Editorial, AI & Corrections Policy

Source material and AI assistance

The platform collects public regulatory announcements and structured datasets. Software extracts and joins selected fields, compares snapshots, and prepares display records. Regulatory summaries, scientific and legal context, and specialist commentary can be generated with AI assistance. An AI-generated factual summary remains a representation of its source, rather than the source itself.

Automated checks validate selected fields, source availability, and article-scoped presentation. They do not establish that every proposition is accurate. Human review is claimed only where a record expressly identifies it. General references to curated catalog material do not mean that each article, match, or subsequent data refresh has received individual review.

The research guide and worked example were prepared with AI assistance and primary-source checks. They are published by PharmaSignal. No personal authorship or substantive review by Saman Salmasi is claimed for the worked example pending his review.

What “verified changes” means

Where this phrase appears in older material, it means that the platform detected a difference between accepted source snapshots, such as a trial status, enrollment count, or patent-listing field. The current interface calls these source changes. Detection establishes a recorded difference, not independent verification of the underlying event. “Observed” dates identify platform detection; they are not necessarily event dates.

Upcoming milestones are calculated from recorded dates and are distinct from observed changes. The first accepted snapshot establishes a baseline. Missing or rejected snapshots can delay detection, and the timeline is not an exhaustive event history.

Impact analysis: method 1.0

The 0–100 impact score is an AI-assisted prioritization aid. The ingestion prompt requests four components, each on a 0–25 scale: scientific novelty, regulatory precedent shift, market relevance, and legal or enforcement exposure. These equal ranges describe the intended composition; they do not represent empirically validated weights.

The interface uses an available numeric total, bounded to 0–100. If the total is missing, it sums the available components and treats absent or invalid components as zero (with each component bounded to 0–25). A zero arising from missing data is not evidence of no significance. The component display identifies missing values separately. The ingestion prompt limits market scores for generic or multi-holder actions and reserves higher values for identified public issuers with confirmed exposure; much of the scoring remains model judgment.

The scale is retained because it is used in existing filtering and publication decisions: the dashboard excludes non-US signals at or below 20, and the static article builder requires at least 35. These thresholds shape coverage. A score is not a probability, a legal risk assessment, a measure of evidentiary reliability, or a prediction of clinical or commercial outcomes. Small numerical differences should not be treated as meaningful precision. This method version documents the inspected pipeline; historical article payloads do not consistently retain a scoring-method version.

Corrections and unresolved records

Use Report an Issue to identify an affected page or record and the supporting source. Reports may concern factual errors, broken links, ambiguous matches, omitted records, or misleading labels. The owner can then compare the report with the original material and determine whether the display, catalog, or processing rule requires correction. Reporting does not automatically alter a source or guarantee a response time.

Substantive corrections and changes to interpretation methods are recorded below. Routine changes in upstream datasets are identified through their timestamps and, where available, the source-change timeline. An unresolved limitation should remain visible while it is being investigated.

Method and correction history

October 9, 2026 · version 1.0

Published the research guide, source register, and application-specific apixaban example. Added contextual evidence and limitation disclosures. Clarified CHMP-stage labels, source-change terminology, and the distinction between listed expiration dates and commercial estimates. Recalculated coverage statistics using explicit definitions. Documented the missing NDA 202155 association in the platform’s apixaban snapshot; the underlying dataset remains incomplete. Repaired eleven source destinations: eight obsolete paths, two malformed Health Canada URLs, and one retired company hostname. These link corrections do not change the underlying source dates or represent a re-verification of historical summaries.