Option A — The General Case Renderer

A short brief on how NZIPL intelligence cases are stored, rendered, and scaled.

The system separates what is drawn (one shared renderer) from what is drawn about (many tiny per-case data files). One renderer, N cases. A “case” is a single technology × country pair (e.g. Solar × Brazil).

1. The principle: renderer / data split

                build_datadriven_proto.R
   pipeline  ──────────────────────────────►   data/*.json   (one per case, ~50 KB)
 (BACI, PC,                                          │
  S&P, sankey)                                       ▼
                                            viz.js  (14 draw fns)
                                                     │
                       ┌─────────────────────────────┼─────────────────────────────┐
                       ▼                              ▼                              ▼
                  report.html                    panel.html                 report_builder.html
                 (print / PDF)              (interactive explorer)            (IR composer)

Cases are pre-aggregated at build time, not computed at view time. A reader just fetches a small JSON and the renderer draws it synchronously (data-in → SVG-out).

2. The data layer — `data/`

scripts/proto/build_datadriven_proto.R reads the real pipeline (cache/bilateral_ds, data/pc/*, S&P firms, dendrite, sankey) and writes:

File	What it holds	Count
`<Tech>_<Country>.json`	one case: `timeline · partners · products · radar · scatter · pc · firms`	253 today
`_tech_<Tech>.json`	country-independent tech-tree edges + value-chain sankey (fetched once on tech change)	11
`_index.json`	switcher manifest: which techs/countries exist, slugs, colors, region sets	1

Each case is a bundle of pre-shaped arrays — exactly what the charts need, nothing more.
Whole-case payload ≈ 50 KB; all 253 cases ≈ 13 MB total.
Per-case files never carry the tech tree / sankey — those live once in _tech_*.json to keep payloads lean.

Build command:

Rscript scripts/proto/build_datadriven_proto.R [--countries all|ISO,ISO]

Default builds the focal + regional sets listed in data/case_registry.csv.

3. The renderer — `viz.js`

One file, ~14 draw functions, each data-in → SVG-out, synchronous — no iframes, no async draw. Give it a target SVG selector + a slice of a case, it draws.

radar · scatter · timeline · treemap · map · firms · sankey · tree
+ setTheme · initGeo · fmtV · highlights

Because every function reads its “ink” colors (axes, labels, land, strokes) from the active theme T, the same code renders the dark interactive screen and the light PDF. The PDF path just calls VIZ.setTheme('light') and redraws. Data colors (value-chain SR palette, HS categories, sankey lanes) are fixed — they read on both backgrounds.

Two copies, by design. panel.html carries its own inline draw code (it predates the extracted viz.js). Renderer changes that must apply everywhere — e.g. the domestic-zoom activity map — land in both viz.js and the inline copy in panel.html.

4. The three consumers

Consumer	Role	Theme
`report.html`	Print / slide path → the PDF deliverable	light on print
`panel.html`	Interactive dark explorer	dark
`report_builder.html`	Selector that composes an IR from chosen panels	dark

All three read the same data/*.json through the same renderer logic.

5. Registries — single source of truth

File	Drives
`data/case_registry.csv`	which countries / region-sets get built (35 rows → 253 cases)
`data/module_registry.csv`	which visuals (the ~14 modules) exist and how they’re labelled

Both the build script and the builder’s selector read these — change the registry, rebuild, and the new case/visual appears across all three consumers automatically.

6. How it scales

The split is the scaling story. The renderer is fixed; cost grows only with data (≈ 50 KB × N countries × 11 techs).

Today: focal + regional countries → 253 cases, 13 MB.
Full grid: --countries all → ~158 × 11 cases. The build script prints the live projection on every run (→ all 158×11 at this rate: ~X MB, ~Y min).

To add a country: add a row to case_registry.csv → re-run the proto build → it shows up in the report, the panel, and the builder with zero renderer changes.

Files at a glance

analysis/_proto_datadriven/
├── data/
│   ├── <Tech>_<Country>.json    # 253 per-case bundles
│   ├── _tech_<Tech>.json        # 11 tech-level side files (tree + sankey)
│   └── _index.json              # switcher manifest
├── viz.js                       # shared renderer (14 draw fns + theme system)
├── geo.js                       # inlined world topojson (renders under file://)
├── report.html                  # print / PDF path
├── panel.html                   # interactive explorer (own inline draw code)
└── README.md                    # this brief

scripts/proto/build_datadriven_proto.R   # the builder
data/case_registry.csv                    # which countries
data/module_registry.csv                  # which visuals
analysis/intelligence/report_builder.html # IR composer

Oriol Vallès Codina