How DoseRank ranks
Every ranking on DoseRank carries its reasoning. This page publishes the full method — how a goal becomes a ranked list, how evidence is judged, what refreshes the data, the policies that keep placement honest, and every change to the model.
From goal to ranked list
DoseRank starts with an outcome, not a brand. Four steps turn a goal like “sleep better” into an explainable, ranked shelf of products.
Outcome matching
Your goal is mapped to a predefined health outcome — never a SKU or a keyword. Every product relevant to that outcome enters the pool, before any availability or relevance filtering.
Ingredient form scoring
Each ingredient form is scored separately for the outcome — magnesium glycinate and magnesium oxide are different answers. Form scores come from the ingredient–outcome evidence map.
Product scoring
The label’s per-serving dose is checked against the studied effective window (dose fit), then blended with the ingredient evidence strength and study confidence into the Clinical score.
Ranking
Products are ordered by the composite DoseRank score — the clinical score gated by product reliability signals: trust, value, and demand. The gate can only trim a score, never boost it.
Composite outcome scoreStage 1 blends the clinical core: a weighted geometric mean of ingredient evidence, dose fit, and study confidence — where study confidence itself is 80% evidence volume and 20% effect size.
Stage 2 multiplies that clinical score by a reliability gate built from trust, value, and demand signals, with the emphasis value > trust > demand. The gate can only trim the clinical score, never boost it — the composite never exceeds the clinical score (composite ≤ clinical), so a product can’t out-market weak evidence.
Composite outcome score
Clinical × Value × Trust
Clinical × Trust
Clinical × Demand × Trust
Every lens re-ranks the same scored set — nothing is rescored to flatter a tab. When an outcome has a thin clinical pool, lens picks are labeled “· early signal”; the DoseRank composite is unaffected.
A fixed vocabulary, defined once
The same seven terms appear on every score card, table, and comparison — no synonyms, no renamed metrics. This is the canonical definition of each.
The overall DoseRank score — a weighted blend of clinical strength, evidence, dose fit, trust and value.
Powered by — clinical score × the reliability gate (trust, value, demand).
How strongly the formula's ingredients and doses match what clinical studies used for this outcome.
Powered by — ingredient–outcome evidence map × dose adequacy.
How much published human research backs the formula's key ingredients for this outcome — both how many studies exist and how strong they are.
Powered by — study count and study strength for the key ingredients.
How closely each key ingredient's dose sits inside the clinically studied range for this outcome.
Powered by — label dose vs. the studied dose window, per ingredient.
How well customers rate this product, with extra weight given to ratings backed by a large number of reviews.
Powered by — customer ratings weighted by review count.
How the price per serving compares with similar products for this outcome — 50 is the typical price, higher means better value for your money.
Powered by — price per serving vs. similar products — median-anchored, 50 = typical.
How strongly this product sells compared with the other ranked products for this outcome.
Powered by — sales strength vs. the other ranked products for this outcome.
What counts as evidence
Scores move on published human research and verifiable product data — nothing else. When the evidence is thin, the ranking says so instead of guessing.
What moves a score
- Human studies only
Animal-only and in-vitro findings never raise a score. Evidence strength comes from published human research — how many studies exist and how strong they are.
- Dose inside the studied window
Dose fit rewards a per-serving dose that lands in the range studies actually used. Beyond that range, extra milligrams stop earning credit — the reward saturates at a bounded cap, so the biggest number never wins by default. Harmful or negative findings subtract from the evidence directly.
- Trust from real customers
Trust reflects how well customers rate a product, with extra weight for ratings backed by a large number of reviews — buyer signals on record, not marketing claims.
How evidence is labeled
- Strong evidence
A solid pool of direct human studies supports the ingredient–outcome link.
- Early signal· early signal
The clinical pool is thin. Lens picks carry this label so a weak signal is never dressed up as a verdict.
- Not enough data
Shown as “Not enough data for this view yet” — a gap is stated, never filled in.
What never moves a score
- Payment from a brand
There is no price for a position, a badge, or a mention.
- Affiliate commission rates
Commission terms are held apart from the ranking pipeline and never enter a score.
- Marketing claims
Front-of-label promises carry no weight. Only the supplement-facts panel is scored.
- Ad spend or popularity alone
Demand appears only inside the Popularity lens, and only multiplied by clinical strength.
Rankings carry their date
Prices, stock, and ratings drift. Every ranking snapshot is stamped, and gaps are named instead of hidden.
Every ranking page shows its snapshot stamp
Rankings are recomputed after each crawl cycle, and the stamp is printed beside the results — never buried in a footer.
Recomputed after each crawl cycle · placement is never paidOnly in-stock products are ranked
Availability is an entry gate — only products currently in stock enter the ranked shortlists. Price and rating history is kept across crawls, so how a listing moved stays on record.
Missing data is named, not filled
When extracted data is incomplete, the page says exactly what’s missing — “The latest extracted data does not include image variants yet.” — instead of estimating.
What keeps the ranking honest
No paid placement
Rankings and placement are determined by the scoring model only — never by payment from a brand. This holds for every surface: score cards, top picks, comparison tables, and search results.
- No sponsored slots, boosted listings, or paid badges
- Brands cannot pay to be re-scored or re-ranked
- Commercial terms never reach the scoring pipeline
How the model changes
No score weight, dose window, or evidence mapping changes quietly. Every change follows the same four-step path.
- 01Proposed with rationale
Every scoring change starts as a written proposal citing the studies or data that motivate it, judged against the evidence standards above.
- 02Review
Independent advisors check ingredient–outcome mappings and dose windows against the underlying research before shipping results.
- 03Versioned release
Approved changes ship behind a new model version, and rankings recompute under it. Each ranking page stamps the snapshot its results were computed from.
- 04Logged and correctable
Every release lands in the changelog below. Found a data error? Report it — corrections are recomputed and noted, not silently patched.
Every change to the model
Clinical blend recalibrated — ingredient evidence 55% / dose fit 30% / study confidence 15%. A bounded dose-overage reward replaced the old overdose decay.
First versioned ranking profile. Clinical-gated composite, the value/trust/demand reliability gate, and lens shortlists.