# Targeted formulation challenge set — 12 September 2026

This supplements the earlier 16-pair ingredient assessment with five provisional source-complete formulation cases and three real near-matches. It is a targeted engineering challenge set, not an untouched holdout, random accuracy sample, clinical validation, or search-completeness measurement.

The source assessor inspected raw registry fields and the current matcher definitions for terminology but did not run the matcher or see predictions for these pairs. Labels and captures were frozen first. All study objects are unedited ClinicalTrials.gov API v2 records. Retained response hashes were checked before extraction; newly fetched record response bytes were captured unchanged. Canadian product rows were decoded from the approved 58,034-row dictionary index after verifying compressed and decompressed hashes against the release descriptor.

| Case | Canadian DIN | Registry study | Provisional source label | Main challenge |
|---|---|---|---|---|
| FORM-01 | 02258595 | NCT00963313 | Formulation supported | 40 mg/0.8 mL adalimumab solution, subcutaneous; explicit syringe contents; arm type is absent but active administration is explicit |
| FORM-02 | 02378604 | NCT03630055 | Formulation supported | 15 mg rivaroxaban oral tablet; separate no-intervention arm |
| FORM-03 | 02378612 | NCT05128591 | Formulation supported | Separate 20 mg reference and 18 mg test tablets in crossover arms |
| FORM-04 | 02274906 | NCT00448669 | Formulation supported | Single oral tablet containing both selected ingredients and strengths |
| FORM-05 | 02274906 | NCT00458393 | Formulation supported | Full chemical names/amounts explicit in one co-formulated oral tablet; trailing acronym order ambiguous |
| FORM-06 | 02316986 | NCT03630055 | Selected formulation not established | Selected 10 mg tablet versus actual 15 mg tablet |
| FORM-07 | 02303922 | NCT02202161 | Selected formulation not established | 50 mg twice daily does not establish a 100 mg tablet; salt/hydrate not explicit |
| FORM-08 | 02274906 | NCT03675815 | Selected formulation not established | Separate TDF tablet and optional FTC capsule, not one fixed-dose tablet |

FORM-05 must not decode the trailing TDF/FTC acronym positionally: its order differs from the preceding fully spelled ingredient order. The chemical names, amounts and co-formulation statement are sufficient for a human source-level formulation assessment; whether a conservative parser can establish the same relationship is the point of this challenge. Do not alter this label after seeing the matcher result. If an independent review identifies a label defect, preserve the original freeze and report a separately versioned adjudication.

The public source assessment includes exact field paths, supporting excerpts, product URLs, registry URLs, capture hashes and remaining identity gaps. Full source bytes, original first-page response captures and product rows are in the private companion directory. The additional fetched NCT02449733 record was screened but excluded because the eligible optional-offer context adds exposure ambiguity beyond the bounded eight-pair selection; its unmodified capture is retained for reproducibility.

To score the set, compare `expected_formulation_match` with a matcher tier of `same_combination` (or a separately justified `linked_product`). Preserve each case's actual tier and reason. Record missing links separately from false promotions. A negative formulation label does not require the record to be unrelated at the ingredient level. There is no independent human approval and no exact Canadian product link in this source-only assessment. Report counts, not a database-wide accuracy percentage.