Grim.Cards Simulation Case Study

Edition: 2026-11-01 · Data snapshot: 2026-11-01 · Published: 2026-11-01

Data download: https://grim.cards/case-studies/editions/2026-11-01.json
License: Creative Commons Attribution 4.0 International (CC-BY-4.0) — https://creativecommons.org/licenses/by/4.0/
Dataset version: 2.5
Permanent edition URL: https://grim.cards/case-studies/editions/2026-11-01
Edition index: https://grim.cards/case-studies

Table of Contents

1. Executive Summary
2. Methodology & Provenance
3. Limitations
4. Dataset Overview
5. Key Findings
6. The Gauntlet: Matchup Results
7. Top Commanders
8. Top Cards by Usage
9. Win-Rate Distribution
10. Winning-Recipe Analysis: Construction vs. Outcome
11. Color-Identity Breakdown
12. Card-Category Insights (heuristic)
13. Most Impactful Cards: Decision Impact (counterfactual proxy)
14. Card Performance Roll-Up: Board Impact (measured proxy)
15. Deck Iteration: Retest Win-Rate Changes
16. Coarse Monthly Trends
17. Game Length / Tempo
18. Matchup Structure: Deck vs. Opponent
19. Per-Deck Appendix (Anonymized)
20. Power Score Composite
21. What We Cannot Conclude Yet
22. Future-Comparison Baseline Table
23. License & Citation

Executive Summary

Between 2026-05-14 and 2026-11-01, Grim.Cards recorded 80,907 simulated games across 4,190 player-submitted decks from 1,787 unique users. Every game pitted a human-built deck against a fixed gauntlet of AI-piloted meta reference decks in AI-versus-AI simulation on the Forge engine. No human piloting is involved; results describe engine behavior on these specific decklists.

The headline number is straightforward: player decks won 43.7% of decisive games overall (wins ÷ [wins + losses], draws excluded; n = 4,190 decks, 80,907 games). Commander decks posted 43.9% (n = 3,990 decks, 77,759 games); Standard decks posted 38.5% (n = 200 decks, 3,148 games).

A sub-50% headline is the structural baseline for a one-versus-the-field gauntlet — the field, in aggregate, is expected to win more often than any single challenger. The number is the foundation for every comparison in this report.

The most decisive finding is not the overall rate but the spread across opponents. In Commander, the same pool of challenger decks won 55.4% against Breya Artifact Combo and only 33.3% against Krenko — Goblin Pressure Test — a 22.1-point swing attributable entirely to the choice of opponent, not to any change in the challenger decks themselves. In Standard, that spread reaches 46.2 points: 58.5% against Temur Harmonizer Combo versus 12.3% against Mono Red Aggro. The specific Gauntlet opponent is a powerful driver of simulated outcomes.

Every statistic in this report carries its sample size. Cohort minimum is 10 unique decks; smaller groups are suppressed or merged. All figures are correlational; no causal claims are made.

Methodology & Provenance

Win-Rate Definition

Win rate = wins ÷ (wins + losses), excluding draws.
This definition is applied consistently throughout the report for every rate figure — overall, per matchup, per deck, per cohort, and per trend month. Draws are not counted in the denominator and do not affect win rate.

Simulation Method

Games are AI-versus-AI simulations on a custom build of the open-source Forge engine. A player-submitted deck is played by the Forge AI against each Gauntlet opponent AI. The process is deterministic given the deck and the random seed; it measures engine behavior on the submitted decklist, not human piloting skill.

Cohort Definition

• Included: Real, human-submitted Commander and Standard decks.
• Excluded: Automated Crucible corpus decks (user_id = '__grinder__') and system sample decks (is_sample = true).
• The automated Crucible corpus is not part of this report's "player deck" narrative.
• Win rates and all outcome statistics are split by format (Commander vs. Standard) and are never pooled across formats.
• Every reported cohort meets the minimum threshold of ≥ 10 unique decks. Cohorts smaller than 10 decks are suppressed.

Data Source

Production Grim.Cards database, read-only replica. Snapshot date: 2026-11-01. Data window: 2026-05-14 through 2026-11-01.

Retests

A "retest" is defined as a deck that received more than one completed simulation. Win-rate change is computed as the per-deck win rate in the latest completed simulation minus the win rate in the first completed simulation for the same deck.

Decision Impact (Counterfactual Proxy)

Decision-impact figures derive from recorded AI decision points. For each card, the metric is the mean of −(best-alternative delta) over all decision snapshots involving that card across the qualifying deck cohort. A positive value means the line actually played beat the engine's own next-best alternative by that margin on average; a negative value means the alternative was better. These figures are a play-quality proxy from replayed decisions — not damage dealt, not kills, not a win count — and they do not imply the card causes wins.

Sign convention: positive = the played line improved on the alternative; negative = the alternative was better.

Board Impact (Measured Proxy)

Card performance figures derive from board-quality deltas recorded around turns a given card was observed in play, pooled across all qualifying decks containing the card. The unit is board-quality points per observation. Positive = the board state improved around turns that card appeared; negative = it declined. This is already presented in positive-is-good orientation and does not require sign conversion. It is a board-state proxy, not a damage or kill metric, and it is not a causal win claim.

Card-Category Classification (Heuristic)

Functional card categories (tutor/search, sacrifice outlets, discard effects, reanimation) are assigned via a keyword heuristic applied to oracle text and pre-existing card-preference flags. The heuristic may mislabel edge cases. All category-based win-rate figures carry a visible [HEURISTIC] label.

