Precon methodology

Upgradeability

Upgradeability is Yama’s judgment of improvements that retain the precon’s main theme. Each commander configuration has its own assessment, including in-box alternatives and supported on-theme commander swaps. A replacement strategy is a separate deck build. Each grade needs a published guide covering that commander, at least two checked nonstock card examples, and written reasons covering card support, commander payoff and restrictions. The guide supplies evidence; Yama assigns the grade. Reviewer grades remain in their original scales. The old unsupported estimates were withdrawn on 27 October 2026.

These are Yama’s ordinal judgments, with no measured win-rate gain attached. A grade belongs to the stated commander configuration and retained main theme. Assess on-theme commander swaps separately; a replacement strategy is a separate deck build. Missing evidence means ungraded. Targeted changes refine the main package; Moderate replaces several packages; Substantial removes much of the shell. Effort describes work, not a budget or Commander bracket.

Commander results and other sources

Playgroup commander results include stock and modified lists, mixed pod sizes and opponents. The raw rate is wins divided by games, not four-player normalised. Partner commanders are separate observations, not a pair's rate. Do not subtract a stock-precon rate to estimate upgrade gain.

Commander counts come from each Playgroup API endpoint, fetched and attributed separately from stock-precon results. We never mix bulk and individual endpoint counts. Zero games has no rate. Samples below 20 games are marked small. No attempt is made to correct these commander rates for pilot skill or selection effects.

EDHTop16 reports competitive tournament entries and conversion rates. Conversion measures advancement, not game wins, and does not describe a stock precon. EDHREC deck inclusion measures card choices, not outcomes. Those measures are not pooled with Playgroup win rates or used as automatic upgrade scores. Dated reviews from named authors provide qualitative evidence. Upgraded-versus-stock trials for these assessments have not been run.

Coverage: 174 packages with 619 stock-compatible commander configurations inventoried. Eighteen configurations have supported upgrade grades; unassessed options may be stronger. Six configurations need banned stock cards replaced, and three require pregame colour-choice input not yet supported by the simulation list format. There are 611 prepared lists including Koma’s outside-commander option, and 575 separately scoped commander observations. The package summary names the highest supported in-box upgrade grade and shows assessed/total coverage; outside commanders cannot supply that grade. A guide using another commander supports that configuration, not the original commander’s grade. New or changed lists need their own matching assessment. Prepared commander lists use separate command zones and identities; they do not inherit the reference list’s simulation results. Upgradeability and commander observations have no weight in the stock meta rank.

Experimental / v1

What the meta rank means

Meta means agreement across evidence sources, rather than a forecast for your local playgroup. We use only stock lists with at least 20 real games and 20 simulated appearances in every published engine. At least four decks must qualify. Unranked decks remain browsable.

We re-rank each source within that same shared cohort. Higher is better: the lowest value scores 0, the highest 100, with tied values given their average rank. For n decks, percentile = 100 × (number below + (number equal − 1) / 2) / (n − 1). Real games use the reported four-player-normalised win rate; each engine uses its fitted strength. The score’s descending order gives the meta rank; tied scores share a rank. Filtering this page never changes the cohort or rank.

The two evidence groups have equal editorial weight. Phase and Forge, when both published, share the simulation half; extra runs from one engine cannot add votes. These weights have not been optimised or validated as a win prediction. Prices, popularity and reviewer opinion have no weight. Sources are dated but not age-weighted, so older results need care.

Engine confidence warnings remain visible and do not change the score or select the cohort: some warnings already use disagreement with real games, so filtering by those warnings would favour agreement by construction. The meta score has no probability interpretation or confidence interval. Large source disagreements and wide simulation intervals remain reasons to inspect a deck.

Why there is no blended win-rate forecast yet

A simulation’s opponents are a closed precon field; real games have different opponents, pilots and selection effects. Averaging those percentages would hide those differences. We show the separate model and observed win rates. A future forecast needs a declared target field, validation against held-out real results, and an interval that includes calibration error. More simulated games alone cannot establish real-world accuracy.

The lab

What simulation measures

Four-player stock-precon pods are fitted with a Plackett–Luce finishing-order model, including turn-position effects. The model win rate estimates a deck’s chance against three random opponents from that run’s field. A 25% baseline represents four equally strong decks. Observed wins are the raw fraction of scored appearances won; they can differ because opponents and seats differ.

Intervals use the published game-cluster bootstrap and their stated coverage level. They describe sampling uncertainty within this simulation, not AI or rules accuracy. Rank ranges can be wide and overlapping. Known engine/AI limitations remain attached to their deck. An unreliable simulation is evidence to investigate, not a reliable power judgment.

Failures and timeouts are excluded, not counted as losses. Turn/action-cap finishes retained by the analysis have no winner. Legacy imports may carry a generic difficulty label without recording the Forge profile; that profile remains unverified unless the reviewed source notes document it. Each engine keeps its own field, scoring, build and dates; percentiles on a shared subset make orderings comparable, but cannot remove differences in their original opponent fields.

Human tables

Real-game evidence

Playgroup.gg supplies attributed aggregate stock-precon results, normalised to a four-player baseline. These are not raw binomial win counts. Its source rules describe stock matching and pilot eligibility; the opponents and pilots differ from our simulation. Yama’s catalog rank is among mapped precons and may differ from the source’s full ranking. Open any deck for its game count and fetch date.

The same real results may already inform simulation confidence labels. They are not an independent validation set for those labels, the meta score or a future calibrated forecast.

Provenance

Lists, prices and publication

MTGJSON supplies stock-list data. Each published simulation is checked against its original index and stock lists. Changed lists lose the old result; missing results remain missing.

Singles replacement estimates use TCGplayer market data via Scryfall. Sealed references use TCGplayer product data via TCGCSV. Each price carries its cached date. Missing copies make a singles total partial; neither estimate proves store availability, a checkout price or resale proceeds. EDHREC links offer upgrade ideas and guides, not measured win-rate improvements.

Publication is deliberate. Imports do not automatically replace the reviewed public snapshots. Only stock catalog data and explicitly reviewed benchmark projections are public. Personal decks, collections, games, experiments and account information remain private. Reading these pages never starts simulations.

Content rights

All rights reserved in Yama’s original software, written analysis and presentation. Third-party material remains with its owners. This notice does not claim exclusive rights in facts, game rules, card lists or ideas. See content rights and reuse.