Game model analysis maps how rules, agents, resources and feedback turn player actions into outcomes. It explains why a strategy dominates, where progression stalls and which mechanic may be behind a difficulty spike. The model should predict what changes when a designer adjusts a rule.
A boss encounter that feels unfair may be driven by hidden probabilities, weak feedback or a difficulty mismatch. Each calls for a different fix. Before changing damage values, identify the choices available to the player, the information they had and what the rules did next. The same logic applies to tabletop prototypes and educational games.

What does a game model represent?
A game model keeps only the elements needed to answer a defined question. Combat balance calls for damage, health, timing, skill and random variation; collaborative learning calls for participant roles, information sharing, choices, feedback and evidence of learning. Textures and dialogue belong in the model only when they affect that question.
The state transition can be written as:
next state = T(current state, actions, rules, external events)
A game state captures relevant conditions: location, inventory, score, remaining time, character health, resources, quest flags or information held by different players. Actions come from human players and non-player agents. Rules constrain what may change. External events cover randomness and other changes outside a player's control.
Change one assumption at a time. If a resource costs one extra unit, does the success rate move? Does lowering an enemy's damage improve perceived fairness or merely prolong combat? Might another tutorial prompt clarify the goal for beginners while interrupting experienced players?
In contrast, game theory focuses on incentives and strategic decisions between interdependent agents. Payoff matrices and equilibrium concepts clarify competitive or cooperative choices. They cannot explain interface friction, narrative meaning, learning outcomes or reported player experience by themselves.
The causal chain: mechanics, dynamics and aesthetics
The Mechanics–Dynamics–Aesthetics (MDA) framework, introduced by Robin Hunicke, Marc LeBlanc and Robert Zubek, distinguishes specified rules, emerging play patterns and the experience players report.
Mechanics → Dynamics → Aesthetics
Mechanics cover actions, algorithms, costs, rewards, constraints and win conditions. Dynamics describe the patterns that emerge when agents interact with those mechanics: competition, cooperation, risk avoidance, resource hoarding, exploration or escalating conflict. Aesthetics concern experienced qualities, including challenge, discovery, expression, narrative, fellowship and a sense of mastery.
In a cooperative game, two players draw from the same energy pool. The shared limit is a mechanic. Deciding whose move matters most becomes a dynamic; that discussion can feel collaborative or frustrating. Raising the pool might ease conflict while removing the reason to coordinate.
A rule diagram cannot establish that players felt cooperative. Check the intended dynamic against observed play and the players' accounts.
The entities and relationships worth mapping
The Game Ontology Project offers a vocabulary for interfaces, rules, goals, entities and entity manipulation. It describes relationships within a game rather than genres. For a behaviour question, the map also needs motivations, resource flows, timing and observable outcomes.
| Entity group | Typical components | Analytical question |
|---|---|---|
| Agents and roles | Human players, avatars, teams, non-player characters, opponents | Who can act, what can they know, and whose interests differ? |
| Goals and constraints | Objectives, victory conditions, failure states, deadlines, rules | What makes an action meaningful or impossible? |
| Game state and time | Levels, phases, turns, checkpoints, cooldowns, progression | Which decisions depend on what happened earlier? |
| Actions and affordances | Movement, crafting, trading, combat, dialogue, available choices | Does the player understand and control the available options? |
| Resources and economy | Health, currency, energy, inventory, experience points, rewards | Where are resources created, transferred, converted and removed? |
| Information and feedback | User interface, indicators, audio, animation, hints, score displays | Can players connect an outcome to the decision that caused it? |
| Chance and uncertainty | Random-number generation, hidden information, variable events | Is uncertainty understood, predictable or unfairly distributed? |
| Experience and meaning | Challenge, agency, immersion, narrative, social relationships | What do players actually experience, rather than what was intended? |
The LUNA Framework extends this view through Frames, Entities, Ludic Agents, Human Agent Motivations, Actions and Activities, Process Segmentation, Human Agent Situations and Agent Setups. Pressing a button is a single action; negotiating with a teammate across a round is a sustained activity. LUNA can distinguish the two.
For learning games, the Game Elements–Attributes Model (GEAM) records design elements and attributes, allowing two games with the same learning objective to be compared despite different interactions.
