Meta Lens

What each number means

Share: an archetype's lists over all published lists in the window. On MTGO that means the top finishers of each Challenge and 5-0 League decks, so it's the share of winning lists, not of everyone who played. Change: share now minus share in the window before, in points. Effective archetypes: how many equally popular decks the field is worth.

effective archetypes = 1 / (sum of every archetype's share squared)

Win rate: match wins over wins plus losses against other archetypes, mirrors and draws left out, from matches where both players' lists are known. On MTGO those are only the published top-eight rounds, so win rates are thin and lean toward players who reach top eight. Shown with a 95% Wilson interval, and hidden under 20 matches:

interval = (p + z^2/2n +/- z x sqrt(p(1-p)/n + z^2/4n^2)) / (1 + z^2/n), z = 1.96

Top 8 conversion: its share of top-eight finishes divided by its share of the published lists that carry a finish (Leagues left out): 1 is the same share in both, 2 is twice its share of those lists. Shown with every published list selected and at 10 or more such lists.

The typical list

The median main deck and sideboard size across the archetype's lists, filled with the most-played cards first, each at the count most lists run. Only lists legal today go into it, whatever the switch below says; an archetype whose every list in the window holds a card banned since has no typical list, and the panel says so.

Legal today

Every list is checked against Scryfall's legality of each card today (the Oracle file the card pages read, dated on the page). A list is not legal today when it holds a card banned since, more copies of a card than a restriction now allows (Vintage: two or more of a restricted card, main deck and sideboard together), or, in Duel Commander, a commander banned as commander since (in the 99 that card stays legal). The date and source of each change come from the format registry behind the format pages (data/formats/registry.json, read from Wizards' announcements and each format's own committee); a card Scryfall marks banned with no dated change on file says "a date not on file". Every such list on file is dated on or before the change that marks it: lists from the day of an announcement are the day's events before Magic Online applied it.

By default those lists are left out of the window's numbers: shares, win rates, top 8, the field, most played cards, the price of the median list and the typical list. The note above the numbers counts them, names each card with its date and source, and has a switch that counts them as they were played. The Change column, the eight-week lines and the weekly chart always count every list as it was played, since they are history: a ban shows where it happened.

lists not legal today in the window = lists in the window (under the filters at the top) that hold such a card

Search the lists and the event pages show every list on file; a list not legal today carries a tag that says which card and since when.

Since a change

Besides the last 7 to 90 days, the window can start on a dated change: the button is the last change to what may be played (a ban, an unban, a restriction), and the list beside it holds every dated change inside the lists on file, releases included (a release is a set with at least one new card legal in the format, its Commander decks and promos counted with it, dated by Scryfall). The window runs from that day to the newest list; the Change column compares it with the same number of days just before the change.

window = [change date, newest list]; the window before = the same number of days ending the day before the change

A change date the format doesn't have (an old link) falls back to 30 days, and the address drops it.

Shares by week

The archetypes with the most lists in the window, week by week since the first full week on file: each week is 7 days ending on the newest list's weekday, and its share is that archetype's lists over all published lists that week under the filters at the top, every list counted as it was played. All the charts share one scale. An unbroken line marks a ban, an unban or a restriction; a dotted one, a release; the numbers above the charts lead to the list of changes below them, each with its date and source. The same numbers are in a table under the charts.

Prices

Each archetype's median list: every list in the window is priced card by card, main deck and sideboard (the commander too in Duel Commander), basic lands left out, and the middle price is shown (the mean of the two middle ones when the count is even). Dollars: each card at its cheapest paper printing, the site's price rule (the same figure Decks uses), TCGplayer market prices via Scryfall. Tix: each card at its cheapest printing on Magic Online, Cardhoarder prices via Scryfall, counting Magic Online-only printings (Vintage Masters, Tempest Remastered, the Masters Editions, MTGO promos) as well as paper ones. A card with no price adds nothing, and the cell and the archetype's panel count those cards.

a list's price = sum over its cards (basic lands aside) of copies x that card's price

median list = the middle of the window's list prices, sorted

Without the Magic Online-only printings, a fifth of Legacy lists and an eighth of Vintage lists from the last 30 days on file could be priced in full in tix (19.4% and 13.0%, October 21, 2026); with them, all of them. The receipts give each format's figure.

New from the latest release

The newest release with a new card legal in the format, dated by Scryfall, and the share of the lists since its release day (under the filters at the top, legal today unless the switch counts the rest) that play at least one card first printed in it. First printed: the card's earliest printing that isn't a promo, else its earliest printing, belongs to the release or its Commander decks and promos. Each card shows the lists that play it, main deck or sideboard, and the archetype with most of them.

Design around it

Every list's cards looked up in the card facts: how many removal spells per deck and how they kill, counterspells, wipes, discard, graveyard and artifact hate (sideboards included where it says 75), and the share of damage-based removal that kills a creature of each toughness. X damage counts as unlimited.

