Cube Doctor · Cardivore

How it works

Survey, simulate, profile, ship.

Every recommendation is earned. The agent reads the metrics before it opens its mouth, names the cube's design contract before it draws conclusions, and re-drafts its own proposals before it ships them.

01

Survey

Static metrics first — colors, curve, rarity mix, removal density, fixing depth. The cheap pass before any drafting.

02

Profile

Detect what kind of cube this is. Arena (conjure / perpetual / spellbook). Vintage powered (Power 9 + fast mana). Storm. Reanimator. Lands-matter. Pauper. The framework that follows is profile-aware, not a one-size guild-pair template.

03

Simulate

500 × 8-seat drafts using adaptive, greedy, and synergy bots. Per-archetype DQI, color contestedness, pick-order timing — all returned as structured metrics the diagnosis cites by number.

04

Propose & verify

Any candidate change must address a measured weakness, clear an oracle-text claim validator, and re-pass the sim. The preservation graph blocks cuts that other parts of the report depend on. Zero changes ship when nothing meets the bar.

What you get

A real diagnosis, not vibes.

Here's an abbreviated preview of the shareable report. Yours will have your cube's name, its detected design contract, the specific lanes the simulator says are healthy or thin, and any cut candidates tied to scryfallIds so you can queue them up in CubeCobra.

Storm Cube · Example

448 cards · 500 drafts · 3m 12s

profile-aware

Format identity

Storm-cube identity: 7 rituals + 3 storm payoffs. Spells-matter ratios are MEANT to skew toward enablers; standard outlet:payoff math doesn't apply.

4.33

/ 5.00 DQI

All 10 archetype lanes simulating ≥ 0.85 — fast mana papers over coordination errors a non-powered environment would expose.

Archetype health

UR Spells

91%

BR Reanimator

88%

Tinker / Artifact

87%

Lands / Depths

86%

Aetherflux Storm

85%

Headline insight

SIGNATURE OBSERVATION

242/448 cards are CMC 0–2.

That's the structural signature of a storm cube: cheap throughput density above what any color-pair lane needs. Fast mana is hiding draft errors at 0.85+ playability across every lane.

Recommendations

VERDICT · NO CHANGES WARRANTED

3 cut candidates generated, 3 rejected.

A cut of Mana Vault was rejected: Mana Vault is a load-bearing dependency for the Aetherflux Reservoir combo line preserved in the brief. The diagnostic stands on its own — no refresh changes shipped.

REVIEW MANUALLY

Lore Seeker (rpl 3.0)

Below the cube's power band floor; flagged by the deadweight scout. Conjure / draft-matters cards in the same band are kept — they're part of the design contract.

What makes this different

Five things most cube tools don't do.

We've shipped these because a cross-provider AI reviewer panel kept flagging the same gaps in earlier reports. Each one is deterministic — no extra cost per run.

Profile detection

Names the cube's design contract before diagnosing. Arena cubes don't get judged by paper-Magic deadweight rules; Vintage cubes don't get scored against guild-pair lanes that don't exist.

Oracle-grounded claims

Every mechanical assertion in the prose ("this is a sac outlet", "lifelink trigger", "haste") is checked against the named card's actual oracle text. Hallucinations get dropped before shipping.

Full-cube gap verifier

When the agent claims "this lane lacks an X-style anchor", we scan the full cube for cards matching the predicate — if your cube has Pillory of the Sleepless and the agent missed it, the gap claim gets retracted.

Preservation dependency graph

If the report preserves Tolarian Academy because of artifact mana, the change engine refuses to cut Grim Monolith. Combo halves of preserved cards are protected by name.

Conservative change gate

When the verification sim is within noise and no finding clears the high-confidence bar, ZERO changes ship. The diagnostic stands on its own. Reviewers consistently rate "0 changes when justified" higher than "10 refresh cuts".

Honest disclosure

Generated 5 candidates, rejected 4? You see all 4 with the rejection reason. No prose contradicts the shipped change set. Walk-back findings are demoted to caveats automatically.

