7 Point Highlander Graph

What does a "finish" mean, and why is it shown as a percentage?

5th out of 200 is not 5th out of 12, so every placement is rescaled to where it landed in its own field: 1 is a win, 0 is last, and 0.5 is what a random finisher would average. Written as a percentage it says the same thing: 0.62 means the deck finished ahead of 62% of the field. That single number lets finishes from events of any size be compared and averaged. A deck is scored wherever a finish ended up on record for it; unscored decks still count as decks played, they just have no finish.

The field a placement is measured against is the number the event published, where it published a usable one. Seventeen of the 139 events on record carry a field this project settled instead, fifteen of those at a floor of 24 that nobody counted. At the four teams events the field counts teams rather than pilots, so teammates share one placement. Those finishes are averaged in like any other, and the head-to-head is the only chart that marks the individual numbers it draws.

Some events published only their top eight, not full standings. At those events the only finishes on record are the good ones, so their average is misleadingly high: 0.91, against 0.51 at an event that published its whole field. Every chart that averages finishes leaves those events out: the metagame landscape's vertical axis, the archetype timeline, a pilot's performance chart, and the player leaderboard. Two charts keep them. The pilot head-to-head plots single placements rather than averages, so there is nothing to distort; hidden gems asks its question inside one event at a time, comparing a card's decks only against the other decks of that same archetype at that same event. Counting decks is a different matter, and meta share, the landscape's sideways axis and card adoption all count every deck the source shipped, brackets included, because those decks were genuinely played.

Where an average finish carries an error bar, the bar is the range that average could reasonably be in, 90% of the time, given how few events stand behind it. Every interval in the app is a 90% one. The spread it is built from is the whole field's rather than the two or three finishes under that one point, so a bar says how thin the evidence is, not how erratic that particular pilot or archetype was.

Where does a year come from?

The source records no date for an event at all. What it does record is when each deck was registered, so an event is dated by the earliest registration among its decks, and a year is the coarsest, safest slice of that: it is what the charts that cut up the metagame are grouped by, rather than anything finer.

Every year axis, the year selector on the Archetypes tab, the label on the Meta tab's own control, and the span in the footer read that one number. The newest year is still filling, so its figures move as events are added. The two timelines are the exception, and place each event on its own registration date rather than in a year.

How is "Meta share over time" calculated?

Each deck counts once, under its main archetype. An archetype's share in a year is its decks that year divided by every deck that year. Small numbers are kept rather than hidden: a thin share is a real sign of an archetype arriving or leaving, and a year with none is a real zero, not a gap.

The "Top 25 / 50 / 75%" control only changes how many lines are drawn. Archetypes are ordered by their share of the most recent year, and the biggest are kept until they add up to that percentage.

The chart opens with the three biggest archetypes at full strength and the rest faded. A faded line is the same real data, not an estimate, and clicking a name in the legend raises that line without hiding the others.

How is the "Metagame landscape" built?

For the year you pick, each archetype is placed by two numbers: how far right it sits is its share of that year's decks, and how high it sits is the average of its finishes. The dot's size is how many separate events those finishes came from.

The two directions count different decks, and the hover shows both. The share counts every deck played. The height uses only events that published full standings (see the finish question above); an archetype with none of those still counts in the year's total but is not drawn.

Only the 25 most-played archetypes of the year are drawn, and the caption says how many the year held in all.

How settled is the "Metagame landscape"?

Each dot carries an error bar: the 90% range its average could reasonably sit in (see the finish question above). Most bars cross the 0.5 line, so most dots could sit on either side of the middle. That is why the caption gives two numbers: how many dots sit above the line, and how many are far enough above that their whole bar clears it. A dot just above the line could easily be there by luck.

More dots sit above the line than below, and that is a real pattern rather than a fault in the data. The 25 most-played archetypes of a year do finish better than the ones the cut leaves out, in every year on record. Taken across the whole field, once the top-eight-only events are gone, the average finish lands within about a hundredth of the middle, so nothing lopsided is left in the source that could account for the tilt. Read it as what it is: a chart of the year's most-played archetypes, which are also the year's better-performing ones.

How is an archetype's "Finishes over time" built?

Each point is one event, placed on the date its first deck was registered (see the year question above). Its height is the average of that archetype's finishes there, usually over just one to three scored decks, so a single point is thin evidence. The hover gives the exact count. Events that published only their top eight get no point (see the finish question above).

