Values that changed
Old coefficients, copied p-values and signs that differ from the agreed analysis output.
Manuscript Results Check Pilot
Do the numbers in your paper match your results?
Check that the numbers in your paper match your R or Python output.
A table can come from an older model. A group label can move while its values stay put. We trace the numbers in your manuscript to the R or Python outputs behind them, then show you where they agree and what needs attention.
Up to 80 agreed numeric claims · R or Python · Human-reviewed findings
An updated model. A partly updated results paragraph.
After adding covariates, you update the coefficient in the manuscript. The confidence interval and p-value still come from the earlier, unadjusted model. Each number looks plausible on its own.
Illustrative example · synthetic data, not a client report
Manuscript: adjusted result
- Coefficient (b)
- 0.24
- 95% CI
- [0.03, 0.51]
- p-value
- 0.028
Output: adjusted model
- Coefficient (b)
- 0.24
- 95% CI
- [−0.03, 0.51]
- p-value
- 0.078
Finding: results from two model versions. The coefficient matches the adjusted model, but the interval and p-value were carried over from the unadjusted fit. The adjusted interval includes zero; the reported interval does not.
Sample results check
Illustrative excerpt · synthetic inputs · not a client audit
Here is how the paragraph above translates into a claim-by-claim record. The agreed source is the adjusted model. Each interval endpoint is checked separately.
Scope: four numeric claims in Results, paragraph 1. Summary: two match at the reported precision; two have discrepancies.
| Claim | Reported | Computed* | Finding | R output reference |
|---|---|---|---|---|
| RES-001Coefficient (b) | 0.24 | 0.240000 | Matches at reported precision | adjusted_result["estimate"] |
| RES-00295% CI, lower bound | 0.03 | -0.026555 | Sign reversal | adjusted_result["ci_lower"] |
| RES-00395% CI, upper bound | 0.51 | 0.506555 | Matches at reported precision | adjusted_result["ci_upper"] |
| RES-004p-value | 0.028 | 0.077613 | Value mismatch | adjusted_result["p_value"] |
*Computed values are displayed to six decimal places here. The coefficient and standard error are synthetic inputs; the interval and p-value use a normal approximation. These are not estimates from a fitted study dataset.
Suggested correction: report the adjusted result consistently: b = 0.24, 95% CI [−0.03, 0.51], p = 0.078. The unadjusted fit in the example source produces the interval and p-value carried over into the manuscript.
Download the illustrative R calculations. The source shows both versions so you can inspect where the copied values came from.
This excerpt demonstrates the report structure. A completed engagement also records the agreed scope, source files, execution details where applicable, reviewer decisions and unresolved limitations.
What the check looks for
Labels and denominators
Results assigned to a different group, conflicting units, or a stated sample size that differs from the output.
Rounding differences
Agreement at the reported precision, small last-digit differences, and discrepancies that exceed the stated tolerance.
Claims without support
Numbers with no corresponding calculation, plus uncertain links that need clarification before they can be confirmed.
How your review works
Agree on the scope
We identify the sections and tables to check, up to 80 numeric claims, and confirm which script and model version produced the intended results.
Trace each number to its calculation
We inspect your Word manuscript and replication package. Where agreed and feasible, we rerun the code; supplied output is identified separately from results reproduced during the check.
Review the correspondence
Nedim Yel, PhD, reviews the proposed links and findings. If two different results round to the same number, the group and variable context must resolve the link. Uncertainty is raised for clarification.
Receive the evidence and next steps
You receive a claim-by-claim record showing the manuscript location, reported value, corresponding output and finding, with unresolved limitations stated.
A record you can work through
- Summary of findings: agreement, rounding flags, substantive mismatches and unsupported claims.
- Actionable discrepancy table: where the number appears, what was reported, what was computed and why the two differ.
- Full claim-to-output ledger: the mapping for every claim in the agreed scope, including script and object origin.
- Review and provenance record: the files compared, recorded execution environment where code was run, reviewer decisions and limitations.
We confirm the delivery format, scope, fee and turnaround before work starts. This pilot checks correspondence between manuscript and output; methodological review can be scoped separately.
What to have ready
A Word manuscript (.docx), the relevant R or Python code, and a clear indication of which analysis version the paper should report. Existing output and dependency information help us assess whether a rerun is feasible.
Start with a description of the project. Please do not send participant-level or sensitive data with your initial inquiry. We agree on access and an appropriate data arrangement first. See our confidentiality approach for file access, handling and retention.
Which check fits your question?
“Can someone rerun this?”
The Reproducibility Check focuses on running one agreed output and finding environment or dependency problems.
Explore the Reproducibility Check →“Does the paper report this correctly?”
The Manuscript Results Check focuses on whether the agreed manuscript numbers correspond to the intended analysis outputs.
Discuss the Manuscript Results Check →Before you start
Does this establish that my analysis is statistically appropriate?
It checks whether manuscript claims correspond to the analysis output. Questions about study design, model assumptions, missing data or interpretation belong in a methodological review. We can discuss that scope in your inquiry.
Will every number in the document be checked?
We agree on sections and tables before starting, with a pilot limit of 80 numeric claims. Text boxes, figures, footnotes and unusual table layouts may need separate handling. The report identifies the scope and exclusions.
What if a number cannot be linked confidently?
We request clarification or the missing calculation. A matching value by itself is insufficient. A claim with no accepted supporting output remains identified as unsupported rather than being marked as verified.
Is this an automatic upload-and-check tool?
This is a scoped service with human review. Send an inquiry first; we confirm the project requirements and file-sharing arrangement before reviewing your materials.
Bring the paper and the output into agreement.
Choose Manuscript review in the quote form and mention the Manuscript Results Check in your concerns. We respond to quote requests within 72 hours.