by Mauricio Castro, Tom Apperley, and Alan Meades

Activity Overview

This case study describes and evaluates a practice-based learning activity in which students use an AI coding agent to port an obscure or out-of-print board game to a browser-based digital format. The activity was initially piloted by Castro, Apperley, and Meades in February 2026, based on previous research by Castro (2024), and Castro and Apperley (2024). Having established proof of concept, we are now exploring what it would take to run this activity at scale across a level 6 cohort, and potentially across multiple institutions, as a structured component of game design education. AI-assisted porting is a rich critical and analytical process as well as an opportunity to develop AI skills and literacy, and is well suited to level 6 learners in game design.

Activity Process

The activity is structured in three sequential phases, each with distinct deliverables. 

Phase 1 is archival preparation. Students select a candidate game from a prescreened library of obscure or out-of-print titles with complete surviving documentation. They digitise all physical components as PNG image files at sufficient resolution for use as game assets. They then produce a structured design document in Markdown format that captures the complete rule system, disambiguates any conflicts or gaps in the original rules, defines all game entities (players, pieces, board spaces, events), specifies win conditions and turn order, and maps the physical component set to its intended digital equivalents. This document is the primary input to the AI coding agent and must be complete and unambiguous before Phase 2 begins.

Phase 2 is AI-assisted implementation. Students install Visual Studio Code and a coding agent extension (such as ChatGPT Codex or Claude Code) with access to their project folder. Working within VS Code, they open the project folder containing the design document, rulebook, and asset files, then prompt the agent to produce a Minimum Viable Product (MVP) implementation plan using the canvas element of the AI app and javascript. The agent is instructed explicitly to interpret rules literally and to flag any ambiguity as a question rather than to resolve it silently. Students review and approve the implementation plan before authorising the agent to generate code, which it does with direct edit access to the project folder. The resulting implementation is tested using VS Code Live Server. Students then run a second agent session in which the agent performs a formal review of the rule system against the generated code, producing a structured conflict log in Markdown, with each issue classified by severity (critical, high, or non-priority). Students examine this log, prioritise critical and high-severity conflicts, and work through iterative resolution sessions with the agent, using both the conflict log and the game files as direct context

Phase 3 is testing, refinement, and reflective documentation. Students play a minimum of five complete games against one or more computer-controlled opponents to verify consistent rule enforcement and identify any remaining behavioural errors. Where errors are found, students add them to the conflict log and return to the agent for targeted fixes. Once stable, students deploy the game as a browser-accessible web application. Alongside the playable output, students produce a critical reflective report that uses the conflict log as its primary source, discussing what the ambiguities and rule conflicts reveal about the design of the original game, its assumptions about player knowledge, and the interpretive gaps that human players routinely navigate through convention. This report is the primary assessed component linking the technical and analytical dimensions of the activity.

Final assessment portfolio: input documentation (phase one), implementation notes (phase two), link to playable game, and critical reflective report (phase three).

Educational Rationale

The activity asks students to treat the porting of an obscure or out-of-print game as a form of heritage practice. Game heritage has historically been challenging games are ephemeral, platform-dependent, and difficult to archive in ways that retain their essential character as interactive experiences. A boardgame rulebook in an archive may be an important and useful document, but it is not a game in any meaningful sense. Out-of-print board games are particularly challenging for heritage institutions: they cannot be stored in digital archives, may be considered too unimportant for many museum collections, and are often too obscure for commercial re-release.

A critical insight from the pilot is that archival labour must precede AI involvement, and that this sequencing is itself pedagogically significant. Before any coding agent is introduced, students must assemble and structure the game’s documentation: digitised rulebooks, component images, a design document that translates the physical game into a format the agent can read. 

At scale, this first phase of the activity, archival and documentary work, represents a meaningful and assessable task in its own right, and one that develops skills rarely foregrounded in conventional game design education: close reading of rules, structured description of game systems, and critical evaluation of what a game’s components actually say versus what players conventionally assume they say.  This maps to CLO1 (Subject Knowledge), as students must apply specialist understanding of game systems to evaluate rule structures, and to CLO4 (Research), since the work requires assessing what the source documentation actually evidences rather than what convention assumes.

Learning Outcomes

The concept of the “playtext,” which we have borrowed from theatre studies and adapted for game analysis, refers to the game’s rules and structures as a kind of script that is always necessarily interpreted in performance, never exhaustively determining it. A rulebook is not an instruction set but a text: it requires reading, contextualisation, and judgment. The activity makes this interpretive dimension unusually visible for students, because the AI agent’s engagement with the source rules foregrounds precisely the gaps and ambiguities that human players routinely navigate through tacit convention. This directly addresses CLO2 (Think & Reflect), requiring students to evaluate how tacit assumptions shape knowledge and practice in game design, and CLO8 (Digital), since it develops critical awareness of what AI tools do and do not understand when processing rule-based systems.

In the pilot activities, the coding agent was instructed to interpret rules as literally as possible and to surface ambiguities as explicit questions rather than silently resolving them. This methodological choice produced a striking result: formal rule review identified multiple conflicts in the source rulebook, some classified as critical. These were not implementation errors; they were ambiguities already latent in the source text. The process of porting thus performed textual criticism, revealing fault lines in the playtext that experienced board game players typically paper over through convention, negotiation, or inattention. For students, encountering this dynamic for the first time is often a genuinely revelation: the game they thought they understood turns out to be underspecified in ways that only become visible when a reader with no tacit knowledge of play encounters the text. This is central to CLO2 (Think & Reflect): students must recognise how their own perceptions and assumptions about a familiar game have been shaping their understanding and revise those assumptions in light of evidence from the text itself.

Rulebooks are generally composed under conditions of anticipated play, their authors assume a reader who will bring common sense, cultural context, and embodied game literacy to the text. An AI coding agent brings none of these things. Its literalism functions as a defamiliarisation device, making the gaps and contradictions in the playtext visible as structural features. This aspect of the activity can be explicitly framed for students as a form of AI supported critical reading practice. The conflict reports the agent produces become primary texts, requiring interpretation and judgment from the student to resolve them in their implementation of the ruleset.  This engages CLO8 (Digital) by asking students to critically evaluate AI-generated outputs rather than accept them uncritically, and CLO7 (Problem Solving) as students must apply specialist judgment to resolve real conflicts between the rule text and its computational implementation. The layered nature of the playtext becomes progressively clearer as the activity unfolds. There is the rulebook text; the agent’s interpretation of that text as code; the student’s own play experience as they test the implementation; and the further divergences that emerge between expected and actual game behaviour during extended sessions. Each layer of interpretation introduces variation, and each round of revision is an attempt to bring these layers into alignment. The game that emerges is not a transparent digital copy of the original; it is the product of a complex negotiation between source text, computational interpretation, and lived play knowledge, what we might call a collaborative reading. This negotiation is itself the primary learning site.  This is the core site of CLO3(Transdisciplinary Perspectives), integrating methods from textual criticism, heritage studies, and game design into a coherent practice, and of CLO5(Applied Skills), where specialist techniques are used to produce a new, functional game artefact.

References

Castro Valdez, M. (2024). Counterforensic Ludology: A Approach to Critical Work with Board Games [Dissertation]. Tampere University. 

Castro Valdez, M. & Apperley, T. (2024). Digitizing Shining path – The Struggle for Peru: Reclaiming and Reconstructing the Peruvian Armed Conflict. Digital Games Research Association, University of Guadalajara, Mexico (1-5 July).