Factuality, source fidelity and usefulness are three different scores

What we found

TrustGrowth published a score of 43 for trustgrowth.ai on 2026-09-06 (TrustGrowth proof, 2026-09-06). The supplied proof does not show that this score separately measures factuality, source fidelity, or usefulness.

That limit matters when you audit AI-assisted content. A draft can contain statements that are factually correct and still fail because it misrepresents its sources, omits important limits, or leaves the intended reader without useful context. In this article, we present factuality, source fidelity, and usefulness as three separate review questions. This is a proposed editorial method, not a measured TrustGrowth result, benchmark, product comparison, or component breakdown of the published score.

Why one overall score can hide three different problems

A single score can summarize something useful when its method and boundaries are clear. But a score alone does not tell you which editorial problem needs attention. The supplied proof does not identify separate measurements for factuality, source fidelity, or usefulness (TrustGrowth proof, 2026-09-06).

Use these three questions when reviewing a draft:

• Factuality: Are the draft’s statements supported by the available evidence?
• Source fidelity: Does the draft represent each source accurately, including its scope, qualifiers, date, and uncertainty?
• Usefulness: Does the draft help the intended reader understand an issue or decide what to do without adding unsupported certainty?

These are proposed working definitions for the review method in this article. Keep them separate because each points to a different editorial action.

A factuality problem may require verification, correction, or removal. A source-fidelity problem may require restoring a qualifier, date, scope, or limitation. A usefulness problem may require clearer context or a bounded next step. Treating all three as one quality judgment can conceal what needs to change.

Do not treat these definitions as evidence that TrustGrowth currently measures the dimensions separately. The supplied evidence supports only the published score and its date.

How a factual draft can still fail the reader

Failure mode 1: The statement is true but the evidence is misrepresented

A source can support a narrow statement while a draft presents a broader one. Consider this a proposed illustrative review pattern, not a measured example: a source describes an observation for a particular date, group, product, or set of conditions. A draft removes the date or scope and turns that observation into a general statement.

The revised sentence might remain broadly true in ordinary conversation. It can still fail the source-fidelity check because the reader is no longer shown the limits attached to the evidence.

When you review a material claim, compare the draft with the source across four points:

1. What exact statement does the source support?
2. What date or period does it cover?
3. What population, sample, product, or situation does it describe?
4. What uncertainty or qualification does it include?

A citation alone does not establish that the wording is faithful. Open the cited source and compare the claim line by line. If the source says “in this analysis,” do not silently rewrite it as “in general.” If it reports an association, do not describe it as a cause unless the evidence supports that change.

A claim can therefore pass a basic factuality check while failing source fidelity. The central proposition may not be false, but the draft has changed how far the evidence appears to travel.

Failure mode 2: The draft cites a source but does not answer the reader’s question

Attribution and usefulness are different review tasks. A draft can name a source, link to it, and accurately repeat its conclusion while leaving the intended reader unsure what applies to their situation or what to check next.

Usefulness does not require a stronger claim. It requires relevant context and a bounded next step. Depending on the draft, that may mean identifying which part of the source applies, naming an unresolved question, or telling the reader which evidence to inspect before acting.

For example, a proposed editorial test could ask whether a reader deciding between two actions can identify the evidence relevant to that decision, the limitation that may change its interpretation, and the next verification step. This is a review question, not evidence of a measured reader outcome.

Do not claim that a particular wording improves comprehension, decisions, or performance unless you have measured that outcome with a stated method. No such outcome evidence has been supplied for this article.

Failure mode 3: A clean summary hides uncertainty

A summary can retain a central finding while omitting the method, date, scope, or uncertainty that limits how the reader should use it. That omission can make the sentence look more useful while making it less faithful to the evidence.

Brevity is not a substitute for fidelity. If an omitted qualification changes the interpretation or responsible use of the claim, keep that qualification close to the sentence it limits.

When you cannot determine whether a limitation matters, record the unresolved question. Do not fill the gap with an assumption. A clean summary should remain bounded by the evidence that supports it.

A practical three-score review

The following worksheet is a proposed editorial method. It is not an established TrustGrowth benchmark, and it does not produce separate measured results for the published score of 43 (TrustGrowth proof, 2026-09-06).

Use the same questions for every material claim:

1. What exact claim is being made?
2. What source supports it?
3. Does the source support the wording, scope, date, and certainty used?
4. What would the reader need to know to use the claim responsibly?

Start by recording the evidence and the gap. Do not begin by assigning a number. If your team later adds numerical ratings, document how those ratings were designed and label them as a draft editorial method unless you have supporting validation. Do not attach thresholds or results that have not been sourced.

Proposed review questions; no measured results are supplied for these dimensions.

Dimension Check Pass condition Failure signal Factuality Can the claim be verified in the cited evidence? The evidence supports the claim as written. The claim is contradicted, unverifiable, or more specific than the evidence. Source fidelity Does the draft preserve scope, qualifiers, date, and uncertainty? The draft represents the source without distortion or omitted limits. The draft broadens, simplifies, or strengthens the source. Usefulness Does the claim help the intended reader understand or decide what to do? The reader gets relevant context and a bounded next step. The draft is accurate but vague, irrelevant, or unusable.

Verify every row against the underlying source before publication. The table is a checklist for editorial inspection, not evidence that the dimensions have been measured separately on trustgrowth.ai.

What the supplied TrustGrowth proof does and does not establish

The supplied fact is narrow: TrustGrowth published a score of 43 for trustgrowth.ai on 2026-09-06 (TrustGrowth proof, 2026-09-06).

The supplied proof does not identify the score’s component dimensions, scoring method, sample or test set, comparison set, independent validation, or relationship to reader usefulness. It therefore does not establish that factuality, source fidelity, or usefulness contributed separately to the score. It also does not show that one of those dimensions passed or failed.

This is one published score. A single published score cannot support product, market, or reader-outcome conclusions. This brief does not establish a benchmark for the three dimensions. Treat the score as a dated fact with a defined boundary, not as evidence for the proposed review model.

Method and limits

The evidence available for this article is one TrustGrowth proof page and one published score dated 2026-09-06 (TrustGrowth proof, 2026-09-06). No separate measurements for factuality, source fidelity, or usefulness were supplied.

The missing information includes:

• component definitions for the published score;
• the score’s calculation method;
• the sample, test set, or assessment basis;
• independent validation; and
• separate measurements for the three review dimensions.

Those gaps limit what we can responsibly say. We can describe the published score and offer a proposed claim-review worksheet. We cannot infer how well a draft performs on any of the three dimensions from the score alone.

This brief does not establish a benchmark for factuality, source fidelity, or usefulness. A single published score cannot support product, market, or reader-outcome conclusions. The proposed definitions and table are editorial tools for review, not observed measurements.

Do this next

Apply the proposed method to one draft:

1. Select one draft and list each material claim.
2. Link each claim to its source.
3. Mark whether the source supports the wording, scope, date, and certainty.
4. Rewrite any claim that is broader or more certain than its source.
5. Add the context a reader needs to use the claim.
6. Record unresolved questions instead of filling them with assumptions.
7. Keep factuality, source fidelity, and usefulness as separate review notes.

Use the three-question review on one draft this week: verify the facts, preserve the sources' limits, and check whether the result helps the intended reader. Record where the draft passes and where the evidence is still missing.

Last measured

Last supplied measurement: TrustGrowth published a score of 43 for trustgrowth.ai on 2026-09-06 (TrustGrowth proof, 2026-09-06). The supplied evidence contains no next re-run date. No next re-run date is supplied.