How we test a claim before publishing it

Measured, not asserted

When Schkovl claims a result, the before and after are shown along with the test that produced them. A capability list is easy to write and proves nothing; a dated measurement that somebody else can re-run is harder to produce and worth more.

The editorial standard behind that is explicit: material factual claims need a source that supports their actual scope, and recommendations, illustrative calculations and hypotheses must be kept separate from measured results. A number with no supporting evidence gets removed or corrected — it does not get promoted into an industry benchmark.

What counts as evidence

The strongest form available to an agency is the same test suite run twice against the live site, with both dates on record and the raw material behind each score independently fetchable.

The NorthGuard Window Film engagement is published on that basis. A baseline audit on 9 July 2026 scored 35/100. The same audit, re-run against the live production site on 10 August 2026, scored 98/100. The duplicate-HTML check that failed across all nineteen routes passes across all nineteen, confirmed by fetching each URL and diffing the raw response rather than by trusting a dashboard. A Schkovl co-founder holds a partner stake in NorthGuard Window Film, so this is not an arms-length client relationship; the full disclosure is on the case-studies page.

What makes that publishable is not the size of the movement. It is that the only variable which changed between the two runs was the HTML shell itself — nobody added content, bought links or increased ad spend in between — and that anyone can fetch the same URLs today and check the current state for themselves.

What a technical audit score is not

Both NorthGuard numbers are technical audit scores from one test suite, run a month apart against the live site. They measure crawlability and indexability. They are not traffic, they are not rankings, and they are not revenue, and Schkovl does not present them as any of those things.

The distinction matters because the defect capped what the site could technically achieve. Removing a cap is not the same as measuring a business outcome. There is no traffic or revenue figure attached to that fix, and none is claimed.

The editorial standard says the same thing in general terms: a technical audit score is not a search ranking or a revenue result, and a study should state its question, dates, sample, method, outcomes, exclusions and limitations.

Where a published engagement involves a party that is not arms-length, the relationship is disclosed on the case study itself alongside the evidence.

What we publish when there is no baseline

Not every engagement produces a comparable number, and the honest response is to say so rather than to manufacture one.

The GlassGuard Window Film case study documents a head-metadata rebuild and six legacy URL redirects. It carries no ranking, traffic or audit-score claim, because unlike NorthGuard there is no before-and-after on record — producing one would have required running the same audit tool against a real baseline that does not exist for that site.

The published metrics there are counts rather than improvements: fifty-four indexable URLs in the sitemap, six legacy service-area pages remapped to their new locations. A count is a fact. An improvement without a baseline is a guess.

Corrections stay visible

Original publication dates are retained. A substantive correction carries an update date and a note describing what changed, and the note stays on the page.

The GlassGuard study carries exactly such a note. An earlier version described the page body as shipping in the initial response; an internal audit found that this overstated the fix, since only the head is prerendered and body content still depends on client-side JavaScript. The correction is dated and left visible rather than quietly rewritten, which is the only version of a corrections policy that can be verified from outside.

Corrections can be requested. Email the company inbox with the article URL, the passage and the supporting evidence. When a material error is substantiated, the affected page is corrected and known syndicated copies are checked.

AI assistance and authorship

AI tools may assist with drafting, analysis or coding. Their output is not a source of evidence on its own, and human authors remain responsible for checking claims and calculations. Material use of generated or synthetic data is disclosed in the affected work.

Authorship follows the same rule. A named byline identifies the person credited for an article; company-authored work carries the Schkovl name; a reviewer is not automatically an author. A personal biography describes that author's role — it is not an implied certification of every topic the article touches.

What we decline to publish

There is no client logo wall and no testimonial section on this site. The case studies page currently carries an empty testimonial list rather than filler, and the portfolio is limited to engagements with dated, re-runnable measurements.

Industry benchmarks with no traceable source do not appear. Neither do projected results presented as achieved ones. Where a historical description depends on platform behaviour or pricing that can change, the linked source is the thing to check before relying on it.

Applying that standard has meant removing numbers from our own published work. Several articles carry dated correction notes recording exactly which unsupported figures were taken out and why.

Where to go next

Last reviewed

The facts on this page were last checked against published Schkovl sources on 2026-09-20.

Everything here is drawn from material already published on this site. If something looks wrong, our editorial standards and corrections policy explains how to have it corrected.