# MRI-003 — The Best Specimens

A local, five-view research prototype within the existing Survival of Fitness static site. No dependencies, remote fonts, analytics, external runtime services, or deployment step.

## Run

From the site root (`survival-of-fitness-site`), run:

```sh
python3 -m http.server 4173 --bind 127.0.0.1
```

Open [MRI-003](http://127.0.0.1:4173/mri/003/). The [static reading](http://127.0.0.1:4173/mri/003/static.html) contains all five views, all 25 seed records, numerical tables, and sources without JavaScript. The interactive page also retains its content when JavaScript is disabled. Evidence links require the supplied research folder to remain alongside `mri/`.

## Files and editing

- `page.html`: authored five-view template. Edit this, then rebuild; do not edit generated HTML.
- `mri003.css`, `mri003.js`: isolated layout and progressive enhancement.
- `data/editorial.json`: name-light display paraphrases, overlapping mechanism tags, and fields supporting each mapping. This is an explicitly labeled editorial display layer, not adjudicated research coding.
- `data/prototype.json`: generated reusable dataset, including unchanged seed rows, all 19 contributors, all 31 source records, numerical transcriptions, preserved blanks, supplied results, and actual input hashes.
- `code/build_site.py`: standard-library generator for `index.html`, `static.html`, and `data/prototype.json`.
- `code/verify.py`: preservation, arithmetic, deterministic-build, data, and local-link checks.
- `FINDINGS.md`, `VERIFICATION.md`: bounded findings and the pre-implementation verification record.

The requested `/research/war-of-the-worlds-part-1/` path was not present. The actual supplied packet is `../../War_of_the_Worlds_Part_I/`; it remains in place and byte-identical to its manifest.

## Rebuild and verify

From the site root, Python 3.10 or later:

```sh
python3 mri/003/code/build_site.py
python3 mri/003/code/verify.py
```

The verifier runs the original `code/check_galton.py` in a temporary copy. It requires exact equality for the summary and rounded arithmetic; only unrounded `normal_integral` and `modern_literal_value` fields allow platform-level floating-point tolerance (`rel_tol=1e-12`, `abs_tol=2e-12`). It checks that the original packet still matches every supplied manifest hash.

The packet’s original reproduction command is `python code/check_galton.py`, run from its own directory. That command regenerates the packet’s `results/`; use the temporary-copy verifier above to preserve the supplied bytes. No analysis parameters have been changed.

## Reading and interaction

1. **Best specimens:** select a publication contributor, then a seed proposition. Discussion and written contributions remain distinct.
2. **Remove the names:** three source-linked pathways summarize difference → explanation → action. Inspect overlapping mechanisms and optionally restore attribution. Original wording and coding remain in the source panel.
3. **Build the representation:** step through source, grouping, approximation, model, and output. Table I’s divided normal curve leads to reproduced class counts; Table II’s paired bars compare the reproduced regrouping with the theoretical normal rates. Select a Table II group to trace its allocation; highlighted chart bars remain whole-class totals.
4. **Where is “best”?** expand each criterion to distinguish what is specified from what is operationalized or established.
5. **Description → direction:** inspect explanation/action connections within selected seed records, including separate Wells and Hobhouse propositions. A concluding observation → representation → interpretation → action graphic distinguishes checked arithmetic from the further criteria and judgment required for application.

The new pathway labels are stored in `data/editorial.json` under `flow_pathways`, with their claim IDs and supporting seed fields. They summarize claims, not observed causal effects. Table I’s curve uses the source’s probable-error boundaries; its drawing is cropped at ±5.2q, while the preserved counts include the complete infinite tails. That crop affects only the drawing, never the calculation. The figures are generated into both reading versions and require no animation or additional runtime library.

All controls use native links, buttons, or disclosures. Navigation supports history and fragment links; source panels use a native modal dialog with Escape and focus return. Reduced-motion preferences are respected. Mobile navigation scrolls horizontally within its own strip; tables have keyboard-focusable scroll regions. Static reading retains named attributions and native disclosures.

## Evidence boundary

This is a presentation of the supplied packet, not a new historical audit. Seed paraphrases are single-pass examples, not a complete or independently reviewed corpus. Original scans, upstream Booth reconstruction, exact Table III replication, and Figure/Table IV verification are absent. The page does not substitute invented data, fit model parameters, rank contributors, or infer ideology from the display categories.

The page is linked locally from `/mri/`. No deployment or publication was performed.