# Meridian specimens · r2

Seven fictional specimens, one uncollected trace, two preparation steps, six declared custody actions and five exact-specimen laboratory results. Sourcebook text and tables are retained; no physical collection or analysis occurred. The original r1 packet is unchanged.

Uses the existing observation 0.1.0, research-entities 0.4.0 and claim-history 0.5.0 contracts at canonical commit `88bcbfcc8b17af30837313aff596cec8e1bde43a`. Build that checkout, then run with Node 22:

```sh
node build.mjs /path/to/disclosureos
```

Canonical links express lineage, predecessors, method references, analyses and exact result specimens. Nominal amounts are sourced observation measurements, with attributed statements about each material. `index.json` associates inventory rows with measurements for the application; it is not an additional standard material-mass field. Amounts are initial allocations, never inferred current remaining balances. Parent and A were split; their amounts are not added to those of their descendants.

The sourcebook's local `analysis` handling label for opening the parent jar maps to canonical `other`, with the original wording retained in input.json and explanatory notes. No completed assay is inferred. Custody begins separately for derived portions; blank/control/A1 do not inherit a fabricated chain. Day-only dates remain day-only.

AI-generated soil/jar illustrations are retained PNG bytes, accompanied by exact prompts and fingerprints in artwork.json. They cannot be deterministically regenerated by the adapter, and supply no mass, size or composition. The seal label is a deterministic drawing of source values. All gallery images describe the parent at collection, before division. No illustration is labeled as an image of an unpictured child or control.

Quantitative copper results retain standard uncertainties. Below-limit results are bounds, not zero. The liquid blank (mg/L) is not plotted with soil concentrations (mg/kg). The mean of three fictional A1 readings is reproducible; the declared uncertainty is not calculated from their spread. The result's method and basis remain inspectable.

Canonical evaluation checks structures and references. Source-table and artifact checks establish local byte consistency, not authenticity, custody verification, mass balance, laboratory adequacy or an origin finding. Rich microscopy/spectrum artifacts, material-type variants and broader custody edge cases remain later catalog items. No npm publication.
