What DeckCortex Actually Does: Sixteen Simulations, an AI That Explains Them, and 367,000 Decklists
Standard, Pioneer, Modern, Legacy, Commander or cEDH — DeckCortex runs sixteen deterministic simulations on your exact list (mana, opening hands, kill clock, combo assembly, power level) and hands the numbers to an AI that writes the primer, the mana audit and the cut list. A full tour of the engine, the AI layer and the 367,000-snapshot tournament corpus behind them, with real lists: from a Modern deck the combo engine correctly cleared to a cEDH pile where it flagged seventeen win lines.
DeckCortex makes one promise on its front page: know how good your deck really is. Behind that sentence sit two systems that deliberately do different jobs. A deterministic analysis engine runs sixteen independent simulations on your exact list and produces hard numbers — mana, speed, consistency, combos, card quality. Then an AI layer reads those numbers, plus live tournament data, and writes the part humans actually want: what's wrong, what to change, and why. The engine gives the AI numbers it can trust, so the advice is data-driven instead of vibes-driven.
And it matters which deck, because DeckCortex isn't a Commander-only tool. It runs the same analysis on a Standard aggro deck, a Pioneer midrange pile, a Modern combo list, a Pauper brew or a cEDH ramp deck — the engine adapts to the format instead of forcing everything through a Commander-shaped lens. As of this writing the platform tracks 367,065 deck snapshots across ten format corpora — 58,017 Modern lists, 44,004 Standard, 36,853 Legacy and 29,039 Pioneer among them, alongside a deep cEDH pool — and every claim below is anchored in that data or in what the app itself documents and displays.
The engine: sixteen simulations, zero vibes
The analysis engine is deterministic and seeded: the same list produces the same numbers every run, results are cached per deck, and the heavy simulations run in a web worker so the app never freezes while it thinks. The simulations execute in dependency order — the mana-source map feeds the consistency model, kill clock and combo assembly feed the game-length model, and so on. Each one exposes a plain-text summary the AI layer can read later.
Here's the full roster, grouped the way the app groups them.
Mana & consistency
Four simulations cover the question every deck lives or dies by. Mana Source Classification sorts every card by how it actually makes mana — basics, duals, tapped lands, fetches, rocks, dorks, rituals — and almost everything else builds on that map. Hypergeometric Distribution is exact draw-probability math, no sampling: the chance of at least k lands in your opener or by a given turn, your color density, and custom "how often do I see one of these N cards" questions. Mana Consistency simulates hands and draws turn by turn to measure real mana per turn, screw and flood rates, and a per-color availability heatmap. And Commander Mana-Basis is the EDH-only layer: it accounts for the free London mulligan and estimates the probability of casting your commander on each turn.
Opening hands
MC Opening Hand deals thousands of simulated openers, applies mulligan logic, and reports your keep rate, mulligan distribution, and how often you make a play on turns one through three. Each hand is scored by the Hand Evaluator, which grades any seven cards from 0–10 on land count, curve, color access, interaction and synergy clusters — and can be calibrated to your specific archetype instead of generic defaults, so a turbo deck's "keep" isn't judged like a control deck's.
Speed, combo & win
The Goldfish Kill Clock plays out games with no opponent interaction to estimate your average kill turn, kill rate and damage curve. Combo Assembly takes the combos detected in your list and simulates how reliably and how fast you actually assemble the pieces, including how much your tutors accelerate it. The Game Duration Model is a Markov chain that combines both into an expected game length, a win probability, and which phase of the game you usually win in. Turn-Win Consistency rolls all of that telemetry into a single score for how dependably the deck closes on time.
Card evaluation
Three simulations judge individual cards. Lift Score is an empirical "do these cards belong together" measure built from scraped co-occurrence data — which cards show up alongside your commander or archetype far more (or less) often than chance predicts. Bayesian Card Inclusion runs inference over tournament win/loss data with role-aware priors and labels every card Core, Flex, or high-risk-review. Synergy Score blends opening-hand fit, lift specificity and an explicit bonus for being a named combo piece into one per-card number.
