How SUPPFAX structures and matches ingredient information.

The record is one exact commercial grade

The central object is a named commercial grade or SKU, with its former names and codes attached to it, not a nutrient category. A buyer’s question is whether this grade meets their specification, and a record about a category cannot answer that.

• An organization is recorded against a grade in a typed relationship: manufacturer, brand owner, marketer or distributor. The roles are separate facts and one organization may hold more than one of them.
• Registered, operating, manufacturing and laboratory locations are separate facts and are never collapsed into one ambiguous location.
• A material-specific fact is never aggregated into a claim about the company. One grade’s MOQ, lead time, result or certification describes that grade.
• Certification scope stays attached to the exact legal entity, facility, method, matrix, analyte, product or grade it was issued for.

Who governs a record

• SUPPFAX creates and governs the canonical record. It is not a supplier-owned listing, and confirming an organization or a relationship does not transfer ownership or publication control.
• SUPPFAX may create a record from defensible public identity information, subject to the publication controls below. Fields with no eligible published information begin as unknown.
• Identity facts and published information are visibly separated. Knowing who makes a grade is not the same kind of statement as knowing what it contains.
• Every organization may report a factual identity error, submit a correction, dispute a published record and submit information for independent review, whether or not it participates commercially, and at no cost. No account is needed: the route is here, it is open to anyone, and a reviewer reads every submission, and replies if a reply was asked for.

Applicability is governed

Marking a criterion not applicable removes it from scope. If a participant could do that, a participant could remove any criterion it did badly on, so it cannot. A non-applicability arrives from exactly one of two places.

▣Applicability: Applicable

This criterion applies to this material.

▢Applicability: Not applicable

Governed: this criterion was removed from scope by a versioned category rule or an administrator-approved, source-linked decision. A supplier can never set it.

◇Applicability: Applicability undetermined

No rule or decision has established whether this criterion applies. Undetermined is not the same as not applicable, and removes nothing from scope.

1. A versioned rule tied to a published methodology version and a category, or
2. an administrator decision that names a rationale and a supporting source.
3. Absent both, applicability is undetermined, which is not the same as not applicable and removes nothing from scope.

The reason is always public. Publishing that a criterion has been removed while withholding why would be a hidden input by another name. Every decision in force publishes its outcome, its rationale, its effective date, the methodology version where a rule governs, and the citation for its supporting source, or an honest statement that the cited document is not public. A revoked or superseded decision governs nothing and is never published. No reviewer is ever named.

How coverage counts are calculated

• Evidenced fields is the number of attributes applicable to that grade carrying at least one current published assertion.
• Unknown fields is the number that carry none. It is printed beside the evidenced count rather than omitted: a record that looks complete because its gaps were hidden is the failure this product exists to correct.
• Not applicable is counted separately and never folded into either of the two above.
• Source-linked fields is the number of published values carrying at least one linked source document, study or authoritative URL.
• On a supplier card you will see counts of published and source-linked fields but no coverage ratio. An honest ratio needs the applicable-attribute denominator for every grade in a portfolio, and calculating that across a whole directory is expensive enough that we would be tempted to approximate it. Rather than publish a number that means something slightly different from what it says, the listing omits it and the supplier profile shows the real per-grade figures.
• None of these is a quality score, and none is combined into one. There is no composite score anywhere in this product, and no column in the database from which one could be assembled.

What kind of value it is

A specification limit is what a grade is guaranteed to meet. A measured result is one laboratory’s reading of one sample. They are never compared as though they were the same number, anywhere in this product.

◑Value kind: Declaration

A stated value. Not a measurement and not a limit.

⌐Value kind: Specification limit

A limit the material is specified to meet. A limit is not a measured result and must never be read as one.

⊙Value kind: Measured result

A value reported for an identified sample or batch, with its method and operator preserved.

¶Value kind: Study finding

A finding reported in a study.

§Value kind: Regulatory status

A status recorded by a regulatory or official source.

∑Value kind: Derived value

Calculated from other source-linked values by a stated rule.

What kind of thing stands behind the claim

◑Source basis: Supplier declaration

Stated by the participant. No source document stands behind it.

▤Source basis: Source document

Supported by a document supplied by the participant.

▥Source basis: Third-party laboratory, as reported

Taken from a laboratory report supplied to SUPPFAX. SUPPFAX did not take the sample, commission the test or verify the result.

§Source basis: Regulatory source

Taken from a regulatory or official source.

¶Source basis: Peer-reviewed study

Taken from a published study.

∑Source basis: SUPPFAX calculation

Derived by SUPPFAX from other source-linked values, by a stated rule.

Evaluation outcome

An outcome may be produced only where a measured result meets or fails a published specification limit on the same grade, in the same unit, by the same method where the limit states one. Nothing else can produce one, and absent that pairing the outcome is not assessed.

