Skip to content

Epistemic Games

Pattern · All patterns
Evidence · 19 claims (7 for, 12 mixed) · 17 studies (4 quant-synthesis, 4 qualitative, 3 review, 3 theoretical, 2 design, 1 causal), q1–q4 · 4 of 17 report an effect size · 17 claims rest on one study

Description and scope

A reusable unit- or course-level policy in which learners take a role inside a simulation of a professional community's work (an urban planner redesigning a street, a consultant preparing an offer for a client, a soldier keeping a squad alive) and learn that community's epistemic frame: what its members know, how they decide what is worth knowing, what they value, and how they judge when an answer is good enough. The policy is: study how practitioners actually reason → build a role and a simulated world that demand the same decisions → give an in-fiction brief → learners act, with mentor characters, tools and consequences supplying guidance → in-fiction feedback on each decision → a debrief that names the frame and connects it to the real practice → repeat with harder and less typical cases, with less mentor support → a final product for a realistic audience inside the fiction. The intended change is in making and justifying the practice's decisions (skill-performance on a complex-skill), in carrying that reasoning to a new case (near-transfer, far-transfer), and in the learner's sense of the practice's values and identity (motivation-affect). Facts the practice needs are learned along the way, not taught first.

The pattern serves two purposes the earlier page named: initiation, introducing novice learners to an unfamiliar frame (Madison 2200), and transformation, built around atypical cases that break an advanced practitioner's current framing.

It sits inside three converted pages and does not repeat their models. Game-Based Learning owns the general relationship for learning through a game: the repeated winning move must be the target reasoning, novices need the strategy shown, and the capability must be checked outside the game. Here the "winning move" is a professional decision justified in the practice's terms. Communities of Practice owns learning by sustained participation in a real community; an epistemic game is a simulated, bounded stand-in for that participation when real participation is not available or not safe. Authentic Assessment owns judging a capability through a realistic performance with a draft-feedback-revise cycle; the final product of an epistemic game can be assessed that way. The measurement problem specific to epistemic games (what a log of play can tell you) belongs to Evidence-Centered Design and the Assessment Mechanic. This page owns the narrower configuration: a role-play simulation of one profession's decision-making, with in-fiction guidance and feedback and a debrief on the frame, run over a unit.

No claim in this wiki tests epistemic games: none compares learners who played one with comparable learners taught the same domain another way, on decisions, transfer or identity. The claims below test or describe parts of the sequence (simulation-based training against no training in health professions, feedback, guidance against unguided discovery, practice-near training content, one role-play organisation simulation described qualitatively) and several limit it (unguided discovery, blanket scaffolding, the difficulty of scoring play). Each is carried to this pattern by extrapolation, and the page says how far. The sequence below is the earlier page's, kept as the default; its doses, move-on rules and the response table are untested design proposals.

Inputs: establish the design brief

Field Record or elicit
Learner and starting observation Before play, give each learner a short realistic case from the practice (a planning request, a client brief, a patient handover) and ask what they would do and why. Record which considerations they raise, which they miss, whether they justify the decision by the practice's values or by rule or preference, and what they already know of the domain. Record whether they are novice to the practice or advanced practitioners. Keep unknowns as unknowns.
The practice Which community's work is being simulated, and what is known about how its members reason: an epistemography, interviews or observation of practitioners, or only a textbook account. Which decisions are characteristic of it, which facts they need, which values decide trade-offs, and what can be safely left out. If no one has studied the practice, the game will simulate a guess at it.
Context and activities Setting (classroom, online-instructor-led, online-self-paced, workplace-clinical, or mixed with some learners remote), days-weeks or term-plus, session length, group size, whether a teacher or practitioner can play the mentor role, devices (full computer or phone only), budget for building the world.
Learner-valued goal Ask what the learner wants from the unit: to find out whether they would like the job, to pass the course, to do the job they already have better, to make something that matters locally. Record disagreement with the designer objective.
Designer objective Which of these the assessment will weigh: decisions made and justified in the practice's terms, the domain content those decisions need, transfer of the reasoning to a new case, or a change in professional identity or values. Name one as primary.
Outcome A task outside the game, at a stated horizon: a new case to decide and justify, scored against criteria drawn from the practice (near-transfer), a case from a different setting (far-transfer), a content test, and, separately, interest or identity measures (motivation-affect). In-game score, time played and enjoyment are recorded but are not the capability.

