Retatrutide Specifications Register: COA, SDS, and Documentation Standards

A retatrutide specifications register keeps three document types apart instead of collapsing them into one product summary: the COA, which reports test results for one named lot; the SDS, which describes the substance as a chemical regardless of lot; and the literature reference, which records what a published trial actually measured. The documentation standard that makes the register usable is narrow. Every entry names the document it came from, carries the date that document was read, and writes a missing field as missing rather than leaving a blank cell.

Which document answers which question?

Document	Scope	Reissued when the lot changes	States a measured purity	States handling and hazard data	Date that matters
COA	One lot or batch	Yes	Yes	No	Test date
SDS	The compound as a substance	No	No	Yes	Revision date
Peer-reviewed trial record	A study population	No	No	No	Publication date
Listing snapshot	One supplier page	No	Repeats what it was given	Sometimes links an SDS	The date you read it

What each documentation tier can and cannot establish. Built from the document types themselves; no trial figures appear in this table.

The rows are not substitutes for each other, and the register should never let one stand in for another. An SDS revised once and reused for years says nothing about the vial in front of you, which is the practical consequence of how COA and SDS scope diverge; the fields a lot document has to carry before it earns a register row are set out in what a COA should report.

How should the register record what a trial measured?

Published retatrutide figures belong in the register as dose-arm rows with the trial attached, not as a single efficacy sentence.

Weekly subcutaneous dose	Body weight, % change at 24 wk (phase 2 obesity trial, n=338)	Total fat mass, % reduction at 36 wk (phase 2 T2D DXA substudy, n=189)
Placebo	-1.6%	not reported in the record
0.5 mg	arm not in this trial	4.9% (SE 1.4)
1 mg	-7.2%	arm not in this trial
4 mg (pooled)	-12.9%	15.2% (SE 3.2)
8 mg (pooled)	-17.3%	26.1% (SE 2.5)
12 mg	-17.5%	23.2% (SE 3.0)

Dose-arm figures as each trial reports them. The two columns are different endpoints (body weight against DXA-measured total fat mass) over different durations in different populations, so the register holds them as separate columns and does not compare the percentages against each other.

The n a register writes down should be the analysed n, not the enrolled n. The DXA substudy in the second column (PMID 40609566) screened 534 people and excluded 253, leaving 534 - 253 = 281 enrolled in the main study. Of those, 189 entered the body-composition substudy, 155 had a baseline DXA scan, and 103 completed treatment with both a baseline and a week-36 scan.

Paired-scan completion is therefore 103 / 189 = 54.5% of the substudy cohort, or 103 / 155 = 66.5% of the participants who had a baseline scan at all. A register cell that records only “n = 189” overstates the number of people who contributed to the endpoint by 189 - 103 = 86. The obesity trial in the first column reports its own enrolment plainly at 338 adults over 48 weeks of once-weekly dosing, with the 24-week weight change as the prespecified primary endpoint.

What does lot-matching require beyond a lot number?

A COA earns a register row when six checks pass, and each one is answerable by reading the document rather than by testing anything.

1. The lot number printed on the COA matches the lot printed on the vial, not merely the product name.
2. The COA’s test date is recorded next to the date you read the listing, so a document reused across restocks shows up as an unchanged date beside a moving one.
3. Identity method and purity method are named separately. Mass spectrometry and HPLC answer different questions, and a document reporting one has answered one.
4. The testing party is named, and the document states whether the work was in-house or third-party.
5. The purity figure is transcribed exactly as printed, including the decimal places and the greater-than-or-equal sign the document used, rather than rounded into the register.
6. Any field the COA omits is written as “not stated on document, checked ”.

Supplier listings for research-use-only retatrutide vary widely in how many of those six they support, and the register’s job is to make that variation legible instead of averaging it away. The gap between this kind of documentation and what a sponsor files for a registrational program is covered in research documentation against trial documentation.

Where do the trial records disagree?

Both trials in the table above run non-monotonic at the top of the dose range, and a register keeps that rather than smoothing it into a dose-response claim. In the phase 2 obesity trial, 24-week body weight change is -17.3% for the pooled 8 mg arms and -17.5% for 12 mg, a within-trial difference of 0.2 percentage points.

The T2D substudy is more pointed: within that single trial, the pooled 8 mg arms show 26.1% total fat-mass reduction against 23.2% for 12 mg, so the higher dose reads lower on that endpoint. Their one-SE bands are 26.1 ± 2.5 = 23.6 to 28.6% and 23.2 ± 3.0 = 20.2 to 26.2%, which overlap across 23.6 to 26.2%. A register line reading “dose-dependent” discards both observations and cannot be reconstructed from the source afterwards.

What goes in a field the listing never states?

An explicit absence, dated. A blank cell is unreadable six months later because it cannot distinguish “the supplier does not publish this” from “nobody checked”, and those two are different facts about the same listing. The same rule applies to the literature tier: if a figure you want is not in the record you actually read, the register carries the gap rather than a number borrowed from a secondary summary. Field-by-field layout for the entries themselves is set out in the register’s core field list.

Record the substudy’s registry identifier, NCT04867785, in the same cell as the figure it supports. The registry record is where a later reader confirms that the week-36 fat-mass endpoint was prespecified rather than selected once the data were in.