Matchup analysis → Method

How this analysis is made

You cannot judge an opinion you cannot inspect. This page explains exactly how every number on this site is produced — and where it is weak.

In one paragraph

Each matchup is analysed by decomposing both primary mission cards into their scoring engines — how much VP comes from holding objectives, how much from tempo actions, how much from end-of-battle bonuses — and then asking which engine is easier to run against the other. The comparison produces an edge score out of 20 where 10 is dead even. Every score is published with the one-sentence reason behind it, a confidence level and a basis, so a reader can reject one judgement without discarding the rest.

The 0-20 scale

A five-point scale forces false precision at the extremes and mush in the middle. Twenty points leaves room to say "slightly favoured" and "clearly favoured" without pretending they are the same thing.

Metric	Type	Meaning
`edge`	Comparative, for side 0	10 = dead even · 11-12 slight · 13-15 clear · 16+ strong. Side 1's edge is 20 minus the value.
`ceiling`	Absolute, per side	What the plan produces when it works.
`floor`	Absolute, per side	What still lands when the plan is disrupted.
`tempo`	Descriptive, per side	Low = scores at the start of your turn, so holds must survive the opponent's turn. High = scores at the end of your turn, so late grabs pay immediately. Neither is better; they demand different armies.
`complexity`	Absolute, per side	Execution and decision load — how easy the plan is to misplay.

How an edge is derived

The analysis starts from structure, not from impressions. For each card we separate three things:

The hold engine — the recurring line that pays for controlling objectives. The decisive question is whether it scales per objective or pays a flat rate for holding "one or more". A per-objective engine rewards board dominance; a flat one makes the second and third objective worth nothing, which changes the whole game plan.
The scoring checkpoint — whether that line is checked at the start of your turn or at the end of it. Scoring at your Command phase means you bank what survived the opponent's turn, which rewards durability. Scoring at the end of your turn means a late grab pays immediately, which rewards burst and reserves.
The conditional lines — kills, actions, markers, raids on the opponent's home objective. These are rated by how easy they are to earn against that specific opponent, which is where most of the asymmetry between two cards actually lives.

The edge is then the answer to a single question: whose engine is easier to run against the other's, once each player is playing well? A card that scores off an action the opponent is directly rewarded for preventing is worth less than its raw numbers suggest. A card whose scoring condition the opponent controls outright — rather than merely contests — is discounted hard.

Confidence and basis

Two separate labels travel with every score.

Basis — theory (derived from card structure), simulation (backed by seeded engine runs), meta (observed results), playtest. Most current ratings are theory; that is stated, not hidden.
Confidence — low, medium, high. A rating is marked low when it depends on information we do not hold, and the reason is written into the rationale.

Where this analysis is weakest

Army lists are not modelled. Every rating assumes generically competent armies. A matchup rated 13/20 can invert with the right list.
Secondary missions are excluded. These notes cover primary scoring only, which is a large but partial share of the game.
Some card mechanics sit on the physical card reverse — marker placement and removal in particular. Where a rating leans on one of those, it is marked low confidence and says so.
Theory is not results. These are structural readings, not win-rate measurements. The intended next step is to raise ratings from theory to simulation by running seeded games: the underlying game engine is deterministic, so any claim tested that way is reproducible by anyone with the seed.

Disagreeing usefully

Because every judgement is atomic — one score, one sentence of reasoning, one confidence label — the useful form of disagreement is specific: name the rating and say which part of the reasoning fails. That is far more actionable than disputing a conclusion, and it is exactly why the numbers are published alongside their rationale rather than as a bare tier list.