1. The answer comes first
The first sentence of a page answers the question in its title, in language you can lift and quote without ambiguity. Nothing is placed above it — no throat-clearing, no scene-setting.
2. Evidence is weighed, not counted
How a study was designed sets how much authority it gets: who was studied, how many, whether there were controls, whether anyone was blinded, what was actually measured, and when. A randomized trial earns a paragraph and a verdict. A case series earns a description and its limits. Ten weak studies do not outvote one good one.
3. A case report stays a case report
One patient can raise a question or describe an individual response. One patient cannot establish a diagnosis for a population, a general mechanism, or that a treatment works. When a study’s own authors say their findings do not generalize, we take them at their word.
4. We say which population a study used
People referred to an environmental clinic are not the same group as people who meet the 1999 consensus criteria. “Chemical intolerance” on a survey is not automatically MCS. Conclusions drawn from one group do not silently transfer to the other, and our pages name the group.
5. Every “it helps” names what improved
Helped is not a result. Coping, fear, sleep, illness perception, symptom intensity, daily function and actual chemical tolerance are seven different outcomes, and a treatment can move one without touching the others. Our pages say which one a study measured — particularly when a program advertises one and measured another.
6. Small uncontrolled studies are not treatment evidence
A handful of people, observed without a control group, can describe something worth investigating. It is not proof that a treatment works, and we do not frame it as one.
7. Exposure reduction gets the weight it has earned
Across the consensus criteria, the survey evidence, the real-life exposure work and decades of patient experience, reducing the exposures that repeatedly make a person ill has broader support than anything else. We say so plainly. We do not inflate it into “the only proven treatment” or demand “complete avoidance,” because the evidence does not say that either.
8. Open mechanisms stay open
TRP receptor sensitization, mast cell activation, autoimmunity, dysautonomia, small-fiber changes, genetics, predictive coding, oxidative stress — each is a candidate for what happens inside the body, and none has been established as the universal answer. We rank the candidates and say why — receptor sensitization is the strongest, because it is the only one tested by provocation in living patients — and we never write “the cause” for any of them, because the evidence does not carry that word.
9. The strongest evidence against us stays on the page
Where a topic has serious contrary evidence, it appears on the page that discusses that topic, described accurately and with its limits — the 2006 provocation review, the 2015 randomized trial of mindfulness-based therapy, the Japanese five-year cohort, the 2024 WHO-commissioned review of radiofrequency exposure. Hiding the other side’s best evidence is exactly what we accuse our critics of doing.
10. No number without a source
Every figure — a sample size, a percentage, a year, a count — traces to a citation. If we cannot confirm the source, the figure does not print, even when it would help our case. We also do not publish running totals that expire the moment the page changes.
11. Verdict voice, where a verdict exists
We write “the weight of the evidence supports” where it does, and we do not hedge it into “some research suggests.” Equally, we do not write “proven,” “only,” or “causes” where the evidence says less than that. Both failures mislead; they just mislead in opposite directions.
12. Withdrawn framings do not come back
Some things this site once published are no longer supported by the evidence, and they are retired rather than quietly softened: MCS as failed detoxification or “toxic load”; MCS as a one-way progressive ratchet; electromagnetic fields as an established cause of EHS symptoms; a 2024 letter to a specialist journal presented as that journal’s position; a genetic panel presented as an explanation of anyone’s MCS.
13. Every source is labeled with what it is
A citation is better than no citation, so where no journal article exists for a point we will cite something else — but never without saying what it is. Four labels, and they appear in the source line itself: peer-reviewed (a journal article, with a PMID or DOI); government or institutional (the EPA, the ADA, the WHO, a national institute); clinical or organizational (a clinic, a patient organization); and community record (a survey, or documented lived experience).
One limit holds the rest together: a source that is not peer-reviewed never carries a contested claim on its own. On anything a critic would dispute — cause, mechanism, whether a treatment works, whether exposure lowers the threshold — the lower-tier source stands beside a journal source, never instead of one. A skeptical clinician who follows our strongest-looking claim to an advocacy page has been handed the argument. On descriptive questions, though — what daily life is actually like, what an accommodation involves, what patients report — a patient organization may stand alone, because on those subjects the organizations often are the primary record.
14. We click the link before we publish it
A citation is not checked until its identifier has been resolved and the paper it returns has been compared to the claim sitting next to it. Correct-looking citation text is not evidence of anything.
Correction, 16 September 2026. The research library behind /science/ and every page’s Key Research block was audited against PubMed and Crossref on this date. Of 86 entries, 77 were kept: 33 carried a DOI that pointed at another paper or did not exist, every citation line was rewritten from the resolved record, and 34 summaries were cut to what the paper supports. Nine entries were removed: two duplicates, one single-patient case report, and six that resolve to no published paper. Every library entry now carries its PubMed or DOI identifier and a register entry, and the build blocks on any that does not.
We learned this the hard way. An audit of every PubMed identifier on this site found four that pointed at real, indexed, entirely unrelated papers — a study of bone cells, one on the evolution of milk secretion, one on gestational diabetes, and one on urinary tract infections — each sitting under a citation whose author, year, title, journal, volume and page numbers were all perfectly correct. Nothing about them looked wrong on the page. They were found by resolving all forty-two identifiers and reading back what each one actually was. Every identifier on this site has now been through that check, and every new one goes through it before it ships.