Research Methods in 2026: The Shift from Transparency to Verifiability

For two decades, the reform agenda in research methodology had a simple slogan: be open. Share your data. Register your study. Publish your code. That agenda worked, in the sense that it became normal. Preregistration is routine in clinical trials and increasingly common elsewhere; data availability statements are near-universal in major journals; reporting checklists sit in most submission systems.

What the last few years have exposed is that openness is a means, not an end. A study can be fully transparent and still be wrong, underpowered, or impossible for anyone else to check. The question now driving methodological policy is narrower and harder: can an independent party actually verify what a study claims?

That question is reshaping practical decisions across the research economy — how protocols are written, which reporting guideline applies, when artificial intelligence may be used in a systematic review, and what a funder or regulator should demand before treating a result as decision-grade. This article maps the current state of that shift for professionals who commission, conduct, review, or consume research: principal investigators, R&D and clinical development leads, methodologists, editors, policy analysts, and evidence users in industry and government.

You will learn what the empirical replication record actually shows, what changed in the major standards during 2024–2026, how the AI question is being governed in evidence synthesis, where preregistration reaches its limits, and which practices are worth adopting now.

---

Executive summary

• Large multi-site replication projects across psychology, economics, cancer biology and, most recently, Brazilian biomedical science consistently report replication rates well below half, with effect sizes markedly smaller on repetition.
• The Transparency and Openness Promotion (TOP) Guidelines were comprehensively updated for the first time in a decade; the update took effect in 2025 and was published in Research Integrity and Peer Review on 5 August 2026.
• TOP 2025 reframes the objective as verifiability of empirical claims, separates seven Research Practices from two Verification Practices and four Verification Study types, and deliberately retires the ambiguous umbrella term "preregistration".
• CONSORT 2025 (30 items) and SPIRIT 2025 (34 items), both published in April 2025, added dedicated open science sections — reporting standards and transparency standards are converging.
• Four major evidence synthesis organisations — Cochrane, the Campbell Collaboration, JBI and the Collaboration for Environmental Evidence — issued a joint position in 2025 permitting AI in synthesis under human oversight, with mandatory disclosure of any AI use that makes or suggests judgements.
• Evidence that prospective specification changes results is strong in trials and registered reports; evidence that it works cleanly in laboratory biology is weaker, because protocols frequently cannot be fully prespecified.
• Transparency is not a substitute for methodological rigour, a point the TOP 2025 authors make explicitly; scoring systems that reward disclosure can invite performative compliance.
• The practical implication for professionals: design studies so that a stranger with your outputs could reach your conclusion — and decide, at protocol stage, which parts of that chain you can realistically deliver.

---

Why methods came under scrutiny: what the replication record shows

The replication literature is often summarised badly, in both directions. It does not show that most published research is fabricated, and it does not show that everything is fine.

What it shows is more specific: when independent teams repeat published experiments under prespecified conditions, a minority of results reproduce by conventional criteria, and observed effects shrink substantially. The most recent large-scale entry is the Brazilian Reproducibility Initiative, whose full analysis appeared as a reviewed preprint in eLife on 10 July 2026. Fifty-six laboratories attempted 143 replications of published Brazilian biomedical experiments using three common methods; 90 replications across 45 experiments passed independent validation. Replication rates ranged from 20% to 44% depending on the criterion applied, and a multiverse analysis across all defensible analytic choices produced a median rate of 25%. Effect sizes were smaller in 41 of 45 replications.

Three features of that project matter more than the headline number.

First, the criterion determines the answer. Statistical significance in the same direction, inclusion in a prediction interval, and researchers' own subjective judgement gave materially different rates within the same dataset. Any single replication percentage quoted without its criterion is close to meaningless.

Second, variability was higher in replications than in originals — substantially so for PCR experiments — which mechanically lowers power and depresses significance-based replication rates. Part of the "failure" is measurement noise, not falsity.

Third, prespecification proved genuinely hard. Protocol deviations were recorded in about 95% of replications, most often for reasons intrinsic to the experimental model or to infrastructure and supply logistics rather than researcher discretion. That is an important qualification to any argument that preregistration alone fixes laboratory biology.

