Full SKILL.md
War Gaming
Tests strategic options against external futures and rival moves — predetermined elements and critical uncertainties, a named four-scenario set with parameter tables, leading signposts with owners, a Value Net and competitor-response matrix, a facilitated or desk war game, and a robustness verdict with an adaptive plan. Use whenever an option's value depends on how rivals respond or which future arrives, or the user says "what will competitors do", "run a war game", "scenario planning", "how does this play out", "what if the incumbent cuts price", "which option holds up", "we need signposts", "red-team the competition". Not for failure modes inside the plan (a pre-mortem), a rival fact base (market-competitive-intelligence), or testing a hypothesis against evidence (hypothesis-stress-testing).
Scenarios describe the world; the response matrix and game describe the actors in it; the verdict says which option survives both. Shell scenario planning (Wack, Schwartz), Courtney's uncertainty levels, Co-opetition's Value Net and PARTS, and Coyne and Horn's response questions are in references/frameworks.md; the timed agenda, role briefs and adjudication rules are in references/facilitation-script.md.
When not to use
- No options to test yet:
strategic-choicefirst. - Rival facts rather than rival moves:
market-competitive-intelligence. - Internal failure modes and kill criteria: a pre-mortem, run separately.
- External research memo:
boardroom-storm-research.
Inputs
Ask for, or infer and label: the focal decision (two to four options including status quo, each with investment, timing and its "what would have to be true" list); horizon and commit date; rival list with facts (revenue, growth, margin, owner pressures, recent moves, stated strategy, balance-sheet capacity), two to four rivals; the Value Net beyond rivals (customers and procurement trends, suppliers, complementors, entrants); a baseline model or at least revenue, gross margin, opex, EBITDA; existing scenarios or forecasts and what they got wrong; for a live game, who plays, how long, who can change the plan afterwards.
Ask one round of questions if the options or the rivals are missing. Otherwise proceed on labelled assumptions in a block after the verdict.
Workflow
-
Focal decision and horizon (
references/templates.md§1). One sentence naming options and commit date. Horizon where consequences have mostly played out (three to five years). Place on Courtney's four levels: level 1 needs a sensitivity table not scenarios; level 4 needs signposts and staged options more than a matrix; this skill fits 2 and 3. Confirm the status quo and what doing nothing costs. Test: which decision would change if the exercise ran? None means stop. -
Predetermined vs uncertain. List driving forces (customer behaviour, technology, regulation, macro, structure, input costs, capital, labour). For each: is the outcome already determined over the horizon, and would a different outcome change the decision? Predetermined elements (demographics, capacity under construction, passed regulation, signed contracts, known cost curves) as a sourced "what we already know" list, five to ten; they appear in every scenario and are often the finding management has not priced. Remaining forces as ranged questions ("share of SME payment volume through embedded platforms in 2029: 15% to 50%"), eight to fifteen; cut the ones that would not change the decision. Test: anything predetermined with a confidence interval is an uncertainty.
-
Two critical uncertainties. Score each candidate on importance (5: moves horizon EBITDA over 15%; 1: under 3%; use the driver tree if it exists) and uncertainty (5: credible experts hold opposite views; 1: firm consensus). Top-right quadrant are candidates. Filters: independence (can move separately; name a mechanism for each off-diagonal cell); decision relevance (each end favours a different option); plausibility (a named credible person believes each extreme). Combine correlated candidates into a composite axis rather than dropping. Done when two axes have labelled ends, a numeric threshold, and a one-line independence argument.
