v17 · Epuyén calibration + smoothed tail + probability audit

MV Hondius Andes Virus Dashboard — Epuyén-calibrated peak tracker + probability audit

This version calibrates the epidemic-wave engine on the 2018–19 Epuyén, Argentina Andes-virus outbreak and explicitly accounts for the long incubation window. v17 also removes the artificial late-tail ripple that could create a small March bump after the wave should be decaying, and it adds a probability audit so the scenario weights are easier to understand.

Anchor: 13 May 2026 · Baseline declared cases/deaths: 11 / 3 · Declared contact universe minimum: 181 · Quarantine/monitoring minimum: 147+

Argentina 2018–19 calibration anchor: Epuyén Andes-virus outbreak

The model is now calibrated to a real human-to-human Andes-virus event, not an arbitrary exponential curve.
calibration
Calibration variableEpuyén 2018–19 valueHow it is used here
Confirmed infections34Defines an observed contained-outbreak final size after strong public-health action.
Deaths11 → CFR ≈ 32%Sets the stress-test fatality anchor; dashboard CFR sliders remain adjustable.
Contacts quarantined142Provides a contact-tracing intensity benchmark: ~4.2 quarantined contacts per final case, or ~7.9 per case at the 18-case intervention point.
Pre-control reproductive numberR ≈ 2.12Used as the missed-containment / extreme H2H growth anchor before control measures bite.
Post-control reproductive numberR ≈ 0.96Used as the post-intervention decay anchor once isolation/quarantine becomes effective.
Control trigger18 confirmed casesSeparates a controlled Epuyén-like response from delayed-recognition scenarios.
Interpretation: Epuyén shows that Andes virus can generate person-to-person chains when symptomatic high-viral-load cases attend social/contact-heavy settings, but also that isolation and quarantine can bend the curve. The dashboard therefore separates four pathways: contained tracing, hidden clusters, missed containment, and extreme out-of-control H2H.

Epuyén calibration and base-case fit

Observed Argentina / Epuyén anchor points against the fitted base-case curve used by the model.
Argentina benchmark
How to read it: the solid lines are the reported Epuyén 2018–19 anchors. The dashed lines are the fitted base-case calibration: pre-control growth uses R ≈ 2.12 per serial interval, then bends toward R ≈ 0.96 after isolation/quarantine. This is the benchmark used for the dashboard’s hidden-cluster/base-case behaviour, not the extreme stress-test curve.

Long-incubation retrace logic

Why a low visible count today cannot immediately rule out missed chains.
lag window
Missed exposure windowWhat could be missedWhen it can become visible
Index/prodromal period around early AprilClose cabin, caregiver, shared-room, cleaning/laundry and medical contacts before risk was fully recognised.Mid-April to early June under a 1–8 week symptom window.
Ship disembarkation and repatriationAirport, transport, flight, hotel, port, baggage, and household contacts not in the formal contact list.May to late June, especially if cases are mild at first.
Healthcare contacts before ANDV suspicionHCWs, specimen handling, waste/laundry, and family contacts of exposed staff.May to July if secondary chains begin before isolation.
Hidden tertiary chainsPeople infected by a missed mild or late-recognised case, rather than directly by a ship passenger.June onward; this is the branch that feeds the missed-containment scenario.
Probability impact: the dashboard discounts early reassurance while the incubation window is open. The posterior does not allow severe scenarios to collapse to zero merely because official counts are still low.
v17 curve fix: if you saw a small increase around late March in the previous version, that was a modelling artefact from deterministic wave roughness after the post-control phase. v17 enforces monotonic tail decay after control, so late bumps only appear if the scenario is still genuinely above-control.
Declared positive cases
11
latest extracted / entered
Declared deaths
3
reported mortality signal
Declared contacts universe
181
147 onboard + 34 prior disembarks
Declared quarantine / monitoring
147+
derived minimum; never forced to 0
Peak new deaths / fortnight
—
extreme H2H peak pace
2-year extreme deaths
—
stress test, not forecast

Epidemic pace: new infections per 2-week period

Default view is indexed to peak so the extreme scenario does not hide the other scenarios.
scale-safe

Scenario probabilities

Epuyén-calibrated planning priors, setting-adjusted priors, visible-data match, and guarded posterior.
probability
Clear interpretation: the red extreme curve is a stress test for missed containment and sustained human-to-human chains. It is not the central forecast. The dashboard keeps it visible because it defines the planning ceiling.
Reality tracking: the latest declared case/death point is compared against each scenario at the same date. If official counts rise faster than the lower scenarios, the posterior match automatically shifts upward.

Mortality pace: new deaths per 2-week period

Shows the top of the wave. This is the “peace/pace” signal missing from cumulative-only views.
peak deaths

2-year cumulative deaths

Cumulative tail over two years; separate from peak pace.
2 years

Reality vs four scenarios over two years

Blue dots are actual declared data. Scenario bands are calculated from the current settings.
auto-tracker

Model controls

These parameters control missed contacts, long incubation amplification, CFR and the Epuyén-calibrated missed-containment / H2H wave.
interactive

Auto-update public scan

Calls the deployed Cloudflare backend and updates actual declared cases/deaths/contacts/quarantine when extracted. Use the 60-day scan to review the full recent public record, not just the latest headline.
/api/latest
Ready. If opened from a local file, the dashboard will call the Cloudflare URL above; when deployed it uses same-origin /api/latest.
No scan yet.
Important: the worker is deliberately conservative. It should not overwrite the dashboard with broad unrelated hantavirus numbers unless they appear in MV Hondius / Andes-virus cruise context.

Peak scenario table

Readable numbers for all four scenarios.

Actual update log

Stored locally in your browser. Auto-scan adds/updates points.

Why the scale problem is solved

One scenario should not visually erase the others.
Indexed-to-peak view
Each scenario is scaled to 100% of its own peak. This shows wave timing and shape, even when absolute magnitudes differ by orders of magnitude.
Log view
Absolute values stay visible without the extreme scenario flattening the lower scenarios against the x-axis.
Selected-scenario view
When you need the true peak value for one scenario, isolate it and read the peak box/table.