Table 1. Selected large-scale replication and verification efforts

Table compiled for OneWise from the primary reports; ranges reflect the fact that each project applied several replication criteria simultaneously. Figures are not directly comparable across projects because sampling frames and criteria differ.

---

From transparency to verifiability: what changed in TOP 2025

The Transparency and Openness Promotion Guidelines, first published in Science in 2015, became the default policy vocabulary for journals: eight modular standards, each adoptable at one of three levels. More than 5,000 journals and organisations expressed support, and the original paper had been cited over 3,000 times by March 2026.

The update — approved by an independent advisory board, in effect from 2025, and formally published on 5 August 2026 — is not cosmetic. It states an explicit objective for the first time: improving the verifiability of empirical research claims. Everything else follows from that.

Three structural changes are worth internalising.

1. Practices, verification, and study types are now separate things. TOP 2015 mixed discrete practices (data sharing), combinations of practices (computational reproducibility), and study formats (replication) into one list of standards. TOP 2025 splits them: seven Research Practices (study registration, study protocols, analysis plans, reporting transparency, materials transparency, data transparency, code transparency), each implementable as Disclosed, Shared and Cited, or Certified; two Verification Practices (computational reproducibility and results transparency), both performed by a party independent of the researchers; and four Verification Studies — replication, registered reports, multiverse analyses, and many-analysts studies.

2. "Preregistration" has been disaggregated. The guidelines now ask which discrete outputs were created, when, and where they were shared, rather than whether a study was "preregistered". This is a meaningful gain in precision: a registered trial with no public analysis plan and a study with a timestamped analysis plan but no registry entry were previously described by the same word.

3. Scoring has been deliberately de-emphasised. Ordinal levels were replaced with categorical classifiers and the TOP Factor journal rating has been discontinued, explicitly to reduce the risk that standards drive performative rather than substantive compliance.

Table 2. TOP 2015 and TOP 2025 compared

Original comparison prepared for OneWise from the TOP 2015 and TOP 2025 source documents.

---

The reporting layer: CONSORT 2025, SPIRIT 2025 and the guideline ecosystem

Reporting standards are the operational surface where most professionals meet these ideas. Two major updates landed in April 2025.

CONSORT 2025, published 14 April 2025 across several journals, supersedes CONSORT 2010 and comprises a 30-item checklist plus a participant flow diagram. SPIRIT 2025, its protocol counterpart, specifies 34 minimum items and was developed through a three-round Delphi process involving 317 participants and a 30-expert consensus meeting. Both added a dedicated open science section; SPIRIT 2025 also added an item on patient and public involvement and aligned with ICH GCP E6(R3) and the 2024 Declaration of Helsinki.

The ecosystem around them is large and still growing: a research-on-research study sampling the EQUATOR Network library found 657 indexed reporting guidelines and extensions as of 30 January 2025. For prediction models, TRIPOD+AI (published 16 April 2024) harmonised reporting for regression and machine learning approaches, and PROBAST+AI (2025) extended risk-of-bias appraisal to the same territory.

The convergence is the point. A decade ago, transparency standards and reporting standards were separate conversations. Now the reporting checklist is one of the mechanisms through which verifiability is delivered — which is why TOP 2025 renamed its old "design and analysis transparency" standard to reporting transparency, defined by adherence to a reporting guideline.

Figure 1 (design brief for OneWise). Title: "The methods assurance stack." Purpose: show how four layers combine to make a claim verifiable. Format: four stacked horizontal bands, bottom to top, with a vertical arrow labelled "verifiability increases" on the left. Band 1 (bottom), Design layer — question, estimand, prespecified analysis plan, power. Band 2, Conduct layer — protocol adherence, deviation log, data management. Band 3, Reporting layer — CONSORT 2025 / SPIRIT 2025 / TRIPOD+AI checklist, flow diagram. Band 4 (top), Verification layer — independent computational reproducibility check, results-transparency check against registration, replication. On the right, a callout box lists the outputs each band produces (analysis plan; deviation log; completed checklist; verification report), with dashed arrows from each band to its output. Caption: "Each layer produces an artefact; verification is only possible when all four artefacts exist."