Privacy

No personally identifying information, raw user or deck identifiers, deck names, or decklists appear in this report. Per-deck appendix rows use rotating pseudonyms (reassigned every edition; no cross-edition tracking is possible) and carry only coarsened derived statistics: Power Score rounded to the nearest 5, win rates and matchup spreads rounded to whole percentage points, and game counts banded. Every published cohort meets the ≥ 10-deck minimum.

Limitations

1. Simulated play, not human play. Results describe the Forge engine's behavior on these decklists. Engine piloting may differ systematically from human play in ways this dataset cannot capture.
2. Correlational only. All construction breakdowns, category comparisons, and card-level figures describe associations in this dataset. None establish causation.
3. Self-selected, non-random sample. Users chose which decks to submit and when. The sample is not representative of all Magic: The Gathering decks or players.
4. Gauntlet opponent selection. Win rates are measured against a fixed set of Gauntlet opponents. Results may differ against other opponent pools.
5. Card-category heuristics. Functional categories are keyword-assigned and may misclassify edge cases.
6. Decision-impact thin data. Counterfactual impact figures rest on small observation counts (as few as 13 recorded decisions in Commander); treat them as directional signals, not precise measurements.
7. Format imbalance. The Standard sample (n = 200 decks, 3,148 games) is substantially smaller than Commander (n = 3,990 decks, 77,759 games); Standard findings carry wider uncertainty.
8. Monthly trend comparability. Monthly cohorts contain different decks; trend figures track submission patterns, not any individual deck's trajectory over time.

Dataset Overview

The dataset covers approximately five and a half months of production activity. Commander dominates the submission pool by volume (3,990 of 4,190 decks, 95.2%) and by game count (77,759 of 80,907, 96.1%).

Key Findings

What follows is the analytical interpretation of this dataset's most important measured results. Every figure is sourced from the data above; section references point to where each finding is substantiated in detail.

1. The Gauntlet opponent matters more than you might expect — especially in Standard.
The same pool of Commander decks won 55.4% against Breya Artifact Combo (n = 3,567 decks, 14,462 games) and 33.3% against Krenko — Goblin Pressure Test (n = 10 decks, 26 games) — a 22.1-point spread across identical challenger cohorts. In Standard, the spread is even starker: 58.5% against Temur Harmonizer Combo (n = 143 decks, 535 games) versus 12.3% against Mono Red Aggro (n = 144 decks, 537 games) — a 46.2-point gap. The implication is plain: reporting a single "win rate" without specifying the opponent can be deeply misleading. The variance decomposition confirms this formally — in Standard, the specific opponent accounts for 37.9% of all matchup win-rate variance, compared to only 22.2% attributable to the deck itself (n = 22 decks, 6 opponents). Commander tells the opposite story: the deck accounts for 48.6% of variance versus 15.1% for the opponent (n = 687 decks, 31 opponents), suggesting that in the broader Commander format, individual deck quality carries more weight against a diverse gauntlet.

2. Most Commander decks cluster near but below 50%, with a thin high-win-rate tail.
The median Commander deck win rate is 42.9% (mean 43.3%; n = 3,842 decks). The distribution is left-skewed: 824 decks sit in the 30–40% bin — the single most populated range — and only 301 decks exceed 70%. This shape is consistent with a challenging gauntlet that separates outlier-strong decks from the field, but most decks converge in a 30–55% corridor.

3. Giada, Font of Hope leads all commanders with qualifying samples in simulated win rate; Meren of Clan Nel Toth is the most-submitted with the lowest rate.
Among the 20 commanders with the largest deck counts, Giada (n = 29 decks, 679 games) posted a 79.2% win rate — the highest of any commander clearing the 10-deck floor. The Ur-Dragon follows at 65.7% (n = 28 decks, 856 games) and Edgar Markov at 63.7% (n = 32 decks, 600 games). The most-submitted commander, Meren of Clan Nel Toth, posted only 35.1% (n = 53 decks, 994 games) — the lowest rate among the top-20 commanders by submission count. Popularity and simulated performance are not the same thing in this dataset.

4. In Commander, land-count bands show a consistent positive correlation with win rate.
Decks in the ≈35% land band (n = 2,481) averaged 44.5% win rate; those in the ≈30% band (n = 496) averaged 37.5%; those in the ≈25% band (n = 84) averaged only 25.7%. High-win-rate Commander decks (>55%, n = 1,029) average 35.9% lands and a mean mana value of 3.28, while low-win-rate decks (<40%, n = 1,545) average 34.8% lands and a mean mana value of 3.03. These are descriptive associations within this sample — not prescriptive build recommendations.

5. Sol Ring appears in 3,450 Commander decks — more than any other non-land card — yet its decision-impact score is deeply negative.
Sol Ring is the most widely played non-land card in the Commander dataset (n = 3,450 decks, 70,181 games, 44.3% win rate for containing decks). Yet its counterfactual decision-impact score is −113.62 points (n = 11 decks, 13 decision observations) — meaning the engine's next-best alternative beat the Sol Ring line by that margin on average across recorded decision points. This is a play-quality signal, not a win-rate verdict, and the observation count is small; it illustrates how a ubiquitous card can still present situations where the engine found better alternatives.

6. Creatures produce the highest board impact by type; battles the lowest.
Pooling card performance across all Commander decks in the cohort (586,193 observations over qualifying creature cards, n ≥ 10 decks and ≥ 25 observations floor), creatures average +3.5 board-quality points per appearance. Battles rank last among types at −7.32 (305 observations, 28 qualifying cards, 19 decks represented). Type buckets pool very different individual cards; these are aggregate descriptors, not verdicts on any specific card.

7. Mono Red Aggro is the Standard gauntlet's wall.
Standard player decks won only 12.3% of games against Mono Red Aggro (n = 144 decks, 537 games) — by far the lowest win rate against any Standard opponent. By contrast, those same decks beat Temur Harmonizer Combo 58.5% of the time (n = 143 decks, 535 games). The 46.2-point spread between best and worst Standard matchup outcome is the defining structural feature of that format's gauntlet.

