In the NDC directory bulk export dated 2026-08-28 (drug-ndc-0001-of-0001.json, 137,521 records), 1,904 product_ndc values appear on more than one record. Each duplicate pair sits under distinct SPL documents (all 1,904 groups have distinct spl_id and product_id), and 1,151 of the groups share the same finished flag, so they are not finished/unfinished splits.
Example, 0409-7338 (Ceftriaxone Sodium, ANDA, both finished=true):
0409-7338_c4d53472-41a6-428e-8935-f2c6117294570409-7338_cc7743b1-4e7b-446a-9ec1-cc234a8bbccf
Reproduction: load the bulk file, Counter(r["product_ndc"] for r in results), count keys with more than one record.
For what it is worth, the identifier column itself is spotless: zero of 137,521 product_ndc values violate the labeler-product digit format, which is better than most registers we measure.
The question: does the directory intend product_ndc to be unique per product? If yes, the 1,904 groups are worth a look (possibly superseded SPL versions both remaining current). If no, a note in the field reference saying consumers must deduplicate on product_id would save downstream users from treating product_ndc as a key.
In the NDC directory bulk export dated 2026-08-28 (
drug-ndc-0001-of-0001.json, 137,521 records), 1,904product_ndcvalues appear on more than one record. Each duplicate pair sits under distinct SPL documents (all 1,904 groups have distinctspl_idandproduct_id), and 1,151 of the groups share the samefinishedflag, so they are not finished/unfinished splits.Example,
0409-7338(Ceftriaxone Sodium, ANDA, both finished=true):0409-7338_c4d53472-41a6-428e-8935-f2c6117294570409-7338_cc7743b1-4e7b-446a-9ec1-cc234a8bbccfReproduction: load the bulk file,
Counter(r["product_ndc"] for r in results), count keys with more than one record.For what it is worth, the identifier column itself is spotless: zero of 137,521
product_ndcvalues violate the labeler-product digit format, which is better than most registers we measure.The question: does the directory intend
product_ndcto be unique per product? If yes, the 1,904 groups are worth a look (possibly superseded SPL versions both remaining current). If no, a note in the field reference saying consumers must deduplicate onproduct_idwould save downstream users from treatingproduct_ndcas a key.