Under the hood

Transparent. Reproducible. Nerdy.

Cube folks don't trust black boxes. Every run carries its configuration. If you're skeptical, the sim is the same one the Cardivore app ships — we just point the agent at it.

Simulator

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

Policy mix

70% adaptive · 15% greedy · 15% synergy

Reasoning stack

A multi-agent pipeline: a main orchestrator picks tools, parallel sub-agents handle per-archetype signpost identification and narration, and a deterministic candidate ranker proposes cuts. Models are picked by capability, not brand.

Validators

45+ oracle-text claim rules · mid-rationale foreign-card check · preservation dependency graph · format-aware finding suppression · framework overlay (guild_pairs / combo_lines / set_tribute / tribal / mechanic_clusters / narrative)

Verification

Proposed changes re-simulated. The conservative gate drops the change set entirely if the DQI delta is within noise and no finding clears the high-confidence bar.

QA panel

A cross-provider AI reviewer panel grades the report corpus. We track the cross-panel mean across batches and only ship Doctor changes when measured quality holds. we tell you when we're not confident

Cost & caps

~$0.10–0.20 per run · 1 cube per IP per 24h · daily global cap so the marketing budget doesn't get drained in an afternoon

Your data

Cube list in, report out. Nothing retained beyond the public report page; reports are sharable by URL.

Run the Doctor

Diagnose your cube.

5-10 minutes. You'll get a shareable report URL when it's done. One free run per IP per 24 hours. Install the app for unlimited runs on your own cubes.

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

Cardivore is a full MTG companion — collection tracking, deck analysis, cube building, card scanning. The Cube Doctor is one of many AI tools inside.

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

Nerdy questions

FAQ for skeptics.

Why should I trust a bot to draft my cube?

You shouldn't, in isolation. The bots are three distinct policies that each draft badly in predictable, human-ish ways. Across 500 drafts that 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 Doctor reads the signal; it doesn't invent it — and every mechanical claim it makes is checked against the named card's actual oracle text before it ships.

What does "profile detection" actually do?

It looks at your card pool — names, oracle text, rarity, density of specific mechanics — and decides whether your cube is built on the standard 10-guild-pair scaffold or on a different design contract. Storm cubes get evaluated against ritual + payoff density, not standard outlet:payoff math. Vintage powered cubes don't get guild-pair lane gap warnings because drafters assemble around busted singletons, not lanes. Arena cubes don't get their conjure / perpetual / spellbook cards flagged as deadweight. Wrong framework is the single biggest reviewer-flagged failure of v3 reports; v4 fixes 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.

Why did I get zero changes?

Because none of the candidates the Doctor generated cleared the bar. Either the verification sim said the change was within noise, the candidates failed oracle-text validation, or they would have cut a card the rest of the report depends on. We'd rather ship "no changes warranted" with a substantive diagnosis than pad the package with refresh cuts. The rejected candidates are surfaced in the report so you can see what the Doctor considered and why each was rejected.

Will it suggest cards I already own?

The Doctor searches within your cube's rough power band, format legality, and detected profile. It flags candidate cuts from your cube and (when warranted) candidate adds from our card database. On Pauper cubes, suggestions stay common-rarity; on Arena cubes, suggestions stay Arena-legal. If the suggestions feel off-pitch, tighten your cube's archetype definitions in CubeCobra before running again.

Is this free forever, or is there a catch?

One free run per day per IP while the budget holds (we cap marketing spend at $50/day). Install the Cardivore app to run the Doctor on any cube you've loaded, unlimited.

What's the sim built on?

A TypeScript port of Cardivore's on-device Dart draft simulator. Uses a custom LCG RNG so runs are reproducible given the same seed.

Ready when you are.

Paste your cube URL — CubeCobra or Cardivore-published — and go. 5-10 minutes later you'll have a shareable diagnosis.

Diagnose my cube →