8. Retesting a Commander deck slightly more often lowered the recorded win rate.
Among 797 retested Commander decks, the average change in win rate from first to latest test was −1.9 percentage points. Of these, 350 improved and 392 declined (with 55 flat). This may reflect regression to the mean, different gauntlet matchup draws between tests, or incremental deck changes — the dataset cannot separate these effects. The Standard retest sample (n = 26 decks) is too small for confident interpretation; the average change was −0.2 percentage points.

The Gauntlet: Matchup Results

Win rate = wins ÷ (wins + losses). All cohorts ≥ 10 unique decks. Split by format.

Commander Matchups

Highest Commander win rate: Breya Artifact Combo at 55.4% (n = 3,567 decks, 14,462 games).
Lowest Commander win rate: Krenko — Goblin Pressure Test at 33.3% (n = 10 decks, 26 games).
Intra-format spread (large-cohort opponents): 55.4% vs. 33.6% = 21.8 points between Breya Artifact Combo and Edgar Markov Vampires (each with 3,565–3,567 decks).

The Aesi Landfall matchup is notable for its draw volume: 2,720 draws in 14,481 games (18.8% of games). This dilutes the decisive-game sample and means the 41.9% rate rests on fewer wins + losses than the raw game count implies.

Note: Four small-cohort matchups (Gisa and Geralf, Kangee, Talrand, Krenko) each rest on exactly 10 decks and 26–35 games. Treat their rates as directional only.

Standard Matchups

Highest Standard win rate: Temur Harmonizer Combo at 58.5% (n = 143 decks, 535 games).
Lowest Standard win rate: Mono Red Aggro at 12.3% (n = 144 decks, 537 games).
Intra-format spread: 46.2 points between the best and worst Standard matchups.

The Standard gauntlet's spread dwarfs Commander's large-cohort spread (46.2 vs. 21.8 points). Player Standard decks collectively won just 66 games out of 537 against Mono Red Aggro — a historically punishing opponent for the submitted pool.

Top Commanders

Commander format only. Minimum 10 unique decks. Win rate = wins ÷ (wins + losses). Ranked by win rate descending.

Most-submitted commander: Meren of Clan Nel Toth (n = 53 decks) — also among the lowest win rates in this group at 35.1%.
Highest win rate among qualifying commanders: Giada, Font of Hope at 79.2% (n = 29 decks, 679 games).
Lowest win rate among qualifying commanders: Vivi Ornitier at 23.5% (n = 30 decks, 646 games).

The contrast between popularity and performance is one of the most striking patterns here. Y'shtola, Night's Blessed is the second-most-submitted commander (n = 42) and sits in the middle of the pack at 41.7%. The top performers by win rate — Giada, The Ur-Dragon, Edgar Markov — are each built on relatively smaller deck counts (28–32 decks), which makes their rates less stable but no less striking. The dataset cannot determine whether these differences reflect commander power, player selection effects (stronger builders gravitate toward certain commanders), or matchup circumstances — all of these may be in play simultaneously.

Top Cards by Usage

Cards ranked by deck count (unique decks running the card). Win rate shown is the win rate of decks containing that card — not the card's own win rate. Basic lands excluded. All cohorts ≥ 10 decks. Correlation, not causation.

Commander — Top 20 by Deck Count

The Commander card landscape is dominated by ramp and fixing staples — Sol Ring, Command Tower, Arcane Signet occupy the top three slots by a wide margin. Removal and protection cards (Swords to Plowshares, Path to Exile, Heroic Intervention, Swiftfoot Boots) appear in the 1,000–1,200 deck range with containing-deck win rates of 46–48%. None of these associations establish that any card causes better or worse outcomes.

Standard — Top 20 by Deck Count

The Standard submission pool is heavily Red-aligned at the top of the card chart — the top 11 most-played cards are all red-spell or red-creature staples typical of aggressive strategies, each appearing in 46–52 decks. The containing-deck win rates for these cards cluster at 45.1–46.9%, consistent with the broader pool performance in favorable matchups. Cards typically associated with Commander gameplay (Sol Ring, Arcane Signet, Command Tower) appear in a subset of Standard submissions (31, 23, and 22 decks respectively) with notably lower containing-deck win rates (29%, 24.3%, 22.4%), which likely reflects the distinct challenges those decks face in a Standard gauntlet.

Win-Rate Distribution

Per-deck win rate distribution, split by format. Win rate = wins ÷ (wins + losses) per deck's completed simulations. Minimum 10-deck floor per bin.

Commander (n = 3,842 decks with completed simulations)

Avg win rate: 43.3% · Median win rate: 42.9% · Min: 0% · Max: 100%

The Commander distribution peaks in the 30–40% bin (824 decks) and has a longer right tail than left. Approximately 65% of Commander decks fall between 20% and 60% win rate, with fewer decks reaching the extremes. The thin 80–100% tail (67 decks total, comprising only 1.7% of the cohort) represents decks that found strong alignment with the gauntlet's weaker opponents or posted extraordinary construction-quality signals — the data cannot distinguish between these.

Standard (n = 173 decks with completed simulations)

Note: The 0–10% bin contains 9 decks — below the 10-deck minimum. It is reported for distribution completeness at the aggregate level but should not be cited as a standalone rate figure.

Avg win rate: 38.1% · Median win rate: 40.0% · Min: 0% · Max: 80%

The Standard distribution shows a meaningful left concentration: 44 of 173 decks (25.4%) posted win rates below 20%. The absence of any decks above 80% contrasts sharply with Commander, where 67 decks cleared that threshold. This likely reflects both the smaller Standard sample and the punishing effect of the Mono Red Aggro matchup on any deck not built to race it.

