Deformulation: rebuilding a marketed product's formula from public evidence

A surprising amount of a product's formulation is disclosed somewhere - spread across jurisdictions, written for different purposes, and never assembled in one place. The raw material for a reconstruction includes:

• Labeling and package leaflets: US labels list inactive ingredients; EU SmPCs list excipients and sometimes their function; some jurisdictions disclose more than others.
• Regulatory review documents: FDA and EMA assessment reports, chemistry reviews, and inspection-adjacent filings reveal composition ranges, process descriptions, and critical quality attributes.
• Excipient databases: FDA's Inactive Ingredient Database caps each excipient's maximum potency per route, bounding what any compliant formula can contain.
• Patents and literature: formulation patents describe the solved problem and its parameter space; patents a product actually practices are findable through Orange Book and litigation records.
• Analytical and dissolution data: published dissolution profiles, stability summaries, and bioequivalence review data expose how the product behaves - the fingerprint the candidate must match.
• International filings: the same product filed in multiple regions often discloses different details in each; triangulating jurisdictions is where most reconstructions are won.

The reconstruction itself is constraint solving: each source rules out part of the possible formulation space until a small region remains - a candidate formula and process consistent with everything the product is publicly known to be.

No honest reconstruction can claim certainty on every figure. A label may name the disintegrant but not its percentage; a patent may describe a range the product narrows to a point; a dissolution profile may confirm the release mechanism but not the granulation liquid. A dossier that presents every number as equally known is doing the buyer a disservice - and a diligence team will catch it.

So each claim in a deformulated dossier carries a provenance tag: verified (established in a cited record), inferred (a reasoned estimate with the reasoning shown), or hypothesized (a candidate value flagged for experimental test). The buyer sees exactly which figures are bedrock and which are engineered estimates, and the first lab batch is designed to confirm the hypotheses - not to gamble on them.

A boundary worth stating: deformulation does not recover the innovator's master batch record. It produces a defensible point inside the region of formulations that match the reference's observable behavior - which is what a licensee actually needs.