How evidence grades work on this site

Last reviewed 9 August 2026

The status strip at the head of a compound library entry opens with a letter, set above the sequence, the storage conditions and everything else the entry has to say. This page is the method behind that letter: what the five bands mean, the rules that decide which applies, where the record is looked for, how often it is looked at again, and two entries worked through from the published literature to the band.

The letter is a statement about a body of published literature. It is not a statement about the compound, and not a statement about the material in a vial. A band near the top means large or late-stage studies have been published; it does not mean they reported anything favourable, and the first worked example below sits at the top of the scale with two randomised trials that both missed their primary endpoints.

What the band describes, and why it is deliberately mechanical

The band records the stage and design of the published record for one named molecule, and nothing else. It is not a score for how good that record is or how convincing anyone found it.

The rejected alternative is worth naming. A quality score would be more informative, and it would also be an opinion — ours — about a compound listed for sale on the same domain. A reader cannot check an opinion, but can check whether a randomised controlled trial exists, because the reference list is on the same page and every identifier resolves to a record. So the letter carries only what is checkable, and everything needing judgement is written out in the evidence-gaps section where it can be argued with. One qualifying citation moves a band; nothing else does.

The five bands

Band E covers two conditions that are not the same thing: a literature that exists and is entirely in vitro, and a literature that does not exist. The scale does not separate them, so the entry says in words which applies.

Band D is a separate rung rather than a fraction of a human one, and the reason is empirical: Perel and colleagues (2007) concluded that “discordance between animal and human studies may be due to bias or to the failure of animal models to mimic clinical disease adequately”, and Sena and colleagues (2010) reported that publication bias “might account for around one-third of the efficacy reported in systematic reviews” of animal stroke experiments.

The rules that decide the band

1. The band attaches to the entity named in the title — not to a parent protein, a precursor or a structural relative. TB-500 is a synthetic seven-residue acetylated fragment and thymosin β4 is a 43-residue protein, so trials of the second do not band the first; the randomised human trials in the NAD+ area administered a precursor rather than NAD+. Where that gap exists the entry states it rather than quietly borrowing the grade.
2. The highest rung with one qualifying study sets the band. Weight of literature moves nothing. A PubMed search for “BPC 157” in title or abstract returned 214 records when this page was last reviewed, 171 of them listing the same author, and that entry sits at band D. Rodent papers do not accumulate into a human trial.
3. The direction of a result is irrelevant. A confirmatory trial that missed every endpoint it declared is still a confirmatory trial and still counts. This is the most misread feature of the scale.
4. A blend is banded on the literature for the blend. It inherits nothing from its components and is never placed above the lowest band any component holds. Component bands are printed separately.
5. Registered and unreported is recorded, not counted. The published record is not the whole record: Turner and colleagues (2008) compared trials registered with a regulator against their subsequent publication and reported systematic divergence, and Anderson and colleagues (2015) examined reporting compliance at ClinicalTrials.gov.
6. An absence is written as an absence. “No study identified” is not “no effect”, nor a safety statement — the point Altman and Bland made in a single page in 1995. One entry here originally read that no UK authorisation existed for a compound; it now says none was identified in the sources searched, because the stronger form was not provable from what had been read.
7. Reporting quality is described, never scored. Where a study omits what CONSORT or the ARRIVE guidelines ask for — randomisation method, blinding, sample-size justification, species and strain — the entry says so. Nothing is deducted, because a deduction is a judgement wearing the costume of a measurement.

What the band is not

• Not GRADE, and not the Oxford levels. GRADE rates the quality of a body of evidence and the strength of a recommendation drawn from it, against criteria Guyatt and colleagues (2011) list as “study design, risk of bias, imprecision, inconsistency, indirectness, and magnitude of effect” — several of which presuppose a question and a named outcome. This library makes no recommendation and names no outcome, so borrowing either scheme’s letters would misuse a defined method.
• Not a safety rating. Nothing about the design of a study is a statement about hazard. Toxicology, where any exists, is recorded separately.
• Not a statement about the material supplied. Identity, purity and endotoxin are properties of a batch established by analysis, and are the subject of a separate page on reading a certificate of analysis.
• Not a forecast. Wong, Siah and Lo (2019) analysed 406,038 clinical-trial records covering more than 21,000 compounds and reported, for one field alone, that “oncology has a 3.4% success rate in our sample”.
• Not a claim that the compound does anything. That is the whole point of the scale.

Where the record is looked for

Vendor pages, forums, content aggregators and machine-generated summaries are not sources and are not cited, even where they are the only thing a search returns. A citation that cannot be resolved is recorded as unresolved, never dropped quietly and never reconstructed from memory.

