Provenpair

Biotech business-development marketplaces increasingly use automated scoring to rank assets, partners or cohorts, yet the operator may struggle to reconstruct why a particular match surfaced. The supplied research reports an approaching European AI governance deadline and confirms horizontal documentation products, but finds no reviewed product focused on evidence for pharma marketplace matching. Whether a particular marketplace use is legally high-risk remains an applicability decision, not a product-generated fact.

Provenpair sits beside an existing matching system and records the source documents actually used, their versions and permissions, the scoring configuration, model release, uncertainty, exclusions and reviewer action for each recommendation. It produces a portable evidence bundle and a technical-documentation draft, while preserving the difference between a source assertion, an extracted feature, a computed score, a reviewer conclusion and a later commercial outcome.

Cryptographic signatures and provenance formats can support origin and integrity; they do not prove that evidence is true, representative, legally sufficient or scientifically valid. Templates can organize a customer's work but cannot classify the system, complete conformity assessment, register it with an authority or guarantee compliance.

The first release should cover one marketplace, one narrowly defined match class and retrospective evidence generation before any live decision support. Marketplace configuration, recommendation, human review, recipient disclosure, technical-file approval, regulatory action and partnership result remain separate.

Product, risk or compliance leader at a biotech business-development marketplace that operates an automated matching layer