✓Evaluation: Within specification

A measured result was compared with a published specification limit on the same material, in the same unit and by the same method.

!Evaluation: Concern

The measured result is reported as a bound, and that bound does not rule out exceeding the limit. This is an assessment outcome, never "no data".

×Evaluation: Outside specification

A measured result falls outside the published specification limit.

–Evaluation: Not assessed

No specification or versioned methodology permits an assessment here, so no outcome is produced. Absence of evidence is never a failure.

Outside specification is an evidenced finding about a sample. It stays visible even when the grade’s own specification satisfies the filter you searched with. A filter is not permitted to hide it.

What was tested matters

Current, expired and undated

◆Freshness: Current

Within its stated validity, and its supporting documents have not expired.

⧗Freshness: Expired

Past its validity or supported by an expired document. It stays visible and traceable; it is not a failure and it is not missing data.

◌Freshness: Undated

Carries no date at all. Evidence of unknown age is not current evidence.

• Expired information stays visible and traceable. It simply stops counting as current: it can never satisfy a requirement, and it is never presented as a failure or as missing data.
• Undated is distinct from current. A claim with no date is information of unknown age, which is a different thing from fresh information, and it is never silently promoted.
• A field’s currentness accounts for the documents behind it. A certificate that expired last month cannot support a claim as current information, even if the assertion itself carries no expiry.

Derived states are never stored

• Unknown is derived from the absence of an eligible assertion. There is no row anywhere in this database whose value is “unknown”, and no invented placeholder is inserted to fill a cell.
• The four search classifications below are derived at query time from the stored facts. They are never stored and they are not a further dimension.
• Coverage figures are counts of records computed on read. None of them is a score, and none is combined into one.

A calculated value carries a derivation record

Where the platform calculates a figure rather than reading one, the calculation is itself a governed record: the versioned calculation rule, the exact input assertions, the governed linkage between them, stage and batch compatibility, the relevant dates and methods, and any rounding or normalization rule.

• Inputs must belong to the same identified test article or sample, the same batch where applicable, the same production stage, and the relevant measurement timepoint.
• A shared document is neither necessary nor sufficient. Two values in one report may concern different samples, and two authenticated reports may concern the same sample.
• Never calculate across different samples, batches, production stages or transformation boundaries.
• A reported figure has no SUPPFAX derivation record, and a calculated figure is never presented as a reported one.

Unit conversion

• A value published in a unit that cannot be converted exactly is left as published and classified unknown, with the reason shown. Mass per volume cannot become mass per mass without a declared density, so it is not converted.
• The conversions performed are only these: 1 %w/w to 10 mg/g, 1 g/day to 1000 mg/day, 1 t to 1000 kg.

Unknown, not applicable, and the difference

■Availability: Evidenced

At least one published assertion exists for this field.

·Availability: Unknown

No eligible published evidence exists. Unknown is the absence of evidence: not zero, not a pass and not a failed result.

Conflicts and disputes

=Consistency: Uncontested

No active reviewed conflict or dispute record touches this field.

≠Consistency: Conflicting

A SUPPFAX-reviewed conflict record states that two or more published assertions materially contradict each other. Both stay visible.

¶Consistency: Disputed

A participant has formally contested this record and SUPPFAX has taken the dispute under review. The record is not removed.

A conflict

An explicit, source-linked record that two or more published assertions materially contradict each other. It must name at least two contradicting assertions and must be reviewed and made active by SUPPFAX before it is public. Two numbers being different is a candidate for review, never a published finding, because supersession is a version relationship rather than evidence of contradiction.

A dispute

A participant’s formal contest of a published record. A supplier can raise one and cannot withdraw the record it contests: the right is to respond, not to rewrite. The disputed state is visible on the field it concerns while the dispute is reviewed. What a dispute names is not currently public: the derived state is, the scope is not. That boundary is a deliberate open decision, recorded rather than presented as settled. Contesting a record needs no account: what that sends is a private submission for review, and it does not itself mark anything disputed. A visible dispute is a separate reviewed record that only SUPPFAX creates.

Expired

Distinct again, and covered under currentness above. Expired information stays visible, satisfies no requirement, and is never rendered as missing or as a failure.

How structured search classifies a grade

For every criterion you set, each grade is classified into exactly one of four states, derived at query time.

✓ Supported: satisfies the requirement

Current published information establishes that the criterion is met. The result card names which record did it and what kind of record it was, so a specification and a batch reading are never mistaken for each other.

✕ Not met: excluded entirely

Current published information establishes that the criterion is not met. Such a grade is not a result at all: not among the matches, and not in the unconfirmed group either.

? Unknown: never satisfies

No current published information can answer the criterion. It never passes a requirement and it is never presented as a failure. Such grades appear only in a separate group, below the matches, and only when you ask for them.

