# How we check what we publish > The method behind every claim on this site: sourcing, evidence labels, review and correction. - Language: English - Page: https://smilehandbook.com/en/methodology/ - Français: https://smilehandbook.com/fr/methodologie/ - العربية: https://smilehandbook.com/ar/methodology/ - All published pages: https://smilehandbook.com/llms-full.txt What follows is the route a statement travels before it reaches any page on this site. We wrote it so that it can be used against us. Find a page that skipped a step described here and you have found a defect worth reporting — and the defect belongs to us. No clinic can buy its way onto this site, and we are neither a clinic nor an agency nor a booking service for clinic treatment or travel; those, though, are statements about intent. Mechanism is what this page covers. ## The stages a claim passes through Every factual statement we publish moves through four stages. Sourcing comes first: before anyone drafts the sentence, someone traces where the claim originates and writes that origin down. Classification is second: the claim gets a label naming the kind of evidence beneath it, and that label sets the ceiling on how strongly it may be worded. Third comes the wording review, where a claim supported only by a single unofficial source may not be phrased as a general truth, and any sentence that would nudge a reader toward treatment or away from it is recast as a description of what is known. Publication is fourth, and the label travels with the claim, so the evidence class is visible to the reader instead of being left to guesswork. When a claim fails at any stage, we do not water it down until it slips through. It either goes, or it stays as an open question with the gap spelled out. One kind of sentence is singled out at the wording review, and it carries a name here: a clinical proposition. A clinical proposition is a sentence a reader can carry into a decision about their own health or treatment; it asserts an effect, a lifespan, a risk or an indication, or it recommends a course of action. Two examples give its shape: "see a dentist if this happens" is one, "this material lasts longer than that one" is another, and both stay clinical propositions whether we quote a named source or say them in our own name. A clinical word sitting inside a sentence about method or procedure does not make one: "this page counts the published reviews on crown lifespan" carries a clinical word and no decision at all, and the same holds for an administrative or pricing rule that names a treatment without asserting anything about its effect. A single clinical proposition is enough to place a page among those that require the practitioner review line described at the foot of this page, and the count stops at the first. The two sentences quoted just above are named here to show the shape, not offered as advice by this page: naming a clinical proposition in order to define it does not make one, which is why this page is not itself placed among those requiring that review line. ## How we choose sources First-hand material beats anybody's summary of it. Our order of preference runs: the regulation or law in its own words; whatever the institution holding the authority publishes itself (health ministries, dental councils, licensing registers); peer-reviewed clinical literature; guidance issued by professional bodies. We do not count journalism, forum posts or marketing pages as evidence for a fact. They can count as evidence that something is being said, which is a different claim and is written differently. When a source is a translation, we say so. Where a regulation has since changed and the older text is what we describe, we name the version. And when we cannot reach a primary source — the registry is offline, or blocks us, or exists only on paper — we log that we could not reach it instead of slotting in a secondary source and calling the two the same thing. ## What the evidence labels mean Regulation marks a statement lifted from a legal text or an official register, checkable at that source. Institutional marks one drawn from a named authority's own publication. Clinical literature marks one drawn from peer-reviewed research, with the usual limits of that research stated alongside it. Our own measurement marks something we gathered and counted ourselves, the method set out. Reported, not verified marks a claim we have watched being made repeatedly without being able to confirm it; that label warns the reader, it cites nothing. Site statement marks a sentence about how this site itself works — a rule, a method, a limit — with no outside source. Absence marks a passage about something we looked for and did not find, or set out to measure and could not; the gap is declared, not dropped. Our evidence audit page counts the passages under each label. ## Rules we follow when we measure something ourselves Three rules carry most of the weight. We choose the sample before we see any result, and we write down what we expect to find, so a disappointing outcome cannot slide quietly into being a different study. We stratify instead of skimming the top of a list — the "five most-reviewed clinics" shortcut is one we refuse on purpose, though we have not measured what any clinic spends on marketing and so make no claim about how saturated that stratum is; what we can say is that skimming it would answer a narrower question than the one we set. And we state the ceiling