The headline above the chart always shows a count, and it hedges itself whenever that count is one luck could easily produce. With a single archetype picked, the count is the events where it finished ahead of the middle of the field (see the finish question above), out of the events it was scored at, which is the number beside its name in the picker. Winning about half the time is what a coin does, and 106 of the 131 archetypes the picker offers are that close, so most of the time the headline says the count could be luck.

Picking a second archetype narrows the chart to the events both attended, and the headline counts how many of those each finished ahead at, hedged the same way. It counts every meeting the record settles, which is every point you can see: one event published just 16 of its 28 finishes, so an archetype can have been there with nothing of its own on record. Its decks finished below every one that event did publish, so the point is drawn at the best it could have been and marked with a ▽, meaning the real finish is that or worse, and the other archetype is counted the winner of that meeting. 67 of the 5,544 pairs the chart will draw carry one. Where neither archetype was scored there is nothing to compare, so that meeting is neither drawn nor counted and both lines break over the date instead.

The chart spans every year in the data; the year selector above it changes the landscape only. If two archetypes were scored together at fewer than two events, the chart says so instead of drawing.

Is a pilot one person, and where do the names come from?

A pilot is whoever registered a deck, held together by the id the source gives them. The name is not the source's: it is recovered from the titles of that pilot's own decks, by taking whichever name appears most often across them.

Names ending in a number, like "Dan S 1" and "Dan S 2", are one place a pilot may not be a single person. The source sometimes files several entries at one event under one id, which no one person can be, so this project separates them into numbered careers. 56 of the 1,263 names offered are one of these, and where one career ends and the next begins is this project's reading rather than a fact on record. Two numbered names sharing a stem may or may not be the same person, so a head-to-head between them is not a rivalry.

It happens the other way round too: one person can be listed as two names. Two spellings are joined only when the names recovered from the titles match outright, so "Josh V" and "Joshua V" stay apart until a maintainer confirms they are one person. 200 of the names offered are currently paired with another that may be the same person, which is a question this project has not settled rather than an answer it got wrong. If you find yourself in here twice, the email and Discord at the foot of the page are the way to say so.

Can I have my name taken off this graph?

Every name here comes from a public decklist. Still, seeing your own results plotted on someone else's graph is a different thing from posting a list. Some people like it, some don't, and I don't think either reaction needs justifying.

Want off? Message me on Discord (alejandrofp92) or email alejandrofuentepinero@gmail.com. I'll pull your identifiable records in the next update, no reason needed.

How is a pilot's "Performance over time" calculated?

For each year, the pilot's finishes are averaged into one score, drawn so higher is better (see the finish question above). A year needs at least two events, otherwise it is left as a gap, and a pilot needs at least two such years before the chart is drawn at all.

The headline "finishes ahead of X% of the field" is the average across all scored years, weighted by events, so a busy year counts for more. It counts every event that published full standings (see the finish question above); the player leaderboard counts only the biggest ones. The two measure different things, and typically put a pilot about 10 places apart in the player leaderboard standings, which rank 139 pilots.

Each year carries an error bar: the 90% range that year's average could reasonably sit in (see the finish question above). The bars are wide and they overlap, because a typical year rests on just three events. Shuffling a pilot's finishes into a random order moves the line about as much as their real career does, so a dip is not a slump and a rise is not improvement. Trust each year's value with its bar, and the career headline, not the shape of the line.

How is the "Head-to-head" timeline built?

It plots the two pilots' finishes at the events they both entered, on the date the event's first deck was registered (see the year question above). Each point is one real placement, not an average, so the top-eight-only events the averaging charts drop (see the finish question above) are safe to keep here. Where only one of the two is on record at a meeting, what happens depends on how much of that event was published. At the one event that published 16 of its 28 finishes, the missing pilot finished below all 16, so their point is drawn at the best it could have been and marked with a ▽. At a top-eight-only event the same reasoning gives an answer too weak to draw: a cut that published a handful of finishes puts the missing pilot below only those few, which is most of the field and says almost nothing, so the line breaks over the meeting instead. Of the 219 pairs affected out of the 44,329 the chart will draw, 164 are drawn and 55 keep the break. A pair needs at least two meetings the record can compare them at, otherwise there is no trajectory to draw and the tool says so: turning up to the same event is not enough if the source scored neither of them there.

A point's hover reads like 3 / 24: the placement, then the field it was ranked against. That second number is the field, not a headcount, and at a teams event it counts teams (see the finish question above). Placements can be shared, by a tie or by teammates, so two pilots sometimes draw the very same point.

