GDnD Wiki

GDnD Wiki

This site is for Ulster University students and staff. Enter the class access password to continue.

Source: Advanced Game Design — A Systems Approach

9 min read

Source metadata

  • Type: Textbook
  • Author: Michael Sellers (Indiana University; lead designer on Meridian 59, The Sims 2)
  • Date: 2018
  • Publisher: Addison-Wesley / Pearson
  • Pages: 456
  • ISBN: 978-0-13-466760-7
  • Structure: 3 parts — Part I: Theory (Ch. 1–4), Part II: Design (Ch. 5–8), Part III: Practice (Ch. 9–12)

Key takeaways

Part I–II (Chs. 1–8): Theory and design fundamentals

  • Central thesis — games are systems: Game design is system design. The failure mode of most novice designers is reductionist thinking — isolating parts rather than understanding how parts interact to produce emergent wholes. Sellers’ entire methodology is built on systems thinking as the designer’s primary cognitive tool.
  • Formal system definition: A system is “a set of parts that together form loops of interaction between them to create a persistent whole.” Parts have state, boundaries, and behaviours; loops connect parts; emergent wholes arise from looped interactions. Hierarchy: parts within parts, systems within systems.
  • Systemic depth and elegance: Depth = multiple levels of hierarchical organisation for the player to explore; elegance = achieving depth from few rules, with self-similarity across levels and few exceptions. Go is the canonical example. Bushnell’s Law (“easy to learn, difficult to master”) is depth + elegance in four words.
  • Second-order design: Game designers do not design the player’s experience directly — they design the system that generates the experience. The designed tokens and rules create a state-space (many paths) not a single path. Emergent gameplay arises from this space, often surprising the designer.
  • Four principal loops: (1) the game’s internal loop; (2) the player’s mental loop; (3) the interactive loop between them (the “core loop”); (4) the designer’s loop — the designer observing the game+player system and adjusting. Core loops can be nested (inner fast, outer slow) or multiple in parallel.
  • Neurochemical engagement: Four primary neurochemical reward pathways map directly to different forms of engaging play: dopamine (novelty/reward), serotonin (security/accomplishment), oxytocin/vasopressin (social bonding), norepinephrine/endorphins (alertness/stress). Sellers argues these should inform the target experience at design time.
  • Psychological engagement (Schaufeli et al.): Engagement characterised as vigor + dedication + absorption. All three are required for deep player investment.
  • Structural coupling (Maturana): Game and player form a mutually adapting system. A systemically designed game creates a broad enough state-space to adapt to the player as the player adapts to it.
  • Yerkes-Dodson arousal curve: Performance peaks at intermediate arousal; too low → boredom; too high → panic/anxiety. Provides neurophysiological grounding for Csikszentmihalyi’s flow channel.
  • Parts taxonomy (Ch. 8): Every game is built from physical parts (objects in game space), non-physical parts (rules, counters, states), and representational parts (visuals, audio, UI). Parts have attributes (name-value pairs) and behaviours. The three attribute orders: first-order (amounts), second-order (rates of change), third-order (rates of rates — acceleration/compounding). The behaviour locality principle: each part’s behaviour should be defined locally and act generically, enabling emergence from simple rules.
  • Feedback timing (Ch. 8): Immediate feedback must occur within 100–250ms for the player to perceive it as responsive. Combine visual and auditory channels; maintain consistency so feedback signals are reliable.
  • Data-driven design (Ch. 8): Separate attribute values from code into spreadsheets or JSON/CSV files, enabling balancing without recompilation. In Unity, ScriptableObject assets serve this role.