Search the lists

The search reads every published list on file for the format, not only the window at the top; its own controls are the only ones that apply. A list not legal today is tagged with the card and the date of the change. A list matches when it passes every filter:

• Cards, up to five. Has means at least the copies you pick (any, 2, 3 or 4 or more) in the main deck, the sideboard, or both added together. Doesn't have means no copies there at all. A name is matched ignoring capitals and curly apostrophes, and a two-faced card by its front name too. A name that isn't in these lists matches nothing, and the page offers up to five names in the lists that contain what you typed, most played first.
• Archetype and event: the exact name, from the lists. Player: the MTGO screen name contains what you type, ignoring capitals.
• Finish: won the event is a place of 1; top N is a place from 1 to N in a Challenge or qualifier; League 5-0 is any League list. Leagues have no places.
• When: the last 7, 14, 30, 60 or 90 days ending on the newest list's date, or everything on file.

Newest first sorts by the event's date, then the event, then the place. Highest finish first sorts by place (no place last), then wins minus losses, then the newest. Each row shows the place and record as MTGO published them, with playoff matches in the record, and the copies of the first card the search asks to have.

Lists that match the cards, and the rest

When the search has a card, the lists that pass every other filter split in two: the ones that meet every card condition (matching) and the rest (the rest). After a "doesn't have" the matching lists are the ones without the card; after "2 or more" the rest can still hold a copy. A placed list is a Challenge or qualifier list with a place; League lists are left out of the last two columns because MTGO publishes only 5-0 League lists.

share = lists in the row / lists that pass the other filters

top 8 rate = placed lists with a place from 1 to 8 / placed lists

match win rate = the row's published wins / (wins + losses), draws left out

Both rates carry the 95% Wilson range above. The difference between the rows' match win rates carries Newcombe's range for a difference of two rates (method 10), and the page says where it sits: higher, lower, or can't tell apart when the range includes zero. What it can't see: MTGO publishes about the top 32 lists of a Challenge, so every record here leans high, on both rows alike; and lists in one event played each other, so their matches aren't independent and the ranges are a rough guide.

Event pages

Every event in the file, newest first, filtered by name and window; each event keeps a steady address made from the last part of its published address (which ends in its date and number). Opening one adds a step to the browser's history, so Back returns to where you opened it. The button above the page names where it goes: back to the search, back to the metagame, or all the format's events.

An event's page counts its published lists, their archetypes (lists, share of this event's published lists, top-eight finishes and the highest place), and the winner (the list with place 1). When MTGO's places and its playoff disagree, the page says so and calls the place-1 list “Listed 1st”. The bracket uses the playoff matches MTGO publishes when both players' lists are known. A match's round is how many matches either player had already played in the event, counted in the order MTGO lists them, so a missing quarterfinal doesn't shift the rest; rounds are named from the last (final, semifinals, quarterfinals). Each round is ordered so the two matches that feed one match of the next round sit side by side. The games are as MTGO published them, and the number beside a name is the player's final place, not a seed.

Standings hold only the published lists (about the top 32 of a Challenge), each with MTGO's place and record, playoff matches included. They don't hold tiebreakers, anyone whose list wasn't published, or those players' places: the link to the event's own page on mtgo.com has those. A League page shows the sample of 5-0 lists MTGO posted; it has no places and no bracket.

When the lists stop

A line at the top of the page, before every control, gives the date of the format's newest list, how many days old it is and where the lists come from. From 3 days old, counted by your device's calendar, it is marked and says the archive has stopped; every window on the page ends on the newest list's day, never on today:

age in days = today - the date of the newest list in the format

The archive these lists come from was updated automatically each day and last changed October 12, 2026 (the commit date of the copy the build read); its own README says it is no longer being actively maintained. Every number on the page stops at the newest list's date. Events played since then are on mtgo.com/decklists.

The field

The field section reads its own window (30, 60 or 90 days) and the archetypes with the most published lists in it (12 unless you pick another number, ties by name), under the sources, event types and finishes set at the top. Mirrors are left out. A match is won or lost by games; a drawn match is shown in the record and left out of the rates. It needs at least 50 decided matches between the archetypes listed, and at least 3 archetypes, before it shows any estimate; below that it says which windows and archetype counts in the format clear it, or that none do.

Recorded: each pairing's match record, both directions pooled, and its games, with the Wilson 95% range above.

Pulled toward 50%: every pairing's match win rate p gets one common prior, Beta(a, a), centered at 50% and worth s = 2a matches. A pairing with w wins in n decided matches is Binomial(n, p), so its wins are Beta-Binomial given s:

log likelihood(s) = sum over pairings of [ log B(w + s/2, n - w + s/2) - log B(s/2, s/2) ], B the beta function

The weight isn't fixed at one value. It runs over 241 weights from 1 to 1,000 matches, evenly spaced in log weight and equally likely beforehand (a flat prior on log s over that grid); each counts by its likelihood, and every estimate, interval and draw averages over them. The page also gives the closest single fit (the grid's top, refined by golden section), the weights within 1.92 log-likelihood units of it (a 95% likelihood range, its ends found by bisection between grid points), and the 95% range of the weight after the averaging.

posterior of weight s_g = L(s_g) / sum over the grid of L; estimate = sum over g of posterior(s_g) x (s_g/2 + w) / (s_g + n)

A cell's 95% interval is the 2.5% and 97.5% quantiles of that mixture of Beta(s_g/2 + w, s_g/2 + n - w), found by Newton's method inside a bisection bracket. The prior is symmetric around 50%, so B against A is exactly 1 minus A against B, interval included.

