07 Game Balance Check

07 — Check calculated balance

Follow the protocol, including its source-loading rules. Inputs through profile.sources: baseline, design goals, changed systems/data, registry 05, and existing calculations; after 07b, include routes, encounters, boards, or generation rules. Without a narrower assignment, review applicable quantitative systems across the agreed product.

If quantitative opposition, resources, randomness, and progression are genuinely absent, document applicability with a source and reopening condition. Do not add an economy or numerical difficulty scale for this stage. Puzzle comprehension and literary pacing may require observation rather than an invented formula.

1. Record target properties and assumptions before calculating. Build a traceable model with each input's source, formula, units, rounding, and scope. Separate accepted invariants, unverified settings, and derived results. Return missing required outcomes to 06; sensitivity analysis based on an explicit hypothesis does not close that gap.
2. For opposition, compare applicable costs, effects, intervals, range, risk, information, and restrictions. Combat may involve damage, defense, and time to victory; strategy, resource exchange, tempo, and position; competitive play, turn order, starting conditions, and permitted asymmetry. Check combinations, infinite loops, and strategies without tradeoffs. Perfect execution does not prove practical playability, and symmetry does not guarantee interest.
3. For resources, list sources, sinks, conversions, capacity, and ownership. Distinguish one-time from repeatable events and personal from shared resources. Check duplicate payouts, free reversible conversion, unrecoverable shortages of required resources, and access to the selected recovery method. Do not introduce a currency sink just to remove permitted surplus in a finite game.
4. For progression and probabilities, check early, experienced, and extreme states, skipped optional rewards, retries, and deviations from the expected path. For random rewards, calculate material spread or tail outcomes and guarantee conditions as well as averages. Account for dependent samples; multiplying independent probabilities does not model correlated events.
5. After 07b, connect critical sequences to node/turn/encounter IDs: entry state → stepwise costs, replenishment, and access → exit → recovery. Check all required authored routes and material procedural route/start classes in the profile. A later resource cannot fund an earlier action; sufficient totals do not prove reachability. A permitted bypass does not make an encounter mandatory. Keep this pass pending until inputs exist, then complete it before 07c wherever resources affect space.
6. Check sensitivity to material assumptions: skill, time, randomness, participant configuration, route choice, and applicable constraints. Recommendations need before/after scenarios, parameters, and side effects. Distinguish numerical dominance from a hypothesis about player behavior. Do not change a product goal merely because it is harder to calculate.

When action results change future opportunities, income, or difficulty, analyze dynamic feedback loops. Record the causal chain, each link's reinforcing or balancing effect, delays, limits, and intended outcome. Carry state through repeated rounds, sessions, or another meaningful period from different permitted starts, including advantage and disadvantage. Check compounding advantage, worsening conditions after failure, oscillation, and intended recovery, especially across multiple reinforcing systems. A document dependency cycle or the absence of infinite conversions does not establish this. Distinguish intentional advantage and permitted accumulation from defects; extra sinks, catch-up assistance, and equal outcomes are not automatic requirements.

For selected events, seasonal progression, reward chains, and player-group analysis, use the applicable service contracts: late entry, missed periods, reward receipt/expiry/carryover, and cohort/period definitions. These clarify accepted mechanics without automatically adding live operations, analytics, or compensation.

Output: balance-review.md in profile.paths.state_dir, plus reproducible calculations in the existing canonical location linked by the profile. Include inputs, assumptions, recalculation method, scenarios, conclusions, and future measurements. A local spreadsheet or small calculation script may verify formulas; it does not authorize building a game prototype.

Completion: an independent Codex subagent recalculates material conclusions from primary data under the subagent review protocol, and open gameplay hypotheses have verification methods. Route fixes to 06, then repeat 05, affected calculations, and 04. The coordinator owns shared logs. Paper results do not establish fun, competitive fairness, or a successful playtest.