Lessons learnt

[FanF]

Here's the place where we can think about doing it better for the next set - always improving :)

[FanF]

Absolute grades reflecting the value of individual cards in a vacuum.

It'd be clearer if the initial grades, instead of being the standard grades we could see on Limited Resources or Lords of Limited, could be grades based on the value of the cards with none of the dependencies fulfilled. That would allow the increments in grade generated by pairing them with other cards to be more reflective of the increase in power that they get this way.

As an example, if we look at the Party keyword: Deadly Alliance is quite commonly considered as a very good removal. For ZNR, we graded the improvement brought by each party member as "less than half a grade" since we based our initial grade at already a good rate.

For the needs of the micro-synergy evaluator, It would be more accurate to grade Deadly Alliance as an average filler (an expensive removal, even if it is playable at instant speed), say grade 2, and get every party member to improve it by half a grade: grade 2.5 for CMC 3B, grade 3 for CMC 2B, grade 4 (i.e. big bomb) for CMC B.

This grading method would provide a finer way in evaluating incremental improvements. For example, while Deadly Alliance improves by "half a grade" for each Party member in play,  it is probably correct to improve Synchronized Spellcraft by only "less than half a grade" every time, which would bring it from a filler grade without anyone in play, to a good filler grade with a complete Party.

[FanF]

Absolute grades need to be displayed to support the estimation of the improvement

There is some confusion on what the grades represent. It could be dispelled by showing how much each card is individually graded, and give examples of higher-graded cards to compare. 

[FanF]

Rating scale

There's an issue with clustering: clustering requires ties between cards to be reflected in the grading, irrespective of the total grade (absolute grade + improvement from the pairing) of a card, which needs to be capped to a maximum (e.g. 5 or A+).

For example, Canyon Jerboa does improve Phylath (or more specifically, its Plants coming when it ETBs) by a quite a bit, in a go-wide strategy. However, Phylath is already an A bomb, so the system based on a rating relative to the absolute grade does not allow to improve it by much more! So the pairing improvement gets rated "by less than half a grade". Generally, when a card is good in the first place, it won't get improved by much by any other card.

The problem comes at clustering stage. A loose connection from a "by less than half a grade" rating between 2 cards won't be strong enough to bring them in the same cluster. Arguably, it'll be the case for all other pairings, since no card will bring more than a marginal improvement. But the relative strength of each connection between these cards is lost in the process.

On the contrary, getting a plain "it improves it a bit" or "it improves it a lot" as input, without mentioning grades, could capture the strength of the connection, and could be adapted later depending on the absolute grade of the improved card.

[FanF]

Rating scale (2)

Another advantage at having a rating scale decorrelated from the gradings would be to provide a list of the best combo pairs of the set, rather than a list of the highest absolute improvements - which is arguably harder to make use of and... so much less fun...!