Milestone 0: Rwanda district-level VCI smoke test

[goldingn]

🤖 Posted by Claude (Opus 4.8), an AI coding agent, from the @goldingn account — not written by a human. Why.

Milestone 0: the first end-to-end run of the package — a district-level VCI
smoke test for Rwanda, built (per DESIGN.md §14) to shake out architecture and
integration problems before the science is signed off. It computes vector control
impact across Rwanda's 30 GADM admin-2 districts for three scenarios: no
intervention (reference), business-as-usual (current net use, all pyrethroid
LLIN), and an alternative (the same use, all dual-AI LLIN).

Illustrative only. Ballpark parameters and a clearly-labelled intervention
stub — the outputs are not a real analysis, and every surface says so.

The kernel is pure base R (zero runtime Imports); sf/terra/geodata and
ggplot2/patchwork are Suggests, used only in data-raw/ and the vignette.
devtools::check() is clean: 0 errors, 0 warnings, 0 notes; 93 test
assertions plus a pinned integration snapshot.

The eight commits are thematic and dependency-ordered: scaffolding → capacity
kernel → host-choice + intervention stub → input contract → dataset → pipeline →
vignette + integration test → session log.

How this was produced

• Issue: Milestone 0: minimal working example (district-level VCI smoke test) #45 (and the Refine Rwanda smoke test example #64 refinement)
• Core prompt(s): build the Milestone 0 smoke test — Rwanda, admin-2, pyrethroid
  vs dual-AI LLIN, three scenarios; test-first kernel following the maths spec;
  prepare district inputs in data-raw/ from provided Africa rasters; a vignette
  with clean ggplot2/sf figures of the inputs and of capacity/VCI.
• Key decisions / trade-offs:
  • Kernel has no runtime dependencies — spatial and plotting packages are
    Suggests (data-prep/vignette only), so R CMD check stays free of unused
    Imports.
  • Intervention mapping is a stub (intervention_effect_stub()), not the
    real Decide how interventions become effects, and how combinations of them compose #17 design — plausible round numbers, loudly labelled, capturing only
    that pyrethroid killing collapses with resistance while dual-AI killing does
    not.
  • Host-opportunity weights derive from the malariasimulation Q0 and
    phi_indoors under an explicit, documented equal-host-availability
    assumption (relates to Choose the final symbol for host-destination fractions #13, Confirm how human landing catch data connect to abundance and biting rate #16) — adopted knowingly, not silently.
  • Abundance layer is m·a, so it is divided by the feeding rate to recover
    adults-per-human m. This surfaced that survival and abundance are decoupled
    here (VCI is invariant to absolute abundance) — recorded as Survival and abundance are decoupled: VCI is invariant to absolute abundance #63.
  • Temperature (Refine Rwanda smoke test example #64): biting_rate() and eip() are illustrative Brière
    thermal responses (Mordecai et al. 2013 / Villena et al. 2022), driven by a
    WorldClim temperature layer, so cool high-altitude districts (Burera,
    Musanze, Nyabihu) now show near-zero capacity rather than an implausible
    moderate value.
  • Tooling: .lintr defers line-width and indentation to Air (the project
    formatter); the NEWS.md heading uses the numeric dev version so R's NEWS
    parser recognises it.
• New dependencies (all Suggests, none at runtime): sf, terra, geodata
  (data prep); ggplot2, patchwork (vignette figures); knitr, rmarkdown
  (vignette builder). patchwork was flagged and approved.
• Session log: dev/sessions/2026-07-24-54f0b094-iss45-milestone-0-smoke-test.md

Fixes #45
Fixes #64

🤖 Generated with Claude Code