Evidence-synthesis assistant estimate

Context

An estimate for a pharma R&D evidence-synthesis assistant. The build is manageable; the cost sits in a scientist-graded eval harness (slow, expert-bound) and in the validation and GxP documentation that regulated R&D requires before use.

The decision

Estimate the validation, not just the build: SME-graded evaluation and GxP documentation dominate a regulated R&D estimate, so size them explicitly — the eval loop is expert-rate-limited, not compute-limited.

What most miss

Teams estimate the retrieval build and forget that in pharma the eval must be graded by scientists and the whole thing documented for validation. That's the majority of the calendar, not a tail.

Stakes

Under-scope SME-graded eval and the estimate assumes a review capacity the science org doesn't have.