[GRAPHIC PLACEHOLDER #1: Side-by-side histogram of Commander and Standard win-rate distributions (10-point bins). Commander: 163, 302, 501, 824, 748, 620, 383, 234, 46, 21. Standard: 9, 35, 12, 46, 27, 27, 12, 5, 0, 0.]

Winning-Recipe Analysis: Construction vs. Outcome

Decks grouped into win-rate brackets and compared on construction attributes. Correlational only — not a build prescription.

Commander Win-Rate Brackets

Commander's high-win-rate decks run 1.1 percentage points more lands than low-win-rate decks on average (35.9% vs. 34.8%) and carry a mean mana value 0.25 points higher (3.28 vs. 3.03). The creature percentage difference is 3.1 points (29.2% vs. 26.1%), while spells trend the other way: low-win-rate decks average 20.5% spells versus 16.2% in the high bracket, a 4.3-point gap. These are aggregate differences across a large and diverse cohort; individual decks vary considerably within each bracket.

Standard Win-Rate Brackets

Standard's high-win-rate decks run 4.5 percentage points more lands than low-win-rate decks (38.0% vs. 33.5%) with a mean mana value 0.36 points lower (2.32 vs. 2.68). The most arresting gap is the artifact/enchantment percentage: low-win-rate Standard decks average 14.1% art/ench — three times the 4.4% seen in high-win-rate decks. This likely reflects a compositional effect: Standard decks leaning on artifact and enchantment packages (often slower, more controlling strategies) struggled disproportionately against an aggressive gauntlet. Note: the high bracket has only 25 decks; treat its averages as indicative.

Commander Type Mix (Full Cohort, n = 3,842 decks)

The average Commander deck in this dataset: lands 35.3% · creatures 27.6% · instants + sorceries 18.5% (instants 10.4%, sorceries 8.1% for the n = 3,842 decks with split data available) · artifacts 10.4% · enchantments 7.4% · planeswalkers 0.8%.

Standard Type Mix (Full Cohort, n = 173 decks)

The average Standard deck in this dataset: lands 36.1% · creatures 29.5% · instants + sorceries 25.2% (instants 14.5%, sorceries 10.8%, n = 173 decks with split data) · artifacts 4.1% · enchantments 4.7% · planeswalkers 0.6%.

Land-Count Bands vs. Win Rate

Commander (minimum 10-deck cohort per band):

Standard (minimum 10-deck cohort per band):

In both formats, higher land percentages associate with higher win rates in this dataset. The direction and magnitude of this association should not be interpreted as a causal effect.

Color-Count Bands vs. Win Rate

Commander:

Standard:

Mono-color and colorless Commander decks post higher win rates in this sample (49.0% and 50.9%) than two-, three-, four-, or five-color decks (41.7–44.0%). For Standard, two-color decks top the group at 41.2%. These color-count correlations are confounded by commander identity, archetype, and submission selection effects; they are purely descriptive.

Color-Identity Breakdown

Win rate of decks including each color (decks may include multiple colors; groups are not independent). All cohorts ≥ 10 decks. Correlation, not causation.

Commander

White and Green are the top-performing color inclusions by win rate in Commander (45.6% and 45.2% respectively), while Blue sits lowest at 41.6%. The difference between the highest and lowest is 4.0 points. Because Commander decks routinely run two, three, or more colors, these groups overlap heavily — a deck including both White and Blue is counted in both rows. The color associations here reflect the total mix of archetypes submitted in each color bucket rather than a property of any individual color.

[GRAPHIC PLACEHOLDER #2: Commander color-identity bar chart. W: 45.6% (n=1,820). G: 45.2% (n=1,808). R: 42.8% (n=1,921). B: 42.2% (n=2,073). U: 41.6% (n=1,780).]

Standard

Standard shows a steeper color spread: 15.5 points separate White (43.0%, n = 100) from Blue (27.5%, n = 44). The Blue cohort is the smallest (n = 44 decks); given the gauntlet's composition — including Mono Red Aggro, where slower Blue strategies performed poorly — this rate likely reflects archetype-matchup interactions more than any property of the color itself.

[GRAPHIC PLACEHOLDER #3: Standard color-identity bar chart. W: 43.0% (n=100). R: 40.6% (n=89). G: 38.7% (n=58). B: 33.7% (n=71). U: 27.5% (n=44).]

Card-Category Insights (Heuristic)

⚠ Heuristic label: Functional card categories below (tutor/search, sacrifice outlets, discard effects, reanimation) are assigned by a keyword heuristic applied to oracle text and pre-existing card flags. The heuristic may misclassify edge cases. All figures are correlational within this dataset.

Win rate = containing-deck win rate. Delta = with-category minus without-category win rate.

Commander

In Commander, sacrifice outlets appear in virtually the entire deck pool (3,829 of ~3,842 decks), making the without-category cohort too small (n = 13) to interpret. Discard effects show the largest negative delta: decks running discard effects averaged 42.7% (n = 3,467) versus 49.0% for those without (n = 375), a −6.3-point gap. This does not mean discard effects cause lower win rates — it is more likely that the 375 decks running no discard effects have a distinct archetype composition that performs better against this specific gauntlet.

Standard

Standard's largest gap belongs to tutor/search effects: decks including them averaged 30.6% (n = 85) versus 45.3% for those without (n = 88), a −14.7-point difference. This is the dataset's starkest category gap. Given that the Standard gauntlet features Mono Red Aggro as a difficult matchup for slower strategies, this gap plausibly reflects archetype selection — slower, tutoring-heavy decks running into a meta they weren't built for — rather than tutors themselves causing poor outcomes. The heuristic label applies; some of these cards may be misclassified.

