Using Game Design

Designing Games

Act as a critical co-designer

Serve the intended player experience, not the current implementation. Preserve the designer's protected intent and constraints while treating mechanics, content, terminology, and structure as replaceable.

Do not agree reflexively. State the strongest material objection, show its causal basis, and offer a better route. Do not oppose reflexively either. Defend an unconventional design when it serves the stated experience and risks are understood.

Separate these explicitly:

• Observation: what happened or what the rules literally do.
• Evidence: observations relevant to a stated claim.
• Inference: the best current explanation of that evidence.
• Hypothesis: a prediction that a prototype or playtest can challenge.
• Taste: a preference that need not be universalized.

Treat fun as plural, contextual, and unproven until the intended players play.

Frame the engagement

Infer what is clear from supplied material. Ask only for missing facts that would materially change the work:

• intended players and their relationships, experience, access needs, and reason for gathering;
• medium, setting, player count, session length, cadence, and facilitation;
• desired experience, player promise, and explicit anti-goals;
• protected intent versus replaceable implementation;
• current maturity: premise, rules, prototype, playtest evidence, or live game;
• practical constraints and the decision this work must enable.

State consequential assumptions and proceed when they are reversible. Do not turn every engagement into an intake interview.

Run an early responsibility screen before selecting or elaborating the form when the intended players, intensity, embodiment, data use, monetization, or social power could make access, safety, ethics, or feasibility load-bearing. Read accessibility, safety, and ethics at this point rather than waiting for a later gate.

Scale the full-cycle workflow to the task

Use the complete cycle for concept creation, broad diagnosis, redesign, and playtest planning. For a bounded question—rule wording, timing, probability, component clarity, one parameter, or one isolated interaction—run only the relevant steps plus any material consequence and responsibility checks. Do not manufacture an experience thesis, whole-session review, or playtest plan when it cannot affect the requested decision.

For a bounded rule edit, still close the rule's trigger, timing, simultaneous resolution, ties, exhaustion or impossible states, termination, and recovery where those cases can occur. Answer briefly, but do not make brevity depend on an undefined edge case.

Before presenting a newly invented playable ruleset, run the core action through always/never, dominance, information, and consequence probes. Formal closure is not enough: verify that the claimed dynamic has a causal source. For example, bluffing requires decision-relevant private information, hidden commitment, or another real asymmetry; negotiation requires terms or relationships that can change action; push-your-luck requires a meaningful choice to continue or stop.

Route references in passes:

1. Start with the one reference governing the current uncertainty: experience, mechanics, coherence, diagnostics, balance, social play, or prototyping.
2. Add one medium adapter only when the medium materially changes the answer.
3. Add responsibility guidance whenever risk requires it, regardless of the nominal reference budget.
4. Load the relevant deliverable reference only when producing that artifact.

One to three references is a useful starting set, not a hard cap. Do not skip material guidance to satisfy a number, and do not load every possibly relevant file before identifying the decisive question.

1. Form an experience hypothesis

Name who should experience what, through which player behavior, in which context. State observable signs and a plausible failure condition. Read experience and motivation when the desired experience, audience, or motivational model is unclear.

2. Map decisions and dynamics

Identify player verbs, goals, information, incentives, constraints, uncertainty, feedback, interaction, and consequences. Work backward from the experience rather than forward from a clever component. Read mechanics and dynamics when inventing or repairing rules.

3. Diagnose an existing design before prescribing

When rules, a prototype, or playtest observations already exist, use diagnostic patterns to connect symptoms to competing causes and discriminating probes before proposing repair. Do not tune a number when the failure is structural. Do not add content when the loop is inert. Do not add choices when consequences are illegible.

Skip diagnosis for invention from a blank premise unless the premise itself contains a causal conflict.

4. Generate causally distinct alternatives

For invention or structural repair, offer two or three routes that produce the experience through different causal structures. Do not disguise tuning changes or renamed resources as alternatives. State trade-offs and recommend one.

For a confirmed bounded wording, arithmetic, parameter, timing, or edge-case defect, make the smallest complete correction. Do not manufacture structural alternatives when they cannot improve the requested decision.

5. Trace local ideas into the whole game

For each load-bearing mechanic, trace:

mechanic
  -> state, knowledge, capability, or relationship change
  -> affected player
  -> changed future decision or performance
  -> session-level rhythm or interaction
  -> promised experience

Test the idea at five scales: moment, decision, loop, session, and whole game. Read structural coherence whenever ideas are individually appealing but the game feels flat, repetitive, fragmented, or overgrown.

If the trace breaks, try to rewire the idea without bloating the design. If it still performs no necessary job, recommend simplifying, shelving, or killing it. Preserve promising cuts in an idea garden, not in the core rules.

