Document checks: Up to twenty named checks behind every verdict, held to the statute and your dose.
Limits from your dose. Every finding, cited.
The read of lot BM-5584: one verdict, the count by group, and what was read, checked and held to. Brennick Compounding is a sample pharmacy; its suppliers, people, lots and figures are invented.
Reading a certificate back is not checking it.
A certificate can be signed, complete and marked Conforms on every line and still fail. Three ways it happens:
An endotoxin limit looser than your dose allows.
A solvent inside the supplier's own limit and over ICH's.
A seller printed where the maker should be.
Veritura reads documents and the public record. It does not test material.
Read, routed, computed, looked up. In the order it runs.
One certificate through the engine: lot BM-5520, Cyanocobalamin USP, sold by Brookmere API Supply.
Every value read, and cited to the line it came from.
Each value is cited to its page and line, with that line quoted word for word. A field the certificate does not state is shown as not stated, never guessed. A page that cannot be read is reported as unreadable.
- Page and line, quoted word for word
- Unreadable, never passed
May it be compounded, and who made it.
Section 503A allows a bulk substance by one of three routes: a USP or NF monograph, a component of an FDA-approved drug, or FDA's bulks list. A certificate that names only a seller holds the lot until someone names the maker.
- FDA's four 503A lists, as printed May 14, 2026
- The maker, found in FDA's establishment registry
- Registered only to relabel or repack does not pass
Limits computed from your own dose.
USP <85> sets the endotoxin limit as K over M: K, the threshold for the route, over M, your largest dose per kilogram in one hour. Record both, and the printed limit is held to the result.
- USP <85> K over M, from your dose
- ICH Q3C and Q3D, with the exposure share
- USP <1111> microbial criteria
![The finding on a sample certificate for thiamine hydrochloride, lot TH-0918, after the pharmacy recorded an intramuscular dose of 100 mg in one hour at 70 kg: failed, endotoxin limit looser than USP <85> allows. The endotoxin result is over the limit computed from the recorded dose: 4.6 EU/mg compared with at most 3.5 EU/mg. K / M = 5 EU/kg/hr / 1.428571 mg/kg/hr = 3.5 EU/mg [M = 100 mg in one hour / 70 kg]. It passes with a result at or under K / M for the pharmacy's own maximum dose](/s2/v8/img/a-ee-endotoxin.webp?v=8c)
FDA's records, and the laboratory itself.
Identity is compared with FDA's substance registry, and a conflict fails the verdict with both figures named. The supplier, maker and laboratory are matched against openFDA enforcement records, FDA warning letters and FDA import alerts. Clear shows only when a source answered.
- Source-confirmed only when the laboratory replies
Eight checks on what it prints. Held to its own figures and ICH Q7.
The first of the four groups reads the certificate against itself: what it prints, and whether its own figures agree.
Each cited to the line it was printed on.
353a(b)(1)(A)(iii)A person signs, not just a role.
ICH Q7 11.43A laboratory and a report number a person could quote back.
ICH Q7 11.44A certificate issued for another pharmacy's order is named.
353a(b)(1)(A)(iii)Inside its retest date, with dates that can all be true.
ICH Q7 11.41A Conforms beside 96.1% against 98.0 to 102.0% is caught.
ICH Q7 11.42The formula against the weight; an assay as is against the water.
arithmeticA distributor's reissue references the original certificate.
ICH Q7 11.44, 17.63175 EU/mg for this dose. The printed limit is held to it.
BM-5520 goes into an intramuscular preparation: at most 2 mg in one hour, for a 70 kg adult by convention.
Arithmetic does not establish USP <85> conformity or replace product-specific limits and method suitability.
Four verdicts, in words a pharmacist acts on.
Every review opens with one of four headings. Anything short of a pass says what failed, why, what passes, and how.
Every check that could run passed, each against the standard it names.
Nothing failed. What could not run is named with what it needs, never counted as passed.
Nothing failed, and some checks need the pharmacist's decision, such as a laboratory not yet asked.
The checks not met are counted and named, each with its standard, what failed and how to pass.
Up to twenty named checks. Six FDA records. Sixty solvents.
A check that cannot run says what it needs and is never counted as passed.
Up to twenty named checks, and what each one is held to.
Which checks apply depends on the document: on 26 published bulk-substance certificates, 11 to 13 applied, 12 on the median one.
