Push‑your‑luck — When to Stop: Comparative Risk Thresholds
EVExpected value — the average outcome you would see across many identical turns; use as the baseline comparison for stop vs push.
TailTail risk — rare, large losses (or wins) that skew outcomes away from the mean; important when a single failure can undo progress.
mEVMarginal EV / threshold — the incremental expected change from taking one more attempt; a stopping rule says “stop if current value ≥ threshold.”div>
Quick comparison
Low‑variance safe mechanic
Predictable gains; lower tail risk
High‑variance push mechanic
Bigger possible jump, steeper failure tail
1 · Problem framing
Across push‑your‑luck turns the essential decision is the same: accept the current payoff or attempt another risky action that reshapes the outcome distribution. Compare distributions — not single numbers. A higher mean (EV) can coexist with a heavier failure tail; the stopping decision depends on whether you value preserving the current quantile (a conservative target) or chasing higher mean over many plays.
Marginal change matters: evaluate the single extra attempt in context of whole‑turn consequences.
Movement 2 — Tools primer (intuitive)
2 · Tools primer
Quick primer: expected value is your arithmetic baseline; variance measures spread; tail risk is where variance concentrates losses. Marginal EV asks: "If I take one more attempt, how does the average move?" Express a stopping rule as a threshold: stop when current value ≥ threshold. Thresholds may be an EV offset (EV + risk premium) or a quantile target (e.g., protect the 25th percentile of outcomes).
Heuristics convert distributional thinking into a single decision point.
Movement 3 — Worked example A: coin‑like discrete push
3 · Example A — discrete coin push
Sketch: a single push yields either a clear advancement or a reset to a lower baseline. Compute marginal EV verbally: weigh the chance of advancing times the gain against the chance of losing the current stake. If the extra attempt's expected upside is smaller than the expected loss (given the probabilities and stake size), stop. If the chance is symmetric (an even coin), the decision hinges on how much the failure reduces tournament progress relative to the gain.
When failures reset to a low baseline, even favorable EVs can be unattractive if you prize survival of the round.
Movement 4 — Worked example B: dice‑pool with partial success and catastrophic failure
4 · Example B — dice‑pool attempt
Dice pool mechanics produce graded outcomes: partial successes accumulate; a critical failure often imposes a severe penalty. The overlay shows two cumulative curves: the conservative option dominates early (higher probability of reaching modest thresholds), while the push curve overtakes at high reward targets. Practical implication: pick the option that dominates at your decision threshold — if you aim to guarantee a modest quantile, prefer the conservative; if chasing rare high rewards, prefer the push.
Crossing points tell you where the preferable choice flips — compute them by comparing cumulative probabilities at candidate thresholds.
Movement 5 — Synthesis: three compact stopping heuristics
5 · Synthesis rules
Three compact heuristics:
- EV + risk premium: set threshold = current EV minus a premium for catastrophic loss; use when rounds aggregate over many plays and mean matters.
- Quantile target: stop when current value secures a chosen quantile of outcomes (e.g., protect the lower quartile); use when preservation matters or failure is costly.
- Context modifier: adjust thresholds by opponent state, remaining turns, or comeback potential — raise the threshold when you can afford to be cautious, lower when behind and variance is rewarded.
Choose the heuristic that matches your objective function (mean, safety, or tournament state), then apply the same rule consistently to avoid ad‑hoc risk swings.
Movement 6 — Design and balance notes (for designers)
6 · Design notes
Designer levers that reshape stopping thresholds: reward magnitude, failure severity, probability skew, and whether results compound between pushes. Increasing reward without changing failure severity encourages pushing; introducing soft failure (partial loss) lowers tail risk and lifts stopping thresholds. Observe players: frequent all‑or‑nothing failures where cautious players win often suggests thresholds are too punitive.
Balance by comparing how design changes shift the cumulative curves and their crossing points at common thresholds.
Movement 7 — Open questions and prompts for playtesting
7 · Open prompts
Prompts to guide exploration (non‑procedural):
- Compare thresholds that protect different quantiles: how does player behavior change when you protect the 30th vs the 50th percentile?
- Track crossing points: identify at which reward targets the push option overtakes safe options for your most common turn states.
- Observe variance aversion in practice: when do players accept lower EV to avoid rare catastrophic losses? Use that signal to adjust failure severity or present clearer affordances.
These prompts are observational — use them to map player incentives, not to prescribe exact rule changes.
Note: worked examples are schematic. The method here is comparative — compute marginal changes and inspect cumulative probabilities at candidate thresholds rather than relying on single‑number heuristics alone.
End of catalog entry.