---

AI in the research workflow: governed, not banned

The most consequential methods debate of 2025–2026 concerns where machine assistance is legitimate. Evidence synthesis moved first, because the labour economics there are brutal: a conventional systematic review can take a year and cost six figures.

In 2025, Cochrane, the Campbell Collaboration, JBI and the Collaboration for Environmental Evidence published a joint position statement simultaneously across their journals, formed a shared AI Methods Group, and endorsed the Responsible use of AI in evidence SynthEsis (RAISE) recommendations. The position is permissive but conditional. Synthesists remain accountable for the synthesis, including the decision to use AI; AI may be used provided authors can demonstrate it does not compromise methodological rigour or integrity; human oversight is required; and any use that makes or suggests a judgement must be fully and transparently reported.

That last clause is the operative one. It converts AI use from a private efficiency decision into a reportable methodological choice, on the same footing as a search strategy or a risk-of-bias tool.

The countervailing risk is already documented. The TOP 2025 authors note emerging evidence that generative tools can be used to mass-produce formulaic papers from public datasets — a pattern documented in 2025 for analyses built on the US NHANES database — and acknowledge the uncomfortable possibility that open data policies could inadvertently make such output easier to generate. This is a genuine tension in current policy, not a settled question.

Table 3. Where machine assistance currently sits in evidence synthesis

Maturity labels reflect current practice guidance rather than any formal certification scheme; positions in this table are contested at the margins and are likely to move.

---

What prespecification can and cannot do

Two bodies of evidence bear on the value of prespecification, and professionals should hold both.

The supportive evidence is strong where designs are simple and stakes are high. An analysis of 55 large NHLBI-funded cardiovascular trials found that 17 of 30 (57%) published before 2000 reported a significant benefit on the primary outcome, compared with 2 of 25 (8%) published after 2000 — a period in which prospective registration on ClinicalTrials.gov became universal for such trials. In psychology, a comparison of 71 registered reports with 152 conventionally published hypothesis-testing studies found positive results for the first hypothesis in 44% of registered reports versus 96% of standard articles. Neither study proves causation; both are consistent with prespecification suppressing selective reporting.

The limiting evidence is equally important. In the Brazilian project, deviations from registered protocols were near-universal and mostly unavoidable, leading its authors to suggest that preregistration in laboratory biology may be more useful after pilot work, when methods can actually be specified. And the TOP 2025 authors state plainly that both strong and weak research can be conducted transparently — transparency is neither synonymous with, nor a substitute for, methodological rigour. Related debates continue over whether many-analysts and multiverse studies genuinely measure credibility or mainly demonstrate that unconstrained analytic freedom produces variable results.

Common misconceptions worth correcting

• "Preregistered" means the analysis was fixed. Often it means only that a registry record exists. Ask for the analysis plan and its timestamp.
• A failed replication refutes the original. It constrains it. Power, criterion choice and inter-laboratory variability all shape the outcome.
• Data availability equals reproducibility. Independent verification of computational reproducibility is a separate practice — one TOP 2025 now names explicitly.
• AI use is a tooling detail. Under current synthesis guidance, judgement-affecting use is a reportable method.

---

Latest developments (dated)

• 16 April 2024 — TRIPOD+AI published in The BMJ, harmonising reporting for regression- and machine-learning-based clinical prediction models.
• 14 April 2025 — CONSORT 2025 published; 30 items; supersedes CONSORT 2010.
• 28 April 2025 — SPIRIT 2025 published; 34 items; new open science and patient-involvement provisions.
• 2025 — PROBAST+AI released in The BMJ for risk-of-bias assessment of AI-based prediction models.
• Late 2025 — Joint AI position statement published simultaneously by Cochrane, Campbell, JBI and CEE; RAISE recommendations endorsed; version 2.1 in development as of 22 September 2025.
• 10 July 2026 — Brazilian Reproducibility Initiative full results posted as a reviewed preprint in eLife (assessed as "important", evidence "convincing"). Not yet a version of record.
• 5 August 2026 — TOP 2025 formally published in Research Integrity and Peer Review; TOP Factor discontinued.

