The most-missed decisions are over-caution: what 2026 Practice Lab decisions show

The findings in brief

• 64% of below-standard decisions were partial, not unsafe (272 decisions, 60 learners).
• The two most-missed decisions both punished reflexive caution, and every miss on each was the same cautious choice.
• Asked why they got a confident answer wrong, more learners chose "I thought my answer was the safer choice" than any other reason (13 of 24).
• 58% of wrong answers that carried a confidence rating were made at 4 or 5 out of 5, yet answers given at the highest confidence were still right most often.
• Time taken to decide did not separate right answers from wrong ones.
• People who left a scenario mostly left before their first decision (68% of unfinished scenarios).

What we measured

The CRLN Practice Lab presents realistic clinical trial situations, one decision at a time, written for a specific role. Each decision is scored on a four point scale: 4 is the strongest response, 3 meets the standard, 2 is a partial response that falls short of the standard, and 1 is unsafe. Learners can also say how confident they are, from 1 to 5, before they see the result.

The population is 91 learners who completed at least one scored scenario between July 13 and October 10, 2026. Five roles had at least five learners: Clinical Research Coordinator (43 learners), Clinical Trial Administrator (24), Clinical Research Associate (21), Research Assistant / Study Coordinator I (13) and Clinical Data Manager (5). Eight other roles had fewer than five learners each; their decisions are included in the totals and never reported on their own.

CRLN staff and test accounts are excluded. These figures are as of October 10, 2026; the counts on our home page are live and keep growing.

1. Most mistakes were partial, not unsafe

Of 272 decisions that fell below the standard, 173 (64%) were partial responses, scored 2 of 4, and 99 (36%) were unsafe, scored 1 of 4. The typical miss was a reasonable instinct applied in the wrong place, not a dangerous act.

Sample: 272 below-standard decisions from 60 learners. Partial misses came from 48 learners and unsafe misses from 38. Range for the partial share, resampling learners: 57% to 69%.

Caveat: Unsafe choices are still more than a third of misses. This finding is about where most errors sit, not a reason to worry less about safety.

2. The two most-missed decisions both punished reflexive caution

The decision missed most often put a new coordinator, trained but not yet on the delegation log, in front of a request to help with ordinary clinic work. 9 of 17 attempts fell below the standard, and all 9 were the same choice: the most cautious one, declining more than the situation required.

The second most-missed decision asked which reporting timeline applies to a serious adverse event reported on a weekend. 6 of 13 attempts fell below the standard, and all 6 were again the same choice: the strictest option applied across the board, rather than the timeline that fits the case.

Sample: Ranked among decisions reached by at least five learners. First decision: 14 learners, with the cautious choice made by 8. Second decision: 11 learners, with the cautious choice made by 5.

Caveat: These are two decisions with small samples. They show a pattern worth testing, not a rate for the profession.

3. Learners said so themselves

When a learner answers wrongly after rating themselves 4 or 5 out of 5 for confidence, the Practice Lab asks why and offers reasons to tap. Of the 24 learners who explained a confident wrong answer this way, 13 chose "I thought my answer was the safer choice". No reason was chosen by more learners: "I misread the question" was chosen by 9 and "I didn't know the rule applied here" by 7.

Sample: 24 learners who tapped a reason after a confident wrong answer, between September 13 and October 10, 2026 (51 reasons in all, 14 of them this one).

Caveat: The question is asked only on some answers, and answering is optional, so these learners chose to explain themselves. A learner can give different reasons on different answers.

4. When people were wrong, they were usually sure

Of 264 wrong decisions that carried a confidence rating, 152 (58%) were made at confidence 4 or 5 out of 5. Forty-eight learners made at least one confident wrong answer.

Confidence still carried information. Decisions made at confidence 5 met the standard 94.7% of the time (1,445 decisions, 73 learners). Decisions made at confidence 1 to 3 met it 78.9% of the time (532 decisions, 39 learners).

Sample: Wrong decisions with a confidence rating: 264, from 58 learners. Range for the confident share, resampling learners: 44% to 78%.

Caveat: More than half of all answers were given at confidence 5, so confident wrong answers partly reflect how often people feel sure. Confidence was not recorded on about 5% of decisions.

5. Speed did not separate right from wrong

The median time to decide was 34.9 seconds for decisions that met the standard and 33.7 seconds for those that did not. The only weaker group was decisions made in under 10 seconds, which met the standard 79.7% of the time against 86% to 90% for every slower group.

Sample: 1,476 timed decisions from 48 learners, made from September 4, 2026, when timing began to be recorded.

Caveat: The under 10 second group is small (123 decisions, 14 learners). Time on screen includes reading, so it is not pure thinking time.

6. People left before the first decision, not partway through

Of 168 scenarios that were started and never finished, 115 (68%) ended before the learner made a single decision. Once people began deciding, most of them finished.

Sample: 168 unfinished scenarios from 128 learners, out of 178 learners who started any scenario.

Caveat: A few of these may still be in progress. We cannot see why people left, only when.

Method and limits

• Small sample. 91 learners, most of them in four roles. No result here is a rate for the clinical research workforce.
• Self-selected learners on one platform. People chose to practice, and practice is not the job. These are decisions made in simulation.
• People who practice more weigh more. A learner who made 50 decisions counts 50 times in a decision rate. The average learner met the standard on 85.6% of their decisions, against 90.1% for all decisions pooled. Learners with fewer than 10 decisions averaged 79.1% (30 learners); learners with 30 or more averaged 92.2% (23 learners).
• Decisions cluster within people, so the ranges shown resample whole learners rather than single decisions (95%, 4,000 resamples).
• The content changed during the period. Answer order was randomized from September 9, new scenarios were added, and scoring moved to the server on October 5. Results before and after those dates are not directly comparable.
• Privacy. Every figure is an aggregate. No group smaller than five learners is reported, and no country breakdown is shown because too few learners have declared a country.
• What we do not publish. Scenario text, answer options and scoring keys stay private, so the practice remains a fair measure. Decisions are described in general terms only.
• A snapshot. These figures describe data as of October 10, 2026, and will change as more people practice.

What sites can do

This is not an argument for less care. It is an argument for teaching the rule precisely enough that people do not have to guess on the safe side.

• Teach the boundary, not only the prohibition. New staff need to know what they may do before delegation as clearly as what they may not.
• Treat unnecessary caution as an error in feedback. Declining work that is in scope, or misjudging which reporting timeline applies, has real costs even when it feels safe.
• Ask people how sure they are, and review the confident misses first. That is where a wrong belief is held most firmly.
• Do not read speed as competence. Fast and slow answers were right at about the same rate.
• Protect the first decision. Most people who leave a scenario leave before deciding anything, so set aside time in onboarding for the first one.
• Report readiness per person, not pooled. A pooled rate is pulled up by the people who practice most.

Try one decision yourself

One realistic situation, one choice, scored with an explanation. No account needed to start.

How to cite

CRLN (Clinical Research Learning Network). The most-missed decisions are over-caution: what 2026 Practice Lab decisions show. October 2026. Data as of October 10, 2026. https://crln-learn.com/findings/decisions-october-2026

Figures on this page may be quoted with the date and the URL. If a figure is corrected, the correction will be listed on our corrections page.