MentalisDraft Sense
← Back

FAQ

Common questions about Draft Sense

What is Draft Sense?

Draft Sense is a real-time draft assistant for Magic: The Gathering – Arena. It watches Arena’s log files while you draft and shows you a recommended pick for each pack, plus a running summary of your pool (colors, curve, creature count, removal density).

How does it work?

The overlay sits on top of Arena and reads the same draft log Arena writes to disk. When a new pack appears, Draft Sense sends the pack contents and your current pool to the backend, which runs them through a graph neural network trained on public draft data from 17Lands. The model returns a ranked recommendation and per-card context scores; the overlay renders them. End-to-end this usually takes under a second once the backend is warm.

How do I enable detailed logs in Arena?

Draft Sense reads MTG Arena’s log files, and the relevant log isn’t written unless you turn on Arena’s detailed-logs setting:

  1. Open MTG Arena.
  2. Click the gear icon (Settings) in the top-right.
  3. Go to the Account tab.
  4. Check Detailed Logs (Plugin Support).
  5. Restart Arena.

This is a one-time setting that persists between Arena launches. It’s the same mechanism 17Lands and most other log-watching tools use.

I installed Draft Sense but no recommendations show up. What’s wrong?

Nine times out of ten this is the detailed-logs setting being off — see the question above. Open MTG Arena → Settings (gear icon) → Account → check Detailed Logs (Plugin Support)→ restart Arena. The next draft should populate Draft Sense immediately.

If that’s already on and you’re still seeing nothing, drop into Discord’s #install-help channel with a screenshot — usually a quick fix.

Is it safe to use? Will I get banned?

Draft Sense doesn’t interact with Arena. No clicks, no inputs, no game-memory access, no calls to Wizards’ servers — it just reads log files Arena writes locally. Same approach as 17Lands, which Wizards has publicly acknowledged.

Which sets are supported?

Draft Sense currently makes recommendations for 32 sets:

(Live list temporarily unavailable; showing the most recent known set list. Status will refresh once the backend is reachable.)

Which draft formats are supported?

Premier Draft, Traditional Draft, Quick Draft, and Pick-Two Draft. For Pick-Two specifically, the backend runs a second inference conditioned on the first pick so the two recommendations make sense together — you won’t get two unrelated splashable bombs.

Does it work for Sealed?

Not currently. Draft is the focus while we keep the model quality high in one format.

Does it work on Mac or Linux?

Windows 10/11 only. Arena itself is Windows-first and almost all early users are there; Mac and Linux builds aren’t on the roadmap.

How does pricing work?

Why does Windows show a warning when I install?

Windows SmartScreen flags the installer as “unrecognized” until enough people download a given version. The installer is signed by us (Joshua Brehm, via Azure Trusted Signing) and runs an independent VirusTotal scan on every release. The install page walks through how to verify the signature and get past the dialog.

What data do you collect?

By default, nothing beyond the minimum needed to serve recommendations (your Clerk-managed sign-in, your draft requests, billing state). If you opt into Share gameplay data on the account page, your draft picks (pack contents, what you picked, what we recommended, final deck) go into the anonymized training corpus we use to retrain the model for future sets. The toggle is off by default; you can flip it at any time, and turning it off stops future collection. See the privacy policy for specifics.

How accurate is it?

A graph neural network trained on millions of real Premier Draft picks from 17Lands. It’s built to model the pool synergy and color signaling that naive heuristics (like always taking the highest win-rate card) miss. Treat it as a strong second opinion, not a verdict — you know your draft history, archetype preferences, and pod tendencies better than any model.

The model recommended X but Y was clearly better. What happened?

Either the model genuinely disagrees with you (sometimes you’ll be right, sometimes it will) or the situation was thin on training data (rare archetype, new set in the first few days). Paste it in Discord’s #alpha-feedback either way — these reports shape the next model version.

How often is the model updated?

The model is retrained as more 17Lands data accumulates and as new sets release. New sets typically come online within 24–48 hours of release — served day-one by the cross-set model, then sharpened as that set’s own data arrives.

How do I update Draft Sense?

On Arena quit, the overlay checks for a new release and updates silently before its next launch. You don’t need to download anything manually unless you’ve turned auto-update off.

Does it work offline?

No. The model runs server-side (the checkpoint is big and the inference is meaningful work), so the overlay needs internet for both pick and deck-build recommendations. Pool tracking still works offline, but you’ll see “network error” on recommendation requests.

How do I report a bug or request a feature?

Everything happens in Discord: #bug-reports for bugs, #feature-requests for ideas, #alpha-feedbackfor “the model picked wrong” reports. The maintainer reads everything. You can also report a bug straight from the overlay — the Report a problem button in the header sends your description plus a recent diagnostic log to support, no Discord account needed.

What’s the alpha tester program?

Unlimited free access for active drafters who give honest feedback. Full details on the alpha tester page.

Still have a question?

Ask in Discord. If it comes up enough, it lands here.