THE REPRESSION TRAP
Instructor notes for Possible States
Created by Yoav Duman / יואב דומן

PURPOSE AND PLACE IN A COMPARATIVE POLITICS COURSE

Students begin with a movement that already exists. Their problem is how an organized campaign can turn participation into political pressure while sustaining its members under repression. This differs from The Square, which explains individual participation, free riding, coordination, and selective incentives. This simulation belongs in a unit on authoritarian rule, contentious politics, and regime change.

The player manages a fictional campaign in Lydora. Its characters, locations, in-world messages, measurements, and decision episodes are invented. Simulated actors represent conflicting interests, not real people. The game is inspired by scholarship; the authors cited below did not design or endorse it.

LEARNING OBJECTIVES

By the end, students should be able to:
1. Explain why visibility and a large rally do not automatically create sustained political power.
2. Trace how noncooperation can affect the institutions, interests, and people on which a ruler depends.
3. Distinguish a campaign's public reach, organizational capacity, endurance, pressure, and human costs.
4. Explain how state repression can damage a campaign, provoke wider opposition, or produce both effects.
5. Explain why the same campaign decision can have different consequences under different structural conditions.
6. Distinguish bounded concessions, a leader's departure, and an enforceable political opening. None establishes consolidated democracy by itself.

PROVISIONAL 45-65 ACTIVE MINUTE SEQUENCE

Timing is an instructional planning estimate, not measured student performance. A central budget is 55 active minutes: orientation and map 6; five episodes 28; changed-setting prediction/comparison 6; final dispatch 10; export and debrief 5. A careful full reading can exceed an hour. Discussion or interruptions extend elapsed time.

For a strict 50-minute class, complete the campaign and comparison, save a checkpoint, then finish the final dispatch as homework or in the next class. For a 65-75 minute class, keep time for a shared debrief. Do not add a countdown or reward speed. Selective rereading, dictionary use, dictation, breaks, and thoughtful disagreement are legitimate.

The learner reads state-sensitive evidence before choosing, makes a directional forecast, and inspects an observed response. Three pivotal decisions require short explanations. First reasoning stays intact; revisions can be appended immediately or later through the Notebook. Its original evidence IDs support the final causal explanation.

The comparison changes external backing for episode 2 only, holding the immediately prior condition and action fixed. New campaigns record a prediction before the alternative is revealed. A zero difference is useful evidence about the model's limited mechanism; it is not a reason to infer that external relationships never matter. This is not a full alternative history.

The final dispatch integrates transfer rather than assigning another essay: use an original evidence ID, trace a dependency, identify a cost and a rival explanation, distinguish actual implementation from democracy, and consider a fictional state-owned economy with guaranteed outside income. An evidence-based statement of uncertainty is acceptable.

WHAT THE MEASURES MEAN

Campaign capacity and endurance represent the ability to coordinate, adapt, and continue over time. Reach concerns connections across different communities and sectors, not just the number of people in one place. Pressure concerns costs and constraints experienced by the governing coalition. Harm records human costs that cannot be canceled by political gains.

The pillars represent different behaviors: economic allies becoming unwilling to carry costs, administrative noncooperation, and security forces showing restraint. These shifts are not interchangeable. A security command's refusal to defend a president does not demonstrate support for democracy or civilian control.

Every numerical measure is an invented teaching index. A value of 60 is neither 60 percent of the population nor a 60 percent probability of success. The engine is an authored model, not a calibrated forecast or a reproduction of NAVCO data. Its cutoffs organize fictional endings; they are not empirical tipping points. Population percentages and the historical 3.5 percent finding must never be read as a guaranteed recipe.

FAIR ASSESSMENT

Grade the report's reasoning, not whether the player secured a favorable political outcome. Suggested rubric: 30 percent accurate causal explanation; 25 percent specific evidence from the simulation; 25 percent comparison of a plausible alternative and contextual constraints; 20 percent thoughtful revision and model critique. Accept defensible disagreement with the model when supported. Word minimums support explanation but do not establish quality, originality, or mastery. An unsuccessful fictional campaign can produce an excellent student analysis.

DEBRIEF QUESTIONS

When did visibility create pressure, and when did it fail to do so? Which institution changed its behavior, through what mechanism? What did the movement lose while gaining pressure? What could a different choice reasonably change, and what remained outside the player's control? Who gained from the final agreement, who remained exposed, and what evidence would distinguish promised change from implemented change?

MODEL LIMITS AND INTERPRETIVE CARE

Nonviolence includes disruption and refusal; it does not mean passivity, politeness, or compliance. Nonviolent methods do not guarantee victory. Repression remains the responsibility of those ordering and committing it. A movement's decisions may affect exposure or political reactions, but never make abuse deserved or necessary. Retreat can preserve people and future capacity. Public sympathy is not automatically available, particularly across entrenched racial, ethnic, or colonial divides.

