Cube Doctor · Cardivore

Run the simulator

Draft your cube 500 times.

~30 seconds for the sim + charts. ~5 minutes if you opt into the AI-written summary at the bottom of the form. Deterministic runs are unlimited. The AI summary is capped at 1 free run per IP per day; install the app for unlimited AI.

How it works

Draft, measure, read.

Three steps, all deterministic. Same cube and same seed produce the same report — every time.

01

Draft

500 × 8-seat × 3-pack drafts on your cube using a deterministic policy mix — 70% adaptive, 15% greedy, 15% synergy bots. Reproducible from a seed; no agents, no inference.

02

Measure

Per-archetype playability, color contestedness, pick-order × DQI, pack-band distributions, role-density triage, top & bottom decks, DQI histogram. All numbers, no narration.

03

Read

Every card charted, every angle. Hover any card name for the image. Optional: layer an agentic AI analysis on top — written cube read plus cuts-to-consider with reasons.

What you get

Numbers, not vibes.

Here's an abbreviated preview of the shareable report. Every chart is built from the simulation output — no narration, no inference. Hover any card on the live report for its image.

Vintage Cube · Example

360 cards · 500 drafts · 36s

deterministic

4.52

DQI

All 10 archetype lanes simulating between 4.10 and 4.60 — tight cluster, no broken lanes.

Archetype playability

UR Spells

4.60

Esper Control

4.60

Orzhov Midrange

4.48

Spells-Matter

4.47

BG Midrange

4.21

Trap quadrant

CARDS PICKED EARLY · DECKS LANDING BELOW AVG

Elspeth, Knight-Errant · P7.6 · DQI 4.24 · 134 MD seats

Drafted seventh on average; the decks running it sim below the cube's 4.52 mean. Sortable per-card tables let you find your own cut candidates without prose.

Wheel predictor

SLEEPER · WHEELS LATE, LANDS IN WINNING DECKS

Lim-Dûl's Vault · P14.7 · DQI 4.64 · 9 MD seats

Dead-last pick on average, but the 9 decks that ran it had the highest DQI of any card in the cube. Drafters are under-valuing it.

+ 13 more deterministic charts · every card in the pool charted, not just the top hits · optional written summary if you check the AI box.

Under the hood

Transparent. Reproducible. Nerdy.

Cube folks don't trust black boxes. The simulator is the same one the Cardivore app ships — TypeScript port of the Dart original, custom LCG RNG so a given seed produces the same draft every time.

Simulator

Monte Carlo draft · 8 seats × 3 packs × 15 cards TypeScript port of the Dart sim

Policy mix

70% adaptive · 15% greedy · 15% synergy deterministic, no model calls

Charts

15 deterministic readouts: power-band bell, traps & sleepers, wheel predictor, MD-conversion, pick variance, P1 leaderboard, power × pick, per-color wheel depth, per-CMC pick band, open-lane frequency, color-pair cannibalization, archetype affinity, co-occurrence threads, DQI histogram, top + bottom decks.

Role-density triage

Each role-ratio finding gets classified actionable / monitor / noise based on lane-health from the sim — no LLM. Healthy lanes' findings demote to "tagging artifact" automatically.

Cost (sim only)

Free to you. ~$0.05–$0.12 of compute per run, charged to us. Deterministic runs are unlimited; AI summary is 1 free run per IP per day. no signup, no card on file

Optional agentic AI

If the box is checked, an agent reads the cube, runs targeted follow-up queries against its card pool, and produces a written interpretation: a cube read, a structural critique, a cuts-to-consider list with reasons, and oracle-text checks on every named claim. Runs can take up to 15 minutes. Limited to 1 per IP per day.

Your data

Cube list in, report out. Nothing retained beyond the public report page; reports are sharable by URL. We don't store account info because there's no account.

Run the simulator unlimited times, on every cube you own.

Cardivore is a full MTG companion — collection tracking, cube building, deck analysis, card scanning, draft archive. The on-device draft simulator is built in.

GET IT ON
Google Play Download on the
App Store cardivore.app →

Nerdy questions

FAQ for skeptics.

Do I have to use the AI?

No. The default flow is sim + charts only — 500 drafts, 15 deterministic readouts, every card on the chart, no LLM involved. The agentic AI analysis is an opt-in checkbox at the bottom of the form: it layers a written cube read, critique, and cuts-to-consider list on top of the same data. Those runs can take up to 15 minutes and are limited to 1 per IP per day. Skip it; the report is still complete.

Why should I trust a sim?

You shouldn't, in isolation. The bots are three distinct policies that each draft badly in predictable, human-ish ways. Across 500 drafts the noise averages out and you can see which archetypes survive, which colors over-draft, and which cards get table-wheeled to death. That's the signal. The report shows the signal; you read it.

What's DQI actually measuring?

Deck Quality Index (0–5). A weighted composite of threat density (creatures 14–17), answer density (4–6 removal), curve efficiency (avg CMC 2.5–3.2 with 8+ early plays), and consistency (color count + fixing). Sanity caps prevent a 2-creature deck from scoring 5. Real cubes top out around 4.2; 3.85+ is "dialed."

What's the sim built on?

A TypeScript port of Cardivore's on-device Dart draft simulator. Custom LCG RNG so runs are reproducible given the same seed. Same code, same cube, same seed = same report. The simulator runs in the browser inside the app; on the public endpoint it runs server-side because it needs Postgres for card data.

What's AI-derived vs hand-coded?

Some of it is AI, some isn't. Per-card role classifications, power levels, and synergy explanations come from a nightly Gemini batch job. The canonical role-ratio targets, the simulator itself, the drafter policies, and every other analyzer are pure code with no model calls. The full split, the ratio table verbatim, the sim's known limits, and the role-taxonomy gaps live on the methodology page.

Ready when you are.

Paste your cube URL — CubeCobra or Cardivore-published — and go. ~30 seconds later you'll have a shareable report.

Run the simulator →