The sharp changes aren’t what count, it’s the “U” shaped changes that count. A sharp change is just something momentarily stopping like, like space dust or something. The “U” shape is the light moving around the spherical, planetary object (the light being bent based on the objects geometry or even atmosphere).
But yes, it is very hard to tell in some, especially the ones that have gaps indicating no planetary object/star death yet count as a planetary body orbiting…
Working as intended?
You are aware, it says explicitly V form in the tutorial? 
I took it 5 times, if it was just a “v” then every spike would be a transit. It said to look for a distinct u curved pattern, made of multiple spikes that from a distance are interpreted as a “u” even if it is just multiple v’s together. Like the sine/cosine based ones.
I came across one weird one but failed to take an image of it. It had 6 transits but did not acknowledge the third as real because it would break the pattern. It was a tainted result (scientist consensus). Think of the first multi-transit tutorial one but with an extra dip that didn’t make any pattern possible but was formed the same as all other transits 
I have been toying with the project Discovery a bit and I got to say that while it is fun, it gives me sometime weird (unjustified ?) results.
Take this one for example:
It says that I failed, when my markers are right on the spot.
Or in this one:
So I am curious a little bit about the validity of the results that we are compared against.
Some explanations about those results would be welcomed to avoid further frustration and ensure the success of project Discovery.
Thanks.
Moving this to https://meta.eveonline.com/t/devblog-exoplanets-the-next-phase-of-project-discovery/8026 as it is the official Project Discovery feedback thread.
Here is another tricky one. I really doubt this “correct” solution is correct. I find it odd for two reasons: the indicated position is exactly half-way (at 14.5d), and there is no dip in the curve at that location.
I’m almost inclined to think that Michel Mayor is throwing us a curve ball to test how we react to spectras that are deliberately wrongfully interpreted. Maybe to test if we are memorizing them or trying to solve them as intended. There are many solved spectras to choose from, but they do reappear from time to time. And gamers are known for coming up with all kinds of clever strategies to get a highscore. Or maybe I just need to put my tinfoil hat back on.
It’s like CCP and the good professor have put a cat in a box. They ask us to identify what is in the box. We determine through smell,hearing, and gentle shaking that it is a cat. Success we think; then they ask us if the cat is dead or alive. Aha, we think, I know this one: it’s both! Good,good, they say, but by using only the tools we give you and for your reward, what color is the damn cat? They then open the box facing only them, take a look, and say, ooh, close, but not the right answer.
Give the EVE community the right tools and decent types of samples and we will rise to the occasion. In its current frustrating form, I’m treating PD more as a form of cardio workout and not an intellectual exercise.
they stated when it was launched that computing was not able to analyze the data effectively.
Probably because their curated samples are so poor.
It’s looking like there are around 1000 curated samples that we’re being graded against. Shouldn’t be too difficult to collect these into a crowd-sourced DB… Then everyone could have perfect accuracy.
You can click on one of the yellow diamonds and see the folded result.
Unfortunately you can’t shift the period around like you can during initial analysis…
And so they think the human eye using limited tools can do what a computer can’t? Interesting…
without the ability to zoom in while folding, i don’t think samples like this one are even possible to solve:
the up axis is so stretched due to the high peaks that the actual transit is not noticeable at all without a better zoom while folding.
Yes, I do something similar to this with underwater acoustics. The department of defense has had an open multimillion dollar contract for anyone that can develop a computer that can perform as well as human analysts for 30 years. Not a single defense contractor has been able to do it.
Some observations:
I made a bunch of posts on the old forums about this project, but I thought I’d come over here and see what’s going on. Overall I’m disappointed in the extreme lack of care shown with some of the evaluation samples. Some are just tricky and that’s fine; it’s always gratifying to find a sneaky transit hidden in the noise. However, many are just completely wrong. Some have completely missed transits that are fairly obvious, some are actually marked but have missed the actual transit by a mile. One may or may not have been a valid transit: the single marked transit on the given data sample was 100% noise with no decrease in average luminosity over the area, and the orbital period was such that any other (actually visible) transits would be just off the edges of the sample (16 days and change, don’t remember exactly). All of these are posted in various places on the old forums as screen captures.
The developers said in their live stream that they’re going to look over these samples again and see what’s wrong with them. I hope that’s true, because the current state of this project is unacceptable. People will just lose confidence and stop participating if they keep getting told that they’re wrong when they’re clearly not.
this project is really messed up and I’m not going to keep helping with it if it tells me that my markings are wrong when they aren’t
I want this fixed and I want credit for ths standings I lost because of it being broken before I do anymore
After a few days of growing pains, I am absolutely loving this new Project Discovery. My accuracy rating is hovering around a comfortable 85%.
But I have absolutely no idea what to make of this sample. It looks less like a star luminosity graph and more like a digital circuit voltage graph. It goes from 75% at the start to peaks of up to 150%? If ever there were a candidate for a Dyson Sphere in progress, it’s this:
I marked that one “No transit”.
Another request:
When you fold, Highligting the folded area pre submission would be helpful. Making it come out, like it does post submission.
So this one may be an example of an eclipsing binary. Eclipsing binaries are weird because they have two dips per orbit: one when the smaller star goes in front of the bigger one and one when the smaller star goes behind the bigger one. I believe the tutorial says that these will have much deeper transits than planets, but that might not always be the case. Large stars are extremely luminous and a much smaller, dimmer star would produce relatively shallow transits. How do you tell whether this is the case? Dunno, I have limited experience with eclipsing binaries. Sometimes one dip will appear shallower than the other, but that might not always be the case either. One of them I’ve seen had this for a few transits at the beginning of the plate but then this effect went away toward the second half.