Aggregate associations do not prove that one tactic caused success. Campaigns, events, and long-term outcomes are different units of analysis. The simulation compresses time and excludes many historical dynamics. Real campaigns often combine methods; isolated unarmed confrontation and organized armed warfare are distinct phenomena. Do not infer that a single incident determines a campaign's character. This activity received fifteen independent AI reviewer assessments using faculty, Gen Z student, graphic-design, UX/product, and learning-design perspectives, followed by a discussion round. These are simulated perspectives, not recruited faculty or students, classroom playtesting, or an endorsement by any named scholar.


EDITION 2: WHAT THE MODEL NOW DEMONSTRATES

Credible exposure can increase attention without changing institutional behavior. Existing audience relationships and structural insulation distinguish the modeled branches. Preparation and tactical pauses can preserve people while some routine cooperation with the ruler returns. These effects are stated in the evidence before committing and explained afterward. Reach represents usable relationships as well as attention; its limited later effects are named rather than a universal turnout threshold.

Business pressure, administrative refusal, and regular-unit restraint remain different behaviors. The presidential guard stays loyal within this scenario; a moving regular-unit index must not be read as guard defection. External patron support also remains fixed during the short horizon. Real patronage can constrain rulers or change, and occupation changes political membership, accountability, and dependencies. The two settings do not cover those histories.

Human costs remain an accumulated ledger. Named consequences record affected households, organizers, volunteers, and essential services. Costs are not charged again through a generic additional endurance penalty, are not moral grades, and are never canceled by a favorable ending. The outcome's claim/observation table separates announcements, implemented gains, and unresolved authority.

The exact authored outcome criteria are available as an optional disclosure. They explain this model; they are not empirical causal coefficients, thresholds, real-world success odds, or student assessment criteria. Counts of possible authored paths are verification coverage, not campaign probabilities.

SAVE, BACKUP, AND REPORT BEHAVIOR

6717 opens the local Instructor studio. It is a convenience code, not authentication or access to student records. Class/exploration routes use the same political rules, and exploration use remains visible in the report. Fonts, text size, contrast, motion, device dictation, and browser-supplied read-aloud remain available.

Autosave is opt-in. Save + pause saves explicitly in that browser. A downloadable progress JSON includes drafts, current phase, first reasoning, revisions, Notebook revision drafts, first comparison prediction, and display/assignment settings. TXT/PDF reports are readable assessment artifacts; they cannot restore a game and do not submit work automatically. Reports retain original evidence, observed consequences, the revealed local comparison, costs, revisions, and model edition. A copyable TXT fallback is available.

Imported progress opens with autosave paused until an explicit Save or an enabled autosave setting replaces the browser checkpoint. Conflicting tabs pause automatic writes. Starting another run retains the open record while the learner visits setup, so unsaved work remains protected. Valid edition-1 histories keep their original frozen numerical rules; new campaigns use edition 2. Unreadable records can be kept as raw recovery text rather than overwritten silently.

CLASSROOM PILOT TO DO NEXT

Recruit actual introductory learners with varied prior knowledge and devices. Ask them to play without instructor coaching and log the first moment they cannot say what to do next, evidence they overlook, unexpected state changes, time at each phase, and successful save/resume/export. Do not treat 'hard' as a defect: ask whether difficulty came from weighing the political dilemma or locating/understanding information.

Use a short new case before and after the activity to check whether learners can distinguish visibility, dependence, actor behavior, and outcome. Code their explanations for a mechanism, evidence, rival account, and acknowledged uncertainty. Review loss/recovery incidents separately from learning difficulty. These checks have not yet been run with human participants.

SCHOLARLY BASIS

Chenoweth, Erica, and Maria J. Stephan. 2011. Why Civil Resistance Works: The Strategic Logic of Nonviolent Conflict. Columbia University Press. Especially chapters 2, 3, 5, and 8.
https://www.ericachenoweth.com/research/wcrw

Chenoweth, Erica. 2021. Civil Resistance: What Everyone Needs to Know. Oxford University Press. Especially chapters 2-4.
https://doi.org/10.1093/wentk/9780190244392.001.0001

Chenoweth, Erica. 2020. "The Future of Nonviolent Resistance." Journal of Democracy 31(3): 69-84. The statistics describe the article's historical sample, not the present game or a continuously updated estimate.
https://doi.org/10.1353/jod.2020.0046

Chenoweth, Erica. 2023. "The Role of Violence in Nonviolent Resistance." Annual Review of Political Science 26: 55-77. Distinguishes organized armed action from the more ambiguous effects of unarmed collective violence.
https://doi.org/10.1146/annurev-polisci-051421-124128

Beck, Colin J., Mlada Bukovansky, Erica Chenoweth, George Lawson, Sharon Erickson Nepstad, and Daniel P. Ritter. 2022. On Revolutions: Unruly Politics in the Contemporary World. Oxford University Press. Jointly authored; especially chapters 2-5 on situated strategy, mixed repertoires, outcomes, and international relationships.
https://doi.org/10.1093/oso/9780197638354.001.0001


BLIND NOVICE CLARITY PASS / OCTOBER 30, 2026

A separate round used five independent AI reviewers taking average first-semester, 18-year-old learner perspectives and three independent faculty lenses. Learner reviewers received only the current screen and made their predictions before the consequences were revealed. They did not inspect source, instructor materials, other reviewers, or future outcomes. These are simulated learner interpretations, not observations of recruited students. Two targeted text retests checked revised explanations; they are not new blind trials or measured learning gains.