---

Practical takeaways

For principal investigators and study leads

1. Write the analysis plan as a separate, timestamped, citable output — not a paragraph inside a registry field.
2. Decide at protocol stage which of the seven TOP research practices you can realistically deliver, and say so; a truthful "not shared, for these reasons" is more useful than an aspirational commitment.
3. In laboratory work, run pilots before registering the protocol, and keep a structured deviation log from day one.

For methodologists and biostatisticians

4. Match the reporting guideline to the design before data collection, not at submission; for prediction models, apply TRIPOD+AI and appraise with PROBAST+AI.
5. Where analytic choices are genuinely contested, prespecify a bounded multiverse rather than a single path — and prespecify how you will interpret dispersion.

For editors, funders and research managers

6. Ask for verification, not just disclosure: an independent computational reproducibility check on a sample of accepted papers changes incentives more than a checklist.
7. Treat AI use in synthesis as a reportable method with a named accountable human, consistent with the four-organisation position.

For evidence users in industry and policy

8. Before acting on a single striking result, ask three questions: what was prespecified, what has been independently verified, and how large is the effect relative to its variability?

---

Key insights

1. The organising concept of methods reform has moved from openness to verifiability.
2. TOP 2025 makes that shift explicit and separates practices, verification, and study types.
3. "Preregistration" is being retired as a term in favour of three named outputs.
4. Reporting guidelines are now a delivery mechanism for transparency, not a parallel track.
5. Replication rates depend as much on the chosen criterion as on the underlying science.
6. Effect-size shrinkage on replication is a more stable finding than any replication percentage.
7. Prespecification demonstrably alters reported results in trials and registered reports.
8. Prespecification is hardest exactly where protocols cannot be fully known in advance.
9. AI in evidence synthesis is permitted, conditional, and reportable — not prohibited.
10. Transparent research can still be weak research; verifiability is necessary, not sufficient.

---

Frequently asked questions

What is the difference between reproducibility and replicability? Reproducibility generally means obtaining the same results from the same data and analysis; replicability means obtaining consistent results from new data collected for the same question. The distinction follows the US National Academies' 2019 usage and is now standard.

What does "verifiability" mean in TOP 2025? It means an independent party can check that reported results are accurate and were not selected on the basis of their findings. TOP 2025 operationalises this through two Verification Practices: computational reproducibility and results transparency.

Is TOP 2025 mandatory? No. It is a modular framework for journals, funders, institutions and researchers to adopt selectively, and its authors state that some standards will not apply to some designs or epistemologies.

Why was the TOP Factor discontinued? Because cumulative scoring of journal policies proved difficult to apply reliably and risked encouraging performative compliance rather than substantive change.

Do I still cite CONSORT 2010? No. CONSORT 2025 supersedes it and the earlier version should no longer be used for new trial reports.

How many items are in CONSORT 2025 and SPIRIT 2025? CONSORT 2025 has 30 checklist items plus a flow diagram; SPIRIT 2025 has 34 minimum protocol items plus a schedule diagram.

Which reporting guideline applies to a machine learning prediction model? TRIPOD+AI, published in April 2024, covers development and evaluation of clinical prediction models using either regression or machine learning; PROBAST+AI covers risk-of-bias and applicability assessment.

Can I use a large language model in a systematic review? Under the 2025 position of Cochrane, Campbell, JBI and CEE, yes — with human oversight, provided rigour is not compromised, and with full reporting of any use that makes or suggests judgements.

Does a low replication rate mean the original studies were fraudulent? No. Publication bias, low power, selective reporting, inter-laboratory variability and honest measurement noise all contribute, and their relative shares differ by field.

What is a multiverse analysis? A systematic examination of results across many defensible combinations of data-processing and modelling choices, reported together rather than as a single preferred path. TOP 2025 lists it as a verification study type.