[GRAPHIC PLACEHOLDER #4: Grouped bar chart of category deltas. Commander: sacrifice 0.0, tutor −2.4, discard −6.3, reanimate −0.9. Standard: sacrifice −6.8, discard −7.5, tutor −14.7, reanimate −3.0.]

Most Impactful Cards: Decision Impact (counterfactual proxy)

⚠ Counterfactual proxy: These figures derive from recorded AI decision points — not damage dealt, not creatures killed, not win counts. A positive value means the line actually played beat the engine's own next-best alternative; a negative value means the alternative was better. The sign convention matches the internal dataset's sign, with the improvement-is-positive orientation applied for readability. Observation counts are small; treat directional signals, not precise measurements.

Metric definition: Decision impact = mean(−best_alt_delta) over recorded decision snapshots involving the card. Positive = played line improved on the alternative by that many counterfactual points on average. Negative = the alternative was better.

Commander — Impact Cards (minimum 10 decks, 10+ observations)

The most striking finding here is the negative scores on some of the format's most ubiquitous staples. Lightning Greaves (n = 14 decks, 19 decision observations) posts −209.16: across the 19 recorded decision points involving this card, the engine's next-best alternative was, on average, 209 counterfactual points better than the line actually played. Sol Ring (n = 11 decks, 13 observations) shows −113.62. This does not mean these cards are "bad" — it means that in the specific situations where the engine faced decisions involving them, it frequently had better alternatives available. The observation counts are thin; these are directional signals from a small window into the engine's decision tree.

Evolving Wilds (n = 10 decks, 13 observations) is the only card in the Commander impact list with a positive score: +25.15, meaning the lines involving it beat the engine's alternatives on average across those 13 recorded moments.

Standard — Impact Cards

The Standard counterfactual impact dataset does not have qualifying cards meeting both the 10-deck and 10-observation floors. No Standard impact figures are published in this edition.

Card Performance Roll-Up: Board Impact (measured proxy)

⚠ Measured proxy: Board-impact figures measure the average change in board quality around turns a given card appeared in play, pooled across all qualifying decks containing the card. This is not damage dealt, not kill count, and not a win probability contribution. Only cards appearing in ≥ 10 distinct decks with ≥ 25 total observations are ranked. Rankings weighted by breadth (more decks and more observations = more trustworthy). Type and color buckets pool very different cards; multicolor cards count toward each of their colors.

Commander — Top 15 by Board Impact (n = 3,968 qualifying cards total)

Sazh's Chocobo leads the Commander board-impact rankings with +604.97 per observation — nearly double second-place Surge Conductor's +286.21. However, Sazh's Chocobo appears in only 19 decks (95 observations), compared to Entish Restoration's 74 decks and 240 observations. Entish Restoration's broader representation makes its +201.59 figure more statistically robust. Maelstrom Wanderer and Steelbane Hydra each appear in 30 decks with 141–149 observations — among the best-supported figures in this table.

Commander — Bottom 15 by Board Impact

Chord of Calling sits at the bottom of the Commander board-impact list with −369.37 per observation across 268 observations in 92 decks — the broadest evidence base on this list. Counterspell follows, with the second-largest sample in the entire qualifying card set: 1,270 observations across 557 decks (−25.25 per observation). The magnitude difference between these two (−369.37 vs. −25.25) is noteworthy: Chord of Calling's board-impact signal is nearly 15 times as negative per observation as Counterspell's, on a much smaller observation base. These are board-state signals — the board declined around turns these cards appeared — not verdicts on the cards' utility or design.

[GRAPHIC PLACEHOLDER #5: Commander top-10 and bottom-10 board-impact horizontal bar chart, with deck count and observation count labeled per card.]

Commander — Board Impact by Card Type

Creatures are positive territory at +3.50 with the largest observation pool (586,193). All other types fall below +1.0, with artifacts (−1.22) and nonbasic lands (−2.84) in negative territory, and battles (−7.32) at the bottom — though battles has by far the smallest sample (305 observations, 28 qualifying cards, 19 decks). Artifact cards appear in 3,222 decks, making their −1.22 figure the best-evidenced negative result by breadth. Type buckets pool radically heterogeneous cards; a high-impact finisher and an uncastable bulk rare both count toward the same type bucket.

Commander — Board Impact by Color Identity

All five colors post positive average board impact in Commander; Colorless is marginally negative (−0.05). Green (+3.67) and White (+3.63) lead, Red (+3.16) follows, while Black (+1.60) and Blue (+1.20) trail. Multicolor cards contribute to each of their colors' aggregates, so these are not independent. Colorless cards appear in 3,222 decks — the broadest representation — with a near-zero aggregate signal.

Standard — Board Impact (14 qualifying cards)

Only 14 Standard cards met the ≥ 10 decks and ≥ 25 observations floor for board-impact ranking.

Top performers:

Standard's board-impact leaders are consistent with a highly aggressive card pool — Eidolon of the Great Revel (45 decks, 826 observations, +5.13) leads, with Goblin Guide and Monastery Swiftspear close behind. Shard Volley posts the steepest negative figure (−4.08, 43 decks, 175 observations), which may reflect the land-payment cost appearing as a board-state cost when the card is played.

Standard — Board Impact by Card Type

Creatures lead at +3.33 in Standard as well, consistent with the Commander pattern. Artifacts are the only negative type (−2.94, 28 decks), which likely reflects the specific artifact cards submitted — primarily Commander-format staples (Sol Ring, Arcane Signet) appearing in Standard submissions where they are less effective.

