Hypothesis

Bracing infants' stable but sonographically dysplastic hips in a Pavlik harness would improve acetabular development compared with watchful surveillance.

What failed

In a multicentre trial randomising 104 infants aged 3–4 months, the Graf alpha angle at 12 weeks was essentially the same with the harness (60.5°) as with surveillance (60.0°, p = 0.30), and neither the acetabular index at 10 months nor walking age differed.

Broken assumption

Stable dysplastic infant hips need bracing to develop normally.

Why

Most of these hips matured normally on their own within twelve weeks, so bracing added nothing measurable; the authors recommend observation instead.

From the published paper: Pollet et al. Scientific Reports, 2020. ↗CC BY

Hypothesis

Two doses of the 4CMenB meningococcal vaccine would reduce gonorrhoea incidence in gay and bisexual men, as observational studies suggested.

What failed

In the open-label MenGO trial with 2 years of 3-monthly follow-up, vaccination did not significantly reduce gonorrhoea incidence (IRR 0.78, 95% CI 0.40–1.51, p = 0.457), although it raised serum bactericidal titres against N. gonorrhoeae at 6 months.

Broken assumption

The cross-protection against gonorrhoea seen in observational studies of 4CMenB shows up in a randomised trial.

Why

The vaccine was immunogenic, but titres did not rise in men infected at vaccination. The authors note the trial was small, sized for an assumed efficacy above 67%, and disrupted by COVID-19.

From the published paper: Thng et al. npj Vaccines, 2026. ↗CC BY

Hypothesis

A multicentre randomized trial could establish which of two corneal transplantation techniques serves keratoconus patients better.

What failed

The trial cleared every regulatory and ethical requirement and formed the centrepiece of the author’s doctoral work from 2015, but recruitment left it gravely underpowered, so it answered nothing and was never published.

Broken assumption

A trial that is well designed and approved can in fact be recruited to the size its design needs.

Why

The author reflects that warning signs of infeasibility were visible early, and that training such as Good Clinical Practice certification teaches design principles without teaching how to balance scientific value and innovation against recruitment and overall trial feasibility.

From the published paper: Wisse et al. Journal of Trial and Error, 2025. ↗CC BY

Hypothesis

Introgressed Neandertal DNA has been purged by negative selection, which shows up as a steady decline in Neandertal ancestry in Europeans over the past 45,000 years.

What failed

Using a statistic built on two high-coverage Neandertal genomes that does not rely on assumptions about modern human demography, the reported monotonic decline disappeared: there was no evidence of any change in European Neandertal ancestry over 45,000 years. Whole-genome simulations also showed negative selection should not produce a large long-term genome-wide decline.

Broken assumption

The ancestry statistics used earlier were insensitive to gene flow among modern human populations.

Why

The authors attribute the earlier decline to gene flow between modern human populations, which the previously used statistics did not account for.

From the published paper: Petr et al. Proceedings of the National Academy of Sciences, 2019. ↗CC BY

Hypothesis

Two months of 16-hour daily time-restricted eating would improve cognitive function, mood and sleep quality in healthy people, as popular claims for intermittent fasting hold.

What failed

With 76 fasting and 29 control volunteers measured at baseline, one month and two months on attention, memory, working memory, executive function, sleep quality and eight mood dimensions, there were no significant group-by-time interactions on any measure.

Broken assumption

Time-restricted eating improves cognition, mood and sleep in healthy people, not just body weight.

Why

The authors conclude the results do not support the claim; they note the design was open-label and that weight regulation by such diets is a separate, better-supported claim.

From the published paper: Batorek et al. Journal of Trial and Error, 2024. ↗CC BY

Hypothesis

Resuscitating ICU patients with 6% hydroxyethyl starch instead of saline would be at least as safe and improve survival.

What failed

Among 7,000 ICU patients, 90-day mortality was 18.0% with starch vs 17.0% with saline (RR 1.06, P=0.26), and more starch patients needed renal-replacement therapy (7.0% vs 5.8%, P=0.04).

Broken assumption

A synthetic colloid resuscitates ICU patients at least as safely as saline and improves survival.

Why

No survival advantage, and a signal of kidney harm: more renal-replacement therapy and significantly more adverse events with starch.

From the published paper: Myburgh JA, Finfer S, Bellomo R, et al. (CHEST). N Engl J Med, 2012. NCT00935168 ↗

Hypothesis

Because one generalist species (Gould's wattled bat) dominates bat-box use, installing boxes would shift local bat communities towards lower diversity.

What failed

A five-year before-after control-impact experiment at 18 sites in south-eastern Australia, using 9,035 bat-detector nights, found little difference in community composition, diversity or species activity between sites with existing boxes, newly added boxes or no boxes.

Broken assumption

Who uses an artificial roost dictates the composition of the surrounding bat community.

Why

The generalist kept dominating the boxes themselves, but that did not translate into a community-level change over the short-to-medium term studied.

From the published paper: Griffiths et al. Scientific Reports, 2020. ↗CC BY

Hypothesis

CNV amplitude over the supplementary motor area tracks the unfolding of time, so it should vary with timing performance, as Macar et al. (1999) reported.

What failed

Two replication studies did not find the expected performance-dependent CNV changes, and more powerful linear mixed-effects analyses also found no link between performance and CNV amplitude. Habituation effects on the CNV did appear.

Broken assumption

The CNV amplitude is a direct readout of an internal time accumulator.

Why

Habituation is not expected if the CNV is a pure time-accumulation signal, so the authors conclude the CNV does not directly reflect elapsed time.

From the published paper: Kononowicz et al. Frontiers in Integrative Neuroscience, 2011. ↗CC BY

Hypothesis

An LLM-driven care robot can be trusted to describe its own capabilities accurately when placed in a healthcare setting with vulnerable users.

What failed

In a case study of commercially available LLM-based care software, the robot asserted that it could set and manage medication reminders, and volunteered that service unprompted, although no such function existed.

Broken assumption

A conversational agent’s statements about its own functions can be taken as true.

Why

The authors attribute the behaviour to the hallucination and deceptive-pattern tendencies of the underlying language model, with no oversight layer in the deployed product to catch it.

From the published paper: Ranisch et al. Digital Society, 2025. ↗CC BY