The revised interface offers first-use definitions by pointer hover, keyboard focus, or tap. The full glossary includes concrete explanations of dependency, external patron, delegated coordination, regular-unit restraint, political opening, and related concepts. Definitions describe meanings rather than ranking strategies. The Notebook has quick episode navigation; help use causes no automatic point deduction.

Forecasts now include unchanged. New runs ask for one institution-specific forecast in the fifth decision, about business pressure, administrative refusal, or regular-unit restraint. This replaces the general forecast at that decision rather than adding another writing task. Earlier saved runs retain their former completion requirements. Wrong forecasts remain legitimate learning material.

Starting indices describe the previous checkpoint. The next update combines events during the episode and the campaign response. Around arrests, a response can reduce damage while net endurance still falls. A matching direction cannot establish the student's causal explanation. The final checkpoint is explicitly a completed period, with no next campaign update.

The comparison explicitly explains replacement finance after the prediction is locked, or why a support-only move has no immediate modeled difference. The final dispatch's checklist organizes the same demands: a decision/report ID and dependency; cost and rival account; implementation versus democracy; and a boundary condition with evidence needed. Do not award credit merely for repeating a scenario assumption as independent proof.

REMAINING MODEL LIMITS TO DISCUSS

Public reach is a composite proxy for attention and cross-group relationships. It cannot tell students whether each audience actually has useful contacts; the displayed contact conditions are scenario assumptions. The first public gathering's small regular-unit restraint response is a stated direct-contact assumption, including under foreign backing. It is not a universal effect of visible crowds.

The overall pressure and institution indices are related, authored measures. They are not independent statistics. Overall pressure can decline during a limited-gains priority while particular institutions continue applying pressure or refusing instructions. The feedback now explains that distinction. Some security restraint persists across this short horizon; the model does not teach that real institutional shifts are irreversible.

Human-cost descriptions do not fully simulate who can veto a coalition's strategy or withdraw consent. That is a limitation to question, not evidence that participants all accept the council's priorities. No extra numerical state variables or outcome bonuses were added in this clarity revision. Complete numerical paths match the prior edition-2 model.


FRESH BLIND REVIEW ROUND: 30 OCTOBER 2026
Five new simulated first-year learner perspectives and three independent simulated faculty perspectives reviewed the version-3 snapshot, then contributed to discussion. No real students or faculty were recruited. The text/controller harness does not measure comprehension, device usability or classroom time. A Notebook text-exporter issue was corrected after the independent runs and is not counted as a game defect.

Mechanism clarifications are added scenario assumptions: patron-backed payments on some firms' protected public contracts continue despite disruption, separately from replacing ruler revenue; willing civil servants withhold paperwork enforcing new association restrictions; on the relevant publicity route, willing intermediaries establish new contacts during the interval. None follows empirically from an index. The local comparison switches the whole backing package, so treasury finance and protected firm payments are not separately identified. Model indices, coefficients, thresholds and outcomes remain authored, uncalibrated teaching devices.

The fifth forecast concerns the period spent pursuing the selected demand before implementation is confirmed; the existing explanation must address the institution forecast. Wrong forecasts remain allowed. Final prompts accept unconfirmed implementation and keep the cost, rival explanation, discriminating observation, democracy distinction and transfer demand. Word checks record completion; an instructor must assess the argument. The reading-time range remains an estimate and may require a saved continuation.


RESPONSE ENTRY AND FACULTY REVIEW
Student response boxes block paste and text drops by default: decision explanations, feedback revisions, Notebook append-revisions, the changed-setting answer, and the final argument. Typed input and normal composition/replacement events remain available. If approved writing or accessibility software inserts through the clipboard, enable Allow paste in student response boxes in Instructor studio and apply the settings. Assignment links, configured offline copies and progress JSON retain that setting. Progress-JSON restoration, instructor instructions, and readable-report copying keep their clipboard workflows.

No clock, countdown, minimum viewing period, or new timing record is added. Decisions stay available when their existing response requirements are met. The political rules, word floors, forecasts, rival-explanation demand, transfer task, and first-reasoning archive are unchanged.

The report and TXT export now begin with a learner-writing summary: choice IDs, first forecasts and explanations, latest appended revisions, changed-setting writing and final argument. The complete evidence, scripted consequences and all revisions still follow. Grade the argument rather than the political ending.

Limits of the supplied feedback: the game checks word count plus at least five distinct words, rather than semantic quality. Forecasts are selected indicators and directions, not free-response forecast textareas. The changed-setting comparison switches a whole support package and can have a zero immediate difference; it does not automatically expose nonsensical writing or isolate treasury finance. Paste blocking is a local entry rule, not proof of original authorship, attention or learning. Manual assessment of the actor, mechanism, evidence status, cost, rival account and transfer remains essential.
