How Scoring Works

Every card in Star Wars Unlimited has a resource cost. The game's designers built a formula into the cost curve: a ground unit should have roughly 2 + (2 × cost) total stat points (power + HP). But in practice, competitive cards consistently trade stats for abilities — the higher the cost, the bigger the trade-off.

Rather than relying solely on the designer's formula, SWU Metrics uses an empirical cost curve derived from real tournament data. We compute the average power + HP of units that appear in competitive decks at each cost point, weighted by how often each card appears. This captures what a "normal" competitive unit actually looks like at each cost — including the ability tax that higher-cost cards pay.

Empirical Curve (Premier)

When tournament data is insufficient at a cost point (fewer than 50 deck appearances or 5 unique cards), the system falls back to the designer's linear formula. Upgrades always use 2 × cost and events use cost in RE.

RE stands for Resource Efficiency. It converts stat points into a normalized scale: 1 RE = 2 stat points. This lets us compare units, events, and upgrades on the same scale despite their different cost formulas.

Delta RE (Δ RE) is the gap between what a card actually delivers and what the cost curve says it should. A card with +1.5 RE gives you 3 more stat points than expected. A card at -1.0 RE is 2 stat points short.

Raw stats are only part of the picture. The AbilityParser reads every card's ability text and matches it against 294 effect patterns. Each pattern assigns an RE value based on the effect's power level.

Keywords like Sentinel, Ambush, Overwhelm, and Restore each have calibrated RE values. Raid and Shielded scale with their numeric values. The parser handles conditional triggers ("When Played", "When Defeated", "On Attack") and applies appropriate multipliers — a "When Played" effect is worth more than a "When Defeated" one because it's more reliable.

Stat Distribution

Not all stat totals are equal. A 5/1 unit has the same total as a 3/3, but the 5/1 dies to any 1-damage ping effect. The scorer applies a small stat skew penalty when the power/HP ratio is extreme (below 0.3). A 5/1 gets a ~0.34 RE penalty; a balanced 3/3 or 4/2 is unpenalized.

Context-Aware Patterns

The parser distinguishes effects that look similar but play differently:

• Capture friendly vs enemy — capturing your own unit (e.g., Escape Pod) is a downside, not removal
• Extra payment costs — "pay 2 resources to take control" deducts the extra cost from the effect's value
• Opponent-assigned vs you-choose damage — "an opponent distributes the damage" is worth less than surgical removal
• Indirect vs direct damage — indirect damage is valued at ~67% of direct, in the debuff tier
• Mass damage effects — "deal X damage to each unit" is scored in the mass removal tier, not as incremental damage

The final Theoretical Δ RE combines: stat delta − skew penalty + keyword value + ability value. This is the card's paper score before any tournament data.

Theory doesn't always match practice. The Adjusted Δ RE blends the theoretical score with real tournament performance data. This is where intangible value gets captured — synergies, meta positioning, and play patterns that no formula can fully measure from card text alone.

For each card, we pull: win rate (how often decks with this card win), meta share (what % of decks run it), average copies (how many copies per deck), and play pattern (mainboard vs sideboard split).

A card that looks average on paper but consistently wins tournaments gets a boost. A card that looks great on paper but underperforms in competitive play gets pulled down. The adjustment is conservative — tournament data nudges the score, it doesn't replace it. Cards with strong intangible value (difficult-to-quantify effects, meta-dependent synergies, game-state interactions) will gradually shift toward their true performance as tournament data accumulates.

How Adjusted Δ RE is calculated

The adjustment uses four components blended into a single score:

Meta weight ranges from 0.1 (no data) to 0.4 (100+ decks, high copy count, mostly mainboard). At low meta weight the score stays close to theoretical; at high meta weight tournament results drive the number. Play pattern RE adds a bonus for high-volume mainboard staples and penalizes cards that mostly sit in sideboards — a card played in 30% of decks at 2.3 copies mainboard earns a meaningful boost, while a card that's 80% sideboard takes a hit regardless of its theoretical score.

Card-level stats can only tell you so much. A card that looks mediocre in isolation might be a 3-of in every winning Boba Fett deck. The scoring engine accounts for this through deck composition analysis.

We import tournament deck lists and compute per-archetype play rates, co-occurrence patterns (which cards appear together), and synergy density — how tightly a card is connected to its most common partners. High synergy density means the card is part of a package (always played with specific partners). Low density means it's a flexible card that slots into many different builds.

The calibrator uses deck-weighted value as its training target: a formula that factors in play rate, average copies, leader win rate, and deck count. This means the system learns from what actually wins tournaments, not just what looks good on paper.

A card's value isn't fixed — it depends on what leader you're playing. Boba Fett makes indirect damage cards stronger. Palpatine makes events cheaper. Sabine rewards cheap, aggressive plays. The Leader-Adjusted Score shows how a card's theoretical value shifts when viewed through a specific leader's lens.

Each leader has a curated synergy profile that defines:

• Pattern multipliers — boost or reduce specific ability effects (e.g., Boba boosts indirect_damage by 1.75x)
• Trait bonuses — flat RE bonus for cards with matching traits (e.g., Mandalorian cards with Bo-Katan)
• Keyword bonuses — flat RE bonus for relevant keywords (e.g., Ambush with Sabine)
• Event cost discounts — leaders that reduce event costs (e.g., Palpatine)

On any card's detail page, you can select a leader from the dropdown to see how the score shifts. This helps with deckbuilding decisions — is this card worth including in your specific archetype?

The scoring engine runs a two-stage calibration against tournament data:

1. Tier multiplier fit — adjusts how much weight each category of effect gets (removal, card draw, stat buffs, synergy density). Uses coordinate descent to find the multipliers that best predict tournament performance.
2. Per-pattern delta fit — fine-tunes individual ability patterns. If "deal 2 damage to a unit" is consistently undervalued, the system learns a positive delta for that pattern.

Calibration is validated with cross-validation and gated on three metrics: Spearman correlation (predicted vs actual tournament value), top-decile precision (do the cards we rate highest actually see the most play?), and zero-meta exceptional (are we over-rating cards that see no tournament play?).

Current Spearman correlation: 0.464 (Premier) / 0.705 (Eternal).

No formula perfectly captures every card's value. Here's what the system handles well and where it falls short:

• Handled: stat efficiency, keyword value, 290+ ability patterns, trigger timing, conditional discounts, extra payment costs, friendly vs enemy effects, mass vs single-target, opponent-assigned damage
• Partially handled: synergy (via tournament adjustment and leader profiles), meta positioning (via play pattern RE), stat distribution (via skew penalty)
• Not captured: complex multi-card combos, bluff value, psychological effects, format-specific tempo advantages, cards whose value is primarily defensive positioning

The Adjusted Δ RE exists precisely because some value is intangible. As tournament data accumulates, cards with hidden strengths gradually shift toward their true performance. The theoretical score is a starting point — the adjusted score is where theory meets reality.