Commander identity
The EDH-specific trio. Power Level scores a Commander deck on a 1–10 scale from mana efficiency, speed, tutor and interaction density, combo potential and consistency, and assigns a tier from Casual up to cEDH. Resilience estimates how well the deck recovers from board wipes and commander removal — role redundancy, cheap recursion, commander independence. And the Saltiness Score rates the deck 0–10 on mass land destruction, stax, two-card combos, extra turns, free counterspells and tutor density, then lists the offending cards per category. That last one is quietly the best pre-game tool the app has: it hands you the exact list of cards to disclose in a rule-zero conversation.
Which of the sixteen actually run depends on what you're playing. The mana, opening-hand, kill-clock, combo-assembly and card-evaluation simulations fire on every format. The three Commander-identity metrics — power level, resilience, saltiness — only run on EDH decks, and the sideboard tooling only appears where there's a sideboard. Hand the engine a Modern list and you get a kill clock and a mana-consistency read; hand it a Commander deck and you additionally get a power level and a salt score. Same engine, format-aware output.
The homepage illustrates a finished report with a Kinnan, Bonder Prodigy cEDH list: power level 8.4/10, kill clock T3–4, mana consistency 92%, salt score 6.1/10, plus an AI primer that opens with the deck's actual game plan — resolve Kinnan early, turn explosive mana into a win around turn three or four. Two of those four metrics are Commander-only; a Standard or Modern report leads instead with the kill clock and consistency numbers every format shares. Either way the shape is the same: hard metrics up top, a paragraph that explains them, not a wall of JSON.
What the AI does with the numbers
The AI layer is built around a simple mechanic: prompts are templates full of `[TOKEN]` aliases, and before anything reaches the model, DeckCortex substitutes those tokens with your real decklist, the engine's results and current meta data. The model answers about your deck — it literally cannot hallucinate your mana base, because the mana simulation's summary is pasted into the prompt.
The built-in roster covers most of what you'd ask a good coach. Generate Primer writes a complete strategy primer — game plan, key cards, lines, matchups — formatted so it renders natively in the deck's Primer tab. Full Deck Analysis does strengths, weaknesses and concrete swaps with reasoning. Mana Audit reviews color requirements against sources and untapped ratios. Win Clock & Combo Reliability reads the goldfish and combo-assembly simulations back to you. Meta & Sideboard positions the deck against what's currently winning; Sideboard Plan Builder goes matchup-by-matchup in formats that have a sideboard. Cut Candidate Review names the weakest inclusions using the inclusion and lift data. Deck Coach walks you through piloting, and Social Profile writes a shareable, flavor-forward profile of the deck's identity for your pod or playgroup. There's even an Inline Narration that explains the role of every card in the list, and a behind-the-scenes Hand Quality Benchmark Advisor that calibrates the hand evaluator to your specific deck instead of generic defaults.
Each run costs mana — the in-app credit — and resolves to one of three model presets, Apprentice, Adept or Archmage, depending on your tier. Long jobs like a full primer run in the background and notify you when they're done. Under the hood the reports come from Google's Gemini models via Vertex AI, and they arrive in your language, not just English.
If the built-in prompts don't fit, Blueprints let you compose your own: a short set of instructions plus checkboxes for the data and algorithms the AI should read — decklist, banlist, current meta, combos, power level, saltiness, whatever the report needs. Archmage users can publish blueprints to a community gallery where anyone can browse them, signed-in users like the good ones, and a blueprint you clone becomes your own editable copy. It's the difference between using a coach and hiring one you trained yourself.
The combo engine, on real lists
Combo detection runs on Commander Spellbook data, and it works on whatever list you hand it — sixty cards or a hundred. What it's really measuring is how your deck wins, and that lands all over the map by format and archetype. The clearest way to show it is to point the engine at real decks near both ends of the range.
