Charniga et al. 2024 checklist
The table below maps each Table 2 checklist item from Charniga et al. 2024 to where (and how) it is addressed in this analysis. ✅ = done, 🟡 = partial / done with documented caveats, ❌ = not done. We have tried to be honest about partial coverage rather than claim more than the deposit supports.
Estimation
| # | Charniga 2024 item | Status | Where in this repo | Note |
|---|---|---|---|---|
| 1 | Adjust for censoring (always), right truncation (when needed), and dynamical bias (when needed); state which adjustments were made | 🟡 | src/model.jl and Model | Double interval censoring at 1-day bins on both endpoints via CensoredDistributions.double_interval_censored. No right-truncation or dynamical-bias adjustment because the 2012 Isiro outbreak is fitted retrospectively in 2026 on the closed Rosello deposit — neither bias applies. |
| 2 | Fit more than one probability distribution and use model-selection criteria; visualise the fit | ✅ | Analysis walkthrough — Family comparison | LogNormal, Gamma, and Weibull all fitted; WAIC ranking + side-by-side posterior-predictive panels per family. Gamma wins. |
| 3 | Convert distribution parameters to summary statistics correctly | ✅ | Model — Common parametrisation | All families share a log-mean / log-shape parametrisation; conversion uses Distributions.jl built-ins (Gamma(k, μ/k), Weibull(α, μ/Γ(1+1/α)), LogNormal(log_median, log_sd)). Documented in a single table. |
| 4 | Stratify or add subgroups when sample size allows and group differences are hypothesised | 🟡 | bdbv_model_stratified and Model | HCW stratification implemented for the delay block (β_*_hcw log-mean shifts on Gamma); HCW, case definition (Probable vs Confirmed), and standardised age in the CFR block. Sex is recorded but not stratified — Limitations explains why (Kratz 2015's community-vs-ETC stratum is not reproducible from the Rosello deposit). |
| 5 | Check model diagnostics (R̂, divergent transitions, ESS) | ✅ | Analysis walkthrough — Family comparison | Per-family table reports max R̂, min bulk ESS, and divergent transition count. Across all three families: max R̂ ≤ 1.002, min ESS ≥ 3,200, 0 divergent transitions. |
Reporting
| # | Charniga 2024 item | Status | Where in this repo | Note |
|---|---|---|---|---|
| 6 | Report measures of central tendency and variability (mean and SD/variance/dispersion) | ✅ | Analysis walkthrough — Headline estimates and — Derived (convolved) marginals | Median, mean and SD reported for every atomic and convolved delay (the convolved-marginals table gained an sd (95% CrI) column once _convolve_delays started returning per-draw SDs). Per-draw shape/scale/SD also exposed in posterior_<family>.csv. |
| 7 | Report key quantiles (e.g. 2.5, 5, 25, 50, 75, 95, 97.5, 99) of the distribution | 🟡 | Analysis walkthrough | The 50th percentile is reported as the median; the 95th percentile is reported for the convolved onset → death and onset → discharge marginals. A full quantile grid (5/25/75/95/99) is not produced — output/ posteriors allow it to be computed downstream. |
| 8 | Report the fitted-distribution parameters | 🟡 | output/ posterior CSV; parametrisation table in Model | The on-page tables emphasise summary statistics (median, mean, 95% CrI). Latent log-mean / log-shape draws are written to the main-latest rolling-release CSV bundle so the underlying Gamma (shape, scale), LogNormal (logmean, logsd), or Weibull (shape, scale) can be reconstructed exactly. |
| 9 | Report uncertainty (90% or 95% intervals) on every estimate | ✅ | Analysis walkthrough throughout | All headline numbers, derived marginals, CFR strata, and log-OR coefficients are reported as median (2.5% – 97.5%). |
| 10 | Report study-sample characteristics (sample size, age, sex, location, route of exposure, vaccination status if any) | ✅ | Analysis walkthrough — Outbreak context; README — Data provenance; Limitations — Data | The walkthrough opens with an "Outbreak context" block tabulating n, onset window, age (median + IQR), sex split, HCW split, case-definition mix, and outcome counts directly from data/linelist.csv. Route of exposure is not recorded in the deposit; no vaccine was available in 2012 (vaccination status not applicable). |
| 11 | Report the epidemic curve and any control measures in place | 🟡 | Analysis walkthrough — Epidemic curve | Weekly onset epi curve with HCW subcounts is rendered live on the docs site. Control measures are mentioned narratively in Limitations (MSF ETC support; rural mining-influenced setting) rather than as a structured timeline. |
| 12 | Provide anonymised / de-identified line-list data and documented code | ✅ | data/linelist.csv; src/; output/ rolling release | The bundled CSV is Rosello et al. 2015's supplementary file 1 (CC-BY 4.0) subset to BDBV — already anonymised at source. All Julia code (model, post-processing, plotting) is MIT-licensed in this repo. The full executable walkthrough is published as the docs site, and posterior CSVs + figures ship as a main-latest GitHub release. |
Items 13+ (Table 3 — incubation period & serial interval)
Charniga et al. 2024 Table 3 adds recommendations specific to incubation-period and serial-interval estimation (exposure-window encoding, transmission-pair confidence, transmission-direction ordering, support for negative serial intervals).
Not applicable here. The Rosello deposit has no exposure dates and no transmission pairs, so neither the incubation period nor the serial interval is estimated in this work. See Limitations — Data for the rationale and the MacNeil 2010 / Zaire-EVD priors recommended for downstream nowcasting or Rₜ work that needs these distributions.
How to update this page
When the model or reporting changes, edit the table directly. Status should err on the side of 🟡 over ✅ whenever a documented caveat exists. The LLM-friendly column layout (Charniga item ↔ status ↔ location ↔ note) is deliberate — it lets future automated reviews re-verify each row by following the link in the third column.