Which analytical framework fits the question?
| Method | Best use | What it cannot establish alone |
|---|---|---|
| MDA | Tracing mechanics through emergent dynamics to intended player experience | Whether actual players felt the intended emotion |
| Game Ontology Project | Cataloguing entities, goals, rules, interfaces and manipulation | How frequently players choose each action |
| LUNA Framework | Describing agents, motivations, situations, framing and activity structures | Numerical balance or observed retention |
| Machinations | Modelling resource flow, economies, progression and feedback loops | Social meaning, accessibility or subjective enjoyment |
| Consalvo–Dutton toolkit | Qualitative analysis through an object inventory, interface study, interaction map and gameplay log | Population-level player behaviour without broader sampling |
| DiGAP | Making game selection, analytical boundaries, coding and reporting transparent | A complete causal explanation without an appropriate analytical model |
| Player Experience Inventory (PXI) | Comparing reported functional and psychosocial experiences | The source of a mechanic's failure without observation or system data |
Mia Consalvo and Nathan Dutton's qualitative toolkit provides four ways to investigate a game without source code: object inventory, interface study, interaction map and gameplay log. They reveal what the player can see and do, without claiming access to hidden rules.
The Digital Game Analysis Protocol (DiGAP) focuses on transparent, reproducible analysis. Its 31 items address the rationale, researcher background, game selection, boundaries, analytical approach, coding, data extraction and reporting. A single tutorial sequence cannot stand for an entire game.
How to conduct game model analysis
- Define a decision, not just a topic. Replace “analyse progression” with a falsifiable question: “Does the early upgrade economy allow players to recover after one failed mission?” Specify the target audience, platform, game version and outcome that would justify a change.
- Fix the boundaries. Decide whether the unit of analysis is a turn, encounter, mission, session or full player journey. Record difficulty settings, player roles, input methods and which game modes are excluded. A single expert playthrough is not representative of a new player cohort.
- Map states, actions and causal links. List the starting conditions, available actions, transition rules, feedback and end conditions. Record the game loop: action → state change → information returned → next decision. Separate a rule's intended effect from an observed behaviour.
- Observe and capture gameplay. Use repeated playthroughs, screen recordings, think-aloud sessions, object inventories, gameplay logs and interaction maps. Where instrumentation exists, capture events for progression, resource flow, failed actions and exits. Document the event definitions before interpreting them.
- Stress-test the model. Try an optimal strategy, a novice strategy and deliberately unusual choices. Vary starting resources, random seeds and difficulty. For a probabilistic system, Monte Carlo simulation can estimate the spread of outcomes across many model runs. Model checking can test specified properties of a formal state system.
- Validate with players. Compare predicted patterns with playtesting, telemetry, interviews and structured measures of player experience. Ask whether players noticed the information needed to make a choice. A strategy that works in a simulation may be invisible or unpleasant in actual play.
- Test a controlled revision. Change one mechanism or a clearly defined bundle of mechanisms. Specify the primary metric, a guardrail metric and the version being compared. Retest the scenarios most likely to expose a trade-off.
Keep causal claims within the tested conditions: with these starting conditions, this strategy has a measurable outcome that the proposed change should alter. Enjoyment, learning and fairness still require separate player evidence.

Worked example: a hypothetical coastal conservation game
Imagine a game about restoring a British coastal wetland. The player chooses where to build paths, protect habitats and spend a limited budget. Its model links actions to environmental effects, visitor access and public support.
Map the state variables, decision rules and feedback loops before testing. A path that attracts visitors may also disturb nesting areas; that trade-off is part of the game system, not a flaw in the mathematics. Compare simulated outcomes with the design goals and explain where the model deliberately simplifies real ecology.
Balance, progression and emergent behaviour
Fairness is not a fixed 50% win rate. Asymmetric roles can be balanced through counterplay, information, timing and opportunity; a cooperative game can be hard without feeling arbitrary. Compare the spread of outcomes with the experience the game is supposed to create.
Resource-flow models can expose specific faults:
- Runaway positive feedback: an early advantage produces more resources, which generates an even larger advantage. Snowballing may erase counterplay.
- Excessive resource sinks: spending, damage or upkeep continually outpaces earnings, leaving players without viable recovery paths.
- Dominant strategies: one action offers a better outcome across nearly every meaningful state, making alternative mechanics decorative.
- Progression dead ends: an upgrade, level gate or difficulty spike requires resources or skills unavailable at that point.
- Opaque randomness: outcomes vary, but players cannot identify the risk they accepted or learn from failure.
Simulations estimate win probabilities, game duration, resource accumulation and outcomes under different strategies. They cannot tell whether someone understood the interface or enjoyed the encounter. A bot with perfect information and instant reactions is a poor substitute for a first-time player.
Players invent strategies the model never anticipated: shortcuts, trading practices, social rules and item combinations. A simulated economy may look stable while an exploit spreads through live sessions. Compare predictions with gameplay logs and telemetry to catch that divergence.