How often the intervals hold the true rate, measured by simulation (the test suite reruns it): the real pairing sample sizes of six windows (Standard 30 and 90 days, Legacy 90, Pioneer 90, Modern 30, Pauper 30), true rates drawn from Beta(s/2, s/2), records drawn from those, and the page's method fitted to each. With true weights of 4, 16, 30 and 60 matches, the cell intervals held the true rate 88.0% to 98.9% of the time; with the true weight drawn from the prior itself, 94.6% to 95.5%. In runs where the closest single fit hit the cap of 1,000, 66.1% to 98.0%: when records happen to look like coin flips, the intervals lean that way too. The field intervals, checked the same way in Standard 30 days and Legacy 90 days, held 88.3% to 95.3%. The intervals this page used before, at one fixed weight held 67.5% to 93.6%. (Measured October 1, 2026, on the match data through September 26, 2026: 97 to 216 pairings with matches per window. The match data is refreshed more often than this is measured, so the page's windows can hold other matches than these.)

Against the field: with x_j deck j's share of the lists of the decks listed (its field weight) and P_ij the estimate for i against j, the mirror counted as P_ii = 50%:

f_i = sum over j of x_j P_ij

Its interval: 4,000 draws of the whole matrix. Each draw first picks a weight from its posterior, then every pairing from its posterior at that weight (j against i as 1 minus i against j); f is computed on each draw and the 2.5% and 97.5% quantiles taken (linear between neighbors). Shares are held at their recorded values. The draws run in a background worker with a fixed seed, so the same data gives the same numbers.

The Nash mix: the game where each player picks a deck and the payoff is P_ij - 50% is symmetric and zero-sum, so its value is 0, and by von Neumann's minimax theorem it has a mix x with sum over i of x_i (P_ji - 50%) <= 0 for every deck j: no deck gains against it. The page solves the linear program

maximize sum(y) subject to (P + 0.5) y <= 1, y >= 0; mix = y / sum(y)

with a dense tableau simplex (Bland's rule, so it can't cycle; pivot tolerance 1e-12). Every mix that meets the condition is an optimum of that program, so the set of them is the face of solutions that reach the optimum: those that leave at zero every column whose reduced cost at the optimum is below zero (by more than 1e-9). From there, two more programs per deck find its smallest and largest weight over that face, and the page shows the range and says when the mix isn't unique. It prints the largest gain any deck has against the solver's mix and against each deck's extreme mixes, and across every draw, beside the bound the check allows (0.001 points). It is solved again on each of the 4,000 draws; a deck counts as in the mix on a draw when its weight is above 0.1%. A deck with no decided match against the others listed is left out of the game: all its estimates are the prior's 50%, so on its own it would be a mix no deck gains against.

One replicator step: each share grows in proportion to its expected result against the field:

x_i' = x_i f_i / f_bar, f_bar = sum over i of x_i f_i (50% here, since P_ij + P_ji = 100% and mirrors count 50%)

with its interval from the same draws. It is one simple model; a step has no set length of time, and it is not a forecast.

Sources: E. B. Wilson (1927), for the recorded ranges; G. Casella (1985), "An Introduction to Empirical Bayes Data Analysis", The American Statistician 39; B. Efron (2010), Large-Scale Inference: Empirical Bayes Methods for Estimation, Testing, and Prediction, Cambridge University Press; A. Gelman and others, Bayesian Data Analysis (third edition, 2013), chapter 5, for averaging over the prior's weight in a Beta-Binomial model; J. von Neumann (1928), "Zur Theorie der Gesellschaftsspiele", Mathematische Annalen 100, for the minimax theorem; P. D. Taylor and L. B. Jonker (1978), "Evolutionary stable strategies and game dynamics", Mathematical Biosciences 40, for replicator dynamics; R. G. Bland (1977), "New finite pivoting rules for the simplex method", Mathematics of Operations Research 2; G. Marsaglia and W. W. Tsang (2000), "A simple method for generating gamma variables", ACM Transactions on Mathematical Software 26, for the draws.

Sources

• MTGO decklists published by Wizards and Daybreak, through the open archive MTGODecklistCache (read October 22, 2026).
• Archetype names: the community rules in MTGOFormatData, matched the way MTGOArchetypeParser does (MIT license); for lists no community rule names, the local rules written here on October 21, 2026 from the lists themselves, listed in the page's receipts with the cards each asks for.
• Legality today: Scryfall's Oracle data; the date and source of each change: the format registry behind the format pages.
• Prices: TCGplayer market prices (dollars) and Cardhoarder prices (tix), both via Scryfall.
• Paper and Arena: off until a source allows reuse. The switch above says why for each.