A request to "build an epistemic game for this course" does not specify these. Ask which practice, who has studied how it reasons, what decision the learner should get better at and how it will be judged outside the game, before choosing the world, the roles or the number of sessions.

Sequence and conditional policy

The default sequence is the earlier page's, with doses added as proposals. Steps 1 and 3 are untested; steps 2, 4, 5 and 6 have a neighbouring claim, carried by extrapolation.

  1. Study the practice (the epistemography). Interview or observe practitioners making the decisions the game will simulate: what they look at first, what they ignore, how they decide an answer is good enough, and which values settle trade-offs. List three to five characteristic decisions and the knowledge each needs. Evidence status: untested; from the earlier page. Interdisciplinary graduate teams rarely noticed that their members differed by discipline in what counts as trustworthy evidence and when a problem statement is settled [~W] (a qualitative video study; it uses "epistemic games" in a different sense, the moves of inquiry, not a designed game), which is a reason to write these standards down rather than assume them.
  2. Build a role and a world that demand those decisions, with the knowledge inside it. Put the facts learners need into mentor characters, documents, citizen letters, maps and tools in the world, not into separate exposition. Match fidelity to the decisions and cues the learner must attend to, not to visual realism. Evidence status: technology-enhanced simulation training in the health professions [+M] produced large gains in knowledge and skills and moderate gains in behaviour and patient outcomes against no intervention (609 studies); it was not compared with equally long practice without simulation, and its subgroup analyses found no consistent design feature that drove the effect. Nurses interviewed after specialty training said the closer training content was to real clinical scenarios, the easier it was to use at work [+W] (one qualitative study, perceptions).
  3. Open with an in-fiction brief. A directive from inside the world (a mayor's request, a client's tender, a mission briefing) frames the whole unit, names the product due and who will judge it. Evidence status: untested; from the earlier page.
  4. Learners act; the world guides them. Mentor characters (or a teacher or practitioner in that role) model one decision in the practice's terms before learners make their own, answer questions in role, and prompt "what would a planner check before deciding that?". Each decision has an in-fiction consequence and an explanation of why. Evidence status: unassisted discovery produced less learning than explicit instruction, while discovery with feedback, worked examples, scaffolding or elicited explanations did better than other instruction [~S] (a meta-analysis of 164 studies, a 112-child experiment and a narrative review; none in a game). Feedback improves learning by a medium average amount, most when it carries information about the task, and over a third of feedback interventions made performance worse [+S] (two meta-analyses, education and work settings).
  5. Adjust support to each learner, then withdraw it. Mentor prompts and modelled decisions are frequent early and fade as a learner justifies decisions unprompted; learners who stall get more. Evidence status: a review argues that serious games give every learner the same scaffolding regardless of expertise, which contradicts the idea of scaffolding [~W] (the authors' synthesis of literature, no data); fading itself is modelled in Scaffolding and Fading.
  6. Debrief on the frame after each session and at the end. Ask learners which decision they made, what they weighed, what a practitioner would have weighed, and where the same reasoning would apply outside the game; name the frame's values explicitly and compare it with the real practice, including what the game left out. For transformation games, centre the late cases on an atypical case that breaks the usual framing and debrief on what failed. Evidence status: one review reports, second-hand, a study in which students given situated instruction with modelling, coaching, scaffolding and reflection outperformed a lecture control on a causality post-test [+W] (no effect size; the original study not read here). In a six-week organisation simulation, student teams' knowledge practices shifted from framing the problem and their expertise early to concretising and reflecting later, and a flip-chart outline turned unfocused talk into structured work [+W] (qualitative diaries and video; no comparison group or learning measure).

