Guide on how to evaluate mechanics

If you went through the following list and are still not quite sure about your mechanic, feel free to ask me and I will try to be of some help.
e.g.: Dredge, Transmute, Buyback
Games of Magic (and more or less every other game, too) are exciting because no two games will be the same. Drawing a card each turn is a mini lottery that could give you right what you need or exactly the opposite. By deck building or effects like scry you can increase your chances, but you can never be sure to have what you want each and every time. Dredge, transmute and, to a smaller extent, buyback circumvent this by allowing you to either find the card you need easily of simply to use a card over and over again. This is horrendously boring and should never be the core of a mechanic.

e.g.: Echo
Keywords have such an important job to sell a set, that a lot of players first look at a set's keywords before looking at any of the cards, because if they know the mechanics, they can get a pretty good idea what the set is about and what to expect of it. If the keyword is purely negative and doesn't even allow for any creative brewing, then people won't get excited by the new set and nobody will look at it, let alone play it. If you have a drawback that, for some reason, appears frequently in your set, then by all means don't keyword it.
e.g.: "Mystic" from GDS2
This one might seem a litte of and there isn't even a real example for a mechanic that does put counter on lands. The reason for this is that mechanics like "Mystic" (a pseudo keyword that puts "Mystic"-counters on lands you control and/or cares about the number of mystified lands you control) get scrapped as soon as a set gets playtested with actual cards. The reason being that lands are the type of cards that are moved the most during a game, as they are tapped and untapped almost every turn. Putting counters on them makes this increasingly difficult and although I realize that there are certain lands that can get counters, these are all at least uncommons or really old so one could argue that wizards wouldn't do them again.
I don't have an example for this but there are a lot of keywords flying around in the forums that are like that. (EDIT: Actually, the former evergreen keywords landwalk and protection from color are good examples or that. Thanks @digiman for pointing that out.) "Keyword - Whenever a creature an opponent controls dies,...", "Keyword - As long as an opponent controls three or more..." or similar things are highly problematic because they give the choice to the opponent whether your keyword does anything at all. Your opponent might simply play a deck that doesn't even have the cards to activate your mechanic, or they can just play around them. Cards that are specifically against something else are good and necessary, but only in sideboards where they wait to counter dominant strategies.
e.g.: Dredge, "Suffuse" from GDS2
"To suffuse a card, exile it. Whenever you would draw a card, you can put a suffused card from exile into your hand instead." This mechanic, exactly like the infamous Dredge, turns the most basic action of Magic - drawing a card - into a complicated decision. If you only have one suffused card in exile or one card with Dredge in your graveyard, it's not much of a problem, but as it is a mechanic, there will probably occur situations where you have four, five or nine cards that you could possibly choose to "draw". This not only wrests the game of some of its most entertaining features but also makes games last much longer without adding interactivity.
Messing with the exile zone is already a heatedly debated topic and a lot of players were more than skeptical when they first saw the Processor creatures of Battle for Zendikar. There, cards in exile (the opponent's exile, mind you) became passive objects to move around for benefit. This mechanic made as carefully as possible and, imho, also the only way possible which did not bring with it serious problems like endless recurring, repetitive games or horrendous complexity. Messing with cards in your own exile zone or making cards in exile zones do something by themselves is definitely a red flag because exile is supposed to be the zone where things go never to be seen again and therefore is highly non-interactive.
There are actually no real Magic examples but people come up with those every time and again. Attacking is not only an integral part of the game, it is also maybe the most challenging thing to learn for a new player. Less experienced players usually shy away from attacking because they are afraid to do something wrong. And that's quite reasonable because there is a lot to be afraid of, especially if you're relatively new to the game: your opponent's creatures which could potentially double block your creatures, the opponent might have a card that punishes you for attacking, or you are simply still a bit intimidated by the complexity of the board situation and you don't want to make a fool of yourself by missing something which costs you your precious creatures. There is, to my knowledge, not a single existing official mechanic that actively discourages attacking, and even borderline cases like Bushido or Deathtouch are rather rare. Imagine a keyword that said "Whenever this creature blocks, you draw a card.". It would be disastrous as people would hide behind their creatures forever and nobody would like to attack anymore to not grant the opponent a bonus. Mechanics should encourage attacking and risk taking, caution comes all by itself.
A lot of people try to make their mechanic extra fancy by creating a counter that indicates something special. That brings with it a bunch of problems. Firstly, there should (and there really should) only be one kind of counter in your set that goes on creatures. That has the simple reason that a board state should be clear and easy to understand on first glance. That gets extremely difficult when the dice or pennies on a creature could either be +1/+1-counters, blast-counters or whathaveyou-counters. This means that you drastically limit your design space for the set if you use any kind of special counter. Secondly, using novel kinds of counters limits the compatibility of your set with others. A mechanic with a unique type of counter is almost always also an insular one because the keyword would otherwise have to appear in other sets too.
e.g.: Sweep, "land deficit" from GDS2
Lands are maybe the most integral part of Magic. Everyone plays lands, everyone needs lands, and having no lands is one of the most unpleasant experiences one can have. Although keywords can work well with lands, landfall being the best example, they should really not push players towards risking their manabases. Sweep makes you bounce a huge number of your lands to your hand in order to do something meaningful, possibly crippling you enough to lose soon afterwards. Land Deficit, which grants a bonus as long as you control fewer lands than an opponent, pushes you towards missing land drops or sacrificing lands. Lands are needed to play with your cards. Mechanics that make you unable to play aren't fun and aren't worth the trouble.