Each of those numbers is marked with a * separately where this project decided it, because they are decided separately. Sometimes the source recorded no placement at all and one was recovered from the deck's own title or from the event's cut: 3* / 24 is such a placement against a field the source counted, and 3 / 24* is a placement the source recorded against a field this project settled instead. A decided number is plotted rather than dropped, since leaving it out would quietly shorten a career, and the mark is there so it can be weighed for what it is.

On a pilot's "Archetype affinity", what do the dot sizes mean?

It shows which archetypes a pilot plays, grouped under the broad class each belongs to (aggro, control and the like). Every dot and every link is measured in separate events rather than decks, so a day the pilot entered several variants counts once.

About half the decks on record carry more than one archetype (3,043 of 5,558), each weighted, and a deck counts only under its main one. So a pilot who played a Grixis deck with a little Storm in it is drawn playing Grixis, not both. Counting every tag instead read half the specialists in the record as generalists, which is the question this view exists to answer. Meta share reads a deck's archetype the same way, so the two name archetypes identically; they still count different things (decks there, events here), so their numbers are not meant to match.

How is the "Player leaderboard" scored?

Only the biggest events count: fields over 64, about the top fifth of all events, so every pilot here is measured on the same kind of event. This is the only chart here that drops events for being small, so a pilot's standing here and their record elsewhere answer different questions and will not always agree (see the performance question above). Top-eight-only events are dropped too (see the finish question above).

A pilot's score is the average of their finishes at those events, nudged toward the average score of everyone on the leaderboard when little evidence stands behind it. That average sits well above the middle of the field, since everyone here cleared a bar to get on it. A pilot with five such events is scored about half on their own record and half on everyone's; one with fifteen is scored mostly on their own. That stops one good weekend from outranking a long, strong record.

To be on the leaderboard a pilot needs five of these events, two of them in the last year, so the leaderboard is who is playing now.

Each point on a line is that pilot's score over everything they had played by that date, and a date they reached with fewer than two of these events is left blank, so a line can begin partway across the chart. A climbing line is a record filling in, not a pilot improving, and the last point is the score the standings rank them on. A recent-form version was tried and showed only noise, so the chart shows the record building up instead.

Why is a pilot who has won majors ranked below one who never has?

Because the score reads how far up the field a pilot finished and nothing else. Places are evenly spaced, so at a major of three hundred decks first counts barely three thousandths better than second, and a win is worth almost exactly what a near-miss is worth. A pilot who finishes near the top again and again will outrank a pilot with a trophy and a thinner record.

A bonus for winning was built and measured, and it predicted a pilot's later results no better than the plain score did, so it was left out rather than added for the look of the thing. Read the standings as a measure of sustained finishing, not as a trophy count.

How settled is the "Player leaderboard" order?

Less settled than three-decimal scores suggest. These are the biggest events, so there are only a couple of dozen of them, and a typical contender has played eight. That is not enough to separate pilots whose scores differ in the thousandths.

The standings put a number on that. The "Rank CI" column (CI is short for confidence interval, which is a range rather than a single number) is built by re-picking each pilot's own results at random, repeats allowed, rescoring the whole leaderboard, and doing that a thousand times over. The column is the band of places a pilot landed in across 900 of those thousand runs. Near the top the bands are wide and overlap heavily, so read the leading group as a group, not as a 1-2-3.

How is "Adoption over time" calculated?

For each year, it is the share of decks that ran the card: decks with the card that year divided by all decks that year. Small numbers are shown rather than hidden and a year with none is a real zero (see the meta share question above).

The board control changes the top of that fraction only. Picking "Side" counts only the decks that ran the card there, still against every deck played that year, so it narrows what counts as running the card but not what it is a share of.

What do the percentages on the "Usage" graph mean?

Every one of them is a share of decks, but they do not all share a base, and they are not slices of one another.

The card itself is labelled with its share of every deck in the graph. A link from the card to a broad class (aggro, control and the like) is the share of that class's decks running the card. A link from a class to an archetype is the share of that archetype's decks running the card, counted across the whole archetype rather than just the part sitting under that class. So a number never adds up to the one above it, and reading down a branch the percentages can climb as easily as fall.