Dose and progression (proposal, untested). Run the unit as eight to twelve sessions of 45–60 minutes over three to five weeks, or a matching block in a longer course. Early sessions (1–3): one characteristic decision per session, a mentor models a decision first, learners decide in pairs, every decision gets in-fiction feedback with the reason, and a 10-minute debrief closes each session. Middle sessions (4–8): decisions combine (budget against ecology against residents' wishes), mentors answer only when asked, learners work alone or in role-differentiated teams (different departments), and debriefs move from "what happened" to "what a practitioner would weigh". Late sessions (9–12): a less typical case, the product drafted, reviewed by a character or person in a judging role, revised and presented; a longer final debrief connects the frame to the real practice. Moving on: a learner moves from modelled to unprompted decisions when they justify two successive decisions in the practice's terms without a prompt, and to the atypical case when they do so on a new case; a learner who cannot name the considerations after two sessions gets a modelled decision and a checklist again, not more free play. These quantities are starting points to record and revise.

Observation Discriminating probe Proposed action, with its uncertainty
Learners play for the score or the fiction and decisions are not justified Pause and ask, in role, "why that option, and what would the residents' association say?" If they can justify when asked, add a required justification to each decision. If they cannot, they may lack the domain knowledge or not see what the practice values: have the mentor model one decision aloud. Untested.
Decisions are fluent inside the game but the outside-game case is weak Give the same decision type with the game's interface and characters removed. If performance drops, the learning may be tied to cues in the game; add debrief that states the reasoning without the fiction, and vary cases. If it holds, the outside task may be measuring something else. Untested.
Learners search the world aimlessly, or copy one another's choices Ask privately what they are trying to decide and what they would need to know. If they cannot say, restore the brief's goal and a modelled decision; this is the unguided-discovery case the guidance claim warns about. If they can, point them to the character or document holding it. Untested.
An experienced practitioner dismisses the scenario as unrealistic Ask which detail is wrong and what a real case would look like. If the detail is wrong, fix it; the epistemography missed something. If the case is atypical by design, say so and debrief on why the usual framing fails there. Untested.
A team's members argue past one another about evidence Ask each member what would convince them, and from which discipline that standard comes. Name the disciplinary difference in the debrief; it is common and usually unnoticed. Whether naming it improves the team's work is untested.

Fitting the design to a situation

The three facts that most change the decision:

