RESEARCH / 01
Useful because the boundary is explicit.
Each substantive guide should answer a buyer question directly, identify the evidence needed for the next decision, and say what that evidence cannot establish.
Official and primary sources come first
Start with the source closest to the claim: the law, official register, regulator or body that issued a standard. Other guides can add context. A vendor or forum post cannot replace an available primary source for a material claim.
Source links and their check date are shown near the relevant guide. A link proves neither that a specific supplier was checked nor that the source will remain unchanged.
Claim, evidence and inference are separated
A source fact stays close to what the record says. Our reading of it, including a checklist or suggested next step, is labelled through the page’s method and limits. A failed lookup or unclear match stays unresolved. It is not a clean result.
We avoid claims that a supplier, factory, certificate, payment or transaction is “safe”, “approved” or guaranteed by a public-record screen.
AI assistance does not replace review
I use AI to help draft, translate, restructure and write code. I remain responsible for the published pages. AI output is not a source and is not cited as evidence.
Automated checks cover specific links, dates and page behaviours. They do not prove every claim has been checked by a person. A translated page does not imply native-language review.
Review and dated updates
Review covers the page’s purpose, readability, links, source support and limits. Automated checks also cover initial HTML and metadata. A modified date records a page change; it does not mean every external source was checked again that day.
Commercial service pages are held to the same evidence boundary as editorial pages. Pricing or availability must be confirmed before fulfilment starts.
Corrections
Email hello@currawongweb.com with the page URL, disputed passage and supporting source. Material corrections update the page and its modified date. We do not silently rewrite an evidence boundary to preserve a sales claim.
For questions about the organisation behind the pages, see About Currawong. For application data paths, see the privacy notice.
The Currawong Evidence Standard, v1
Four rules, published 21 August 2026. You may quote or cite this standard with its version and date. It sets the requirements for our pages and reports; it is not an independent audit of compliance.
- ES-1 · Source hierarchy
- A material claim traces to the highest available tier: the official register, statute or standards text first; a licensed commercial republication second, named as such; vendor and forum statements are context, never the basis of a material claim.
- ES-2 · Mandatory query date
- Every fact drawn from a record carries the date it was read — per field, not per report. A fact without its query date is treated as unverified: it is re-read or removed, not carried forward.
- ES-3 · Gap disclosure
- What could not be checked is stated next to what was. A failed lookup, an ambiguous match or an unavailable source appears in the output as a stated limit — never silently dropped, and never converted into a clean result.
- ES-4 · Fact, inference, decision
- What the record showed (fact), what we infer from it (inference) and what the buyer then decides (decision) are labelled separately. No fact is inflated into a verdict, and no verdict is presented as a fact.
Citation form: “Currawong Evidence Standard v1, 21 August 2026, currawongweb.com/editorial-policy/#evidence-standard”. Corrections to the standard itself follow the same corrections path above and increment the version.
Whether the published pages follow the standard
The scan below counts evidence markers and nearby limits notes in the English pages. It does not verify each underlying fact or establish full compliance with ES-2, ES-3 or ES-4. Counted 6 September 2026.
| Standard | Result |
|---|---|
| Pages where every evidence marker has a query date | 119 of 119 |
| Pages with an evidence marker | 119 of 125 |
| Distinct measurement sets behind them | 38 |
| Pages with a limits note beside the evidence | 118 of 119 |
What this check can and cannot tell you matters as much as the number. It verifies that a date is present and that a limits statement exists. It cannot verify that the date is honest or that the limits are the right ones. A machine can enforce the shape of the standard; only reading can enforce its substance.
A page without a measurement marker still needs review. Policy and service terms should state the rules that apply to the user; unrelated research tables do not make those rules more credible.
Whether the sources we cite are still reachable
Requiring a source on every fact is worth little if the source has since gone. On 6 September 2026 we requested every outbound link in the body of our own pages and recorded what came back. This measures our citations. Whether a reader can open a Chinese government portal is a separate and dated study.
| Result | Links | What it means |
|---|---|---|
| Returned a page | 170 of 219 | 77.6 per cent. The citation still resolves for a reader. |
| Of those, returned only after a redirect | 48 | The address we published has moved. The document is still there. |
| Refused an automated request | 36 | Answered 403 to our script. Concentrated in standards and DOI hosts, which block automation by policy. |
| Genuinely gone | 1 | One provincial PDF, withdrawn by the department after we read it. The page citing it says so and links to the department instead. |
| Did not answer, or answered with another error | 12 | Four returned nothing at all. Eight more came back with a server or gateway error. Neither result tells us the document is gone. |
| Government hosts among the above | 74 of 93 | 10 of these refusals were government hosts declining automation. |
Read the 403 row before reading the headline number. A refusal aimed at scripts is not a dead link. Those pages open normally in a browser, so the share a reader can actually reach is higher than 77.6 per cent. We report what the script got back. The alternative is to keep retrying until the number looks good. There was no retry and no second attempt from a different network.
What this cannot tell you: whether the content behind a working link still says what we said it said. That is what the query date on each fact is for, and re-reading is the only way to settle it. The full result, link by link, is published as a dataset.