What is a many-analysts study? A design in which multiple independent teams analyse the same dataset to answer the same question, revealing how much conclusions depend on analytic choice.

What is a registered report? A publication format in which the introduction, hypotheses and methods are peer-reviewed and provisionally accepted before results exist, so publication does not depend on the outcome.

Do registered reports produce fewer positive findings? In psychology, one comparison found positive results for the first hypothesis in 44% of registered reports versus 96% of conventional articles. Interpretation is debated, but reduced publication bias is a leading explanation.

Where should I register a non-clinical study? Common options include the Open Science Framework registries for social, behavioural and life sciences, and PROSPERO for systematic reviews; clinical trials use ClinicalTrials.gov or another WHO-recognised primary registry.

Is preregistration useful in qualitative research? Its applicability is contested. TOP 2025 explicitly notes that some standards suit hypothesis-testing quantitative designs better than other epistemologies, and leaves applicability decisions to disciplinary experts.

---

Glossary

• Analysis plan — A document specifying the statistical analyses to be performed, ideally timestamped before data are examined.
• CONSORT — Consolidated Standards of Reporting Trials; the reporting guideline for randomised trials, updated in 2025.
• Computational reproducibility — Independent confirmation that reported results can be regenerated from the shared data and code.
• EQUATOR Network — International initiative maintaining a library of health research reporting guidelines.
• Estimand — A precise definition of the quantity a study aims to estimate, including population, endpoint and handling of intercurrent events.
• GRADE — A framework for rating certainty of evidence and strength of recommendations.
• Many-analysts study — Multiple independent teams analysing one dataset to expose analytic dependence.
• Multiverse analysis — Reporting results across many defensible analytic specifications.
• PROBAST+AI — Tool for assessing risk of bias and applicability in prediction model studies, including AI-based models.
• RAISE — Responsible use of AI in evidence SynthEsis; recommendations for AI use across the synthesis ecosystem.
• Registered report — Two-stage publication format with peer review before results exist.
• Results transparency — Independent confirmation that findings were not selectively reported.
• SPIRIT — Standard Protocol Items: Recommendations for Interventional Trials; updated in 2025.
• TOP Guidelines — Transparency and Openness Promotion Guidelines; policy framework updated as TOP 2025.
• TRIPOD+AI — Reporting guideline for clinical prediction models developed with regression or machine learning.

---

References

Academic papers

Faggion, C. M., Jr., & Kohl, C. B. S. (2025). Exploring ethical elements in reporting guidelines: Results from a research-on-research study. Research Integrity and Peer Review, 10, 20. https://doi.org/10.1186/s41073-025-00180-0

Grant, S., Corker, K. S., Mellor, D., Stewart, S. L. K., Mayo-Wilson, E., Cashin, A. G., … Nosek, B. A. (2026). TOP 2025: An update to the Transparency and Openness Promotion Guidelines. Research Integrity and Peer Review, 11, 40. https://doi.org/10.1186/s41073-026-00223-0

Kaplan, R. M., & Irvin, V. L. (2015). Likelihood of null effects of large NHLBI clinical trials has increased over time. PLOS ONE, 10(8), e0132382. https://doi.org/10.1371/journal.pone.0132382

Nosek, B. A., Alter, G., Banks, G. C., Borsboom, D., Bowman, S. D., Breckler, S. J., … Yarkoni, T. (2015). Promoting an open research culture. Science, 348(6242), 1422–1425. https://doi.org/10.1126/science.aab2374

Scheel, A. M., Schijen, M. R. M. J., & Lakens, D. (2021). An excess of positive results: Comparing the standard psychology literature with registered reports. Advances in Methods and Practices in Psychological Science, 4(2), 1–12. https://doi.org/10.1177/25152459211007467

The Brazilian Reproducibility Initiative. (2026). Estimating the replicability of Brazilian biomedical science [Reviewed preprint]. eLife, 15, RP111001. https://doi.org/10.7554/eLife.111001.1

Reporting guidelines and standards

Collins, G. S., Moons, K. G. M., Dhiman, P., Riley, R. D., Beam, A. L., Van Calster, B., … Logullo, P. (2024). TRIPOD+AI statement: Updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ, 385, e078378. https://doi.org/10.1136/bmj-2023-078378

