lin-0241

20.5.3 Product and review foundries

A product-review foundry declares restriction ports for product identity, review corpus, rating metadata, theme surface, claim evidence, and negative-review rebuttals. A broader product foundry can then join those charts to store experience, assortment, brand promise, and corporate context.

This decomposition prevents three common collapses. Reviews for adjacent products are not evidence for the focal product merely because their names are similar. A frequent review theme is not automatically a population-level fact. And a brand promise cannot be inferred from consumer sentiment without a declared bridge to corporate evidence. Product identity, source provenance, corpus coverage, and negative evidence remain first-class obligations.

Experiment: frozen trustworthy-foundry audits

The GLP-1 atlas records 3,376 extracted events, 191 local predictive-state contexts, and 190 restriction checks; the Adobe 10-K workflow records 46 candidate claims, of which 28 are supported and 18 retained with qualifications. These are descriptive artifact audits demonstrating retained provenance, scope, and obstruction states. They are not population-level estimates of medical, financial, or product-answer correctness.

Foundry	Unsafe global collapse	Localizing cover	Characteristic gate
GLP-1 research	One verdict across trials, safety reports, populations, and mechanisms	Population, intervention, outcome, evidence level	Regime-compatible clinical claim with explicit qualifier
Corporate 10-K	Fluent strategy narrative detached from numbers and risks	Statement, risk, management, strategy, market	Source anchor plus exact unit, period, and rebuttal
Product research	Aggregate sentiment transported across products, stores, and brand claims	Identity, reviews, ratings, themes, store, brand	Product-local evidence and retained negative-review rebuttal

Table 20.3 Trustworthiness depends on the domain-specific cover and admission law, not on a universal confidence threshold.