-
Four scenarios (
references/templates.md§4). Per cell, a 150–300 word narrative in the past tense from the horizon year: what happened, in what order, why it made sense to actors at the time. Predetermined elements in all four. State of each rival written into each. Two-to-four-word names carrying the logic, never optimistic/pessimistic; every scenario good for somebody. Parameter table per scenario (growth, price, channel mix, inputs). Name the base case; no probabilities on the others. Test: someone should say "that could never happen" about at least one; nobody saying it means the ends are tame. -
Signposts (
references/templates.md§5). Per axis, three to five observable events that move belief toward one end: leading not lagging, observable from outside the client's own results, cheap to monitor, with a threshold and a date, an owner, a source, a review cadence, and the action if tripped. Test: could an analyst check it next Tuesday? If it needs the client's annual results, it belongs inkpi-architecture. -
Value Net and response matrix (
references/templates.md§6–7). Value Net: customers, suppliers, competitors, complementors; note whose payoffs change under each option. Response profile per rival: objectives, assumptions about the industry and about us, current strategy, capabilities and constraints. Per rival per option, Coyne and Horn: will they react at all (do they see it, does it hurt on their metrics, do they have bandwidth); which responses will they consider (their history is the base rate); which will they choose (best payoff on their metrics, not ours). Cell: most likely move, second, lead time, damage to our option, our best counter. Classify sequential vs simultaneous; in sequential games reason back from their best reply; note where a credible commitment changes their calculus. PARTS: add a player, raise added value, change a rule, signal, redraw scope. Test: would the rival's CEO recognise the profile as fair? Caricatures (irrational, passive, all-powerful) are wrong. -
War game. Teams: home, one per key rival, a market team (customers, channels), control adjudicating; strongest thinkers on rival teams. Three rounds, one period each; round 1 opens with the home option announced; rounds 2–3 open with published state and an inject drawn from signposts or narratives to test the options' conditions. Briefing packs: scenario, own response profile, payoff metrics, rules, legal moves (price, capacity, product, channel, M&A, partnerships, signalling). Control resolves with the model and a stated rule; scoreboard before each round. Play the base case with a mid-game inject, or once per decision-relevant scenario. No live game: a desk game, each team's moves written in their voice on their payoffs, labelled as such. Outputs: move log, scoreboard, surprise log. Test: the home team lost at least one round, or the rivals were played weakly.
-
Robustness verdict (
references/templates.md§9–10). Option-by-scenario payoff table adjusted for observed rival responses. Three lenses: count (best or acceptable in how many scenarios); minimax regret (best achievable minus achieved; smallest worst regret); downside (worst case against the client's stated tolerance: covenant, dividend, cash floor). Classify survivor (acceptable in all four), bet (excellent in one or two, poor elsewhere), dominated. Adaptive plan: no-regret moves now, options bought cheaply (right of first refusal, pilot, partnership), signpost-triggered decisions with dates. Verdict in one sentence containing "unless". Test: no "unless" means a forecast.
Output
The verdict sentence in the first 150 words. Then, each once, 2,000 to 3,500 words:
- Focal decision sheet
- Predetermined elements (sourced) and critical uncertainties with thresholds
- Four scenario narratives with parameter tables and base case
- Signpost table
- Rival response profiles and competitor-response matrix
- War game design, move log, scoreboard, surprise log (or desk-game equivalent, labelled)
- Payoff table, three lenses, option classification
- Adaptive plan with dated triggers
No method vocabulary, framework authors, skill names or the worked example's client in the deliverable. Full example: references/worked-example.md.
Quality bar
Run silently; fix, do not report.
- The decision sentence names options and a commit date, and something would change.
- Every predetermined element has a source; none has a confidence interval.
- Axes are independent, decision-relevant, and each extreme has a named believer.
- Scenario names carry logic; every scenario is good for somebody; no probabilities.
- Every signpost is leading, external, thresholded, dated and owned.
- Every matrix cell has move, timing, damage and counter; profiles are fair.
- Home team lost a round; surprise log is non-empty.
- Three lenses applied; classification consistent with the table.
- Verdict contains "unless"; adaptive plan has dated triggers.
- No leaked method vocabulary.
Anti-patterns
- Scenarios without a decision in view.
- Forecasts disguised as predetermined elements.
- Optimistic/base/pessimistic instead of two independent axes.
- Probability-weighting the scenarios so the one that matters gets ignored.
- Lagging signposts read from the client's own results.
- Rivals modelled as irrational, passive or omnipotent; rivals assumed to see everything.
- A game the home team wins every round.
- A verdict without "unless".
Hand-offs
References
War-Game Facilitation Script
1,641 wreferences/facilitation-script.md7 sub-sectionsFrameworks: Scenario Planning and Competitive Game Theory
4,683 wreferences/frameworks.md10 sub-sectionsTemplates and Checklists
1,511 wreferences/templates.md26 sub-sectionsWorked Example: Oakmere Payments
6,955 wreferences/worked-example.md11 sub-sections