What a supplier specification commits to

Two suppliers send paperwork for what is nominally the same material. Each packet holds a specification sheet and a certificate of analysis. The certificates are easy to compare, because they carry numbers and numbers invite comparison. The specifications look like formalities. That reading is backwards. The certificate describes one lot that has already been made and already been measured; our note "How to read a certificate of analysis" sets out what that document claims and what it does not. The specification is the only document in the packet that says anything about the next lot.

The difference is one of tense. A specification is forward-looking: these are the tests, these are the methods, these are the limits that material must meet to ship under this name. A certificate is backward-looking: this is what was measured, on this lot, on this date. The two are not two views of one fact. A certificate cannot be evaluated on its own, because pass and fail are properties of the specification, not of the measurement. Held up alone, a certificate is a list of numbers with no verdict attached.

A specification row has three parts

ICH Q6A, the guideline on specifications for new drug substances and new drug products, describes a specification as a list of tests, references to analytical procedures, and appropriate acceptance criteria. Research-use reagents are not sold under that framework, and naming it here is not a claim that any supplier works to it. The three-part structure is the part worth borrowing. A specification commits row by row, and a row commits only when it carries all three of the following. Remove any one and the line stops obliging the supplier to anything.

• The test. What property is being determined, named precisely enough to be unambiguous. "Purity" is not a test; it is a category that different methods populate differently, which is the subject of a separate note in this library.
• The method. The procedure by which the test is performed, identified well enough that a second laboratory could reproduce it — a pharmacopoeial chapter, a published procedure, or an in-house method identified by number and revision. "HPLC" is an instrument class, not a method.
• The acceptance criterion. A limit or a range, with a direction and units. "Conforms" is acceptable only where the test is genuinely pass or fail, such as an identity test against a reference, and even then the reference and the pass condition should be named.

Above the rows, the header matter carries its own commitments. Which product or grade the specification governs. A revision identifier and date. Who issued it. Whether the supplier undertakes to notify you before revising it. A specification with no revision number cannot be shown to have changed, which means you cannot tell whether the document in your file is the document the supplier is currently working to.

The rows that commit to nothing

Weak specifications rarely make false claims. They fail by omission, and the omissions are formatted to look like content — same table, same font, same column widths, no criterion. Watch for these.

• Typical values, sometimes labelled nominal, representative, or characteristic.
• Rows marked "for information only", "report only", or "not specified". The test is performed and the number is printed, but no limit governs it.
• Limits with no method. A number without a procedure is not testable against, because a different procedure can return a different number on the same material.
• Criteria qualified as "or as agreed", "target", or "guideline". These are negotiating positions, not commitments.
• Tests inherited from a monograph where the monograph and its edition are not named.

When the certificate names a test the specification never did

This is common and is not by itself a problem. It has three ordinary explanations, and they are worth telling apart before you decide what the extra line is worth.

1. The supplier tests beyond its own specification as routine practice and reports what it has. Generous, and it commits to nothing. Ask whether the test will appear on future certificates.
2. The result comes from a different document — a customer-specific specification, an internal control limit, or a monograph the supplier works to but has not listed. Ask which document sets the criterion, then ask for that document.
3. The test was added in response to a problem, and the specification has not caught up. This is the most informative case. Ask when the test was introduced and what prompted it.

The reverse asymmetry matters more and is easier to miss. A specification row with no corresponding line on the certificate means a criterion exists and there is no evidence it was checked on that lot. Ask whether the test was performed elsewhere, performed on a periodic rather than per-lot basis, or carried forward from an incoming raw material. Periodic and skip-lot testing are legitimate arrangements. They belong on the face of the specification, declared, rather than being inferred by a buyer who noticed a gap.

What "pass" depends on

A certificate that declares a lot conforming is making a judgement, and the rules behind the judgement are usually not printed. ISO/IEC 17025:2017 addresses this directly at clause 7.8.6, on reporting statements of conformity: where a statement of conformity to a specification is provided, the laboratory is required to document the decision rule employed and to apply it, taking account of the level of risk associated with it. A decision rule is the laboratory's stated policy on how measurement uncertainty is handled at the limit. A result sitting on a limit can pass under one decision rule and fail under another, from identical data. Two suppliers can both print "conforms" and mean materially different things by it.

Rounding is the same kind of hidden convention. Whether a reported value is compared to the limit before or after rounding, and to how many figures, is a rule someone chose. USP General Notices section 7, Test Results, treats this as a topic in its own right: 7.20, Rounding Rules, states that observed or calculated values are rounded off to the number of decimal places that is in agreement with the limit expression before they are compared to it.

So is the question of who performed the test. Other industries have codified this. EN 10204 classifies inspection documents for metallic products by whether the declaration rests on non-specific inspection, whether it is validated by someone independent of the manufacturing department, or whether an external inspector signed it. Peptide materials have no equivalent scheme, but the question transfers intact: who ran the analysis, and were they independent of the people who made the lot. A certificate that names a contract laboratory is answering that question. One that does not is leaving it open.

Comparing two packets

Practically, in this order.

1. Line up the two specifications first and leave the certificates face down. The certificates will pull attention toward whichever supplier reports the more flattering number, which is a fact about one lot.
2. Count committed rows on each: test, method, criterion, all three present. Put typical values and information-only rows in a separate column and do not count them.
3. Read the method column. Compendial or published references, and in-house methods identified by number and revision, are stronger than a bare technique name — and note that pharmacopoeias themselves allow alternative procedures, so "we use an equivalent method" is a real answer that still needs to be written down.
4. Now open the certificates and check the mapping in both directions: every specification row present on the certificate, every certificate line traceable to a specification row.
5. Ask both suppliers for their change-notification terms in writing. A specification you are not told about the revision of is a snapshot, not an agreement.
6. Ask both suppliers the same narrow question about the same row — which method, which revision, what decision rule. Compare the answers. That comparison is usually more informative than the paperwork was.

The supplier whose paperwork looks cleaner is sometimes only the supplier whose specification commits to less. A short specification is easier to meet than a long one, and a certificate crowded with information-only rows can carry fewer obligations than a plain one with four properly formed lines. What you are buying, on the documentation side, is the set of statements someone is prepared to be held to about material that does not exist yet.

The documents cited below are the sources of the definitions and expectations used in this note. Citing them describes where the ideas come from. It is not a statement that Northlake, or any supplier, conforms to them.