Public COA purity drift: tracking 12 suppliers over 8 weeks

I track the same thing. Lot prefix changes are the tell. Saw one vendor switch from A23 to B24 but keep exact 99.1% area. Suspicious.

Exact same purtiy across lots? Rare. Ask for chromatogram.

Analytical variance can hit 1% on older HPLC columns. Without method details, drift is not proof.

Method details matter, sure, but a 1.2% jump with unchanged method? I'd want to know column lot and injection count, plus whether system suitability was run. Public COAs rarely include that.

I've started scoring suppliers on whether they show method number alongside chromatogram and residual solvent or heavy metal data. Vendors that omit method number usually omit other QC stuff too. That's more useful than one purity number.

Documentation completeness score is the only metric I trust.

Lot prefixes changing with identical COA layout is common. I paste PDF metadata into a sheet; creation dates sometimes predate the batch.

PDF metadata is slippery. Some labs re-save scans, which resets dates. I tried that and got false positives. Better to compare elemental impurity profiles or NMR if you can. Also check whether the COA has a unique report ID that matches a lab portal. If the supplier just emails a JPEG, I discount it.

That said, two vendors in my sheet have clean metadata and still show purity drift, so it's not a single smoking gun.

What's your threshold for drift? I flag over 0.8% when method number is unchanged. Above that, I ask for raw data before I even consider a vendor.

I was team “normal analytical variance” until I split one lot into two vials and sent them to the same lab—came back 98.7 and 97.5, so a 1.2% swing can absolutely happen from prep or column temp alone. What made me side-eye the jumpy vendors in my own 10-week log wasn’t the purity number, it was the gap between date tested and manufacture date: the tight 98.4–99.0 crowd sat under 45 days, while the drifters were 6–9 months out and their chromatogram images looked screenshot-y with no method number.

If you’re not already pulling retest date, storage condition, and whether the method or column ID changed between lots, I’d start there—that’s where recycled reports usually show their seams.

One thing I added to my own sheet: whether the COA purity is “as is” or corrected for water/residual solvents. That alone can swing a headline % by >1 even with the same method. I also save the chromatogram page, not just the summary table—peak shoulders or extra small peaks showed up before the top-line number moved.

Are the 3 drifting vendors the ones reporting area-normalization only? That’s the pattern I’d check first.

the headline purity is the least interesting column imo. when i was logging this stuff i started tracking the single biggest impurity peak instead — had a run where purity sat at 98.7 the whole time but the main related-substance went 0.4 → 0.9 → 0.3 across three lots sharing a prefix.

that reads like a different synth route, not sloppy hplc.

do your COAs break out individual impurities or just hand you the one number? if it's just the headline, the drift you're chasing is probably the tip.

Airport-brain thought: I started adding a “prefix-change” column and cross-checking the COA test date against the batch date. If that lag suddenly jumps from 3 weeks to 4 months, I treat the drift as supply-chain, not analytical.

Had one stretch where a 1.3% bump lined up with a different fill site in the fine print. Not proof, just a pattern. Do you log test-date lag? That’s been more predictive for me than purity alone.

My next column would be method metadata, not purity: column lot, mobile phase pH, analyst initials, integration threshold.

I’ve seen a “drift” vanish once the lab switched HPLC columns and the peak got sharper. Also run a variance ratio—between-lot Δ vs pooled within-lot SD.

If the jumpers only cross 3x on prefix changes, that’s signal. If it’s ~1.5x, you’re measuring the lab’s mood.

Do your 3 jumpers share a testing lab, or is that column still blank?

I’m brand new here, week one, so apologies if this is obvious. The one column I added to my tiny tracker is COA revision/issue date vs lot date.

I saw a couple listings where the same lot prefix had an older COA and a newer one with a ~0.8% higher purity, which would look like drift if you only sort by batch date. Do you log whether a COA was reissued or amended? If so, does the jump shrink when you use only the latest revision per lot? Just trying to learn.

Random thought: do you log the gap between manufacture date and COA date? I did that for a couple of protein tubs (not the same thing, I know) and the lots with a >6-month lag were the oes with weird clumping/texture. Dorm fridge temp swings probably didn’t help either.

If your 3 jumpy vendors also have long COA lags, that might point to storage/shipping drift rather than the assay itself. Or I’m just avoiding my biochem homework.

In the trials of my own spreadsheet, the method tab is weirdly soothing.

ngl late-night spreadsheet goblin here. one thing i’d add: are you logging whether the lot prefix is the API/fill lot or just the vendor’s internal lot? i

Tbh data calm take: I’d stop staring at consecutive-lot deltas and plot each vendor’s purity against lot sequence instead.

A 1.2% jump can be analytical noise; three lots drifting the same direction is a signal. Do the 3 jumpy vendors have fewer lots or more method changes? Also, are COAs summary-only or full chromatograms? I log “has chromatogram y/n” and it cut my false alarms way down.

Random check that saved my spreadsheet sanity: sort by testing lab, not vendor. I had two suppliers whoe “drift” vanished once I split in-house vs third-party COAs — HPLC area % isn’t apples-to-apples if method details shift.

Also, do the 3 jumpy vendors have COA PDFs that suddenly change layout, logo, or signature block? That’s usually my tell they switched labs or contract manufacturers. Typing this from an airport gate, so grain of salt, but I’d plot lab ID before blaming the lot prefix.