Purpose: Use this checklist to prevent low-quality scenario work. It preserves the practical traps, the bias layer, and the quality gate Helm should apply before trusting scenarios.
Contents
SCN-01..SCN-10- Bias impact
- Scenario quality checklist
- Helm quality gate
Scenario Planning Pitfalls
Anti-Pattern Catalog
| ID | Pitfall | Failure mode | Fix |
|---|---|---|---|
SCN-01 |
Searching for the “correct” future | Treats scenarios as prediction instead of decision support | Use “plausible” and “challenging,” then ask what action each scenario demands |
SCN-02 |
Asking the wrong question | Question is too broad, too narrow, or detached from decisions | Anchor the scenario question to a live investment or policy choice |
SCN-03 |
Mixing trends and uncertainties | Uses deterministic trends as scenario axes | Separate fixed trends from true uncertainties |
SCN-04 |
Four versions of the same worldview | Scenarios do not force different choices | Require materially different decisions per scenario |
SCN-05 |
Data dominance, no imagination | Spreadsheets replace creative future recombination | Use data to inform stories, not to flatten them |
SCN-06 |
Lifeless narrative | No actors, weak causality, no memorability | Add concrete actors, motives, and causal links |
SCN-07 |
Decision-makers excluded | Scenarios do not change executive thinking | Involve decision-makers early in question and uncertainty selection |
SCN-08 |
Missing “So what?” | Scenario creation stops before strategic implications | Spend equal time on action implications |
SCN-09 |
One-off exercise | Scenarios are published and forgotten | Build early-warning indicators and review cadence |
SCN-10 |
Ignoring emotional impact | Stakeholders resist because threatening futures are emotionally unmanaged | Normalize discomfort and make implications explicit |
Bias Layer
| Bias | Distortion | Mitigation |
|---|---|---|
| Confirmation bias | Preferred scenario appears “most likely” | Assign a devil’s advocate |
| Anchoring | First scenario dominates evaluation | Start from multiple anchors |
| Overconfidence | Ranges are too narrow | Run a pre-mortem |
| Availability heuristic | Recent events dominate scenario salience | Review historical analogs |
| Framing | Wording shifts scenario evaluation | Reframe the same scenario from multiple angles |
| Groupthink | Agreement pressure reduces diversity | Use anonymous voting / Delphi-style input |
Scenario Quality Checklist
Design Quality
- The focus question is tied to a real decision.
- Scenario axes reflect genuine uncertainty, not fixed trends.
- The scenarios force different decisions.
- Each scenario is internally coherent.
Narrative Quality
- Human actors and incentives are visible.
- Causal chains are explicit.
- The story is understandable to non-specialists.
Process Quality
- Decision-makers participated.
- Strategic implications were derived.
- Early-warning indicators were defined.
- A refresh schedule exists.
Bias Check
- A Red Team or devil’s advocate reviewed the set.
- “Most likely” language was avoided.
- At least one uncomfortable scenario was included.
Helm Quality Gate
| Gate | Rule |
|---|---|
| Post-generation review | Run the full checklist on every scenario set |
| Bias penalty | If bias checks were skipped, subtract 0.2 from confidence |
| Refresh warning | If no update schedule exists, add a warning flag |
| Simulation use | Only feed high-quality scenarios into SIMULATE and ROADMAP |
Integration
Use these rules with:
simulation-patterns.mdfor short-, mid-, and long-horizon scenario generationstrategic-calibration.mdto track scenario quality over timestrategy-monitoring.mdto bind scenarios to live signals and assumption drift
2026 Generative-AI Scenario-Building Guardrails
GenAI now generates scenarios at near-zero marginal cost (per wargaming-simulation.md's GenWar / CSIS-Futures-Lab notes and the Scaling Intelligent Agents in Combat Simulations line of work). Three failure modes are now standard checks:
| Failure | Signal | Mitigation |
|---|---|---|
| LLM "consensus future" | All N generated scenarios converge on the same trend (often the median pretraining-data view) | Force the model to defend the minority scenario; require historical analog citations |
| Hallucinated probabilities | Model attaches confident P=...% without evidence |
Strip model-emitted probabilities; reassign via human-assessed Delphi or Magi panel |
| Narrative homogeneity | Same protagonist archetype across scenarios | Specify divergent actor roles (incumbent / disruptor / regulator / non-consumer) per scenario before generation |
Treat GenAI scenario output as Tier 4 (external default) per data-inputs.md — usable with disclosure, never authoritative.