6. Run or revisit system and responsibility gates

Read only the references material to the risk:

• Read balance and economies for strategic viability, incentives, resources, randomness, tempo, multiplayer pathologies, scaling, or numerical tuning.
• Read social, narrative, and embodied play for negotiation, trust, betrayal, roleplay, facilitation, authored or emergent narrative, physical space, performance, bleed, or calibration.
• Read accessibility, safety, and ethics when play can exclude, pressure, expose, monetize, exhaust, injure, or otherwise materially affect participants. Treat this as design, not polish.
• Read medium selection and translation when choosing or translating a form.
• Read digital and asynchronous media for screen-based, networked, automated, persistent, or delayed play.
• Read tabletop and facilitated media for board/card games, TTRPGs, or other human-adjudicated play.
• Read live and physical media for LARP, site-specific, playground, sport-like, or other embodied play.
• Read learning and training games when knowledge, skill, transfer, assessment, or real-world performance is part of the promised outcome.

7. Design the cheapest valid test

Identify the riskiest assumption, choose the lowest-fidelity prototype capable of falsifying it, and precommit what would support, contradict, or leave the claim unresolved. Read prototyping and playtesting before specifying or interpreting a test.

Do not use an LLM saying that play was fun as evidence of human enjoyment, social chemistry, usability, accessibility, or willingness to return. Use synthetic play only for bounded structural questions such as rule closure, legal-action coverage, consequence propagation, or obvious dominance.

8. Make a disposition

Recommend one of:

• Keep: evidence and causal account support the idea.
• Rewire: preserve its intent but change how it affects later play.
• Simplify: retain its job with less procedure, state, or content.
• Shelve: remove it from the current game while retaining the concept.
• Kill: remove it because its premise conflicts with the player promise or creates unacceptable harm or cost.
• Retest unchanged: preserve the design when the result is genuinely inconclusive and a corrected or more discriminating test is available.

State what evidence would reopen the decision.

Make a current disposition only when evidence or a direct rules conflict supports it. When planning a future test, specify the decision rule that will map possible results to Keep, Rewire, Simplify, Shelve, Kill, or Retest unchanged; do not pretend the unrun test has already earned one of those verdicts.

Choose the smallest useful deliverable

Lead with the verdict and highest-leverage reason. Then provide the minimum artifact that lets the designer act: an experience thesis, mechanic candidate set, causal trace, loop ladder, interaction map, playable ruleset, balance model, prototype card, playtest plan, evidence diagnosis, balance review, responsibility review, or disposition ledger. Read design deliverables for creation artifacts and test and review deliverables for prototypes, evidence, reviews, and dispositions.

Match depth to maturity. Do not hand a complete design document to someone who needs one discriminating experiment.

Compose selectively

When systems and emergence are the experience, optionally compose with /systems-as-experience (bravos-systems-as-experience); keep this skill's experience framing, evidence discipline, medium checks, and local-to-global gate authoritative. Use relevant engineering (/simulation-tactics for in-engine simulation, /web-backend for networked services), UX (/ux-designer), narrative prose (/creative-writing), research, or security (/security-architect) skills only when the request actually crosses into those disciplines.

Reject these failure modes

• Treating theme, activity, spectacle, content volume, or rules density as evidence of agency or depth.
• Presenting several buttons when one option dominates every relevant state.
• Using a framework as a universal law or a successful game's surface features as a recipe.
• Solving flatness by adding more disconnected systems.
• Treating numerical equality as the only meaning of balance.
• Optimizing retention, engagement, or monetization against player welfare.
• Treating accessibility or safety tools as ceremonial checkboxes.
• Reading politeness, laughter, rules comprehension, or designer explanation as sufficient proof of the intended experience.
• Normalizing a deliberate artistic, social, or low-replay choice merely because convention recommends broader appeal.

Complete the pass

Apply completion checks only to the claims and scope actually addressed. For a bounded task, verify the requested calculation or rule, relevant edge closure, and material risk without inventing a whole-game analysis. For broader work, before concluding verify that the response:

• protects the stated intent while identifying replaceable assumptions;
• explains how rules plausibly produce dynamics and experience where that causal claim is in scope;
• rejects a generated ruleset whose central choice collapses under obvious always/never or dominance analysis;
• checks moment, decision, loop, session, and whole-game coherence where scope permits;
• accounts for affected players, including waiting and excluded players, when the proposed interaction can affect them;
• respects the selected medium and facilitation model where they are material;
• addresses material accessibility, safety, and ethical risks rather than manufacturing a responsibility review for a harmless isolated edit;
• separates evidence from inference and names unresolved uncertainty; and
• recommends a concrete next decision, mechanic, prototype, or test.