  • Whether learners are new to the practice or already in it. Novices need modelled decisions, frequent in-fiction feedback and a gradual hand-over; practitioners need atypical cases that break their framing. If unstated, ask: have the learners ever made this kind of decision for real, and how would a practitioner rate their first attempt?
  • Whether anyone has studied how the practice reasons. Without an epistemography the game rehearses a textbook's idea of the job. If unstated, ask: who are the practitioners we can interview or watch, and what decisions do they say are hard?
  • What the outcome is and how it will be judged outside the game. Decisions justified in the practice's terms, content knowledge, transfer and identity are different targets and need different measures; a log of play is hard to score reliably. If unstated, ask: after the unit, what case will learners be given, and who will judge their answer against what?
If the brief says… Then change… Basis
Learners are novice to the practice Open each session with a mentor modelling one decision aloud; one decision type per early session; feedback with the reason after every decision; fade prompts only when a learner justifies unprompted. Guided against unguided discovery [~S], carried from classroom and laboratory studies to games
Learners are advanced practitioners (continuing education, in-service) Skip the early modelled sessions; build the unit around two or three atypical or breakdown cases; debrief on what their usual framing missed. Untested; from the earlier page's "transformation" games
Learners are child or young adolescent Choose a practice whose decisions are visible and concrete (planning a park, running a shop); shorten sessions to 30–40 minutes; keep an adult in the mentor role; check the content separately. Untested proposal
The goal is facts or a procedure, not judgement Do not build an epistemic game; teach directly with practice and feedback, and use a short simulation only for applying it. Untested; from the earlier page's constraint ("overkill for simple factual or procedural learning")
adult-workplace learners in workplace-clinical training Use cases drawn from the learners' own workplace, a practitioner as mentor, and an outside-game task that matches a real work task; record transfer to work separately from in-game performance. Simulation training in the health professions [+M] (against no training); practice-near content [+W] (nurses' perceptions)
online-self-paced, no teacher Mentor characters must carry all guidance: a modelled decision before each new decision type, required written justification, scripted feedback with reasons, and a written debrief prompt at the end of each session. Expect weaker debriefs than with a person. Untested proposal; feedback [+S] carried to software feedback
Mixed mode: some learners in the room, some joining remotely, or phone-only Keep the world playable in a browser on a phone (documents and decisions as text and images, no heavy 3D); give remote learners a named role with their own documents so the team must consult them; run the debrief in a shared document everyone writes in. Untested proposal
One teacher or practitioner for many learners (30 or more) Run in teams of four with differentiated roles (departments); the teacher plays one mentor character for the whole class at set points, and teams check decisions against a written checklist from the epistemography between visits. Untested proposal; the earlier page's role differentiation
single-session only (one workshop) Drop the full game: one case, one role, a modelled decision, one learner decision with feedback, and a 15-minute debrief on the frame. Do not expect a change in identity from one session. Untested proposal; Communities of Practice notes norms take a term to form
Learners with low confidence or low literacy Give the world's documents as short texts with audio and images; start with low-stakes decisions whose consequences can be undone; let learners justify orally or by choosing among stated reasons. Untested proposal
The final product goes to a real audience (a council, a client, a community group) Keep the fiction's audience real: practitioners or residents read the final plan and respond; share their criteria at the brief; build a draft-review-revise step before the presentation. Untested here; see Authentic Assessment and Authentic Audiences and Purposes
The game will be used for high-stakes grading Grade the final product and an outside-game case against public criteria with two raters, not the play log; if the log is scored, play long enough to accumulate evidence and keep few, broad dimensions. Reliability is hard to define in epistemic games [~W] (an argument); longer play separated simulated learner types better [~W] (simulated learners, not real ones)
Interdisciplinary teams Ask each member at the start to state what evidence would convince them and when a framing is "done"; revisit it in the debrief; give the team a shared visible plan (flip chart, board). Disciplinary differences went unnoticed [~W]; a shared outline turned unfocused talk into structured work [+W]; both descriptive, neither tests the move
Little budget or time to build a world Use a paper or document-based simulation (briefs, letters, maps, a budget sheet) with a teacher in role, rather than software; spend the budget on the epistemography. Untested proposal; the simulation synthesis [+M] found no consistent design feature, and fidelity was not among those it recorded

Six of the fourteen rows cite a claim (novice learners, workplace training, online without a teacher, high-stakes grading, interdisciplinary teams and a small budget), each carried beyond the learners or setting it studied as its Basis cell says; the other eight (advanced practitioners, children, factual goals, mixed mode, one teacher for many, a single session, low confidence or literacy, and a real audience, the last resting on converted neighbouring pages) are untested proposals or the earlier page's guidance.

