Up front: this is a validation check on the Battle Lab simulator, comparing it against 8 weeks of real tournament data. Battle Lab is a rules engine that plays both decks out with a deterministic in-house policy (no LLMs, no machine learning, just a play algorithm I wrote), so the question is always: does what it says line up with what actually happens? Here's one way of measuring that.
What Battle Lab is for: matchup testing. Real win-rate data shows what happened in tournaments and games. Battle Lab lets you test any archetype against any other archetype directly — including matchups that may not have shown up enough in real events to draw a conclusion from.
The check: as a sanity test, I compared Battle Lab's archetype ranking against the real win-rate ranking from 8 weeks of Meta Monday data (MSA), using Spearman rank correlation — a measure of how well two orderings agree:
- 1.0 = perfect ordering (the sim ranks the archetypes in exactly the real order)
- 0 = random ordering (no relationship)
Result: average ρ = 0.54. Battle Lab was positively correlated with real results across all 8 weeks — not one week where it pointed the wrong way. To me, that suggests the simulator is capturing a real signal while also letting you explore specific matchups on demand.
Takeaway
Meta Monday tells you what happened. Battle Lab lets you test what you want to explore next. The two agree often enough (ρ = 0.54, positive every week) that when the sim ranks a matchup, it's worth listening to — and when a matchup hasn't happened enough in real events, the sim is the only read you've got.
Try it here: gundambay.com/battle-lab. Happy to share the method or rerun anything if you think something's set up wrong.