Rmerge and CC1/2 are both reported across resolution shells, but they summarize different structures in repeated diffraction observations. Rmerge pools deviations from reflection means. CC1/2 correlates independently formed half-dataset means across reflections. Their values cannot be substituted or placed on one universal quality scale.
The guide works a small residual example and a separate half-set correlation example, then tests multiplicity, shell boundaries, split stability, systematic error and paired refinement. The central decision keeps a fictional outer shell provisionally even though Rmerge is high because other evidence indicates reproducible signal and improved downstream modelling.
A careful cutoff record also preserves how observations were split, whether Friedel mates were merged, how many unique reflections entered each shell, whether completeness is directional, and whether the same evaluation reflections were used in paired refinement. These details prevent a correlation generated by shared batch effects from being treated as independent evidence. They also show why a percentage residual can rise as intensity weakens or multiplicity grows without implying that every outer-shell observation is worthless. The aim is not to select the statistic that looks best. It is to ask what variation each statistic summarizes, how stable it is under reasonable reprocessing, and whether including the shell improves a model under a fair comparison. When results disagree, keep the disagreement visible and investigate shell size, split hierarchy, scaling, anisotropy, outliers and systematic error before fixing the boundary. A reader should be able to reconstruct both the arithmetic and the decision rule from the saved receipt rather than infer policy from a rounded table.
The examples are educational and do not validate a structure, prescribe universal cutoffs, replace processing documentation or substitute for expert crystallographic review.
This specialist guide belongs to the X-ray Crystallography root. It also connects to The Gold Standard of Education | Learning That Transfers into Life and the How X Works hub. The worked cases are original teaching examples; their numbers are stipulated so that the reasoning can be checked.
eduKateSG · X-ray Crystallography
Find your next learning step
Read merging residuals and half-dataset correlation as different statistics, then justify a cutoff with shell definitions, counts and downstream checks.
ROUTE 1 · CHAPTERS 1–5
Define the quantities
Fix the object, unit, denominator and reference condition.
ROUTE 2 · CHAPTERS 6–10
Work the central example
Carry the calculation through a complete, checkable case.
ROUTE 3 · CHAPTERS 11–15
Audit the evidence
Test sensitivity, measurement choices and rival explanations.
ROUTE 5 · CHAPTERS 21–24
Report and continue
Repair shortcuts, state limits and follow verified routes.
Chapter contents
Define the quantities · Chapters 1–5
1. Name the statistics before ranking shells
2. Write the Rmerge numerator and denominator
Work the central example · Chapters 6–10
6. Complete a small Rmerge example
Audit the evidence · Chapters 11–15
Transfer the reasoning · Chapters 16–20
16. Transfer to a high-resolution cutoff
17. Transfer to merging crystals
Report and continue · Chapters 21–24
21. Repair lower Rmerge always means better
22. Repair positive CC1/2 proves correctness
CHAPTER 1 OF 24 · Define the quantities
1. Name the statistics before ranking shells
Begin with the quantity rather than the convenient number. Rmerge is a normalized residual among repeated intensities, while CC1/2 is a correlation between independently formed half-dataset averages.
Work the teaching case in a visible order. A fictional outer shell has Rmerge=0.78, multiplicity8 and CC1/2=0.42.
Make the reasoning independently checkable. Write each statistic’s inputs and aggregation before judging quality.
Test transfer by changing one condition while holding the others still. Double multiplicity with the same underlying signal and noise.
Keep the claim boundary visible. The numbers do not share one universal good-to-bad scale.
Before accepting the result, predict what should change and what should remain fixed. Apply this check: Write each statistic’s inputs and aggregation before judging quality. Then run the changed case: Double multiplicity with the same underlying signal and noise. Keep both outcomes visible.
The evidence receipt for “Name the statistics before ranking shells” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
CHAPTER 2 OF 24 · Define the quantities
2. Write the Rmerge numerator and denominator
Keep observation, transformation and conclusion on separate lines. Rmerge sums absolute deviations from each reflection mean and divides by summed intensity.
Work the teaching case in a visible order. For two reflections with repeated stipulated intensities, calculate deviations before pooling them.
Make the reasoning independently checkable. Recompute from the observation table rather than a rounded summary.
Test transfer by changing one condition while holding the others still. Multiply every intensity by a common positive factor.
Keep the claim boundary visible. The ratio is scale invariant but remains sensitive to multiplicity and weak denominators.
Build an audit receipt with input, unit, model, calculation, observation and limit. Use this independent check: Recompute from the observation table rather than a rounded summary. For transfer, Multiply every intensity by a common positive factor. State why the conclusion moved or stayed.
The evidence receipt for “Write the Rmerge numerator and denominator” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Use the short answer only after fixing the boundary and units. CC1/2 requires splitting observations into two groups and correlating the independently averaged reflection intensities within a shell.
Work the teaching case in a visible order. The teaching shell uses paired half means for six reflections.
Make the reasoning independently checkable. Verify that no observation appears in both halves.
Test transfer by changing one condition while holding the others still. Change the split seed and repeat the correlation.