Choosing configurations from evidence

  • Simulation of the work. Simulation-based education improves outcomes [+M]: a systematic review and meta-analysis of 609 studies (35,226 trainees) of technology-enhanced simulation for physicians, nurses, dentists and other health professionals, against no intervention, found large pooled effects on knowledge (g = 1.20), time and process skills, and product skills, moderate-to-large on behaviours, and moderate on direct patient outcomes (g = 0.50); heterogeneity was large, 405 of the studies were single-group pre-post designs, and subgroup analyses found no consistent interaction with curricular integration, distributed practice, feedback, mastery learning or repetitive practice. It is the strongest evidence that practising a profession's tasks in simulation beats not practising them; it does not compare simulation with equally long non-simulated practice, does not test role, narrative or identity, and comes from health professions training. Carry it to "adult trainees practising the decisions of their profession in simulation learn more than with no such practice", no further.
  • Guidance inside the world. Unassisted discovery produces less learning than explicit instruction, while discovery enhanced with guidance outperforms other instruction [~S]: one meta-analysis (580 comparisons, d = 0.38 favouring explicit instruction over unassisted discovery; 360 comparisons, d = 0.30 favouring enhanced discovery), one experiment with 112 third- and fourth-grade children, and a narrative review arguing the advantage of guidance recedes only with enough prior knowledge. This is the boundary of the pattern: a rich world left to explore without modelled decisions, feedback and debrief is the unguided case, which is what the earlier page meant by "guided immersion, not unguided exploration".
  • Feedback on each decision. Feedback improves learning [+S]: across 435 studies the average was d = 0.48, rising to d = 0.99 for high-information feedback and falling to d = 0.24 for reinforcement or punishment; a second meta-analysis found d = .41 on average, with over a third of interventions lowering performance, worse the more feedback drew attention to the self. For a game: an in-fiction consequence with the reason it followed is high-information feedback; a points total is closer to reinforcement.
  • Practice-near content, as perceived. Training content close to real clinical scenarios is easier to use at work [+W]: interviews with neurosurgical specialty nurses, who said case analysis and simulated emergencies helped them link knowledge to action. It reports perceptions in one qualitative study; it does not measure transfer.
  • What learners do in a role-play simulation of a profession. Organization-simulation knowledge practices support interdisciplinary learning [+W]: 84 students in a six-week course in which mixed-discipline teams prepared an offer for a simulated HR-consulting client; diaries (1,083 coded units) showed six knowledge practices, problem framing and expertise roles early, concretising and reflection late; video of two teams showed a flip-chart outline making a sidelined proposal usable; unclear leadership and time pressure were the commonest reported barriers. One course, no comparison group, no learning measure (its i2 code rests on no printed statistic). It is the closest description in the wiki of an epistemic-game-like unit in use; it shows what to observe, not that the unit works.
  • Situated instruction against lecture, second-hand. Situated instruction outperformed a lecture control on a causality post-test [+W]: a review reports Hendricks (2001), whose treatment followed a situated model with modelling, coaching, scaffolding and reflection; no effect size, and the original study is not recorded here.
  • Limits on measurement. Reliability and measurement error are hard to define in epistemic games [~W]: Rupp et al. (2010) argue that complex tasks induce dependencies across observations, indicators pass through several layers of human judgement, and rich process data may be distant from the interpretations wanted. In a simulation study with simulated, not real, learners, a statistic separated learner types better the longer the game was played [~W]. Together: do not grade high-stakes decisions from a short log of play.
  • Limits on adaptivity. Serious games give all learners the same scaffolding [~W]: a review's argument, with no data, that embedded support in serious games ignores learners' differing needs.

Do not rank epistemic games against case-based learning, direct instruction or work placement from these claims: none compares them, and none measures the cost of building a world. Do not read enjoyment, time played or in-game success as the capability. A precise abstention names the missing comparison (an epistemic game against equally long case-based practice with the same feedback, on an outside-game decision task at a stated delay) and the local observation that would stand in for it.

The evidence by question

Further evidence, not yet read against this model

The earlier page cited no claims. These claim pages concern how play in an epistemic game can be scored; they bear on assessment design, not on whether epistemic games produce learning, and none studied real learners.

