiki/Farnsworth-Munsell_100_hue_tes...

In particular, I suspect the DACs and gamma buffers used to derive the LCD driving voltages may have some non-monotonic effects, which would definitely affect the test --- i.e. if increasing the numerical R, G, or B value actually decreases the corresponding colour intensity, even slightly, of the light emitted.

I did this a few years ago on a work LCD monitor. It was much more difficult than on a U2410 at home, on which it was mostly pretty easy. But even that was more challenging than on a GDM-FW900, on which the colors and the gradient were crystal clear, with no need to do any swapping.
I think those colors have been chosen very carefully because I have 0 issues rearranging them on poor quality TFT display of macbook air 11. The seem quite distinct to me.
You don't know what a poor quality TFT display is then ;)
Damn I got 0, but now that I think about it I am pretty sure my monitor is not accurate, things tend to be oversaturated and reddish.
you nailed the color test but failed the reading comprehension test
Exactly my thought.

Also, finally some use for my recently purchased monitor with decent color calibration out-of-the-box.

Gamut matters more than accuracy here. If the monitor is skewed blue, or has darker reds, or something like that, the test remains valid. In general, asking for more blue will lead to a bluer pixel.

The exception is when a monitor is already maxed in blue, but that is about the gamut (range) of the monitor, not its accuracy.

(comment deleted)
> Alternatively: how good is your monitor's colour accuracy?

Mine is a terrible decade old TN panel with banding on most coloured gradients... I got a perfect score so it doesn't seem to measure that much fidelity.

It's not a matter of accuracy, but precision i.e. low delta-E doesn't matter here. Having bad greyscale tracking / non-flat gamma are what would primarily make this difficult but almost any modern screen is pretty good in these areas.
A messsge blocks the site (without ability to remove) to tell me to rotate my phone - but how I use my device is not anyone’s business.
oh look we got a rebel over here
Best Score for your Gender -2147483648 Worst Score for your Gender 2147483647

Hum... I thought 0 was the lowest score :)

(Made 2 by the way, but maybe the night shift on my phone have something to do with it)

I got 2 also with a night shift -- redshift -t 6500k:3900k
They mention that 0 is the perfect score, though I'm not sure how the highest and lowest they mention relate to one another (let alone how someones score could be that far off), but I also got a score of 2 and also in Night Shift mode. My weakest area was 17 (a greenish-teal like slice of the spectrum), which would make some sense in how yellow/orange relate to the colors green and blue.
It would be interesting if they factored in time spent.

I'd also be interested to know if people have different strategies for solving these. I first arranged them into left and right halves, then went through and tweaked the order until I was happy. Toward the end, I did a couple reversals just to see if it looked better or worse. In all cases it looked worse and I switched it back. Final score was 0.

I'm curious what you'd be looking for in time spent. It seems to me that it's just a true or false. I personally did it intermittently while I was gaming on my other monitor. If anything, I'd say adding a confidence weighting on each color scale would gather more data.
Just did a rough ordering and then went through swapping each to see if that improved matters, and ended at a score of 0. I wonder if years of amateur photography and editing has helped, similar to grinding through IQ tests.
The swapping technique seems powerful to me. I do these with a type of bubble sort algorithm which works surprisingly well.
Doubtful, I did the same and got a 0. No photography experience.
I did basically the same, and also ended up with 0. Likewise I didn't end up finding any problems swapping adjacent blocks, but it did make it very clear that the order was correct.
I used the same method, got the same result(s).
Score 6

i got some problems with Ishihara color tables though (but only tried the wikipedia ones)

Zero, but I cheated and just utilized the negative image cast by a crappy TN LCD monitor, viewed from the side, to discern the hue.
it's bugged

Best Score for your Gender -2147483648

Not to brag but as 35 I got 0 on $49 phone. Exupery would be so proud of me.
Perfect score and I am supposedly slightly colorblind in the blue/green range.
This particular color test is tricky because it's not a simple linear progression. After finishing the test, I converted the RGB values to HSV and found irregular jumps in hue (sometimes 3 degrees, sometimes 6) along with nonlinear changes in saturation and value levels. The irregularity probably makes it a better test, but it's annoying because when it's "perfect" it still looks ragged. :-)
> The irregularity probably makes it a better test, but it's annoying

This made me take a look at what they're doing and the sRGB vals map fairly close to their Munsell counterparts. The value has been fixed which is nice for this kind of test, but yeah non-linear hue progression. Probably helps as a discernment test considering how much closer the center colors are to each other.

