Destinations ?
Generating scenario…
| Parameter | Value | Controls |
|---|
Destination readiness is computed as a weighted average of seven factor scores. Three factors — Security, Local Authority Consent, and Willingness — are designated gatekeepers and receive double weight, reflecting their role as non-substitutable prerequisites: a destination that is physically dangerous, legally inaccessible, or whose host community has explicitly refused to receive evacuees cannot be made viable by strong food supply or shelter alone. Willingness was added to the gatekeeper set because an unwilling host is as disqualifying in practice as a blocked checkpoint — modelling it as a standard (weight-1) factor previously let strong shelter/food/medical scores mask an unviable site.
Unknown vs. Unwilling: for the Willingness factor specifically, a blocked status is displayed as "Unwilling" rather than the generic "Blocked" — a confirmed refusal by the host community — which is a materially different epistemic state from "Unknown" (willingness has simply not been assessed yet). Both still score 0 in the readiness formula, but only "Unwilling" is a confirmed exclusion; "Unknown" is an intelligence gap that AI-assisted assessment could close.
| Status | Score | Rationale |
|---|---|---|
| Operational | 1.0 × conf | Fully functional; confidence-scaled |
| Partial | 0.5 × conf | Degraded function; confidence-scaled |
| Blocked | 0.0 | Non-functional; gatekeeper triggers cap |
| Unknown | 0.30 × (0.5 + 0.5·conf) | Epistemic conservatism — see §2 |
Uncertainty operates at two levels:
Factor-level confidence — each factor carries a base confidence value (0–1) generated at scenario creation, reflecting how well-assessed that factor is. The global Field Uncertainty slider scales all confidences multiplicatively:
Unknown status treatment encodes a key epistemic distinction: a known unknown is less dangerous than an unknown unknown. A factor assessed as "unknown" with high confidence (we are sure we don't know) scores higher than the same status with low confidence (we are not even sure our ignorance is accurate):
This models the real-world asymmetry in which a field team that reports "we could not assess food supply" provides actionable information, whereas a report produced under conditions where the team itself was unreliable provides less.
For each (evacuee group, destination) pair, 500 Monte Carlo runs simulate the outcome distribution arising from field uncertainty. In each run, every factor's assessed status is independently subjected to a perturbation test:
Success in a single run requires all three conditions:
The reported success rate is the proportion of runs meeting all three criteria. Standard deviation across runs quantifies outcome uncertainty — high SD indicates a volatile scenario where small real-world changes could flip the outcome.
Groups are first sorted by urgency (immediate → urgent → can-wait), then assigned greedily in that order. Each group is placed at the highest-scoring destination with sufficient remaining capacity. This prevents capacity contention, ensures critical groups are prioritised, and may result in an UNASSIGNED status if all sites are full.
Weights reflect the priority ordering in humanitarian evacuation planning: protection quality first, physical fit second, operational burden third, population needs fourth.
All data is generated from a seeded pseudo-random number generator using the Park-Miller Linear Congruential Generator (LCG):
Using a seeded RNG means every scenario is fully reproducible by seed number — a critical requirement for thesis research where peer reviewers must be able to regenerate results exactly.
Destination capacity follows an exponential distribution (λ = 3, mapped to [200, 5000]) to reflect the realistic skew of humanitarian site sizes: many small sites, few large camps. Factor status distributions are weighted differently for gatekeepers (more likely impaired in active conflict) versus standard factors.
Groups are drawn from eight ecologically valid archetypes grounded in ICRC field taxonomy and IHL protected-person categories. Each archetype has fixed vulnerability, special-needs, and urgency properties; only group size is randomised per scenario.
| Archetype | Vuln. | Needs | Urgency |
|---|---|---|---|
| Elderly & mobility-impaired | 5 | mobility | immediate |
| Families with children | 4 | — | urgent |
| Wounded & medical cases | 5 | medical | immediate |
| Unaccompanied minors | 5 | medical | immediate |
| Pregnant women | 4 | medical | urgent |
| Unaccompanied adults | 2 | — | can-wait |
| Mixed general population | 2 | — | can-wait |
| Journalists & aid workers | 2 | — | urgent |
The vulnerability_match sub-score in §4 rewards destinations with operational medical capacity when a group has medical or mobility needs — operationalising the IHL principle of proportionate protection for the most vulnerable.
Assumptions
Scope & intended use
Math.random() is not called anywhere, ensuring full reproducibility.Parameter reference — all values are in the PARAMS constant at the top of the script and are also displayed in the ⚙ Model Parameters panel above.
This tool uses no real-world datasets — every destination, evacuee group, and factor status is generated synthetically by the seeded RNG (§5). Nothing here is imported or fitted from real evacuation records. The sources below inform parameter design — which factors exist, which are gatekeepers, and where thresholds originate — not the data itself.
| Source | Used for |
|---|---|
| Park & Miller (1988), CACM 31(10) | Park-Miller LCG — the seeded random number generator underlying every synthetic value (§5) |
| ICRC (2013), Violence and the use of force | IHL framework grounding the Security gatekeeper factor |
| Additional Protocol I (1977), Art. 12 & 58 | Protection of medical units (Medical capacity) and precautions against attack (Security) |
| IASC (2007) Guidelines | Host-authority consent framework grounding the Authority consent gatekeeper factor |
| SPHERE Handbook (2018, 4th ed.) | Minimum standards — 3.5m²/person shelter, 15L/person/day water — behind the Shelter and Food & water factors |
| UNHCR (2007), Handbook for Emergencies (3rd ed.) | Source of the 40% minimum readiness viability threshold (§3, SUCCESS_MIN_READINESS) |
Full citation detail is in Evacuation_Simulator_Methodology.docx §14 (References) in the project repository. This table is a summary of that document, checked directly against it — the methodology doc is authoritative if the two ever diverge. Note the Willingness factor's gatekeeper promotion (§1) and the 500-run Monte Carlo default (§3) postdate methodology doc v1.0 and are not yet reflected there.