Illustrative design instance and observation record

Invented illustration, not a tested unit. A Year 9 geography class (30 adolescent learners, one teacher, classroom with four students joining remotely on phones during a period of illness) takes a four-week, ten-session unit modelled on Madison 2200: each team of four are planners asked by the mayor to redesign a run-down high street, with a budget, residents' letters, a traffic count and a flood map. The teacher interviewed two council planners first and listed four decisions they called hard (where to put parking, how much green space, whose letters to weight, what to do about flood risk). Because the learners are novice and there is one teacher, the table changed the default: each early session opens with the teacher, in role as a senior planner, deciding one case aloud; teams have differentiated roles (transport, environment, housing, residents' liaison), and each remote student holds one role's documents in a shared online folder, so the team must consult them; every decision is entered with a one-line justification and gets a written consequence from the "council". By session 6 the senior planner answers only when asked; in session 8 a flood the teams did not plan for breaks most designs; sessions 9–10 are revise-and-present, to two actual council officers who read the plans. A student says she wants to find out whether planning is a job she would like; the teacher's objective is justifying land-use decisions with evidence. They agree to score a new outside-game case (a different town's park) four weeks later, separately from an interest questionnaire. The simulation claim covers adult health professionals against no training, not this population, setting or horizon; these are local choices to observe, not settled ones.

Record: practice and the decisions studied → role, world and documents → starting case and stated reasons → decisions made, in order, with justifications and the guidance given before each → consequences and feedback shown → debrief statements about what a practitioner would weigh → the atypical case and the response to it → final product and the audience's response → outside-game case at the stated horizon, scored against criteria from the epistemography → candidate interpretations (frame used, rules followed without the frame, playing for the score, content missing) and their basis. Preserve the learner's stated purpose and any change in it. An agent may summarize this record but must not fabricate missing inputs or assign numerical probabilities of having acquired the frame without a calibrated model.

Elements and limits

Simulations, role-playing, feedback, debrief, assessment mechanic and the shared-understanding moves of epistemic games (a related sense of the term: the inquiry moves teams use). Principles and theories the pattern draws on: Game-Based Learning, Learning Embedded in the Core Mechanic, Communities of Practice, Scaffolding and Fading, Authentic Audiences and Purposes, Situated Learning and Self-Determination Theory.

This pattern is scoped to learning one profession's way of deciding through a guided role in a simulated world over a unit. Factual or procedural goals, single sessions, and practices nobody has studied need other configurations (direct instruction, case-based learning, cognitive apprenticeship in a real setting). The policy supports observation and design reasoning; whether an epistemic game produces better decisions on new cases than equally long case-based practice with the same feedback, for which learners, at what cost and at what horizon, remains to be tested.

Claims

  • Evidence-Centered Design — the assessment-design method Rupp, Gushta, Mislevy and Shaffer (2010) worked out against epistemic games specifically, which is how a game about professional judgment yields anything scoreable
  • Assessment Mechanic — the repeated in-game activity that elicits that evidence, designed so the log discriminates between the constructs the epistemic frame is made of
  • Game-Based Mastery Learning — both use game structures to sustain engagement, but mastery learning targets repeatable skill drills with progression gates, while epistemic games target situated professional judgment through role and narrative stakes

Examples

  • Madison 2200 — players act as urban planners redesigning a downtown pedestrian mall, learning urban ecology as a byproduct of planning decisions
  • Full Spectrum Warrior — players think and act as a professional soldier, with military doctrine distributed between the player and AI-controlled squad members
  • SimCity integrated into a high-school economics unit, where students build and defend a functioning city and encounter economic principles through the consequences of their design choices

Key Sources

Page metadata
Field Value
type pattern
id epistemic-games
title Epistemic Games
description A reusable policy in which learners take a role in a simulation of a profession's work and learn its way of deciding; expected to improve decisions justified in the practice's terms where the practice has been studied, mentors model decisions and feedback and debriefs are given, and the outcome is judged outside the game; no claim in this wiki tests epistemic games.
status review
grain_size unit
author David Williamson Shaffer
canonical True
generated by claude/unspecified on 2026-10-05
trust unverified — no one has confirmed this page's content

Frontmatter is the page's machine-readable record — see the Schema & Guide.