Start with a fair deck. Run the Modern Boros Energy list this site covered in June through the engine and it finds zero complete two-card combos — correctly. That deck wins with creatures and reach; there's no infinite loop, and the engine doesn't invent one. Most of the constructed ladder lives right here: the aggro and midrange decks of Standard, Pioneer and Modern get checked and mostly come back clean. That zero is a real answer, not a blank — it says the deck's plan is combat, so its clock and consistency are the numbers that matter, not a combo bracket.
Now the other extreme. In cEDH, lists are built to trip over a win, and the same engine that cleared Boros Energy lights right up. Blue Farm, the format's most-played deck at 7.7% of our 26,503-list three-month sample, is combo-compact: the engine flags four complete win lines in the #1 Turbo Tuesday list. They're the two families you'd expect. Demonic Consultation names a card that isn't in the deck, exiles the entire library, and Thassa's Oracle turns the empty library into the win — "if X is greater than or equal to the number of cards in your library, you win the game." The other family is Underworld Breach plus Lion's Eye Diamond plus Brain Freeze: escape from the graveyard, three mana from the Diamond, and a storm-copied mill that ends the table. Across 2,029 Blue Farm lists in our corpus, Underworld Breach sits at 99.94% inclusion — the deck without it isn't the deck.
Point the same engine at the Kinnan list that took #6 at the 62-player Battle by the Bay, and it flags seventeen complete two-card combos in the same hundred cards.
The lines the deck is built around: Kinnan plus Basalt Monolith. The Monolith taps for , Kinnan's static — "whenever you tap a nonland permanent for mana, add one mana of any type that permanent produced" — makes it four, and the untap ability nets a mana every cycle. Infinite colorless, two cards. Freed from the Real does the colored version on any blue-producing dork: to untap, tap for more than under Kinnan, profit. And the most popular line the engine found isn't even a Kinnan line — it's Hullbreaker Horror plus any cheap rock for infinite mana and storm.
A combo count is a signal, not a verdict — and it spans the whole spectrum, from a Modern aggro deck's honest zero to seventeen loops in a hundred cards. Your Pioneer enchantment-combo deck or Legacy storm list lands somewhere in between, and the Combo Assembly simulation goes further than the raw count: it measures how reliably and how fast you actually put the pieces together, and how much your tutors speed it up. DeckCortex weighs all of that — plus speed, interaction and tutor density — to assign a deck's final 1–5 bracket itself.
The data layer: 367,065 decklists and counting
Everything above runs on your list. The other half of the platform is what your list gets compared against — a scraped, continuously updated corpus of real decklists. Pulled from our own database today (2026-07-07):
| Format | Tracked snapshots | |---|---| | cEDH | 108,728 | | Modern | 58,017 | | Standard | 44,004 | | Pauper | 41,538 | | Legacy | 36,853 | | Pioneer | 29,039 | | Duel Commander | 16,972 | | Vintage | 16,891 | | Premodern | 14,833 | | Commander (precons & casual) | 190 |
That corpus powers the Staples browser: the most-played cards in any of ten formats, over a one-, three- or six-month window, as decklist-appearance share (not field share — the label matters, and the app is explicit about it). Pick your format and it answers "what am I actually up against right now" — the Modern tab alone draws on 5,601 decks as of this writing. The top of a list is rarely the surprising part; in cEDH over the last six months, across 55,937 snapshots, Sol Ring leads at 98.0%, Command Tower at 87.5%, Lotus Petal at 86.5%. The rows that earn the feature are the ones intuition misses — and that's exactly where splitting by format beats a single global popularity list.
Take the starkest example the data holds. Gemstone Caverns sits at 85.7% of cEDH lists — fourth in the entire format — while on EDHREC (external data) it shows up in roughly 5.1% of casual Commander decks. Same card, same rules text, two different worlds. A global "how popular is this card" number would blur those into a meaningless average; per-format data keeps them apart, and it's the reason a Modern player and a cEDH player see completely different staple lists for the same game. Mystic Remora and Rhystic Study tell the same story from the card-draw side: near-definitional in cEDH lists, quiet role-players outside them.
