12 rows out of 188,556

What a reference page is actually showing

Every compound page in this portfolio carries a list of records drawn from one query. The query is a phrase search of titles and abstracts on the compound's canonical name, run against Europe PMC, sorted by citation count, and then cut to at most 12 rows. A compound with fewer than 5 usable records gets no study page at all.

That is a defensible way to build a reference list. It favours precision over recall, and the alternative, an unsorted sample of a large result set, would put unrelated papers on the page. What it is not is an index of the literature, and nothing on a compound page says which of the two a reader is looking at.

Counted across all 77 compounds, the queries returned 404,706 records. The corpus carries 902 of them: 0.22% of the pooled result. The pooled figure is dominated by the largest query, so the per-compound median matters more, and it is 5.5%.

Coverage is rows carried divided by records returned, for that compound's own query on the date it ran. Glutathione is the extreme case: the query TITLE_ABS:"Glutathione" returned 188,556 records and the corpus carries 12.

The distribution, not the average

Coverage across the 77 compounds spans four orders of magnitude, from 0.0064% to complete, so an average would describe no compound in the set. The buckets below are the honest summary: 20 compounds show under one per cent of what their query returned, 47 show under ten, and 9 show half or more.

The depth is set by a ceiling, not by the literature

71 of 77 compounds carry exactly 12 rows. That is the pipeline's cap, so for all but 6 compounds the length of the reference list is a configuration constant. A reader counting twelve citations on a page about a well-studied molecule and twelve on a page about an obscure one is reading the same number twice.

The sort order has a second consequence. Records were ranked by citation count, which favours older, heavily cited work: 18 of the 77 compound sets carry nothing published after 2019, and 5 carry nothing after 2014. The oldest is Somatostatin, whose newest carried record is from 2003.

At the other end, one compound is covered completely. AOD9604's query returned 11 records and the corpus carries 11 of them. The difference between that page and the Glutathione page is not effort or editorial judgement. It is how many records exist under the name.

Why the gap tracks the name, not the compound

The widest gaps all belong to molecules that occur in the body and have a literature that mostly has nothing to do with the research-chemical market: Glutathione, vasopressin, somatostatin, oxytocin. A phrase search on those names returns decades of physiology, oncology and critical-care work. The narrowest gaps belong to research codes, where almost every record under the name is about the substance being sold under it.

So the gap is partly a measure of ambiguity in the name rather than of thoroughness in the index. That cuts both ways for a reference site: a small gap can mean good coverage, and it can equally mean there is almost nothing to cover.

What the gap cannot be read as is a count of missing peptide studies. The 188,544 records this corpus does not carry under Glutathione are not 188,544 omitted papers about a research chemical. They are records that mention the phrase.

What would make a reference list checkable

No other peptide reference site publishes the query behind its citation list, so no comparison is possible here and none is implied. What is possible is to state the requirement, which is cheap and which this portfolio now meets: publish the query, the number of records it returned, the cap applied, and the date. Three numbers and a string turn a list of citations into something a reader can audit.

The 77-row dataset below carries all four for every compound in this corpus. The methods page states the rules this publication applies to its own counting.