Sources and methodology.

Standfirst

Every quantitative claim, every quoted statement, every attributed framing on a Candid Creative page carries a named source, a publication date, a verifiable URL, and an explicit confidence label.

The three buckets
01
Confidence labels

We use three labels, applied per claim, visible on the page.

  • Verified. Primary source quoted — a study, official documentation, court record, or named expert speaking on the record. Most claims on this site sit here.
  • Industry consensus. Multiple independent corroborating sources, none of which is the originator. We name the convergence — not just one of the corroborators.
  • Single-source. One source only. We use the label as a flag — single-source claims are eligible for publication but we say so on the page and we lean on them less.

Claims that meet none of these bars do not appear on the site. Speculation, when it appears, is labelled as such and kept out of factual paragraphs.

Acknowledged sources
02
Where the rules come from

We did not invent any of this. The standards we use are adapted from:

  • BBC Editorial Guidelines, §3.2.2 (sound evidence, well-sourced, thoroughly tested), §3.2.3 (attribution required when uncorroborated), §3.3.4 (preference for named on-the-record sources).
  • Reuters Handbook of Journalism on anonymous sources as the weakest form of evidence.
  • Society of Professional Journalists Code of Ethics on identifying sources clearly.
  • Wikipedia's verifiability policy — "verifiability, not truth."
  • FTC reasonable basis doctrine (1984), which makes unsourced objective claims in commercial speech legally actionable.
The exclusion list
03
Stats we will not quote
  • "A 1-second delay reduces conversions by 7% across all sites." Traces to Akamai/SOASTA 2017 retail data. Defensible as a 2017 retail benchmark; not a universal law.
  • "Amazon loses 1% revenue per 100ms of latency." Reportedly Amazon internal research from roughly 2006. No published methodology.
  • "Google made Core Web Vitals twice as important." Not in any Google primary statement.
  • "53% of mobile users abandon in 3 seconds" applied to 2026 baselines. Original study is from 2016. Quote as historical context, not as a current number.
  • "Faster site = better AI Overview citation rate." No defensible primary research as of May 2026. Severe CWV failure suppresses citation, but "good → great" does not lift it.
When we get one wrong
04
Corrections

We publish corrections at /corrections. For material corrections, a notice persists at the top of the affected page for at least 90 days. For retractions, the original URL serves the retraction notice, with the original text preserved in an annex below it.

To report an error: email corrections at this domain with the URL, the specific claim, and (where possible) a link to a source that supports the alternate version. We aim to acknowledge within two business days and to resolve, or to explain why we disagree, within ten.

Sources & methodology · v1
Corrections log →

Candid Creative