---
title: "Deck guide | Moenotes"
description: "What the deck page optimizes, what “proven optimal” means and how the search proves it."
canonical: "https://bdon.moe/de/tools/deck/guide/"
markdown_url: "https://bdon.moe/de/tools/deck/guide.md"
locale: "de-DE"
data_version: "en:1a6e354e47c1a5e0e89d4d91bf21dbd7,hk-tw-mo:d7fe53227ffb35badd4e1e7f29e1f789,jp:1.1.0.600/6c240b61b49f61ead4b9f8769fdb68a9,kr:44710a0fa3738ae089ed8740d79a9927"
---

# Goals, proofs of optimality and search strategy

The deck page finds the 5 teams in your collection with the highest goal value. This guide explains what it optimizes, what “proven optimal” means exactly, and how the search proves it without trying every team.

## What is optimized

The CP goal defaults to CP only. Its optional secondary priorities compare expected CP first, then expected PT and badges in the chosen order. This option uses Average performance and applies to ordinary Live modes and ordinary Skip. Challenge Live and Challenge Skip spend CP. Badge resource and reward-data availability are required. The engine supports this combined objective in its lottery-free skill domain and reports unsupported skill conditions explicitly.

Event badges use the current event’s badge item. For each possible live result, the engine selects the single reward row for that exact result grade, then calculates the badge amount for the chosen consumption. Average performance averages these amounts; Theoretical maximum takes the highest reachable amount. Normal and challenge scenes each use their own event reward group. Each grade must contain exactly one reward row with probability marker 10000; other reward shapes are unsupported.

Alongside live events, Simulated Event Bonuses let you define custom attribute, character, band or card-specific member PT and Snap badge bonus rules before official event data is published, so you can plan formations and resource budgets ahead of time.

Each goal is a number the search makes as large as possible: for Gekisou Live, Gekisou Arena (Arena Live) and Free Live, the score on the chosen song and difficulty; for event points, the event points per live; for highest power, the team's power. Free and Challenge Live, including their event payoff goals, offer Average performance or Theoretical maximum. Gekisou modes use Average performance and let you specify section placements (1st–5th). Power and skipped lives have a single value, so this choice is not shown.

Multi-song optimization and rule overrides: Gekisou Arena loads active schedule rules (allowed band, song attribute and three Gekiso missions) automatically and supports switching between One shared formation (proving the single formation with the highest expected mean score across the selected song pool) and Per-song optimal. All-song searches in Free and Mission Live likewise offer per-song optimal or shared-formation mean optimization, plus optional manual overrides for song attribute and the three Gekiso mission types.

Average performance (default): at the start of each live the members' skill order is shuffled. We treat the 120 orders of the five members as equally likely, simulate a whole live for each order (skills, Snap effects and, in Gekisou Live, the gekisou sections and rank bonuses) and average them. Where a skill or mission draws a lottery, the draw is averaged with the game's probabilities.

Theoretical maximum ranks Free and Challenge Live teams by the highest goal value reachable across skill orders and random skill outcomes with nonzero probability. Switching keeps your collection, song, difficulty and declared play conditions. Only a complete search proves the maximum. Reaching the value requires the corresponding random outcomes and is not guaranteed every live; the option is enabled only when the engine supports the current conditions.

Play: by default the play is the theoretical best, all Perfect, with every Just-eligible note in Gekisou Live hit Just. Under “Play assumptions” you can set Great %, Just % or a periodic Miss interval (every N notes), and in Gekisou modes specify a steady-state JUST window lower bound (±ms) to filter out formations that rely on ultra-tight windows. It describes a declared play, not a prediction of how you will play.

A team is a leader, 4 other members and the Snap paired with each. Swapping the positions of the 4 members does not change the score, so the positions in a result are only for display.

## Candidates

The candidates D are every legal team in the active Box mode: those allowed by the game's formation rules (including Arena band regulations) and by your leader lock, required/excluded members or Snaps, No-Snaps toggle and explicit member–Snap bindings under “Card restrictions”. Each choice of leader, members and Snap pairing is a separate candidate.

You can switch with one click between My Card Box (supporting manual entry, account sync and screen-recording video recognition import) and Full-Catalog Max-Practice Theoretical Mode (evaluating every catalog card at max level, trained, Rank 5 and Skill Lv.5 to inspect theoretical ceilings and gacha targets). Missing values in My Card Box are listed for you to fill in first; when a result does not cover every card you own, the page says so.

