Evacuation Readiness & Uncertainty Simulator — Explainer

? What is this tool?

The Evacuation Readiness & Uncertainty Simulator is a conceptual research instrument developed as part of a PhD project on civilian protection in armed conflict. It models the problem of assigning evacuee groups to destinations under conditions of incomplete, unreliable, or absent field intelligence — a situation routinely faced by humanitarian organisations and civilian protection actors.

The tool does not use real locations or real data. All scenarios are generated synthetically from a seeded random number generator, meaning any scenario can be reproduced exactly by entering the same seed. This reproducibility is a core requirement for academic peer review.

1 The core problem

When civilians need to be evacuated in a conflict zone, decision-makers must identify suitable destinations quickly. But the information available is almost always incomplete: some sites may not have been assessed, assessments may be outdated, and field reports may be unreliable. The central question this tool explores is:

How does uncertainty in field intelligence degrade the quality of evacuation decisions — and where would better information have the highest impact?

The simulator makes this abstract problem concrete by quantifying uncertainty, propagating it through a readiness model, and running Monte Carlo simulations to show how likely a given assignment is to succeed given what we know (and don't know).

2 How to use it

The header bar contains all the controls you need to generate and explore scenarios.

Click Generate Scenario to create a new random scenario. Adjust the uncertainty slider at any time — it recalculates instantly without regenerating the underlying data.

3 Readiness factors

Each destination is assessed across seven factors. Three are designated gatekeepers — they receive double weight and, if blocked, impose a hard cap of 20% readiness regardless of how strong other factors are. This reflects the principle that a destination that is physically dangerous, legally inaccessible, or whose host community has explicitly refused to receive evacuees cannot be made viable by good food supply or shelter.

4 The uncertainty model

Uncertainty operates at two levels in the model:

• Base confidence — each factor carries an individual confidence value (0–100%) generated when the scenario is created. This reflects how well-assessed that particular factor is at the time of the report.
• Field Uncertainty — the global slider scales all confidences multiplicatively. At 50% uncertainty, every factor's confidence is halved.

The model encodes a key epistemic distinction: a known unknown is less dangerous than an unknown unknown. If a factor is assessed as "Unknown" and we are highly confident in that assessment (we are sure we don't know), it scores higher than the same status with low confidence (we are not even sure our ignorance is accurate). This reflects the real-world asymmetry where a field team that explicitly reports "we could not assess food supply" provides more actionable information than a report produced under conditions where the team itself was unreliable.

5 Monte Carlo simulation

For each (evacuee group, destination) pair, the tool runs 500 Monte Carlo iterations. In each run, every factor's assessed status is subjected to a perturbation test: with a probability proportional to uncertainty, the status shifts one level (e.g. operational → partial). This simulates the distribution of what real conditions might look like given our imperfect knowledge.

A run counts as a success only if all three conditions are met:

• Readiness ≥ 40% (minimum viability threshold)
• Destination capacity can physically absorb the group
• No gatekeeper factor is blocked

The success rate is the proportion of runs that succeed. The standard deviation tells you how volatile the outcome is — high SD means small real-world changes could flip the result, which is itself a warning signal for decision-makers.

The uncertainty curve chart in the simulator shows how each group's success rate degrades as uncertainty rises from 0% to 100%, making it easy to identify which assignments are robust versus fragile under information degradation.

6 Assignment algorithm

Each evacuee group is assigned to the destination with the highest composite score, combining four components:

• Readiness (40%) — the weighted average readiness score for that destination
• Capacity fit (30%) — whether the site can absorb this group's size (capped at 100% to avoid over-rewarding excess capacity)
• Proximity (20%) — closer destinations reduce operational burden and transit risk
• Vulnerability match (10%) — a bonus when medical capacity is operational and the group has medical or mobility needs

The weights reflect humanitarian evacuation priority ordering: protection quality first, physical fit second, operational burden third, population-specific needs as a tiebreaker. Groups are sorted by urgency (immediate → urgent → can-wait) and assigned greedily in that order, deducting each assigned group's size from the destination's remaining capacity before the next group is placed. This models real capacity contention between groups — a group can be left UNASSIGNED if every viable destination is already full, which is flagged in red in the outcome table and alerts panel.

7 The research argument

The simulator is designed to support a specific thesis argument: that AI-assisted field assessment would improve outcome predictability in civilian evacuation planning. The tool makes this argument legible by showing:

• How much success rates drop as field uncertainty rises
• Which factors, if assessed, would most reduce outcome variance (the "highest information value" finding in the insight panel)
• How gatekeeper uncertainty is disproportionately dangerous — resolving a gatekeeper unknown simultaneously unlocks or eliminates whole destinations

The practical implication is that limited intelligence-gathering or AI-assisted verification capacity (satellite imagery analysis, sensor fusion, field report NLP) should be directed toward high-weight unknown factors first — not spread evenly — to maximise decision quality under time and resource pressure.

8 Data sources & references

This tool uses no real-world datasets. Every destination, evacuee group, and factor status is generated synthetically by the seeded RNG described in §5 — there is no live feed, import, or historical case data behind any number on screen. What the standards below inform is parameter design: which factors exist, which are gatekeepers, and where thresholds like the 40% viability floor or the 3.5m²/person shelter norm come from.