The Radon Connect trust standard
A list of promises is worth nothing. Anybody can publish one, nobody can check it, and the promises published most often — vetted, trusted, best — are the ones kept least.
So each of the 12 commitments below carries the mechanism that enforces it, and most of those mechanisms are build-time checks rather than intentions. You cannot audit our sincerity. You can read what a check does, and a check that fails the build is a promise somebody would have to deliberately delete rather than quietly forget.
We will not tell you a request was delivered unless it was
No confirmation message claims a provider has your details on the basis of the form having submitted successfully, your postcode resolving, or your result being high enough. Those were all, at some point, the thing that wrongly stood in for delivery.
How it is enforcedOne function is permitted to authorise delivery language, and it takes the delivery outcome and nothing else. Confirmation copy declares whether it asserts delivery, and a build-time check compares that declaration against the outcome — a mismatch fails the build. A mutation test deletes the check and confirms the suite catches it.
We will not describe providers as vetted, screened, or trusted
Those words describe a completed process. Where we have completed one we will say what it was, what it covered, and on what date. Until then the words are not used, because the alternative is a claim that sounds specific and means nothing.
How it is enforcedA prohibited-claims scanner runs over every built page and fails on the phrases that assert a verification we have not performed. It also scans its own source, so the prohibition list cannot pass by matching itself.
We will not rank, score, or recommend one provider over another
No best-of lists, no star ratings, no "top rated", no scoring rubric. We hold no performance observations on any business, so a ranking would be a preference presented as a finding.
How it is enforcedThe question builder has no scoring field by construction, and the schema scanner rejects review and rating structured data anywhere on the platform. Both are checked on the built output rather than on intent.
We will not publish a price, an average, or a statistic we did not observe
No cost benchmark, no "homeowners typically pay", no invented response times, no fabricated demand or conversion figures. Where a number appears it comes from a named source or from our own dataset with its observation count published — including when that count is zero.
How it is enforcedDatasets carry provenance and a real observation count, and an instrument with no observations reports N=0 rather than being quietly omitted. The claims scanner fails the build on price-benchmark phrasing.
We will not stamp a page as updated when it was not
The "last updated" date on any page is derived from its actual revision record rather than set to today at build time. A page nobody has touched in a year says so.
How it is enforcedThe editorial record has no writable updated field. The date rendered is computed from the revision log, so refreshing it requires actually logging a revision.
We will show you what we cannot determine before we show you a result
The limits of a personalized output appear beside or before it, never as a disclaimer underneath. A limit read after a conclusion has formed changes nothing.
How it is enforcedThe limits block renders inside the tool runner above the result, and every library entry carries it. The published list of what we cannot determine is required to be longer than the list of what we can — if that balance ever inverts, the build fails.
We will never present an unknown as a failure
Device calibration not established, chain of custody unknown, conditions unrecorded — these are the ordinary condition of a home test kit, not marks against your house. They are rendered as open questions, with the same visual weight as anything else and no warning styling.
How it is enforcedUnknown items are structurally forbidden from carrying a value, and the evidence card component gives the unknown group the same weight as the established group. A QA gate checks the built pages for failure styling on unknown items.
We will not tell you to throw away a measurement
Every entry in the testing library says what your result still supports, whatever went wrong. Whether a result should be set aside is a judgement for somebody who can review how it was produced — not a conclusion to reach from a web page, and not one worth reaching from a page that cannot see your device.
How it is enforcedThe library framework requires a "what this does not settle" section on every entry and refuses one without it, then scans the prose for discard language and fails the build on it.
We will not invent a provider, a licence, or a coverage area
No placeholder companies, no illustrative contractors, no example licences. Where we have no provider in an area we say the area is editorial, and the request is recorded rather than routed.
How it is enforcedMarket state is derived from the provider registry rather than declared. A market asserted as active with no assignable provider reports as blocked and names the reason, instead of being believed.
We will not tell you what the law requires of you
Requirements vary by state and by transaction, and most states require disclosure rather than action. Where we describe a requirement it belongs to a named authority and is cited so you can check it — including when the answer is that no requirement exists.
How it is enforcedRegulatory statements are held in the state registry with citations attached, and the claims scanner fails on language that converts guidance into obligation.
What you type into a tool stays in your browser
Every decision tool and the question builder run entirely client-side. Your result, your foundation, and what you ticked are never transmitted, because there is no server to receive them.
How it is enforcedThe site is a static export with no runtime backend other than the single form endpoint. A bundle gate checks that the content registries never reach the browser, and the tools have no network call to make.
We will tell you where this platform stops
At the point where somebody has to look at the building. Whether your test was run correctly, whether your result is accurate, whether a system would work in your house, whether your home is safe — none of these is answerable from a description, and we say so rather than approximating.
How it is enforcedEvery library entry ends with what genuinely requires somebody on site, and the framework refuses an entry with that section empty.
What this standard does not promise
- That we are right. Enforced checks stop specific failures; they do not make a judgement correct. Where we are uncertain we say so, and the page-level confidence blocks are where that lives.
- That a provider will be good. We hold no performance observations on any business and will not pretend otherwise. What we can do is tell you what to ask and what a complete answer contains.
- That we have no commercial interest. We do. Contractors pay when a request is routed to them, which is stated plainly on how we work. The response to a conflict of interest is disclosure and structural limits, not a claim of independence.
- That this list is complete. It is what we have committed to and can currently enforce. Additions are more useful than assurances that nothing is missing.
How to check any of this
Most of it is visible from the outside. The list of what we cannot determine is longer than the list of what we can, and you can count both. Every outcome a request can reach is published before the form rather than after it. Every page carries its sources, its revision record, and what it does not claim.
Where a claim rests on something you cannot see — a provider licence, a laboratory record, an authority’s current position — we say who holds it and how you would check it yourself, which is the honest substitute for asking you to take our word.
When you are ready to talk to somebody
Describe your situation and we will tell you honestly whether we can route it to a provider in your area today, or whether we cannot.