Summer Movie Wager — Bayesian Outcome Visualizer

Interactive Julia project that models the final Top 10 domestic box-office ranking of Summer 2026 as a probability distribution, then derives Summer Movie Wager scores and win probabilities from it.

Live viewer: snoble.github.io/smw

See docs/brainstorms/2026-07-13-summer-movie-wager-bayesian-brainstorm.md for the design decisions.

Setup

Requires Nix. From the repo root:

nix-shell
julia --project -e 'using Pkg; Pkg.instantiate()'

The first instantiate can take 10+ minutes (Turing + Makie precompile).

Update weekly data

1. Edit data/weekly_2026.csv — add a new row per film with the latest cumulative domestic gross and date.
2. Optionally update data/films_2026.csv registry fields (released, notes, etc.).
3. Re-run the notebook (or scripts/smoke.jl); posterior draws cache to output/.

Run tests

nix-shell --run 'julia --project -e "using Pkg; Pkg.test()"'

With line coverage (writes lcov.info, fails below 95%):

nix-shell --run 'julia --project scripts/coverage.jl'

Launch the notebook

Pluto cannot load a local package through its built-in package manager. The notebook activates this project's environment on startup.

nix-shell --run 'julia --project -e "using Pluto; Pluto.run()"'

Then open notebooks/smw2026.jl.

If you see UndefVarError: Slider not defined, the notebook is using a stale buffer — close the tab and reopen the file from disk (CairoMakie also exports Slider; we alias PlutoUI's as UISlider).

Smoke run (no Pluto)

nix-shell --run 'julia --project scripts/smoke.jl'

Static browser viewer

The viewer conditions the full 2026 slate in-browser: fitted historical type priors and raw interval observations are shipped as a versioned input bundle (no posterior draws). A Web Worker runs nested quadrature + progressive Monte Carlo (1k preview → 20k final). A Julia/WasmTarget numeric kernel is compiled for the same ABI (site/wasm/smw_kernel.wasm); the worker currently uses a JS fallback for draws and reports wasm binary present when the module loads.

Design notes: docs/brainstorms/2026-07-13-julia-wasm-bayesian-engine-brainstorm.md.

# Regenerate site/smw2026_data.json after model or data changes (no NUTS)
nix-shell --run 'julia --project scripts/export_viewer_data.jl'

# Recompile the WasmTarget kernel (Julia 1.12 local toolchain)
.julia_versions/1.12.6/bin/julia --project=wasm scripts/wasm_spike.jl

# Open in a browser
./scripts/serve_wasm_site.sh
# → http://127.0.0.1:8765/smw2026_wasm.html

Commit the regenerated site/ directory; GitHub Actions deploys it to GitHub Pages on every push to main.

Season-total model: already banked + positive remaining gross (geometric decay pinned to the latest observation; theater-tail soft constraint near end of run).

Data files

File	Role
data/films_2026.csv	Candidate universe (registry + optional opening_prior_m for unreleased)
data/weekly_2026.csv	Cumulative gross + theaters + source over time
data/exclusions_2026.csv	Pre-window titles excluded from the wager
data/picks_2026.csv	Player Top 10 + dark horses
data/historical/films_history.csv	Training trajectories (synthetic placeholder for now)
site/smw2026_wasm.html	Fully-static interactive viewer
site/smw2026_data.json	Versioned input bundle: priors, intervals, picks (generated)
site/smw_engine_worker.js	Progressive inference Web Worker
site/wasm/smw_kernel.wasm	WasmTarget-compiled numeric kernel