P R I S M

Method changelog

Material changes to what Prism measures, how it reasons, and what readers see.

This is not a source-control log. It records changes that alter the epistemic contract, evaluation, public evidence or interpretation of an edition. Minor repairs, typography and ordinary maintenance do not appear. A change to an already published report belongs on Corrections as well.

17 JULY 2026 · RESEARCH RECORD

Prospective evaluation and publication types introduced

Prism now distinguishes working papers, prospective protocols, calibration notes and results. Evaluation Protocol 01 freezes selection, error taxonomy, measures and decision gates before the next editorial-2 run. The existing canaries are explicitly described as predecessor observations rather than validation of the current contract.

17 JULY 2026 · EDITORIAL-2 CONTRACT

Model-judged membership; code-derived breadth

The principal editor now supplies one account list and the article IDs it judges to carry each account. Code derives broad, clustered and sole-source circulation. The model no longer chooses a prevalence band or writes counts.

Every comparative item carries priority and why it matters. Optional chronology/referent orientation and a tightly capped largely-unasked-question category were added. Both may remain empty and receive explicit scrutiny from the critic and final article audits.

17 JULY 2026 · EVIDENCE CONTRACT

Evidence links bind to stable publication targets

Each public evidence use now names one exact target such as a headline, map field or report paragraph. Code validates the target and requires the excerpt to occur within it. Legacy excerpt-only reports remain renderable, but new runs cannot satisfy an evidence link with the same words found elsewhere in the publication.

17 JULY 2026 · READER CONTRACT

Orientation, synthesis and comparative anatomy separated

A short high-priority coverage brief now precedes the analytical report. The complete comparative map follows the essay. This gives readers a rapid orientation without making the database-shaped anatomy the medium in which the report must think.

16 JULY 2026 · PIPELINE SIMPLIFICATION

Prose-first briefs and final raw-article audits

Source briefs became the editor’s primary input; complete quote catalogs moved into a separate verification context. A sparse correspondent hotdesk replaced routine recursive questioning. Every article now audits the near-final publication once, and only major or critical findings open a narrow repair followed by affected-article recheck.

16 JULY 2026 · LOCAL HARNESS

Model conversations made non-authoritative

Run, work-item and artifact records became resumable by complete input hash. Validated result files land before database completion commits. A missing provider thread can therefore be reconstructed from frozen inputs instead of losing paid work or making conversation state part of the evidence chain.

CURRENT PUBLIC CONTRACT