Every check, in the methodologyThe certificate itself 8 checks
- Names the substance and the lot
- The substance and lot, each cited to its line 21 U.S.C. 353a(b)(1)(A)(iii); ICH Q7 11.41
- Signed for the quality unit
- Who signed, or that the sign-off names only a role ICH Q7 11.43
- Issued by a named laboratory, with a report ID
- The laboratory and a report number a person could quote back 353a(b)(1)(A)(iii); ICH Q7 11.44
- Addressed to this pharmacy
- A certificate issued for another pharmacy's order, named, where one is printed 353a(b)(1)(A)(iii)
- Dated, possible, and inside its retest or expiry date
- The date that passed, or the dates that cannot all be true, cited ICH Q7 11.41 and its glossary
- Every result inside its own limit
- Each result beside its printed limit, and a Conforms the number contradicts, such as 96.1% against 98.0 to 102.0% ICH Q7 11.42
- Its own figures agree with each other
- The formula against the weight; assay as is plus water against the whole sample Arithmetic on what it prints
- A distributor's reissue references the original
- What the reissue leaves out, and the ask that closes it ICH Q7 11.44 and 17.63
The substance 4 checks
- May be compounded under 503A
- The route that applies, a USP or NF monograph, a component of an FDA-approved drug or FDA's 503A bulks list, and the four lists, 216.23(a), 216.23(b), Category 1 and Category 2, as printed May 14, 2026 353a(b)(1)(A)(i); FDA's 503A lists
- States the USP or NF monograph it meets, where one exists
- The line that claims it, or a decision owed to the pharmacist 353a(b)(1)(A)(i)(I)
- The original manufacturer is named, with its address
- The maker, or a hold until someone names it 353a(b)(1)(A)(ii); ICH Q7 11.43
- That manufacturer is registered with FDA
- The FEI and the registered operations; registered only to relabel and repack fails 353a(b)(1)(A)(ii); FDA DECRS
Published limits 4 checks
- Endotoxin limit no looser than the dose allows
- K, M and the computed limit beside the printed one. K is 5 EU/kg/hr for a parenteral route and 0.2 for intrathecal; M is your maximum dose per kilogram in one hour USP <85>, K over M from your dose
- Residual solvents inside their limits
- Each solvent against its limit in ICH Q3C's table of 63; with a dose, its share of the daily exposure ICH Q3C; USP <467>
- Elemental impurities inside their limits
- Cadmium, lead, arsenic and mercury for the route; with a dose, the exposure share ICH Q3D(R2); USP <232>
- Microbial counts inside the criterion
- Total aerobic count and total yeasts and molds against their maximums USP <1111> Table 2; USP <61>
Outside sources 4 checks, and the receipts
- No recent FDA enforcement record matches the supplier
- Which sources answered, and any record from the last 24 months matched for a person to judge; older records are listed with their dates openFDA enforcement; FDA warning letters
- Identity agrees with FDA's substance registry
- Name, UNII, CAS, formula and weight; both figures on a conflict FDA GSRS
- Agrees with every earlier certificate for the lot
- The two certificates side by side, neither named wrong; an independent laboratory's report on the same lot, and a result repeated from another lot, shown the same way Every document seen
- The laboratory confirms it issued the report
- The request, drafted for you and opened in your own mail, the reply, and where the reply is kept; only the laboratory's recorded reply marks a report source-confirmed The laboratory's recorded reply
- How other suppliers print the same tests
- Beside each result, the review shows how other certificates for the same substance, read anywhere in Veritura, print that test's specification, with how many distinct suppliers and certificates printed each. It never names a supplier or a pharmacy and never reproduces USP's text. A specification printed unlike the rest stands out before you rely on it.
- A receipt for every public lookup
- Every lookup keeps the source address, the query, the time, the result and its limitation. A lookup that could not run is shown as not run, with the reason, and never as a zero that looked like a check. No match in a public source is a fact about the sources that answered, never proof of absence. FDA's 503B outsourcing facility list is checked against a daily capture, with the capture's date kept.
- FDA's 503A lists
- The codified 503A lists are re-read from the eCFR every 24 hours; FDA's Category 1 and 2 lists are a dated capture, May 14, 2026.
- After the read: the hourly watch
- Lots and suppliers already on file are re-matched against openFDA enforcement records on a monitoring cycle, hourly in production. A match holds the lot and marks every Passport minted from it, and an identical re-match never reopens a person's dismissal.
The arithmetic, worked through Endotoxin, Q3D, potency
- Route
- Intramuscular. A parenteral route, so K is 5 EU/kg/hr under USP <85>
- Largest dose in one hour
- 2 mg
- Body weight
- 70 kg. The pharmacopeial adult convention, flagged as an assumption until a real weight is given
- M
- 2 mg over 70 kg, in one hour: 0.028571 mg/kg/hr
- K / M
- 5 EU/kg/hr over 0.028571 mg/kg/hr: 175 EU/mg, the endotoxin limit for this dose
- BM-5520 as received: no endotoxin line
- Your call “A limit was computed from the dose and the certificate states none to compare.” The fix it offers: ask for the endotoxin result and its limit.
- A certificate printing 500 EU/mg
- Not verified 2.857 times the limit. “Stated limit 500 EU/mg versus computed limit 175 EU/mg. Stated is 2.857x the computed requirement.” A result could sit inside the printed limit and still be above what your dose allows.
- A certificate printing 10 EU/mg
- Passes this check At or under the limit computed for this dose and route, so the printed limit does not understate the requirement.
- The same arithmetic for ICH Q3D
- Lead at 4.0 ppm is over the Option 1 concentration for a parenteral route, yet at 2 mg a day it is 0.008 ug/day, 0.16% of the parenteral PDE, and the review prints that line so the pharmacist decides with the numbers in hand. Arithmetic does not establish USP <85> conformity or replace product-specific limits and method suitability.