· Not applicable: neither satisfies nor fails

A governed ruling that the criterion does not apply to this grade. It does not count as a match. A grade whose vegan-suitability criterion had been ruled inapplicable must not answer “must be vegan”, and a grade whose oxidation index is not measurable for its matrix must not answer a limit on that index. A filter may accept a governed non-applicability only where that specific criterion declares in advance that it does, with a stated reason. No filter in this release does.

What a threshold actually reads

• A minimum is met by a declaration or a guaranteed-minimum specification at or above it, or by a measured result at or above it, labelled as a measurement.
• A maximum limit is not a value. It cannot satisfy a minimum, and it only rules one out when the cap itself falls below the threshold. Otherwise it says nothing, and saying nothing is unknown rather than a miss.
• Information that exists but is no longer current cannot support a requirement. It is reported as lapsed, which is neither missing nor failing.

The criteria you can search on

This is the complete list. A criterion not on it cannot be applied, and a URL cannot introduce one.

• Ingredient family
• Source type
• Chemical form
• Physical format
• Minimum EPA
• Minimum DHA
• Minimum combined EPA + DHA
• Maximum peroxide value
• Maximum TOTOX
• Declared vegan-suitable
• Current certification · absence is not a denial
• Authorised market
• Has an openable source document · absence is not a denial
• Dosage-form compatibility · absence is not a denial
• Samples available
• Maximum order quantity accepted
• Maximum lead time

The neutral ordering rule

Results are ordered by this rule and nothing else. It is published here from the same declaration the query is built from, so the two cannot disagree.

Ingredient results

1. How many of your stated requirements the record is supported as meeting. A requirement met by current published evidence counts; an unknown one never does.
2. How closely the record matches the words you typed, across the material’s name, SKU, grade, summary and former names. It is zero for every record when you have typed nothing, so it changes nothing unless you searched.
3. How many of the attributes you asked about carry current published evidence at all, so a record that answers your question, even unfavourably, sorts above one that is silent.
4. Then alphabetically by exact commercial name.
5. Then by record identifier, which is unique, so the order is total and the same search always returns the same page in the same sequence.

Supplier results

1. How closely the organization matches the words you typed, across its name, registered legal name and summary. Zero for every record when you have typed nothing, so it changes nothing unless you searched.
2. Then alphabetically by name. With no search term this is the whole order: the directory is a list, not a league table.
3. Then by record identifier, which is unique, so the order is total and the same search always returns the same page in the same sequence.

Omega-3 products

• Name, A to Z: By brand name, then product name, A to Z. the default
• Most EPA+DHA per unit first: Softgels, capsules, tablets and gummies are compared with each other; per ml and per g each come after, on their own. “At least” sorts on its lower limit, and so does a range. “At most”, figures that differ and unknown are not sorted: they come last, A to Z, each with its reason. A product whose packs count different units, such as softgels and capsules, is not sorted either: it comes last, with its reason. Plant omega-3 is never sorted on EPA+DHA. sort=epa-dha in the address bar
• Most EPA per unit first: Softgels, capsules, tablets and gummies are compared with each other; per ml and per g each come after, on their own. “At least” sorts on its lower limit, and so does a range. “At most”, figures that differ and unknown are not sorted: they come last, A to Z, each with its reason. A product whose packs count different units, such as softgels and capsules, is not sorted either: it comes last, with its reason. Plant omega-3 is never sorted on EPA. sort=epa in the address bar
• Most DHA per unit first: Softgels, capsules, tablets and gummies are compared with each other; per ml and per g each come after, on their own. “At least” sorts on its lower limit, and so does a range. “At most”, figures that differ and unknown are not sorted: they come last, A to Z, each with its reason. A product whose packs count different units, such as softgels and capsules, is not sorted either: it comes last, with its reason. Plant omega-3 is never sorted on DHA. sort=dha in the address bar
• Price per 1,000 mg EPA+DHA, low to high: The price per 1,000 mg EPA+DHA uses only euro prices seen in the last 7 days at shops in the Netherlands, and products without that figure come last, each with its reason. sort=price in the address bar
• Price per unit, low to high: A pack’s price ÷ the units in it, from euro prices seen in the last 7 days at shops in the Netherlands; where several shops or packs have one, the smallest. Softgels, capsules, tablets and gummies are compared with each other; per ml and per g each come after, on their own. A product whose packs count different units, such as softgels and capsules, gets no single figure. Products without that figure come last, A to Z, each with its reason. sort=price-per-unit in the address bar

The last term of each rule is a unique record identifier, so the ordering is total: the same search returns the same page in the same sequence, every time.

Portfolio size is not a default ordering factor

The supplier rule used to place organizations with more published grades above organizations with fewer, immediately after text relevance. We removed it. Portfolio size is not supplier quality and it is not relevance. And under SUPPFAX’s planned pricing an organization would pay to activate each record individually, so ordering by it would track commercial spend closely enough to become a ranking that money moves: neutral by the letter, purchasable by effect. Passport activation is not built yet, and no organization can activate anything on SUPPFAX today.