Standard — Board Impact by Color

Only four color groups meet the qualifying threshold for Standard. Red leads at +3.12 (51 decks, 7,508 observations). Green (15 decks) and Black (14 decks) sit at minimum cohort; treat their figures with appropriate caution.

Deck Iteration: Retest Win-Rate Changes

A "retest" = a deck with more than one completed simulation. Win-rate change = latest sim win rate minus first sim win rate per deck.

Commander (n = 797 retested decks)

On average, Commander decks that were retested posted a win rate 1.9 points lower in their latest simulation than in their first. More decks declined (392) than improved (350). This pattern is consistent with regression to the mean, variations in which Gauntlet opponents were drawn between tests, or deck changes made between tests — the dataset cannot isolate which factor dominates. It does not imply that iterating on a Commander deck makes it worse.

Standard (n = 26 retested decks)

The Standard retest sample (n = 26 decks) is too small for confident interpretation. The −0.2-point average change is effectively flat at this sample size. No further sub-group analysis is published for Standard retests in this edition.

Coarse Monthly Trends

Test volume and average win rate by calendar month and format. Monthly cohorts contain different decks each month; this tracks submission patterns, not individual deck improvement. Win-rate figures suppressed for months with no game data in the dataset.

Commander Monthly Trend

Commander test volume expanded dramatically from under 200 simulations per month through August to 1,858 in September and 2,778 in October. The win rate moved from 38.3% in June to 47.1% in November — a +8.8-point increase. This trend tracks who submitted decks each month, not any single deck improving; October–November may reflect a shift in the composition of submitted decks toward higher-performing builds. May shows no game data in the dataset. June game count is low (60 games) and its rate carries wider uncertainty.

Standard Monthly Trend

Standard monthly volumes are modest throughout the window. After sitting at 31.7–33.3% in July–August, the Standard win rate rose to 42.0% in September before easing back to 39.8% in October. No November Standard data appears in the monthly breakdown. Given the small monthly cohort sizes in Standard (20–70 decks per month), month-to-month swings reflect compositional change in who submitted that month rather than any trend in deck quality.

[GRAPHIC PLACEHOLDER #6: Dual-axis line chart of Commander and Standard monthly win rates (2026-06 through 2026-11), with simulation counts as bar overlay. Commander: Jun 38.3%, Jul 42.8%, Aug 41.6%, Sep 43.1%, Oct 44.9%, Nov 47.1%. Standard: Jul 33.3%, Aug 31.7%, Sep 42.0%, Oct 39.8%.]

Game Length / Tempo

Average game length measured in turns. "Avg turns" = mean across per-match average turn counts; shortest and longest reflect the mean of per-match minimum and maximum game lengths within each deck's simulation set. Correlational and descriptive only.

Commander games average 10.2 turns, with typical variation spanning from 8.3 turns (shortest games) to 12.1 turns (longest), across 3,569 decks. Standard games are slightly shorter on average at 9.5 turns, ranging 7.3–11.6 turns, across 145 decks. The similarity in game length between the two formats is notable given their structural differences; within this gauntlet environment, the Forge engine converges outcomes at comparable speeds across both formats. Game length varies by matchup and deck archetype; this report does not publish per-matchup or per-archetype game-length breakdowns, as the per-cell sample sizes are not verified to meet the minimum cohort floor for those cuts.

Matchup Structure: Deck vs. Opponent

This section examines how much of the variation in simulated win rates is explained by which deck is playing versus which Gauntlet opponent it faces. Four complementary views are presented: a Power Score band × opponent win-rate grid, a matchup-inversion rate, a variance decomposition, and per-opponent Spearman correlations between Power Score and win rate. All findings are descriptive decompositions of simulated results. No causal claims are made.

Power Score Band × Opponent Win-Rate Grid (Commander)

Three Power Score bands (terciles) crossed with five high-volume Gauntlet opponents. All cells ≥ 10 decks and ≥ 5 games. Win rate = wins ÷ (wins + losses).

The grid tells a consistent story: across every opponent, the High PS band outperforms the Low PS band by a wide margin. The gap is largest against Breya Artifact Combo (76.4% vs. 34.8%, a 41.6-point spread) and smallest against Edgar Markov Vampires (51.0% vs. 18.0%, a 33.0-point spread). Even within the High PS band, win rates vary substantially by opponent — from 51.0% against Edgar Markov Vampires to 76.4% against Breya Artifact Combo — confirming that the specific opponent still shapes outcomes significantly even among the highest-scoring decks.

A Standard Power Score grid is not published in this edition; the Standard per-cell sample sizes do not meet the minimum cell floor for a reliable three-band × five-opponent cross-tab.

[GRAPHIC PLACEHOLDER #7: Commander PS-band × opponent heatmap. Rows: High/Mid/Low PS. Columns: Breya, Derevi, Aesi, Atraxa, Edgar. Values: 76.4/59.2/34.8, 69.8/53.7/34.8, 59.2/42.6/26.4, 55.4/37.6/20.9, 51.0/33.8/18.0.]

Matchup-Inversion Rate

Share of comparable deck pairs whose strict win-rate preference order between two meta opponents is reversed. Higher = the specific opponent reshuffles outcomes more. All cells meeting the per-cell game floor.

In Commander, 37.2% of the 1,322,706 comparable deck pairs reversed their preference order between the two opponents in a given pairing — meaning that of two decks which both strictly preferred opponent A over opponent B, more than one-third flipped that preference in the opposite direction for at least one other opponent pairing. In Standard, the figure is 29.4% across 1,561 comparable pairs. These rates confirm that the Gauntlet opponent is not merely a scale adjustment applied uniformly across all decks — it genuinely reshuffles outcomes in a format-dependent way. The higher Commander rate (37.2% vs. 29.4%) is consistent with the greater diversity of Commander deck archetypes creating more opponent-specific performance profiles.

