Every battle score on this site is computed at specific IVs on both sides. That makes any single number - a battle rating, a W-L record, a “wins the 1-1” claim - a point estimate: true for the two exact spreads that were simmed, and silent about what happens when either side’s IVs move. IV robustness is the follow-up question: does this outcome survive the IVs? It splits into the half you can’t control (which spread did the opponent happen to roll?) and the half you can (exactly which of your spreads win this fight?).
This guide explains the machinery behind every robustness surface on the site - the per-opponent robustness numbers in the dives, and the matrix / cheat-sheet / per-pair pages built for Worlds 2026. The worked examples below are a 2026-08 snapshot of the Worlds pages; the method is general and will show up on future surfaces (limited cups, post-rebalance re-checks) in the same shapes.
For a given pair of Pokemon in a given shield scenario, checking every competitive IV combination on both sides can only end three ways:
Most matchups are safe or dead. The robustness surfaces exist to find the IV-decided minority and then show you where the line falls.
The classic dives have carried one for a while: the per-opponent robustness figure asks “do we beat this opponent regardless of which good IV it rolled?” - a win is only credited if your spread beats the opponent across its whole competitive IV cohort (top 512 spreads by stat product), per shield scenario. An exact 500-500 tie counts as not beaten. That is the conservative direction: it protects you from the opponent’s roll, which you can’t see and can’t control.
The Worlds-style pages flip the lens around as well: for one opponent spread (or its whole cohort), which of your 4096 spreads win? Both directions come from the same underlying data.
The robustness data is a set of outcome planes: a win/loss bit for every (your spread) x (their spread) x (shield scenario) combination, simmed exhaustively. Nothing is sampled or interpolated - when a page says “192 of the top-512 spreads win this”, that is a count over actual battles, and hovering a cell shows exact spread counts and score margins. Bait strategy changes outcomes, so planes are baked for bait ON and bait OFF separately and shown side by side. (The bait toggle applies to the focal side only; in the “no-bait” line the opponent still baits, which is why the two directions of a no-bait pair describe different battles and need not mirror each other.)
“Every IV combination” needs a competitive cut, and one cut is not enough:
On the focal side, summary tables use two probe spreads - the rank-1 stat-product spread and the max-attack spread within the top-512 - as representative builds. Probes are a screen, not a proof: a pair can be “IV-decided in probe slices only”, and the per-pair detail pages recompute everything from the full 4096-spread grid rather than trusting the probes.
Each pair gets a 3x3 shield grid in PvPoke’s battle-matrix layout: rows are your shields (0/1/2 top to bottom), columns are the opponent’s, so “2-1” means you keep two shields and they keep one. Each cell holds one letter:
The hub matrix is the same information as color only; its links open the text-carrying cheat sheets, so nothing depends on color alone.
For an IV-decided cell, the next question is how the outcome varies across your spreads. The per-pair pages plot the share of the opponent’s cohort you beat as a function of your stat-product rank, in bands of 16 spreads. A flat curve means your IVs barely matter; a cliff means a hard cutoff - some stat crosses a damage tier - and the page names the boundary. As a snapshot example, the Aegislash (Shield) vs Altaria page’s 2-2 reads: 0 of Aegislash’s top-512 spreads beat every top-512 Altaria, 275 beat none, and the bands in between climb from 0.4% to about 6% of Altarias beaten - IV-decided, but shallowly: no Aegislash spread makes the matchup safe.
Damage in this game is stepwise: a fast or charged move does a fixed integer of damage until a stat threshold flips it to the next integer. That makes the IV question exactly solvable - no simulation needed for this part:
The pages compute these cutoffs in closed form and print them as ladders (“attack >= X reaches the 4-damage Snarl”). One honesty subtlety the tables keep straight: “your spread beats their particular spread” and “your spread is guaranteed against their whole cohort” are different quantities - a cutoff that wins the per-spread fight can still lose to a bulkier roll from the same cohort. Ladder rows are labeled with which claim they make. The IV explorer runs the same closed-form ladders in your browser: enter IVs and a level, and it lists which cutoffs you reach or hold against every meta entry.
Robustness analysis is exactly the place where a pipeline can quietly overclaim, so the shipped pages carry their own checks:
The Worlds pages are pinned to the pre-Worlds game data (Worlds runs on the older battle system), but nothing in the method is Worlds-specific: wherever a matchup is close enough that a roll of the dice on either side’s IVs could flip it, this is the machinery that finds the line and tells you which side of it your Pokemon is on.
Battle engine is a Python port of PvPoke; all game data from PvPoke by Empoleon_Dynamite (MIT license). This project would not exist without it.
Part of the PvP dive site. Explainers regenerate from current dive data every publish, so numbers stay in sync with the methodology. Last regenerated 2026-09-28.