For data managers, statisticians, programmers and medical writers
Trial data passes through four teams, and each is trained in its own layer. TrialFluent is a complete, fictional phase 3 study where you follow any value from the form at the clinic to the number in the report — and see what each team did with it, and why.
Which one are you?
The orange dot is a value moving through the study. Whatever your role, you see it at one, two or three of these stops. The friction lives in the others.
Follow one value through all five →Why it matters
None of these is a mistake anyone would be blamed for. Each is what happens when one team can't see what the next one needs — and each is a built-in defect of the guide study that you can open and walk through.
The rescue medication page needs a start date because, eighteen months earlier, a statistician chose a strategy for that intercurrent event. A gap there doesn't just raise a query — it changes which values count in the primary analysis. The data manager cleaning the form has no way to know that.
What it costs: the primary result rests on a field nobody prioritised.
Subject 0104-1003 has two HbA1c results in the baseline window — one at screening, one on Day 1. The SAP says the one closest to the target day counts; the other stays in the dataset, flagged out of every analysis. The site did everything right. There was never a query. One result still vanishes from the number — by a rule written before the study began.
What it costs: an argument at lock that neither side can win, because they're reading different documents.
A transposed body weight — 46.8 instead of 84.2 — is a possible human weight, so no edit check fires. A person finds it three weeks later, in a listing sorted by change from baseline — after the interim data cut. The statistician who designed the checks never saw the listing; the data manager never knew the interim was already frozen.
What it costs: an interim and a final that disagree, and a paragraph in the CSR explaining why.
"LS mean change −1.14" goes into a sentence. What model produced it, which subjects it includes, why one subject's Week 26 value isn't in it — the writer has the table, not the trail. The programmer has the trail, and nobody asked. Review comments go back and forth for a week.
What it costs: a CSR cycle spent reconstructing what one trace shows in a minute.
CDISC documents its layer. ICH documents its layer. Every training provider teaches a vertical. Nobody teaches the seams — and nothing lets you watch the data cross them. That's what TrialFluent is for.
How it works
Not a course and not an encyclopedia — a complete trial with the explanation attached, so you learn the other layers by looking at what they actually did.
MERIDIAN-3 is a fictional phase 3 study — 390 subjects, 17 forms, SDTM, ADaM, 49 outputs, a CSR — generated so that every layer is consistent with every other. Any number can be opened. Any cell can be traced to the field someone typed.
Open the primary result →Fifteen stages from objectives to submission, 101 topics beneath them, and the fifteen handoffs between teams written up as pages of their own: what crosses, what "done" means on each side, and what breaks when the assumption is wrong.
Read the lock handoff →Say which team you're on and the same material is arranged for you: the layers next to yours explained in your terms, the handoffs you stand on with your side named, and the tour steps where your own artefact crosses.
Choose your role ↓And forwards
The study is served in two states — the database as the sites left it, and as it was declared final. Switch, and every downstream artefact switches with you while you stay on the same record.
Subject 0311-1076's Week 12 weight passed every edit check: 46.8 kg is a possible human weight, and the range check allows 35–200. A person found it three weeks later, in a listing.
| USUBJID | VSSEQ | VISIT | VSDTC | WEIGHT (kg) | Check |
|---|---|---|---|---|---|
| 0311-1076 | 20 | Week 8 | 2024-06-03 | 84.6 | passed |
| 0311-1076 | 25 | Week 12 | 2024-07-01 | 46.884.2was 46.8 | passed |
| Week 12 | Placebo | Sematiglizin 25 mg |
|---|---|---|
| Change, mean (SD) | −0.41 (3.533)−0.10 (1.011) | −0.50 (0.980) |
| LS mean change (SE) | −0.40 (0.20)−0.09 (0.09) | −0.50 (0.20)−0.50 (0.09) |
| Difference vs placebo (95% CI) | −0.10 (−0.65, 0.44)−0.41 (−0.66, −0.16) | |
| One transposed weight triples the placebo SD, doubles every standard error, and the confidence interval crosses zero.Corrected: the standard errors halve and the treatment difference excludes zero. Served from the study, both states. | ||
Start from your side
Same study, same pages — arranged as a reading order for your team: the layers next to yours, the handoffs you stand on with your side named, and the tour steps where your own artefact crosses.
What statistics does with what you send — and why the SAP reached into your CRF eighteen months before you saw a query.
Start here → I work inStatisticsHow the CRF got designed, what a query actually involves, and what the writer does with your outputs.
Start here → I work inStatistical programmingWhere the specifications you program from came from, and where your outputs go after you deliver them.
Start here → I work inMedical writingThe analyses you describe, explained from the estimand down to the table cell — so the review cycle gets shorter.
Start here →New to all of it? The same material is orientation: take the tour first, then the stages in order →
The tour and the trace are free, no account. No cohort, no deadline. Plans for the full study →