Part III (Chs. 9–12): Practice

  • Game balance — methods (Ch. 9): Four balancing methods: designer intuition, player-based (playtesting), analytical (data), and mathematical modelling. All complementary; none sufficient alone. Tumbleseed case study: 59% of players never reached the first checkpoint — detectable by analytics, preventable by earlier playtesting.
  • Transitive vs. intransitive balance (Ch. 9): Transitive systems balance parts against each other (rock-paper-scissors — each beats half and loses to half, requires an odd number of parts). Intransitive systems balance via cost/benefit — some parts are simply better, but they cost proportionately more. Intransitive balance underpins progression systems.
  • Progression curves (Ch. 10): Six curve types with distinct properties — linear, polynomial, exponential, logistic/sigmoid (S-curve; most psychologically satisfying for player progression), piecewise-linear, and near-arithmetic (NAP). Most games combine sigmoid for overall arc, polynomial for item power, linear for secondary attributes.
  • Attribute weight coefficients (Ch. 10): Assign weight coefficients to each attribute; total weighted cost should be equal across all items of the same tier. The mathematical core of intransitive balance.
  • Hedonic fatigue (Ch. 10): Repeated rewards become less satisfying; progression systems must escalate intensity or vary reward types. Punctuated (milestone-based) progression is more engaging than continuous at macro scale.
  • Economic balance (Ch. 10): Three failure modes — inflation (resources arrive faster than spent), stagnation (too few resources), arbitrage (exploitable price differentials). Habitat doll/crystal ball case study is the canonical arbitrage failure. Albion Online’s power = 1.2^tier and rarity = 3^tier is the canonical deliberate market design.
  • The iron equation (Ch. 10): LTV > eCPU + Ops. Lifetime Value must exceed the cost of acquiring the player plus operational costs. Retention (D1/D7/D30) is the leading indicator of LTV; FTUE is the critical design battleground.
  • Game Outcomes Project (Ch. 11): Tozour (2014), ~300 completed games. Top success factors: (1) clear/compelling/consistent/shared vision, (2) focused development, (3) cohesive teams, (4) limited crunch, (5) swift conflict resolution. Limited crunch improves both quality and team satisfaction.
  • Sellers’ three team principles (Ch. 11): Integrity (doing what you say), Flexibility (adapting without losing coherence), Communication (information flowing freely and accurately). Teams are hierarchical systems — dysfunction should be diagnosed at the structure level, not the individual level.
  • Studio hierarchy (Ch. 11): C-suite → VP/GM → Executive Producer → Producers (1:10 ratio) → Designers / UI/UX / Programmers (by specialisation) / QA / Art / Audio. Producer span of control: approximately 10 team members per producer.
  • Find the fun fast (Ch. 12): Before art, story, or UI, close the interactive loop: player action → system response → player observation. This must be engaging before anything else is built. Keep prototype code separate from game code — prototype answers questions; production code ships products.
  • Wizard of Oz prototyping (Ch. 12): A human operator manually performs the role of an expensive system (AI, procedural generation) while a participant tests as if it were automated. Tests the interface and response structure before implementation.
  • Playtesting discipline (Ch. 12): Observe silently. Use a playtest script. Posttest Likert-scale surveys for quantifiable attitude data. Directional analysis: classify each problem by direction (easier/harder, clearer/subtler) and check for tester agreement. Players diagnose symptoms accurately; they rarely correctly identify causes or solutions — collect diagnoses, not prescriptions.
  • Pitching (Ch. 12): Elevator pitch (30 seconds, tweet-length concept statement). Pitch meetings use an iceberg approach — present just enough to intrigue, not everything you know. Goal of a first meeting is not a deal but a follow-up. “Fast no” is better than a slow maybe: a clear rejection is more valuable to the developer than months of ambiguous interest.

Notable claims / quotes

“A system is the integrated whole that arises out of independent, interacting parts. Those parts have their own internal state, boundaries, and behaviors by which they mutually affect each other.” — Sellers, Ch. 2 summary

“Game design is second-order design. The game’s design as expressed in its tokens and rules creates the specification for a state-space, not a single path.” — Sellers, paraphrasing Salen & Zimmerman, Ch. 6

“A game that is balanced avoids having one narrow dominant winning path or strategy or creating situations where one player has an inherent or insurmountable advantage.” — Sellers, Ch. 9

“Such games are often described as ‘easy to learn, difficult to master’ (known as Bushnell’s Law, after Atari founder Nolan Bushnell).” — Sellers, Ch. 2

“[Vigor] is characterized by high levels of energy and mental resilience while working, the willingness to invest effort in one’s work, and persistence even in the face of difficulties.” — Schaufeli et al. (2002), quoted by Sellers, Ch. 4

Relevance

Primary source for:

Substantial additions to:

  • game-balance — Sellers’ transitive/intransitive framework, four-method taxonomy, and Tumbleseed case study
  • internal-economy — Ch. 10: inflation, stagnation, arbitrage failure modes; Habitat case study
  • game-analytics — Ch. 10: iron equation, FTUE, D0/D1/D7/D30 retention, analytical balance
  • prototyping — Ch. 12: find the fun fast, interactive loop test, Wizard of Oz, thinking aloud
  • playtesting — Ch. 12: silent observation, Likert-scale surveys, directional analysis, players diagnose symptoms
  • flow — Sellers’ Yerkes-Dodson and structural coupling provide physiological grounding
  • fun-as-learning — Sellers’ neurochemical model provides biological mechanism for Koster’s thesis
  • mda-framework — Sellers’ systems model works below MDA at the mechanics/parts level; they are complementary

Open questions raised

  • Sellers’ formal system definition focuses on loops — but some game systems (e.g. pure narrative games) have few explicit feedback loops. Does the definition fit walking simulators?
  • The neurochemical model implies that designers should consciously target specific neurochemicals in their design. Is this a useful design tool in practice, or too reductive?
  • Sellers argues that second-order design is the distinguishing feature of good systemic games. Is there a meaningful category of games that achieve great engagement without second-order design (e.g. highly crafted linear experiences)?
  • The iron equation requires LTV projections, but LTV is only measurable after months of live operation. What early proxy metrics most reliably predict LTV?

Systems and design foundations:

Parts, balance, and progression:

Analytics:

Production and practice:

Supporting:

Graph View

Backlinks