How we grade evidence and risk

Sources last reviewed October 2026

Every compound page in Roicel shows two labels: an evidence rating for how much controlled human research exists, and a risk tier for how hard the compound is on the body at the doses people typically report. This page explains how both are assigned so you can judge them yourself.

Evidence ratings

The rating describes the quantity and quality of human research, not how well a compound works. A compound can have strong evidence that it does very little.

Each entry links to the study or review it is based on. Where a figure is quoted (for example kilograms of lean mass gained), it is the figure from that study at the dose the study used, which is also shown. We prefer randomised trials and meta-analyses in humans; surveys of users are labelled as surveys and never raise a rating.

Risk tiers

Tiers compare compounds with each other, not with doing nothing. They weigh five things at typical reported doses: liver strain, cardiovascular load (blood pressure, hematocrit, heart structure), lipid effects (HDL and LDL), hormonal suppression and recovery time, and mental-health effects (sleep, mood, aggression). Dose and duration change the picture, and no tier means "safe".

Each compound's page states in one sentence why it sits in its tier. If you think a tier is wrong, email the study that shows it and we will look.

Half-lives and estimated levels

The levels chart is a first-order model: each dose decays with the compound's half-life and the curves add up. Every half-life figure carries a confidence label. Label means it comes from the prescribing information. Study means a published pharmacokinetic paper. Estimate means no human data exists and the figure is inferred from the ester, used only so the planner can draw something. Real levels vary with absorption, body weight, injection site and individual metabolism, which is why the app says bloodwork is the only real measurement.

Safety checks

The checks on the Home tab are fixed rules, not judgement calls: for example a 17α-alkylated oral running longer than six weeks, two orals at once, a hematocrit at or above 54%, or liver enzymes above twice their upper limit. The thresholds come from endocrine-society guidance on testosterone therapy and standard lab ranges. They are deliberately conservative.

The AI bloodwork read

When you upload a report, an AI model transcribes the markers, values and reference ranges, and you confirm them before anything else happens. The explanation it then writes uses the confirmed values, your plan, your age and sex if set, and your earlier results. It is written to explain, not to diagnose: it never tells you to change a dose, and any value in a dangerous range is answered with "contact a doctor" rather than a suggestion. The file itself is never stored. The model and date are shown on every result.

Vendors

A compound page links to a vendor only if we have ordered and tested that vendor's product ourselves. Ratings, tiers and wording are written before any link is added, a vendor cannot pay to be listed, and if a link earns a commission the page says so. Stock and price shown are a snapshot with its date.

Review cadence and corrections

Research entries are reviewed together; the date is shown on each research page. Compound directory entries carry their own review date. When a reader points out an error with a source, the entry is corrected and the date updated. Write to [email protected].

Independence

Roicel has no sponsor, no advertiser and no vendor with a say over content. The person who writes the entries is the same person who answers the email above.