Measure player behaviour and player experience separately
Game telemetry records observable events: session starts, actions, tutorial completion, purchases, resource gains, resource spending, level failures and exits. The GameAnalytics metric definitions distinguish daily active users (DAU), session counts, playtime, progression, retention and virtual-currency flows. Cohort analysis groups players by a shared starting period or event, while funnels examine movement through defined steps.
Day 1 and Day 7 retention track returns on specified days after a defined start. The reporting platform's retention definition and day boundary matter: a session logged under UTC can fall on a different date in the UK. Churn and session length show different patterns of use, neither of which proves boredom or satisfaction.
| Hypothesis | Primary evidence | Guardrail or interpretation check |
|---|---|---|
| The first objective is unclear | Time to first successful action and observed goal comprehension | Do extra prompts slow down confident players? |
| The upgrade economy traps unsuccessful players | Resource balance and recovery rate after failure | Does the change eliminate meaningful risk? |
| One strategy dominates | Outcome distribution across strategies and player skill levels | Are alternatives understood and available? |
| A cooperative mechanic creates teamwork | Interaction patterns, observed communication and player accounts | Are quieter players excluded from decisions? |
| An applied game teaches its target concept | Defined learning task and delayed transfer or recall where suitable | Does improved in-game scoring reflect genuine understanding? |
The Player Experience Inventory (PXI) provides a structured measure that connects design choices with reported experience. Its five functional constructs are ease of control, goals and rules, challenge, progress feedback and audiovisual appeal. Its five psychosocial constructs are meaning, curiosity, mastery, immersion and autonomy. The full validated instrument is preferable to selectively inventing substitute survey questions and treating their scores as equivalent.
Self-determination theory distinguishes autonomy, competence and relatedness. Rewards might bring players back without offering meaningful choice. For an applied game, engagement is an intermediate signal: the real outcome may be learning, transfer or better decisions.
What can invalidate a model?
- Sampling bias: volunteer playtesters, experienced developers and returning players may behave differently from the intended audience.
- Unobserved context: fatigue, language, social setting, device performance and controller familiarity can change the experience without changing any formal rule.
- Instrumentation errors: missing events, duplicate session starts, inconsistent identifiers and changing event definitions distort behavioural comparisons.
- Confounded changes: simultaneous alterations to rewards, tutorial text and difficulty make the cause of any improvement unclear.
- Overfitted simulation: a model tuned to one strategy or small set of random seeds may fail when players discover a new approach.
- Inaccessible interaction: small touch targets, information conveyed only through colour or sound, and fixed control schemes can appear to be player-skill problems.
The Game Accessibility Guidelines cover controls, captions, information conveyed through colour and difficulty options. When players cannot operate an interface reliably, a model may mislabel an access barrier as a skill gap.
Record the game build, platform, participant profile, observation period, included modes and assumptions alongside contradictory evidence. Success across many encounters does not establish months of retention or transfer to another cultural setting.
Game model analysis for UK studios and researchers
For a UK audience, a useful model separates design assumptions from evidence. Document data sources, accessibility constraints, privacy considerations and the limits of any real-world analogy. Testing with varied players helps identify rules that appear fair on paper but feel confusing or exclusionary in play.
What a defensible game model analysis delivers
Six artefacts make a game model analysis auditable:
- A scope statement: the research question, audience, game build and boundaries of the analysis.
- A system map: agents, entities, goals, resources, actions, information, state transitions and feedback loops.
- A behavioural hypothesis: an explanation of how a rule produces a predicted pattern of play.
- An evidence record: model runs, gameplay observations, relevant telemetry and player-experience findings, with their limitations visible.
- A proposed intervention: the smallest rule, interface or progression change that addresses the diagnosed mechanism.
- A validation rule: the primary outcome, guardrail measure and evidence threshold used to decide whether the revision worked.
If a model predicts that a cheaper upgrade will improve fairness but players still avoid it, the contradiction matters. Check what information players saw, how feedback arrived and whether the model omitted a competing action.
Frequently asked questions
Can an existing game be analysed without access to its source code?
Yes. Repeated play sessions, object inventories, interaction maps, screenshots, gameplay logs and interviews can build an external model. Hidden probabilities, server-side rules and behaviour outside the observed sample remain uncertain without further evidence.
Does game model analysis require simulation?
No. Simulation suits questions about resource accumulation, game duration, strategy outcomes and randomness. Cultural meaning, player motivation, accessibility and the clarity of feedback still call for observation and interpretive methods.
How is a game model different from a game design document?
A game design document specifies intended features. A game model connects selected rules and states closely enough to explain or test behaviour. A document might call an upgrade balanced; a model exposes the costs, benefits and player strategies against which that claim can be tested.
