An evidence tier is a short label attached to every claim about a supplement, describing how dependable that claim is before it feeds the app's recommendation ranking. Tiers compress thousands of pages of heterogeneous literature into a discrete signal that a human can act on quickly. The same compound can carry different tiers for different outcomes — creatine is robust for strength, mechanistic for cognition, anecdotal for hair — and Unfair keeps those tiers separated per claim rather than per ingredient.
The four tiers
Unfair uses a four-level ladder. Each step is a material jump in data quality, not a smooth gradient.
| Tier | What it means | Typical source | Ranking weight | Example claim |
|---|---|---|---|---|
| Anecdote | One or more users reported it; no controlled data | Forum posts, self-experiment logs | 0.15× | "Tongkat ali raised my morning mood" |
| Mechanistic | Plausible biological pathway; animal or in vitro data only | Cell lines, rodent models, biochemistry | 0.35× | "Berberine activates AMPK, so metabolic benefit is plausible" |
| Preliminary human | One or two small human trials, often open-label or unreplicated | Pilot RCTs, open-label cohorts | 0.6× | "A 60-person trial found a sleep-onset improvement with apigenin" |
| Robust human outcome | Multiple replicated RCTs or a credible meta-analysis | Pooled RCTs, registered-replication literature | 1.0× | "Creatine monohydrate 3–5 g/day improves strength output" |
The ranking weights above are the default multipliers Unfair applies when a claim contributes to a recommendation score. A tier change shifts the recommendation — it does not remove the ingredient from the library.
Why tier is separate from effect size
Tier answers "how sure are we this works at all," not "how much does it work." A compound can sit at the top tier with only a small effect size — tens of trials agreeing on a 5% improvement. Another compound can claim a large effect at a low tier — one dramatic study that replication will almost certainly deflate. Both signals matter independently, and Unfair displays them as two separate chips on the recommendation card rather than collapsing them into a single star rating.
What pushes a claim up or down
Claims move between tiers as the literature updates. Unfair's evidence quality metadata pipeline watches for a handful of signals that justify a tier change.
- Replication. A second independent RCT that agrees in direction promotes a preliminary claim toward robust.
- Publication context. Registered trials and pre-registered protocols count more than post-hoc subgroup findings at the same sample size.
- Population match. A trial in postmenopausal women does not elevate a tier for 25-year-old men at the same weight as an age-matched trial would.
- Dose alignment. If the only supportive trial used 1200 mg but the shelf product contains 150 mg, the real-world tier for the user's product is lower than the study tier.
- Conflicts of interest. Manufacturer-funded single-site trials are held to a stricter replication bar before they lift tiers.
A claim can also drop. When a well-powered trial fails to replicate, Unfair demotes the tier rather than averaging the conflicting results.
How tier drives the recommendation ranking
Tier feeds directly into the recommendation ranking. The effect is visible in three places. First, low-tier ingredients need to pass a higher adherence threshold before they generate a confident recommendation — a randomized controlled trial-backed compound rises faster. Second, lower tiers trigger softer language ("may help," "some evidence suggests") while the top tier uses direct language. Third, low-tier claims contribute less to any stop-or-swap logic — a mechanistic-only claim cannot outweigh a robust-human-outcome concern when the two disagree.
Common tier mistakes users make
- Treating one study on the internet as robust because the effect size was large.
- Trusting mechanistic data from rodents as if it were human trial data.
- Assuming a compound's tier for one outcome extends to a related outcome it has never been tested on.
- Over-weighting anecdote from a trusted podcast host; the source does not change the tier.
How this appears in Unfair
Each ingredient card in the library shows tier per claim, with the linked citations one tap away. The recommendation ranking uses tier as a direct multiplier on score, so a user who filters for "robust human outcome only" sees a visibly shorter, higher-confidence stack. Review cycles never promote a compound's internal rating above its source tier, even when a user's own logs look positive.
Clinical safety note
A high evidence tier is about population-level reliability, not personal safety. Strong evidence for an average benefit does not mean the compound is safe in the presence of a specific medication, pregnancy, or condition. Safety-critical decisions should route through a clinician regardless of how confident the tier looks.