## Definition of “proven optimal”

For CP with secondary priorities, first restrict D to teams with the globally highest expected CP. Within that tier, rank by the selected secondary reward order, then the usual tie-breaks. Return up to 5 teams from that tier, even when fewer than 5 qualify. A timeout leaves this result unproven.

“Proven optimal” means: among the complete set D of legal candidates, the results listed are exactly the first min(K, |D|) teams ordered by goal value (descending), then power (descending), then canonical ID order, where K is the number of teams the page returns (5). Team choice, ranks and the order of ties are all settled.

The proof applies to the selected objective under the declared play conditions: the probability-weighted average for Average performance, or the highest reachable value for Theoretical maximum. It does not turn a possible maximum into a guaranteed outcome.

The proof does not evaluate every team. Each candidate has either been compared exactly with the results or been ruled out by a sound upper bound: the best value it can reach is already below the exact value of the K-th team. This is neither sampling nor an approximation.

If D is empty, “proven” means no team is feasible under these conditions.

This is an algorithmic proof relative to the declared scoring model and relies on the model, the bounds and the program being correct. Cards you have not entered, wrong progress values and mistakes in an actual play are outside the computation.

## “Not proven” and the best so far

“Not proven” means the exact search has not finished proving optimality; the results so far may or may not be optimal.

This happens when the search reaches the time limit you chose or you stop it. The page lists the best teams found so far, each computed exactly, together with two numbers: the best so far and the value the optimum is at most. No team exceeds the second, so the true optimum lies between the two; the smaller the gap, the closer the result is to the optimum.

To get a proof, choose a longer time limit or “No limit” and run again.

## Search strategy

The number of teams grows very fast with the collection, so simulating each one is not feasible. The search is a branch and bound: it builds candidates step by step (the leader first, then members and Snaps) and bounds the best value each partial team can still reach. Once that bound is below the exact value of the current K-th team, no team in that branch can enter the results and the whole branch is pruned.

The bounds come in two layers. Partial teams are bounded from precomputed tables, which is fast and only ever overestimates; only complete five-member teams are simulated over a whole live. In Average performance, the 120 orders and lottery draws are first enclosed in intervals and refined to exact values only when the intervals cannot separate the ranks.

The higher the cutoff, the more is pruned. The search finds strong teams early and computes them exactly to set the cutoff, then raises it each time it finds a better team.

Candidates are split by leader into disjoint groups that are searched separately and merged. The split does not change the result: the same input always gives the same teams and ranks.

At the time limit the search stops and reports the best teams so far and an upper bound over the remaining candidates, the two numbers shown with “not proven”.

## Reading a result

A CP result with secondary priorities retains the sorting order used for that calculation and shows expected CP, PT and badges with their exact fractions. Changing the current input marks it as a previous result until recalculation.

Each result shows the goal value, the power, the formation in the game's own style, and for each position the member card, its progress and the paired Snap. Shared-formation runs additionally list each pooled song's power, mean score and min–max order span.

A goal value is sometimes shown as an interval that contains the exact value, for example while a skill with a lottery has not been refined yet.

When an Average performance calculation finishes, each team lists the orders with the lowest, median and highest score among the 120. A Theoretical maximum result shows a skill order for the highest selected goal value; the corresponding random skill outcomes are still required. The objective label belongs to that calculation. Changing the objective marks it as a previous result until you recalculate.

Best timing ceiling and early-hit notes: when an ordinary Live's optimal skill order gains points from hitting cumulative-Snap step-completing notes one 60 Hz frame (~17 ms, still PERFECT) early, the skill-order card shows the Best timing ceiling, a visual counting-window strip and a per-step breakdown of timestamps, chord counts and score gains; hit the first notes right after activation as usual.

Single-card upgrade impact ranking: in My Card Box mode, open a team's upgrade evaluation to simulate exact goal gains for each unmaxed member and Snap under Max Level, Skill Lv.5, and Full Max (Rank 5 + Skill Lv.5), ranked from highest to lowest return.

## Source code

The Deck Builder runs on the Haneoka branch-and-bound WASM engine (public in haneoka-gakuen/ournotes-team-builder on GitHub, paired with MoeNotes' exact event-reward patch), with earlier reference components also public in empty-sekai/ournotes-deck.