Variance Decomposition

Games-weighted share of total per-(deck, opponent) cell win-rate variance attributable to a deck main effect, an opponent main effect, and a residual. Descriptive decomposition only — not a significance test.

The two formats point in opposite directions. In Commander (687 decks, 31 opponents, 3,225 cells), the deck main effect accounts for 48.6% of win-rate variance across cells, while the opponent main effect accounts for only 15.1%. The residual — interaction effects and noise — is 36.3%. The Commander Gauntlet, with its broad 31-opponent diversity, sees individual deck quality as the primary driver of outcome variation.

In Standard (22 decks, 6 opponents, 106 cells), the opponent main effect (37.9%) exceeds the deck main effect (22.2%), with a 39.9% residual. With a narrower five-opponent gauntlet and a smaller, less diverse deck pool, the specific opponent a Standard deck faces plays a proportionally larger role in determining its win rate. These decompositions describe simulated results in this dataset under this gauntlet; they do not generalize to other opponent pools or deck populations.

Per-Opponent Spearman Correlation: Power Score vs. Win Rate

Spearman rank correlation (ρ) between a deck's Power Score and its win rate against each opponent, over decks clearing the per-cell game floor. Correlation, not causation.

Commander

All five Commander opponents show strong positive correlations between Power Score and per-matchup win rate (ρ ranging 0.66–0.75). Breya Artifact Combo has the highest at ρ = 0.75 (n = 660 decks); Aesi Landfall the lowest at ρ = 0.66 (n = 538 decks). The consistently high correlations across opponents mean that, in this dataset, Power Score rank and simulated win-rate rank align substantially — but far from perfectly. The 36.3% residual variance in the decomposition above quantifies what Power Score does not explain.

Standard

Standard's Spearman correlations are more varied — and in the case of Mono Red Aggro, essentially flat at ρ = −0.05 (n = 21 decks). This is the clearest evidence in the dataset that Power Score does not consistently predict win rate in Standard: against Mono Red Aggro, a higher Power Score carried no rank advantage. Jeskai Control shows a weak positive correlation (ρ = 0.30), while Dimir Midrange (ρ = 0.79) and Temur Harmonizer Combo (ρ = 0.68) show the stronger relationships seen across Commander opponents. The small Standard sample (n = 21 decks per opponent) means these correlations carry substantial uncertainty.

These correlation figures involve Power Score, which is an internal Grim.Cards composite. All correlations here are descriptive; they identify how much the internal metric's rank ordering lines up with simulated outcome rank, in whichever direction the data points.

Per-Deck Appendix (Anonymized)

Privacy notice: Pseudonyms (C-001 through C-681, S-001 through S-021) are randomly reassigned every edition. No pseudonym tracks to the same deck across editions, and no pseudonym links to any account, deck name, or decklist. Published values are coarsened: Power Score rounded to nearest 5, win rates and matchup spreads rounded to whole percentage points, game counts banded. Only derived statistics appear — never deck names, card lists, or user identifiers.

Commander Appendix Summary (n = 681 decks, minimum 20 games per deck)

The per-deck appendix spans a wide range: the lowest recorded win rate in the published rows is 0% and the highest is 90%. The median of 45% sits slightly above the full-cohort mean of 43.3%, reflecting that the appendix floor of 20 games per deck skews toward decks with more simulation data. The median matchup spread of 16 percentage points (std dev of per-opponent win rates, for decks that faced ≥3 Gauntlet opponents) indicates that the typical Commander deck's results vary by about 16 points across its different matchups — enough variation to distinguish matchup specialists from generalists in many cases.

Outlier callout — widest overperformance vs. Power Score expectation:
Deck C-507 (PS band: 10, games band: 20–49, 5 opponents faced) posted a 29% win rate against an OLS-expected win rate of 13% for its Power Score — a residual of +16 percentage points. This is the widest positive residual in the Commander appendix: the deck substantially outperformed what its Power Score alone predicted.

Outlier callout — widest underperformance vs. Power Score expectation:
Deck C-326 (PS band: 60, games band: 20–49, 1 opponent faced) posted a 45% win rate against an OLS-expected win rate of 59% — a residual of −14 percentage points. This deck underperformed its Power Score expectation by the widest margin in the published rows.

Both residuals are computed from an OLS regression fit to the published per-deck appendix rows (coarsened Power Score vs. coarsened win rate). They describe departure from the dataset's own internal trend line, not from any absolute standard.

[GRAPHIC PLACEHOLDER #8: Commander per-deck PS vs. win-rate scatter. X-axis: Power Score (coarsened to nearest 5). Y-axis: Win rate (whole points). 681 data points. OLS trend line overlaid. Outliers C-507 (+16 residual) and C-326 (−14 residual) labeled.]

Commander Per-Deck Rows (Selected; full table at data download)

The table below presents all 681 published Commander deck rows. Fields: anonymized ID, coarsened Power Score, games band, win rate (whole points), number of opponents faced, and matchup spread (std dev of per-opponent win rates, for decks with ≥3 opponents; null otherwise).

Full table with all 681 rows is available in the JSON data download at https://grim.cards/case-studies/editions/2026-11-01.json under perDeck.commander.decks.

Standard Appendix Summary (n = 21 decks, minimum 20 games per deck)

The Standard appendix has 21 published decks — exactly at the minimum threshold for publishing this section. The median matchup spread of 24 percentage points is higher than Commander's 16 points, reflecting greater outcome variation across Standard's five-opponent gauntlet per deck. The win rate range (10%–70%) is narrower than Commander's (0%–90%), consistent with the smaller Standard sample and the absence of extreme outliers.