An archetype is drawn under whichever class its card-running decks mostly sit in. Its share counts the decks whose main archetype it is, the same decks meta share counts, so a deck carrying that archetype as a lesser tag beside a stronger one is not in either term. Counting every tag instead would describe a pool of other engines wearing this archetype's name: Golgari, for one, is tagged on 200 decks and is the engine of 12.

What does a percentage on the "Co-occurrence" graph mean?

With one card picked, a partner card's percentage is how often the two turn up in the same deck and in the same board: decks running both in Main, or both in Side, out of all the decks running the picked card in either. Because the top of that fraction insists on one board and the bottom does not, the percentage reads lower than the share of decks that simply run both cards somewhere, sometimes far lower.

"Cards to show" keeps the partners with the highest of these percentages. For a single card that is the same order as the decks they share with it, since every one of them is measured against the same total.

With two cards picked, the hub in the middle is the decks running both of them, and every partner hanging off the hub is a share of that hub. The two links from the picked cards into the hub are the exception: each is a share of that card's own decks, so the hub's count is the wrong number to read them against.

What makes a card a "Hidden gem"?

Everything is measured inside one archetype. Put that archetype's scored decks in order by finish (see the finish question above), and the best 20% of them are its best decks. A card then has to pass two tests. It has to be rare in the archetype: in at least 8 of its decks, but no more than 5% of them. And it has to turn up in so many of that archetype's best decks that luck alone would manage it no more than one time in a hundred.

That second test is asked inside each event, comparing the card's decks only against the other decks of the same archetype at that same event, which is why the top-eight-only events are safe to keep here. An archetype needs 160 scored decks before the question can be asked at all, which today rules most of the format out.

Nothing is compared across archetypes, so a card is never a gem just because its archetype wins a lot. The only question is whether the archetype's own best decks are the ones running it. That is also why the table is grouped by archetype rather than ordered best to worst down the page: a single ranking would invite exactly the comparison between archetypes the rule refuses to make. A card can be a gem in two archetypes at once; that is two separate findings.

The graph shows each gem's archetype on one side and, on the other, every one of the best decks that run it, each openable on Moxfield. The decks drawn around a card match its "In top 20%" column exactly. Decks of one archetype sit close together because many of its best decks run the same one card; a deck drawn between two cards runs both.

Does it matter which board a card is in?

Not to the rule. A deck counts once whether the card sat in Main, in Side, or in both, so "Decks running it" is not a count of Main play alone. Every card on the list today is a Main deck card, but a card the archetype's best decks bring alongside in Side would count the same way, and by this measure that is worth what playing one in Main is worth.

It is worth knowing which before acting on a gem, since a card the best decks bring to a matchup is different advice from one they play every game. The link between a deck and a card says which board it was in.

How settled is a "Hidden gem"?

Less settled than a printed list suggests. A card only counts as a gem if luck alone would produce its crowding no more than one time in a hundred (see the hidden gem question above). But every rare card of every archetype is put to that test, which is more than a thousand chances for a coincidence, so even a bar of 1% lets some cards through on luck alone: about one of the three on the list today, and nothing says which ones. The list cannot be checked against later results either: a gem that works stops being rare, so a card still looking like a gem a year on is a card nobody acted on.

The odds are discounted for the pilots behind a card rather than counted per pilot. The same pilot is often behind several of a card's decks, and one pilot's decks rise and fall together, so a card in seven decks by three pilots is charged as about five and a half pieces of evidence rather than seven. The discount is deliberately partial: counting once per pilot instead was tried, and it left nothing on the list at all.

A card's deck count is not all evidence either. Because the question is asked one event at a time, the card's decks at events where none of that archetype's decks reached the best-20% cut could not have landed there, and so say nothing either way. A card in seventeen decks with four in the best 20% is usually a shorter story than that sounds: most of the seventeen never had the chance, and the odds rest on the few events that could have gone either way.

The table says what each gem rests on. "Decks running it" is how rare the card is in that archetype, "In top 20%" is how many of those landed in the archetype's best decks, and "Pilots" is how many people are behind them, which matters because a finish follows the pilot more than any card. A card carried by one pilot is kept and labelled, not hidden. Read a gem as a card worth trying, not as a card proven good.

Why can I not filter the hidden gems?

Because there is nothing left to narrow. The rule is strict enough that the whole format produces well under a dozen gems across a couple of archetypes, which fits in one picture, so the tab draws all of them at once. It recalculates as decks are added, and if it ever finds more gems than the picture can hold, it draws the ones least likely to be luck and the caption counts what is shown.