The meta view is live, too. The homepage's Top Meta Right Now widget aggregates top-cut share from placed tournament decks — 777 of them as of this writing, sourced from MTGO Challenges and MTGTop8 — switchable across nine constructed formats. On the Pioneer tab today: Red Deck Wins at 11.2% of the top cut with a 22% conversion rate, The Rock at 10.3% and 32%, Greasefang Parhelion converting at a nasty 36%. Cards and archetypes carry trend arrows, and the AI's meta tokens read the same feeds — a "what's rising, what's falling" over the last six months is one alias away from any report.
Poke a real tournament deck
Every scraped list is browsable, and the snapshot view is where the data layer gets tactile. Take the Boros Aggro list that won MTGO Challenge 64 on June 21 — piloted by Relieve, built around Ajani, Nacatl Pariah / Ajani, Nacatl Avenger and the frankly rude Voice of Victory, whose Oracle text reads "your opponents can't cast spells during your turn."
Open it on DeckCortex and every card carries its inclusion rate across the archetype, computed from 762 Boros Aggro decks over three months, with rise and fall arrows: Blood Moon shows 75% in the maindeck slot, Voice of Victory 82%, the core four-ofs at 100%. You get the mana curve, the land/spell ratio, colored-pip breakdown, an estimated deck price ($956.99 for this one, with per-card TCGplayer and Cardmarket prices on hover), and the tokens the deck needs to bring. Then it gets playable: the Sample Opening Hand panel deals simulated seven-card draws from the exact mainboard — this list averages 2.68 lands in the opener — and one more click puts you in a full Playtest goldfish mode. Duplicate any list into your own library, export it, or diff it against another build with Compare. It's the difference between reading a decklist and interrogating one.
Where the numbers come from
DeckCortex is open about its sources, and the attribution list doubles as a map of what feeds what: Scryfall for card text, printings and banlists; MTGTop8, TopDeck.gg and MTGO results for tournament data; EDHREC for Commander popularity and co-occurrence; Moxfield for decklist aggregation and one-click imports; Commander Spellbook for combo detection; MTGJSON for precons. The engine's simulations run client-side on that foundation, deterministically. The AI reports run on Gemini via Vertex AI and are grounded in the engine's output — the FAQ's phrasing is exact: the engine gives the AI numbers it can trust.
Formats covered: Commander, cEDH and Duel Commander, the constructed ladder from Standard through Pioneer, Modern, Legacy and Vintage to Pauper and Premodern, plus Arena's Historic, Timeless and Alchemy in both Bo1 and Bo3. Commander-only metrics like power level, resilience and saltiness only run where they make sense, and sideboard tooling only appears in formats that have one.
If you want to see it on your own deck — whatever you play — it's free to start, and a Moxfield import is one click. Run the sixteen simulations, read the report, and find out what your list has been hiding: the engine found zero win lines in a Modern aggro deck and seventeen in a cEDH ramp pile, and both answers were right.
Data notes: corpus counts and cEDH staple shares are DeckCortex's own scraped data (probe run 2026-07-07; staples window 2026-01-08 to 2026-07-07, n=55,937 cEDH snapshots, decklist-appearance share — not field share). Blue Farm/Kinnan combo counts, shares and conversion rates are from our cEDH State of the Format (three-month sample, n=26,503; combo counts per specific list, engine-flagged). Casual-Commander comparison figures are EDHREC (external). Live-site figures (Top Meta widget, staples browser, snapshot page) are as displayed on 2026-07-07 and move with the data. Prices are Scryfall paper aggregates, not live store prices. A combo count is a signal, not the final 1–5 bracket — DeckCortex assigns the bracket itself.