Outlier callout — widest overperformance vs. Power Score expectation:
Deck S-001 (PS band: 70, games band: 20–49, 5 opponents, spread: 29 pts) posted a 70% win rate against an OLS-expected win rate of 67% — a residual of +3 percentage points. With a sample of only 21 decks, the OLS line is too underpowered for confident residual interpretation; this callout is published for structural completeness.

Outlier callout — widest underperformance vs. Power Score expectation:
Deck S-020 (PS band: 45, games band: 20–49, 5 opponents, spread: 25 pts) posted a 43% win rate against an OLS-expected win rate of 45% — a residual of −2 percentage points. The same caveat applies: the Standard sample is too small for residual analysis to carry meaningful weight.

[GRAPHIC PLACEHOLDER #9: Standard per-deck PS vs. win-rate scatter. X-axis: Power Score (coarsened). Y-axis: Win rate (whole points). 21 data points. OLS trend line overlaid. S-001 and S-020 labeled.]

Power Score Composite (Secondary & Caveated)

⚠ Power Score is an internal Grim.Cards composite indicator — not an objective win-rate measurement and not a universal assessment of deck quality. It is presented here as a single secondary section for reference only and is not used to rank, judge, or draw conclusions about decks anywhere else in this report. The grade distribution below reflects the composition of the player-submitted deck pool in this snapshot; it is not a performance verdict.

Power Score combines multiple internal signals into a composite numeric score; the formula and weightings are Grim.Cards proprietary. Score bands map to letter grades.

Grade Distribution (player-submitted decks with Power Score)

The submitted pool skews toward the lower grade bands: D is the single most common grade (1,440 decks), with C close behind (956 decks). Only 22 decks receive the S grade — the top tier. This distribution reflects the self-selected submission pool and the Power Score's internal calibration; it does not describe the broader population of all Magic: The Gathering decks.

Power Score's relationship to simulated win rate is documented in the Matchup Structure section above (Spearman ρ ranging 0.66–0.75 in Commander per opponent; −0.05 to 0.79 in Standard per opponent) and in the per-deck appendix scatter plots. The correlation is meaningful but leaves substantial variance unexplained — particularly in Standard, where one Gauntlet opponent (Mono Red Aggro) shows essentially no correlation between Power Score rank and win-rate rank.

What We Cannot Conclude Yet

The dataset supports the figures above; it does not support the following claims, which would require additional data, different methodology, or future editions:

1. Causal direction of any construction correlation. Whether land count, creature percentage, color identity, or card-category presence causes higher or lower win rates — rather than merely correlating with them in this submitted deck pool — cannot be determined from this data.

2. Human-piloting win rates. All figures describe AI-versus-AI engine behavior on these decklists. How human pilots would perform with the same decks, against the same opponents, is outside this dataset's scope.

3. Generalization beyond this gauntlet. Win rates are measured against a fixed Gauntlet. Performance against different opponents, in different tournament environments, or against human opponents may differ substantially.

4. Individual deck trajectory over time. Monthly trend figures track different cohorts of decks each month. This report cannot show whether any specific deck improved or declined over time.

5. Why specific commanders outperform or underperform. The top commanders by win rate (Giada, The Ur-Dragon, Edgar Markov) may owe their rates to deck quality, player selection effects, favorable gauntlet matchups, or some combination. The data cannot separate these.

6. Card causation. Containing-deck win rates for specific cards, board-impact figures, and decision-impact figures describe associations and engine behavior — not whether any card "causes" wins or losses.

7. Decision-impact reliability at low observation counts. Four cards appear in the Commander impact table with 13–19 decision observations each. Figures this thin are directional signals only; they may shift substantially as more observations accumulate.

8. Standard retest patterns. With only 26 retested Standard decks, no reliable conclusion about retest behavior in that format is possible.

9. Cross-edition deck tracking. Pseudonyms in the per-deck appendix are randomized every edition. No deck can be tracked across editions by a reader; no trend at the individual-deck level is derivable from comparing editions.

Future-Comparison Baseline Table

Machine-readable baseline metrics for this edition. Intended to enable future editions to report what changed. All rates are win rate = wins ÷ (wins + losses), excluding draws.

A machine-readable version of this table is available in the edition JSON at the scope, formats, matchups, retests, tempo, and matchupStructure keys: https://grim.cards/case-studies/editions/2026-11-01.json

License & Citation

License: This dataset and report are published under the Creative Commons Attribution 4.0 International (CC-BY-4.0) license. You are free to share and adapt the material for any purpose, provided appropriate credit is given.

Suggested citation:

Grim.Cards. Grim.Cards Simulation Case Study, Edition 2026-11-01. Dataset version 2.5. Published 2026-11-01. Permanent URL: https://grim.cards/case-studies/editions/2026-11-01. License: CC-BY-4.0. Data download: https://grim.cards/case-studies/editions/2026-11-01.json

Publisher: Grim.Cards
Dataset version: 2.5
Snapshot date: 2026-11-01
Temporal coverage: 2026-05-14 / 2026-11-01
Canonical URL: https://grim.cards/case-studies/editions/2026-11-01
Data download: https://grim.cards/case-studies/editions/2026-11-01.json
Edition index: https://grim.cards/case-studies
Determination method: AI-versus-AI simulations on a custom Forge engine build; win rate = wins ÷ (wins + losses), excluding draws; cohort = real human-submitted decks, Crucible and sample decks excluded; minimum published cohort = 10 unique decks.
Correlation statement: All construction breakdowns, card-level figures, and category comparisons in this report describe statistical associations in the Grim.Cards simulation dataset. None establish causation.

End of Edition 2026-11-01 · Grim.Cards Simulation Case Study · Dataset Version 2.5