- Each lot's potency factor
- Context for a person, never a verdict. Worked out from the lot's own certificate: the assay, and the water or loss on drying when the assay is printed on a dried or anhydrous basis, each cited to its line. When the certificate does not say which basis, both readings are shown and neither is chosen. It is computed on every read, and the pharmacist confirms it on the compounding record before anything is weighed.
The file itself, and across your records 17 signals
- Signals for a person, never verdicts
- Integrity checks run on every file and speak in signals for a person, never verdicts. The engine never calls a document forged or fake.
- Seventeen document-integrity signals
- Seven read the file itself: edit history, an unexpected producer, a text layer that disagrees with the page, font anomalies, unnatural number patterns, a dead or mismatched verify link and an unknown laboratory. Ten compare it with your other records: the same file, report ID and lot collisions, one lot number on two substances, signature and stamp reuse, the same embedded image, a template mismatch, and the same rendered page or picture of it. Matches to FDA enforcement, warning-letter, import-alert and registration records are outside-source checks, made against the six FDA records.
- Seven findings from the PDF's own bytes
- Created after the date it claims, modified after creation, a creation date in the future, an image editor as producer, an incremental edit stack, a creator and producer that disagree, and chromatography claimed with no trace attached. A high contradiction holds the lot.
- Thirteen signs of a generated document
- From language-model boilerplate and unfilled template placeholders to an impossible phone number, a ZIP code in the wrong state and uniformly round values. A file whose own C2PA content credentials say it was AI-generated holds.
- The supplier's own history
- Each document is compared with the supplier's earlier documents across 11 structural features, from field order and label words to date style and email domain, so a new template shows the day it arrives.
- Reuse, across your workspace and beyond
- The same file, report ID or lot where it should not be is raised. Strong identifiers are also compared across every Veritura account; you see a count of the others, never who they are or what they hold.
- What is accepted
- A PDF, a photo (PNG, JPEG or WebP) or a plain-text certificate, up to 8 MB. The file is identified by its bytes, not by its name or the type the browser claims, so random bytes named .pdf are refused. Word and Excel files, SVG, HTML and empty files are refused, and the refusal says why in words. The same file read twice is recognised and not filed twice.
- Kept as received
- The original is kept as received, with its SHA-256 fingerprint, and every decision is kept with who, when and why.
- How the page is read
- A PDF with a text layer is read directly. The read runs at temperature 0 and is cached on the file's fingerprint, so the same bytes give the same answer. A text certificate is read first; the whole check then runs with its public lookups. A page is cited by page and line only when its page is known, and never estimated; a plain text file is cited by line alone.
- The four lanes every document reports on
- Reconciled against the issuing laboratory, checked against the standard, checked against the file itself, and checked against every other document seen. Each lane says ran, not applicable or could not verify, with its reason. No lane goes quiet, and none borrows another lane's result.
- The Second Reader
- Where it is enabled, a Second Reader re-reads each check. Its correction lands only if the line it quotes is found in the document. A field it contests holds the lot for a person. It can never release, issue or notify.
What the engine never says
- No fraud calls
- It never calls a document forged, fake or fraudulent; integrity checks speak in signals for a person. A similarity is not a fraud determination.
- No endorsements
- It never says a supplier is approved, a product is safe or a facility is inspected.
- No accuracy rate
- It never quotes an accuracy rate of its own reading. A missing document is not a failed test.
Not checked: USP monograph acceptance criteria. Checked against 21 U.S.C. 353a, FDA's 503A lists, ICH Q3C, ICH Q3D, USP <85>, USP <1111>, FDA DECRS, FDA GSRS and openFDA: published standards and public FDA records, no affiliation or endorsement implied.
Questions, answered plainly.
No result is held to a monograph's acceptance criteria, because that text is licensed, not public. Veritura checks whether a monograph exists for the substance and whether the certificate states it meets it, and it computes or reads the published chapters and guidelines: USP <85> from your dose, USP <1111>, ICH Q3C and ICH Q3D.
The dose-based checks say they could not run and name what they need: the route, the largest dose in one hour and the patient weight. They are never counted as passed. Without a weight, the 70 kg adult convention is used and flagged as an assumption.
The check says so. Clear is shown only when the sources answered, and every lookup keeps its address, query, time and result. After the read, lots on file are re-matched against openFDA enforcement records on an hourly cycle.
A PDF with a text layer is read directly. Scans and photos go through a page-by-page vision reader where it is enabled, and a scan needs a focused second read before the product relies on it. A page that cannot be read is reported as unreadable, never as clean.
No. Verified means every check that could run passed on the paperwork. Veritura does not test material, and no score or grade appears on anything a patient sees.
Your workspace. Strong identifiers such as a file's fingerprint are compared across Veritura accounts so reuse can be caught, and you see only a count of the other accounts involved, never their identity, supplier, file, lot or order.
Still have questions? Talk to a person.