How a citation becomes a fact

An identifier is a draft until it has been resolved against the record it points at. A verification pass over the first batch of entries found four classes of error, every one in text that read as authoritative: an author list spliced from a different paper by the same lead author, with title, journal, year and pagination all correct; a regulatory status stated backwards and attributed to the wrong body in the wrong year; an anti-doping classification naming two categories that are mutually exclusive by definition; and two studies described as not stating the species of the model, where the species appears in the title of one and the first line of the other’s abstract. None would have been caught by proofreading, and two sat inside tables designed to look definitive.

Worked example at the top of the scale

The SS-31 (elamipretide) entry carries band A. Three records bear on it:

• Reid Thompson and colleagues (2021) reported a randomised, double-blind, placebo-controlled crossover trial in 12 participants with a genetic disorder of mitochondrial cardiolipin metabolism, followed by an open-label extension. Of the randomised phase the authors write: “In part 1 neither primary endpoint was met.”
• Karaa and colleagues (2023) reported MMPOWER-3, a 24-week randomised, double-blind, placebo-controlled trial in which 218 participants with primary mitochondrial myopathy were randomised 1:1. The primary endpoints were change from baseline at week 24 in distance walked on the six-minute walk test and in total fatigue on a symptom assessment scale; the reported differences were −3.2 m (95% CI −18.7 to 12.3; p = 0.69) and −0.07 (95% CI −0.10 to 0.26; p = 0.37). Neither was met.
• Shirley (2026), in Drugs, records the grant of a first regulatory approval for a medicinal product whose active substance is this molecule.

Band A fires on two independent limbs and either alone would be enough. Both are statements about the existence and design of evidence, and neither is a statement that anything worked: two randomised trials are on the record, both missed what they set out to show, and the band would be identical had they missed by wider margins. That no UK authorisation was identified, that an approval elsewhere is not a UK authorisation, and that the material described is not the approved product are all recorded separately, on the regulatory row rather than the evidence row.

Worked example at the bottom of the scale

The KLOW entry carries band E. KLOW is a blend designation for a preparation containing four separately named peptides. It has no International Nonproprietary Name, no CAS registry number, no PubChem record and no monograph in any pharmacopoeia, because it is a mixture rather than a substance.

A PubMed search for “KLOW” in title or abstract returned 32 records on 9 August 2026, and every one was opened. Not one concerns a peptide preparation: the set includes near-infrared spectroscopy of muscle oxidative capacity, radiolabelled dopamine-receptor ligands, ferroelectric domain walls in single crystals and a coastal-aquifer model. The string is matching author surnames and tokenised phrases such as “high serum K–low dialysate K”.

The nearest published experiment covers two of the four components together: Biçer and colleagues (2026) reported a four-arm study in 32 male Sprague-Dawley rats — control, each of two peptides alone, and the two combined — and reported no additional effect from the combination relative to either peptide alone.

So the band is E under the second of its conditions: not in vitro only, but no published study of the named preparation at all. Rule 4 governs in both directions here — the blend inherits nothing from its components, and could not sit above the lowest of their bands if it did — and neither route would answer the question a blend actually raises, which is what the combination does. That is the question nothing has asked.

Review cadence, and what the last-reviewed line means

Every entry carries a Last reviewed date and a note of what changed. The date moves only when the content changed — not for a formatting pass, a re-render or a no-op edit. A freshness stamp that advances on a page nobody re-read is a false statement on a page whose whole value is that it does not make them.

Two triggers govern a re-read. Scheduled, at least annually: the anti-doping line alone requires it, because the Prohibited List is revised each year and takes effect on 1 January, so a line accurate in December can be wrong in January with nothing on the page having been touched. Event-driven: a regulator publishes a decision naming the compound, a registered trial reports, a cited record acquires a correction or retraction notice, or a reader writes in.

A live example of the second, and of why the two rows are kept apart: on 23 July 2026 the United States Food and Drug Administration’s Pharmacy Compounding Advisory Committee voted — narrowly, and against the position taken by the agency’s own reviewers — to recommend that several peptides including BPC-157 be made available to compounding pharmacies. The votes are not binding, the agency decides, and nothing about them changes the position in the United Kingdom. What the vote did not do is move a band, because a committee vote is not a study. It changed a regulatory-status line and left the letters where they were.

When a band is wrong

The band is the most falsifiable thing published here. It rests on the presence or absence of one kind of citation, the reference list is printed on the same page, and a single qualifying record moves it. Anyone holding a citation that ought to change a band is invited to send it; the contact routes are on the about page. Corrections are made on the page, dated, and stated in the last-reviewed note. A method that can be checked but whose failures are edited out of the record is not a method.

