Funnel A candidate-discovery run

_Generated by pipeline/rdh_diff.py in 26s. Data snapshot: UNII 4Aug2026 · Drugs@FDA 12 Oct 2026 · Orange Book 12 Oct 2026 · EMA 12 Oct 2026 · Health Canada 12 Oct 2026 · Swissmedic 12 Oct 2026 · PMDA 12 Oct 2026 · GSRS relationship graph._

What this is: active substances approved by a foreign regulator whose active moiety is absent from the entire US approval record — i.e. the population A1 (never approved in the US) that § 360ff makes PRV-eligible. This is a screening list, not a clearance; downstream filters (505(b)(1) fileability, rare-pediatric evidence, FTO, orphan-exclusivity conflict) still apply. The matcher is measured, not asserted: python3 pipeline/test_accuracy.py --real replays labeled known-answer drugs through this exact code path.

Funnel

US exclusion set

• 188,948 active-ingredient rows from Drugs@FDA + Orange Book + openFDA NDC (71,309 NDA/ANDA/BLA rows), expanded to every salt/spelling/acid-ate variant (moiety.py) and to GSRS prodrug/salt-parent/active-moiety relatives two hops out (gsrs.py). The NDC pass covers moieties present in Drugs@FDA under a spelling whose UNII does not intersect the foreign name's.

How to read this

The orphan-flagged, multi-regulator, INN candidates are the highest-value rows: an INN exists (mature molecule), more than one stringent regulator approved it (real drug, real dossier), and at least one called it orphan (already in rare-disease territory). Start manual review there.

Top orphan-flagged candidates

Dossier-only — named, never-US, and no authorisation in any register loaded here (funnelA_dossier_only.csv)

These moieties reached a stringent regulator but the application was withdrawn, refused, or sits at opinion stage in every register this pipeline loads. They are not foreign approvals on this evidence and must never be counted as such. Two cautions, in opposite directions:

• They may still be approved somewhere unseen. Only EMA centralised, Health Canada, Swissmedic and PMDA-FY are loaded, so a drug licensed in Japan, Australia, or via an EU national/mutual-recognition route lands here wrongly. Known cases are flagged AUTHORISED_ELSEWHERE; that list is hand-built and certainly incomplete. Wiring in EMA Article 57 and a Japanese source is what actually fixes this.

• A refusal is a real negative signal. These are still genuine repurposing leads — a full dossier and human clinical data exist — but a refusal may reflect a safety or efficacy problem that would also sink a US filing. Read each one's refusal grounds before spending on it.

Needs human review (funnelA_review.csv)

These rows are neither kept nor dropped: the data on hand cannot settle them. Each carries the specific open question.

Known residual error classes (do not skip)

• US OTC-monograph marketing is NOT a disqualifying approval. s.529(a)(4)(B)(i)(I) turns on a prior approval in an application under 505(b)(1), (b)(2) or (j). A drug marketed under an OTC monograph has no such application, so it does not bar a voucher. The exclusion set is therefore correctly NDA/ANDA-based, and adding OTC-monograph moieties to it would introduce false NEGATIVES. (The earlier version of this note claimed the opposite and used aspirin as the example; aspirin also holds NDAs, so it was doubly wrong.) What OTC status does change is commercial: no exclusivity runway and existing competition. openFDA drug/ndc marketing_category is the field that tells them apart.

• Pre-1962 approvals may not count against a candidate. In the Sun Pharma litigation over phenobarbital (SEZABY) the D.D.C. held that a pre-1962 NDA which became effective by FDA inaction and was later 'deemed approved' was never approved in the ordinary sense, so it did not bar a rare-pediatric voucher. 740 applications in this Drugs@FDA snapshot have an original approval before 10 Oct 1962, and 70 distinct active moieties are approved ONLY through them. This diff still excludes all 70. If that holding stands, they are a separate candidate population rather than exclusions — a district-court ruling, so treat it as a live legal theory, not settled law.

• Pre-2000 US withdrawals are missing from the bulk files. Of 30 hand-checked drugs with genuine FDA approvals later withdrawn, 13 are absent from Drugs@FDA + Orange Book — every one withdrawn before ~2000 (terfenadine, astemizole, mibefradil, temafloxacin, dexfenfluramine, encainide, flosequinan, zomepirac, benoxaprofen, ticrynafen, nomifensine, methaqualone, phenformin). Post-2000 withdrawals are retained. A withdrawn approval still disqualifies, so this is a real false-positive channel; measured 2026-08-29, none of the 13 currently appears in any loaded foreign register, so the hole is latent rather than active. US_WITHDRAWN_SUPPLEMENT patches one case (betahistine) and is not a substitute for a 21 CFR 216.24 feed.

• Prodrug/conjugate exclusion runs via GSRS (two hops along ACTIVE MOIETY / PRODRUG / SALT-PARENT edges) and catches the isavuconazole-class leaks the string layer measured. It depends on the GSRS graph being complete for the substance; a missing edge is a missing exclusion. Two such gaps are known and handled explicitly: the stereoisomer layer (GSRS holds no enantiomer/racemate edge for zopiclone/eszopiclone or enclomifene/clomifene) and GSRS_GAP_PRODRUGS (tegomil fumarate -> monomethyl fumarate). Newly-INN'd substances are the likeliest next gap, so prodrug-looking names still deserve a manual glance.

• EEA national approvals are now loaded, with one caveat. EMA Article 57 (xEVMPD) covers every nationally-authorised EEA product, closing the gap that made a drug licensed in 25 member states but never centrally read as 'never approved abroad'. The caveat: unlike the centralised medicines report, Article 57 has no status column — it is marketing-authorisation-holder-submitted data, so presence is taken as authorisation and there is no 'refused' or 'withdrawn' to filter. It is therefore labelled as its own source, 'EU national (Art57)', so it can be discounted separately from a regulator decision this pipeline has actually read.

• Non-EEA approvals are still invisible. A drug licensed only in Japan, Australia, Korea or Brazil whose sole EMA record is a refusal still lands in the dossier-only bucket. KNOWN_AUTHORISED_ELSEWHERE hand-corrects the cases found by audit (anamorelin, garenoxacin, plitidepsin, lisuride, ximelagatran — Article 57 now covers perflubutane on its own). That list is hand-built and certainly incomplete; KEGG DRUG and the MHLW price list are the cheapest routes into Japan, and TGA/MFDS/ANVISA remain unfetchable from this environment.

• Japan is effectively absent, and Switzerland nearly so. Only PMDA FY2025 (140 rows) is loaded and 139 of them fail to resolve, so Japan contributes 1 moiety. Swissmedic resolves 102 of 2,970 rows (3%) because the list is in Latin pharmacopoeial genitives ('abacaviri sulfas'). The unresolved mass is Swiss, not Japanese — 2,868 rows against 139. KEGG DRUG (rest.kegg.jp/list/drug) carries Japanese drugs under Latin-script names with JAN tags and is the cheapest route into Japan without building a katakana transliteration layer.

• This diff does NOT test the § 360ff(a)(3) rare-pediatric gate, FTO, or existing orphan exclusivity. pipeline/eligibility.py walks all ten statutory criteria per candidate and labels which a human must decide; it never returns 'eligible', only 'nothing has ruled it out yet'.