Every score should show its work.

01 · FOUNDATION

The executable component ledger.

The foundation prior is inspired by the inspectable, trait-based approach documented in Decks of KeyForge AERC and SAS methodology, but RDS is an independent Riftbound model with different rules, components, coefficients, and grades. It converts printed Might and expected rules-text outcomes into one value scale, then subtracts the full Energy and colored-Power burden. Power is not free: recycling matching Runes also removes future Energy tempo. Popularity and collection ownership never change this stable card grade.

Power burden = 0.55 × (1.15×Power + 0.25×Power×(Power−1))
Energy burden = 0.55 × Energy

Foundation = clamp(1.5, 9.8,
  type baseline
  + printed strength
  + trigger-activated expected strength
  + effect value + card advantage
  + board control + battlefield control + resource value
  + flexibility + role utility + reliability
  − Energy burden − Power burden − high-cost risk
  − explicit costs and drawbacks
)

Each rules-text effect first enters its own bounded component and then receives a trigger activation prior based on its real timing window; play effects are reliable while Attack, Defend, Conquer, Hold, and Deathknell effects are conditional. These are declared design priors, not match-data claims. Inside a deck, actual enabler density, opening-hand overlap, Rune reliability, and curve position replace generic optimism. Time Warp is expensive alone, but gains contextual value beside acceleration and a Hold/scoring engine that can convert the extra ready-and-score window.

Requirements change realized value.

Conditional cards are displayed as standalone floor → this deck → supported ceiling. Equipment requirements are strict: generic Gear never counts as Equipment. The deck-context activation also checks actual compatible copies, a Unit carrier when attachment is required, printed Energy ceilings, and whether kill or sacrifice payoffs compete for a limited supply. Detaching keeps the Equipment in play, so repeated detach lines pay re-equip timing and resource friction rather than pretending the card was consumed. Access is estimated without replacement from the drawable library, so adding support improves reliability with diminishing returns instead of switching a payoff on at an arbitrary threshold.

02 · COMBINATIONS

Pairs and three-card packages remain explainable.

Pair affinity requires complementary traits, an explicit tag/card reference, or another reviewed mechanical connection. A shared region or legal domain is not treated as a combo. Three-card experiments are derived as 65% of their three pair edges plus a bonus only when the three cards close a real setup → resource → payoff chain. Resource collision, target competition, redundant top-end, and missing setup remain negative. In Deck Lab, every eligible completion is evaluated against the selected deck on the user’s device, including quantity-permitted repeated-copy pair and three-card multisets.

• Synergy uses opening/early-draw reliability and diminishing returns.
• Antisynergy covers orphan engines, dead payoffs, curve/role gaps, competing trash consumers, early colored-Power collisions, and Rune demand.
• The curve prior follows Riot’s guidance: at least nine 2–4 Energy units and six interactive spells/cards in a 40-card starting point, while limiting 6+ Energy saturation.
• Legend, Chosen Champion, Battlefield, copy-limit, ban, and domain rules remain separate from card strength.

03 · DECK ENGINEERING

Strategy first, quantities second.

The generator starts with a Legend/Champion plan, win condition, and executable enabler → payoff packages. Curve and card-type targets support that plan; they do not replace it. The Chosen Champion is set aside, so normal access estimates use a 39-card drawable library. Opening, partial-mulligan, and early-draw probabilities are shown as model assumptions rather than tournament outcomes.

Three copies are the default for must-see core cards, reliable early plays, signature cards, and combo pieces without redundant access. Two copies suit conditional setup, redundant effects, and expensive cards that remain useful in multiples. A one-off is an exception with a reason: searchable/tutored access, a narrow finisher or matchup tool, or a high-cost effect the plan cannot afford to draw twice. Current Core Rules make a named Unique card a mandatory one-copy case, not a strategic preference. Singleton penalties disappear when one of those explanations is mechanically supported. After structural quotas are met, the generator folds discretionary one-offs above four into a higher-utility card with the same curve, type, and interaction footprint; the requested anchor, Chosen Champion, and last Legend-identity link remain protected.

Riot’s official primer supplies the stable 40-card, early-unit, interaction, unit-presence, and Rune/Power starting constraints. The Mobalytics deckbuilding fundamentals are labeled as a community heuristic: seven to nine zero-Power 2-drops, three copies for cards a deck wants every game, and fewer copies for situational cards. Token engines and other reliable board generation can satisfy unit presence without blindly forcing the same raw Unit ratio on every archetype.

04 · DECK LEDGER

Foundation + synergy − antisynergy + fit + meta.

A deck’s ledger exposes each component and links every named card to its permanent card page. Card-by-card rows apportion the deck result back to the selected copies and list their strongest connections. Candidate recommendations compare the deck before and after adding a card, so “best fit now” can differ from “highest standalone grade.”

Candidate fit is decomposed into 30% foundation, 25% pair/triple package, 20% curve repair, 10% missing role, 10% Rune reliability, and 5% Legend/Champion/Battlefield context. Owned, not-owned, and unrestricted are hard filters rather than hidden score bonuses.

The final deck ledger keeps Foundation, Synergy, Antisynergy, Compatibility, Deck Structure, and bounded Meta as separate conserved components. The visible total is their signed sum; no hidden server-side score replaces it.

05 · TRUST BOUNDARIES

Browser-side theorycraft, versioned evidence.

Interactive scoring runs in the browser rather than consuming a scoring server. Deterministic shortlist, pair, and triple work uses the visitor’s CPU; card surfaces use compositor-friendly transforms and lazy rendering so the GPU handles presentation rather than pretending rules-text evaluation is a graphics workload. Public meta values stay neutral unless a complete, approved, version-matched calibration snapshot passes coverage and coefficient caps. Raw play rates, win rates, matchups, and source decklists are not published by this app.