References

1. Guyatt G, Oxman AD, Akl EA, et al. GRADE guidelines: 1. Introduction—GRADE evidence profiles and summary of findings tables. J Clin Epidemiol. 2011;64(4):383–94. Model: methodology guidance. PMID 21195583 · doi:10.1016/j.jclinepi.2010.04.026
2. Guyatt GH, Oxman AD, Vist GE, et al. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. 2008;336(7650):924–6. Model: methodology guidance. PMID 18436948 · doi:10.1136/bmj.39489.470347.AD
3. OCEBM Levels of Evidence Working Group. The Oxford Levels of Evidence 2. Oxford Centre for Evidence-Based Medicine, 2011. Model: methodology guidance. cebm.ox.ac.uk
4. Schulz KF, Altman DG, Moher D; CONSORT Group. CONSORT 2010 statement: updated guidelines for reporting parallel group randomised trials. BMJ. 2010;340:c332. Model: reporting standard. PMID 20332509 · doi:10.1136/bmj.c332
5. Percie du Sert N, Hurst V, Ahluwalia A, et al. The ARRIVE guidelines 2.0: updated guidelines for reporting animal research. PLoS Biol. 2020;18(7):e3000410. Model: reporting standard. PMID 32663219 · doi:10.1371/journal.pbio.3000410
6. Perel P, Roberts I, Sena E, et al. Comparison of treatment effects between animal experiments and clinical trials: systematic review. BMJ. 2007;334(7586):197. Model: systematic review of animal experiments and human trials. PMID 17175568 · doi:10.1136/bmj.39048.407928.BE
7. Sena ES, van der Worp HB, Bath PMW, Howells DW, Macleod MR. Publication bias in reports of animal stroke studies leads to major overstatement of efficacy. PLoS Biol. 2010;8(3):e1000344. Model: meta-research on rodent studies. PMID 20361022 · doi:10.1371/journal.pbio.1000344
8. Wong CH, Siah KW, Lo AW. Estimation of clinical trial success rates and related parameters. Biostatistics. 2019;20(2):273–286. Model: analysis of clinical-trial records. PMID 29394327 · doi:10.1093/biostatistics/kxx069
9. Turner EH, Matthews AM, Linardatos E, Tell RA, Rosenthal R. Selective publication of antidepressant trials and its influence on apparent efficacy. N Engl J Med. 2008;358(3):252–60. Model: meta-research on human trials. PMID 18199864 · doi:10.1056/NEJMsa065779
10. Anderson ML, Chiswell K, Peterson ED, Tasneem A, Topping J, Califf RM. Compliance with results reporting at ClinicalTrials.gov. N Engl J Med. 2015;372(11):1031–9. Model: meta-research on trial registration. PMID 25760355 · doi:10.1056/NEJMsa1409364
11. Altman DG, Bland JM. Absence of evidence is not evidence of absence. BMJ. 1995;311(7003):485. Model: statistical note. PMID 7647644 · doi:10.1136/bmj.311.7003.485
12. Reid Thompson W, Hornby B, Manuel R, et al. A phase 2/3 randomized clinical trial followed by an open-label extension to evaluate the effectiveness of elamipretide in Barth syndrome, a genetic disorder of mitochondrial cardiolipin metabolism. Genet Med. 2021;23(3):471–478. Model: human randomised crossover trial, n = 12. PMID 33077895 · doi:10.1038/s41436-020-01006-8
13. Karaa A, et al. Efficacy and safety of elamipretide in individuals with primary mitochondrial myopathy: the MMPOWER-3 randomized clinical trial. Neurology. 2023;101(3):e238–e252. Model: human randomised controlled trial, n = 218. PMID 37268435 · doi:10.1212/WNL.0000000000207402
14. Shirley M. Elamipretide: first approval. Drugs. 2026;86(3):377–383. Model: regulatory approval review. PMID 41335372 · doi:10.1007/s40265-025-02269-8
15. Biçer O, Adanir O, Güleryüz Y, et al. Effects of BPC-157 and TB-500 on Achilles tendon healing in rats: a histopathological and biomechanical study. Jt Dis Relat Surg. 2026;37(3):822–837. Model: rodent — 32 male Sprague-Dawley rats, four arms. PMID 42542926 · doi:10.52312/jdrs.2026.2951
16. Medicines and Healthcare products Regulatory Agency. A Guide to What is a Medicinal Product, MHRA Guidance Note 8. Model: regulatory document. gov.uk
17. World Anti-Doping Agency. The Prohibited List. Revised annually, in force from 1 January. Model: regulatory document. wada-ama.org · UK Anti-Doping, 2026 Prohibited List: summary of changes
18. United States Food and Drug Administration. Bulk Drug Substances Used in Compounding Under Section 503A of the FD&C Act, and the interim policy establishing categories 1 and 2. Model: regulatory document. fda.gov
19. Reporting of the Pharmacy Compounding Advisory Committee meeting of 23 July 2026. STAT News, 23 July 2026. Model: contemporaneous reporting of a regulatory meeting. statnews.com