of our own tools, because a tool limit that goes unreported turns into a false finding. ## A worked example — counting 2,115 review stars Across six Istanbul dental clinics we gathered 2,115 publicly visible reviews and counted the star ratings ourselves. Five stars accounted for 93.85 percent of them. Saturation that heavy traps the corpus-wide average mathematically: whatever the remaining reviews say, it cannot land outside roughly 4.75 to 4.94. The bound is a property of the corpus rather than of each clinic within it — three of the six clinic averages sit outside it, and the margins are nowhere near equal: one far below at 4.20, and two above it much more narrowly — 4.97 a little further out, 4.95 only barely. Each of those averages we counted review by review, rounded only for printing. Between our recomputed averages and the publicly displayed ones the gap never exceeds 0.052; on the platform's one-decimal display five of the six land on the same figure, and on the sixth ours slips a notch — 4.9 against a displayed 5. Read that as a check on our collection rather than a finding about the clinics, and bear in mind that we do not know when the platform computed the figure it shows. We then read by hand 50 one-star reviews — every one-star review in the corpus with enough text to say what happened. At least seven describe a failure that appeared after a delay, and at least seven describe losing contact with the clinic after returning home. We also looked for the signatures of fabricated reviews and did not find them: the median five-star review ran 187 characters, template-like phrasing showed up in 6.3 percent, and reviewer-farming patterns — a single account reviewing more than one of the six clinics — showed up in 14 of 2,085 reviewer profiles. We matched those accounts by display name rather than by a platform account id, since the platform surfaced no account link we could use. The error runs both ways: different people sharing a name are read as one account, and one person posting under different names is read as several. So 14 is imprecise in both directions rather than a ceiling. Because we did not find it, we do not write that clinics are commissioning fake reviews. Absence of evidence is reported as absence of evidence. We do not publish the raw dataset; in this analysis we name no clinic, quote no review verbatim, and name no reviewer. Leaving the clinics unnamed here is not a promise that they cannot be identified: the reviews are public, matching them against other records could point to a particular clinic, and we do not guarantee that it cannot be done. ## How we declare what we could not measure We set out to test whether reviews arrive in bursts. We have not weighed the coordination signals available to us against one another, so we do not call this one the strongest; we chose it because our own collected data could in principle carry it. We could not. Our tool stopped at roughly 490 reviews per clinic, and 43.9 percent of the dates came back only as a year. The honest sentence, then, is that timing clustering was not measurable under this design — not that no clustering exists. That distinction is the entire point of this page. ## What does not exist yet A named medical editor has been appointed. Content pages now end with a line naming the dentist who wrote the page and the medical editor who checked it. Behind that line stands a written sign-off from the writing dentist, dated 6 September 2026. That approval was given in conversation on 1 September 2026; it was not put on paper until the later date, and our corrections log carries that correction. It was a collective approval, given for the site as a whole rather than page by page. Our approval line runs page by page instead: its first signature records were written on 3 September 2026, against pages that were still drafts that day. Our approval-line page sets out the steps — a page carrying a clinical proposition is read by the medical editor, and a dated signature record is then either written against that page or withheld, bound to the text of the clinical passages it approves. Every such page has now come out the other side: 41 pages carry at least one clinical proposition, and each of the 41 holds a signature record — 2 dated 3 September 2026, 3 dated 5 September 2026, 20 dated 6 September 2026 and 16 dated 10 September 2026. What does not exist yet is the rest of it. One person signed all of them, the medical editor, which is 11 of the 11 English clinical pages, 14 of the 14 French ones and 16 of the 16 Arabic ones, a concentration of 100 percent in each of those three lines; our ceiling tightens by itself as approvers are added and binds on nobody until the second one is on record, so the figure is printed here rather than left to be worked out. We publish no credential class for either name, because we hold no document to support one. And no page a reader can open here prints a date of review: the date sits in the signature record, not on the page, so the byline still says a page was checked but not when. Our corrections log records these very changes; every entry in it is dated and the log only grows, so its length is a statement about the age of the log, not about our accuracy. ## Sources used on this page - _calisma/plan-olcum/puan-band-olcum.json - _calisma/yorumlar/*.json - _calisma/bir-yildiz.json - _calisma/bir-yildiz.json · plan/16-YORUM-METNI-OLCUMU.md:355,360 - _calisma/plan-olcum/sablon-izi-olcum.json