[Possibly unordered] 10R 5/6, 2.5yr 5/6, 7.5YR 5/4, 5YR 5/6, 7.5YR 5/6, 10YR 5/6, 2.5Y 6/6, 5Y 6/6 [0]https://github.com/germ/munsellScript

C'mon HN. Yeah there is no backend validation on the min and max scores. Doesn't mean you have to hack it.
Referring to the broken min max scores? Should put median though.
Yeah, that was lame. I mean the test was mildly interesting, but they could have provided a few more stats and not just the min and max for my gender. Maybe if they had decided to make a graph they would have thought about form validation and acquired useful results.
Proud owner of a score of 16. I'm red-green colorblind with f.lux running.
6 with night shift on

0 with night shift off (0 is perfect score)

Interesting how much the yellow shift impacts the ability to discern color difference. My night shift slider is only about a quarter of the way up.

Good reminder to not do color critical work with night shift on.

Also got a 6 with night shift on. Haven't tested without yet, but I was having difficulties with a few because of what I felt was the night shift.

EDIT: Just tested without night shift and got 0. Very iteresting :)

I got a score of 74 without doing anything.
Figures. 0 is the perfect score.
A similar thing from different people in which you have to match colors before a timer runs down: http://color.method.ac/
The time addition makes it very stressful for me, and that makes the game a very different experience than the article's one.
I've got perfect score even though I'm regularly unable to name simple colors correctly
Same here... looks like distinguishing colors doesn't mean you can be bothered to name them.
I got a prefect score and I didn't really try. I'm not sure their game is working correctly.
Well, that was different, haven't actually seen one of these. Thanks for sharing. Ps: Best Score for your Gender -2147483648 Worst Score for your Gender 2147483647

:/

Yeh - I saw that.. someone's been having a play!
Reminds me of the Android game "I love hue".
I barely see these color differences, but swapping pairs clearly shows what’s wrong. It is like a bubblesorting by hand.

Score 0 (iphone se screen)

> swapping pairs clearly shows what’s wrong. It is like a bubblesorting by hand.

I found exactly the same, looking at the overall gradient only gets you so far. I wonder if the magnitude of the statistic of 1 in 255 women and 1 in 12 men have more to with how many people figure out how to effectively sort things without being explicitly told than it has to do with colour - perhaps if an equivalent subtle sorting test was done with something other than colour it would reveal a similar statistic.

The difference is definitely genetic. Colorblindness is generally caused by recessive genes that are part of the X chromosome. That means that the proportion of men who are colorblind is basically equal to the proportion of X chromosomes with one of these recessive genes. Women have two sets, so they need the gene on both chromosomes in order to exhibit colorblindness. This means that their proportion of colorblindness is roughly the square of that of men (actually less because there are different types of colorblindness).
Yes but i'm talking about the statistic having a potentially invalid baseline because of the method of testing gives some people an advantage separate from physical ability to perceive colour depending on how they complete the test.
Could we not also correct by simply telling the participants this technique (I used it too and have the same experience of only being 100% sure after bubble sort)
Yes, but it's probably more reliable to just construct a test that forces them to compare in a pairwise sort, than try to explain how to do it.
the trick for me was to compare i and [i + 2] rather than i and [i + 1]. If you're not sure which card is which, rather than swapping them, move one of them so there's a card between it and the other card.
(comment deleted)
> Best Score for your Gender -2147483648 > Worst Score for your Gender 2147483647

Who did this? Come on... own up.