Practice Lab Math and Rules

Practice Lab numbers must be explainable. A player should be able to see whether a number came from exact probability math, a seeded simulation, or a simplified practice rule.

The Short Version

Practice Lab currently uses MTG-informed math, not AI guessing. Some values are exact counts from your deck. Some are Monte Carlo simulation estimates from repeated shuffled runs. Results from Run 100 simulations can move between batches because each batch uses fresh seeded shuffles.

A 100-run batch is useful for a quick deck-feel check. It is not final tournament truth. Exact odds should be labeled as exact. Simulation results should be labeled as estimates with the run count and seed.

Exact Odds vs Simulation Results

Exact odds use formulas. For opening-hand questions such as "what are my odds to see at least three lands," the correct tool is hypergeometric probability: the deck size, number of desired cards, and cards drawn determine the answer without shuffling samples.

Simulation results use repeated trials. Practice Lab shuffles the actual deck, draws hands, plays a simplified solo sequence, records what happened, and then reports percentages from those trials. If 72 of 100 simulated openings meet the keep rule, the result is 72% for that batch.

Current 100-Run Simulation Assumptions

The current Practice Lab trainer report uses a seeded Monte Carlo batch. For each run it shuffles the deck, draws an opener, then advances a simplified solo sequence for up to ten turns.

The practice sequence can play a land when available, count available mana, cast cards the simplified engine considers castable, and record stalls, dead cards, land drops, mana use, ramp, and removal availability.

The current sequence does not model an opponent, combat, sideboarding, the stack, priority passes, replacement effects, all card-specific rules text, or optimal strategic play.

What The Metrics Mean

Opening keep rate is the percentage of simulated opening hands that met the current keepable land window. Today that window is a simplified land-count rule, not a full London mulligan strategy.

Land drop by turn 3 is the percentage of runs where the simulation made the land drop on turn 3.

Average mana turn 4 is the average available mana recorded on turn 4 across the simulated runs.

Stalled turns are turns where the simplified sequence neither played a land nor cast a spell.

Dead-card rate measures nonland cards left in hand at the end of the test window that were never seen as castable by the simplified sequence.

Flood and screw are opening-hand signals. Flood means the opener had six or more lands. Screw means the opener had one or fewer lands.

Early ramp and early removal are role checks. They count simulated opening hands that included cards tagged as ramp or removal.

MTG Rules Compliance Status

Practice Lab must never imply rules coverage it does not have. The current trainer report is rules-aware for basic deck, land, mana, and casting flow, but it is not a full Comprehensive Rules engine.

A result is safe to use as a Practice Lab estimate when it depends on shuffled hands, land count, mana availability, simplified casting, or card role tags.

A result should not be called fully rules-compliant until it is owned by the ScryDuel rules engine, runs through legal-action checks, records accepted or rejected actions, and can replay from seed with the expected state and projection.

How To Read The Report

Use Exact Odds when you need formula-backed probability. Use Simulation Results when you want to know how the deck behaves across many practice starts.

If a number looks surprising, run a larger batch later, inspect the assumptions, and compare it with exact odds where an exact formula exists.

Last reviewed: 2026-06-29.