The count has not been hidden. It is still shown on every supplier card, and you can still order by it deliberately. The difference is that you have to ask. The available orderings are:

• Relevance, then A–Z: text relevance, then alphabetically by name. The neutral default. the default
• Published materials (most first): published portfolio size, largest first. A count of published records, not a measure of quality and not an endorsement. Never the default. sort=portfolio in the address bar

Three things stay distinct, and this is the reason the ingredient rule keeps a term the supplier rule dropped. Information about what you asked for is a legitimate default ordering signal: a record that answers your question, even unfavourably, is more useful than one that is silent. Overall coverage is a visible metric and never a default ordering factor. Portfolio size is a visible fact and never a default ordering factor. What makes the first of those fair is that publishing costs nothing: every organization, paying or not, may submit evidence for independent review at no charge, into the same queue, under the same criteria, and payment can never buy faster review, approval or publication.

The supplier rule also has no “matches your filters” term, deliberately. Filters on that directory are absolute: an organization that does not match one is not in the results at all, so every organization you can see already satisfies every filter you set. A sort term for it would be the same value for every row and would imply a distinction the query never draws.

Corrections and supersession

• A correction inserts a new assertion that supersedes the previous one. The previous value stays queryable and appears in the record’s version history with the reason for the change. A published assertion cannot be edited or deleted; a database trigger enforces it.
• A superseding assertion must concern the same grade and the same attribute as the one it corrects, so the chain cannot be used to swap one fact for another.
• Supersession is a version relationship. It is not evidence that two values contradict each other. That requires a reviewed conflict record.
• A supplier can correct, dispute and add information. It cannot silently erase a valid unfavourable record.

How text became structured data

This is recorded on every value, and it is not a measure of how strong the information is. A hand-entered value and an extracted one are the same kind of claim; only the route differs.

⌨Processing method: Entered by a person

A person typed this value from the source.

⌁Processing method: AI-extracted, reviewed

A model read the source and proposed this value; a person reviewed it before publication. How text became data, not how strong the evidence is.

⌁Processing method: AI-inferred, reviewed

A model inferred this value rather than reading it directly, and a person reviewed it before publication.

∑Processing method: System-derived

Computed from other stored values by a deterministic rule.

Where a record is in review

Only published records are public. A participant may submit information and may never govern it: setting a review state, an applicability decision or a source-basis label is a SUPPFAX act, enforced by the database rather than by process.

○Review state: Draft

Not submitted.

◔Review state: Pending SUPPFAX review

Submitted and waiting for review.

◑Review state: Changes requested

Returned to the participant with review notes.

●Review state: SUPPFAX reviewed and published

SUPPFAX reviewed the source-to-claim mapping, its scope and its presentation. It does not mean SUPPFAX tested anything.

×Review state: Rejected

Not published.

▣Review state: Archived

Superseded or withdrawn, and retained so history is never destroyed.

What a model may and may not do

• A model proposes. It never publishes. Generated output points back to approved sources and exposes what is missing.
• A model may propose a value and a scope for review. It may not authenticate a document, a sample identity, a chain of custody, whether a sample represents the marketed product, or whether a laboratory is independent.
• Where the subject or the production stage is ambiguous, that is recorded and never guessed.

The commercial firewall

The principle is that payment only ever buys tools, never a better truth. No paid tool exists yet.

• No ordering is for sale, and none ever will be. Supplier payment, plan, tier or commercial relationship is not an input to ordering, to comparison results, or to any information state.
• This is structural rather than a policy. No plan, payment, tier, subscription, sponsorship, promotion or commission column exists anywhere in the database, and affiliate links live in a separate, private part of the database that no ordering query can reach. An automated test asserts each, and another asserts that no such field appears in the queries that select or order results.
• There is no advertising and no sponsored placement, and paid participation is never shown as a public trust signal. SUPPFAX may earn a commission when you buy through a link marked “Affiliate link”. It never changes the facts shown or the order, and shops without one are listed the same way.
• We do not accept payment to change, remove or soften a published record. A supplier’s remedy for a record it believes is wrong is a correction or a dispute, both of which leave a trace.
• Commercial state lives in private schemas that the public ordering path cannot reach, through neither a database dependency nor an application import. Both are asserted by tests.

What is published today

• No record is published on SUPPFAX for business yet.
• The correction, submission and dispute process is live and open to anyone, without an account and at no cost. What is still outstanding before a named real organization, laboratory or commercial ingredient may be published is the responsible legal entity and the published terms that name it. The production publication path fails closed until an authorized administrator deliberately opens it, and that decision is an audited record rather than a setting.
• A finished consumer product is a further step again, behind its own public and versioned methodology.