Hopewell, S., Chan, A.-W., Collins, G. S., Hróbjartsson, A., Moher, D., Schulz, K. F., … Boutron, I. (2025). CONSORT 2025 statement: Updated guideline for reporting randomised trials. BMJ, 389, e081123. https://doi.org/10.1136/bmj-2024-081123

Moons, K. G. M., Damen, J. A. A., Kaul, T., Hooft, L., Andaur Navarro, C., Dhiman, P., … Collins, G. S. (2025). PROBAST+AI: An updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods. BMJ, 388, e082505. https://doi.org/10.1136/bmj-2024-082505

Chan, A.-W., Hopewell, S., Collins, G. S., Hróbjartsson, A., Moher, D., Schulz, K. F., … Boutron, I. (2025). SPIRIT 2025 statement: Updated guideline for protocols of randomised trials. The Lancet, 405(10490). https://doi.org/10.1016/S0140-6736(25)00770-6

Position statements and organisational guidance

Center for Open Science. (2025). TOP Guidelines. https://www.cos.io/initiatives/top-guidelines

EQUATOR Network. (2026). Library for health research reporting. https://www.equator-network.org/library/

Flemyng, E., et al. (2025). Position statement on artificial intelligence (AI) use in evidence synthesis across Cochrane, the Campbell Collaboration, JBI and the Collaboration for Environmental Evidence 2025. Campbell Systematic Reviews, 21, e70074. https://doi.org/10.1002/cl2.70074

Thomas, J., Flemyng, E., Noel-Storr, A., Moy, W., Marshall, I. J., Hajji, R., et al. (2025). Responsible use of AI in evidence SynthEsis (RAISE). Open Science Framework. https://doi.org/10.17605/OSF.IO/FWAUD

Government and institutional sources

National Academies of Sciences, Engineering, and Medicine. (2019). Reproducibility and replicability in science. National Academies Press. https://doi.org/10.17226/25303

UNESCO. (2021). UNESCO recommendation on open science. https://doi.org/10.54677/MNMH8546

---

Publishing notes for OneWise editors

Primary keyword: research methods Secondary keywords: research methodology 2026, verifiability in research, TOP 2025 guidelines, CONSORT 2025, SPIRIT 2025, preregistration, registered reports, evidence synthesis AI Long-tail: what is the difference between reproducibility and replicability, how many items in CONSORT 2025, can you use AI in a systematic review, does preregistration reduce positive results Entities: Center for Open Science, EQUATOR Network, Cochrane, Campbell Collaboration, JBI, Collaboration for Environmental Evidence, TRIPOD, PROBAST, eLife, National Academies

Internal links suggested: Research › Research Methods hub; explainer on systematic reviews and meta-analysis; guide to statistical power and sample size; explainer on clinical trial phases; AI › AI in Science category; Data › Data Management and FAIR principles.

External authority links: cos.io/initiatives/top-guidelines; equator-network.org; spirit-consort.org; tripod-statement.org; cochrane.org; nap.edu/catalog/25303.

People Also Ask candidates: What are research methods? What is the replication crisis? What is preregistration in research? Is AI allowed in systematic reviews? What replaced CONSORT 2010?

Schema recommendations: Article with about and mentions for named standards; FAQPage for the FAQ block; DefinedTermSet for the glossary; BreadcrumbList for Research › Research Methods.

Featured-snippet targets: the reproducibility vs replicability definition pair; the CONSORT/SPIRIT item counts; Table 2 (TOP 2015 vs TOP 2025).

Image alt-text suggestions: "Diagram of the four-layer methods assurance stack from study design to independent verification"; "Comparison table of TOP 2015 and TOP 2025 standards structure"; "Chart of replication rates across four large-scale replication projects".

Verification note: figures for replication rates, checklist item counts and publication dates were checked against the primary sources listed above; the Brazilian Reproducibility Initiative result is a reviewed preprint and should be re-checked before any future republication of this article.