e.g.: Affinity for artifacts, "Tithe" from GDS2
"Tithe <number> - You may cast this card without paying its mana costs as long as you have <number> or more <color> cards in your graveyard." Mechanics that allow free casting aren't necessarily unfun but always have the extremely high risk of being used by players in unfair ways. Even a keyword like Cascade, which brings as much randomness as possible, could be "abused" by searching for the same card over and over - like Hypergenesis for instance. Adding a single mana which can't be reduced away like done with Emerge or Delve really does a lot and can put a dangerous mechanic back into the safe zone.
e.g.: Replicate
Sometimes, a mechanic simply wants too much and its potential is so big that it needs to be nerfed. Unfortunately, people get interested in sets not after they evaluated each and every card and studied its potential, but by looking at a set at first glance and finding something that appeals to them. Powerful cards - or better: cards that look powerful - are key to selling a set, and that's why cards should rather looks powerful but aren't than the other way 'round. Replicate is a good example for a keyword that is reasonably costed and sounds interesting, yet looks like complete garbage in most cases. This is hardly the keyword's flaw, but rather the fault of the human attention process, but nonetheless you should keep this in mind and design accordingly.
e.g.: Chroma, Devotion
Draft and Sealed are not only really fun, but they are also the way a lot of new players come in contact with Magic. To make it a great experience not only for the hardcore players who study the appropriate draft guides, a high variety of different strategies should be offered by the set. Normally, this is not the nbr. 1 concern of mechanics, but they can get into the way of this goal sometimes. A good way of crippling the number of draft archetypes is pushing for monocolored decks. The mentioned Chroma or Devotion didn't pose a problem because they were either seldom to find on cards or embedded into a web of other different archetypes with completely different mechanics. They could, however, easily be a problem if overused. Imagine a set with Devotion as the main mechanic (and without hybrid cards). There would be exactly five decks to build in a draft with this set: the mono white deck, the mono blue deck, the mono black deck, the mono red deck and the mono green deck. This would be boring very soon and although pushing mono colored decks isn't inherently wrong, it's something which has to be done with care.
e.g.: Splice onto Arcane, Energy, Zendikar-block Allies
Sometimes a mechanic is so novel that it needs a specific and new environment to shine. Thus, it can happen that the mechanic works well within its block but hardy outside of it. That's what is meant by being insular. A mechanic is parasitic, on the other hand, when the mechanic gets better and better the more cards you have with this specific mechanic on it. Insular and parasitic keywords are a part of Magic and not really a problem. The troubles begin, however, if a mechanic is both at the same time. Tribal cards are usually parasitic as you tend to just stuff your deck full of all your Zombie or Goblin lords to make it stronger. This is no problem as Zombies and Goblins reappear quite frequently in Magic sets. The original Allies from Zendikar and Worldwake however posed the problem that they only cared about other allies and those only existed in these two sets. This made the use of them so narrow that Wotc changed the Ally ability to affecting all creatures you control instead in the later Battle for Zendikar block. A trick that designers should keep in mind when they find their mechanic to be parasitic or insular.
e.g.: “Ransack“ from GDS2, Madness
Mechanics like these can be tricky. They appear to be harmless little tricks but in truth they need so much support that they have the potential to completely warp a set around them. This doesn't have to be a bad thing and can in fact lead to interesting and outstanding designs, however, a keyword like Ransack, which cares about permanents being sacrificed, really requires space and commitment to work properly. It is also advised to find ways beyond the ordinary in which the respective action can happen (for discard that would be: discarding as a cost, looting or wheel effects.).
e.g.: Radiate, “Blight“ from GDS2
Magic has mostly gone away from having cards that benefit each player equally, like done with the first „lords“ which simlpy gave all Goblins +1/+1, including those you didn't control. This is, in my opinion, a good thing, because having one of your cards randomly buffing one of your opponent's creatures feels weird, is unflavorful (in most cases) and can be the source of embarrassing mistakes, especially for inexperienced players. In fact, the most negative thing about the Blight mechanic I can think of is its tendency to feel awkward as soon as an opponent plays Blight-cards, too. For those who don't know, Blight works like a delayed destroy effect and works the following: “ - Sorcery – Blight target creature. (Destroy all permanents with blight-counters on them, then put a blight-counter on target creature.) As soon as an opponent brings their own Blight cards, they suddenly have to destroy their own permanents in order to play their cards – and the same is then true for you. Radiate, the Boros mechanic of original Ravnica, is already weird as soon as an opponent happens to play one or more colors that you play. Suddenly, your spell buffs your opponent's creatures too or you involuntarily destroy your own enchantments. If your mechanic has the potential to lead to awkward situations if both players use it, then you might want to reconsider it.
e.g.: Unleash, Bushido, Ingest
You see, not every mechanic can be super interesting – and not every mechanic has to be. Actually, the vast majority of keywords is probably suitable to play an important role in one set and that's it. Some mechanics, like Morph or Energy, can carry a whole block and more, but most can't and some even have such a small design space that they appear only once in a set and play only a small role there. This doesn't make these mechanics bad and, quite the contrary, can be quite helpful if you want to design mechanics for numerous factions as done in Khans of Tarkir or Ravnica block without having one keyword grab too much attention. Just make sure that your keyword can carry the weight you put on it.