Investigation
30 of 77 carry no human-trial record
The tier on each compound in this corpus is a count of record types, not a verdict. Counted across all 77, it says that a catalogue presents compounds with a substantial trial literature and compounds with none in exactly the same format.
Published 13 October 2026Subject: this portfolio's corpus, 77 compounds, 902 recordsCounted 13 October 2026
- Across 77 compounds: 23 human-trial, 16 limited-human, 24 animal and in vitro only, 13 preparation or mixture, and 1 whose records state no model.
- 30 of 77 carry no interventional human-trial record in their indexed set.
- Counted by record rather than by compound, interventional trials are 24% of the 902 records. The largest group, 391 records, carries no metadata that settles the model at all.
- The ladder is ordered, so an earlier test pre-empts a later one: 8 compounds with at least one trial record are labelled preparation or mixture, because no chemical database record exists under the name.
- 50 of 77 are withheld from indexing in the reference publication pending named qualified review, and 0 currently have a reviewer recorded.
What the tier is, and what it is not
Every compound in this corpus carries a derived label called an evidence tier. It is not a grade. It is the output of a short ladder of tests run over counts of record types, it is marked INFERRED in the data, and it ships with a sentence stating the counts that produced it. It says nothing about whether a study was well conducted, nor about what any of them found.
The ladder is reproduced in full below, because a label is only checkable if the test behind it is visible. Counted across all 77 compounds, the distribution is the finding: the corpus holds 23 compounds whose indexed records include three or more interventional trials, and 30 with no interventional trial record at all.
| Tier | Compounds | The test, run in this order | Withheld from indexing |
|---|---|---|---|
| Human trial | 23 | three or more records whose own publication type is an interventional trial | 23 |
| Limited human | 16 | one or two such records | 16 |
| Animal and in vitro only | 24 | no interventional trial record, at least one rodent record | 3 |
| Preparation or mixture | 13 | PubChem returned no compound record for the name | 8 |
| Model not established | 1 | records carry no subject heading that states a study model | 0 |
Tiers with no members are omitted. The last column counts compounds the reference publication builds but does not publish as indexable, pending named qualified review: 50 of 77 in total, and 0 have a reviewer recorded.
Counted by record rather than by compound
A tier describes a compound. The records underneath it tell a blunter story. Of the 902 records the corpus carries, 218 are interventional human trials by their own assigned publication type, which is 24% of the total.
The largest single group is none of the three models. 391 records, more than any other category, carry no subject heading that settles whether the work was done in a dish, in an animal or in people. The pipeline records those as unsettled rather than guessing from the title, which is the right call and also a hard ceiling on how much any tier derived from them can mean.
| Model | Records | Share |
|---|---|---|
| Interventional human trial | 218 | 24% |
| Rodent | 292 | 32% |
| In vitro | 1 | 0.11% |
| No model stated in the metadata | 391 | 43% |
Model is derived only from the subject headings and publication types the indexer assigned to each record, never from its title or abstract.
The ladder is ordered, and the order shows
The tests run in sequence and the first match wins, so an earlier test can pre-empt a later one. It does. 8 compounds that carry at least one interventional trial record are labelled preparation_or_mixture, because PubChem returned no compound record for the name and that test sits above the human-trial tests.
Those 8 are ACE-031, Cerebrolysin, Cortexin, Dulaglutide, Follistatin-344, Maridebart cafraglutide, Retatrutide, Somatropin. The per-compound record counts are in the dataset. The label in those cases is partly a statement about how a chemical database indexes a name, not about what has been studied under it, and a reader who takes the tier at face value gets that backwards.
This is the argument for the basis string. Every tier in the corpus carries one, and it is the basis, not the label, that can be checked.
Why the distribution matters more than any one label
A research-chemical catalogue presents a compound with twelve trial records and a compound with none in the same row, the same typeface and the same confident tone. This corpus holds both: 23 compounds in the top tier, and 30 of 77 with no interventional trial record anywhere in their indexed set.
Where the derived label indicates a compound a reader could mistake for an approved medicine, the reference publication holds the page back rather than publishing it: of 77 compounds, 50 are built but not indexable, and none has a named qualified reviewer recorded. That is a disclosure about this portfolio, not a claim about anyone else's.
The per-compound table is in the dataset rather than on this page. This publication reports distributions; presenting one compound's tier, its record counts and its basis on a page of its own would be presenting that compound's evidence, which is the line this publication does not cross.
What this cannot show
- No full text was read. Nothing here states what any study concluded, or whether any trial was well designed, adequately powered or independently funded.
- A record counted as an interventional trial is one the indexer tagged as such. The tag says a trial was reported; it says nothing about its size, its quality or its result.
- The 391 records with no stated model are not records of no model. They are records whose assigned metadata does not settle one, and reading the full texts would move an unknown number of them into the other three rows.
- The 77 compounds are this corpus's selection, not a sample of any market. The distribution describes what was collected here and cannot be generalised to peptides as a category.
- Counting records favours well-indexed compounds. A compound with a large literature and a twelve-row cap contributes the same twelve rows as one with twelve records in total, which is the subject of the companion investigation on index depth.