Keep the claim boundary visible. One split can fluctuate, especially when a shell contains few reflections.
Ask a second reader to reproduce the decision from the stated evidence. They should be able to apply: Verify that no observation appears in both halves. Next, Change the split seed and repeat the correlation. Preserve any mismatch as evidence.
The evidence receipt for “Build independent half datasets” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Trace the mechanism in the order another reader can reproduce. Rmerge tends to increase as more observations contribute chances for deviation, even when averaging improves the merged estimate.
Work the teaching case in a visible order. Compare multiplicity2 and8 simulations under one noise model.
Make the reasoning independently checkable. Report multiplicity beside every merging residual.
Test transfer by changing one condition while holding the others still. Add observations without changing per-observation precision.
Keep the claim boundary visible. A larger Rmerge can coexist with a better determined merged intensity.
Turn the claim into a falsifiable test. First, Report multiplicity beside every merging residual. Then change one consequential condition: Add observations without changing per-observation precision. Name the observation that would count against the explanation.
The evidence receipt for “Keep multiplicity visible” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Treat a plausible result as a prompt for a check rather than proof. Correlation asks whether reflection-to-reflection variation is reproduced across halves; a residual measures disagreement around means.
Work the teaching case in a visible order. The central shell can retain positive half-set correlation while its percentage residual is large.
Make the reasoning independently checkable. Plot half means and inspect leverage.
Test transfer by changing one condition while holding the others still. Add one extreme bright reflection.
Keep the claim boundary visible. A high correlation may be driven by range and does not rule out systematic error.
Before accepting the result, predict what should change and what should remain fixed. Apply this check: Plot half means and inspect leverage. Then run the changed case: Add one extreme bright reflection. Keep both outcomes visible.
The evidence receipt for “Separate correlation from residual size” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Begin with the quantity rather than the convenient number. Use observations100,110 for reflection A and20,30 for B; means are105 and25.
Work the teaching case in a visible order. The absolute-deviation sum is20 and the intensity sum is260, giving Rmerge≈0.0769.
Make the reasoning independently checkable. Recalculate each reflection contribution.
Test transfer by changing one condition while holding the others still. Add third observations105 and25.
Keep the claim boundary visible. This toy shell is too small for a real cutoff decision.
Build an audit receipt with input, unit, model, calculation, observation and limit. Use this independent check: Recalculate each reflection contribution. For transfer, Add third observations105 and25. State why the conclusion moved or stayed.
The evidence receipt for “Complete a small Rmerge example” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
CHAPTER 7 OF 24 · Work the central example
7. Complete a half-set correlation example
Keep observation, transformation and conclusion on separate lines. Use half means A=[100,80,60,40,20] and B=[98,82,59,43,18].
Work the teaching case in a visible order. Their Pearson correlation is very high because the ranks and spacing broadly agree.
Make the reasoning independently checkable. Centre both lists and reproduce the covariance calculation.
Test transfer by changing one condition while holding the others still. Add random noise to the weakest two reflections.
Keep the claim boundary visible. A worked correlation illustrates mechanism, not a recommended threshold.
Ask a second reader to reproduce the decision from the stated evidence. They should be able to apply: Centre both lists and reproduce the covariance calculation. Next, Add random noise to the weakest two reflections. Preserve any mismatch as evidence.
The evidence receipt for “Complete a half-set correlation example” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Use the short answer only after fixing the boundary and units. Shell membership changes the reflections pooled into both statistics and can change an apparent cutoff.
Work the teaching case in a visible order. Move the resolution boundary so twenty borderline reflections enter the outer shell.
Make the reasoning independently checkable. Report shell limits and counts with the result.
Test transfer by changing one condition while holding the others still. Use equal-count shells instead of equal-width shells.
Keep the claim boundary visible. Two tables cannot be compared when their shells differ silently.
Turn the claim into a falsifiable test. First, Report shell limits and counts with the result. Then change one consequential condition: Use equal-count shells instead of equal-width shells. Name the observation that would count against the explanation.
The evidence receipt for “Change a shell boundary” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Trace the mechanism in the order another reader can reproduce. Multiplicity-aware alternatives adjust the residual for repeated observations or precision of merged means.
Work the teaching case in a visible order. Apply the declared factor sqrt(n/(n-1)) to a constant-n teaching shell.
Make the reasoning independently checkable. Keep the exact convention used by the processing program.
Test transfer by changing one condition while holding the others still. Let n vary by reflection.
Keep the claim boundary visible. Named R statistics are not interchangeable labels.
Before accepting the result, predict what should change and what should remain fixed. Apply this check: Keep the exact convention used by the processing program. Then run the changed case: Let n vary by reflection. Keep both outcomes visible.
The evidence receipt for “Compare Rmeas and Rpim” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
CHAPTER 10 OF 24 · Work the central example
10. Use paired refinement as a downstream check
Treat a plausible result as a prompt for a check rather than proof. A resolution decision can be tested by refining comparable models against nested datasets and evaluating them on common reflections.
Work the teaching case in a visible order. The teaching comparison uses a full dataset and a truncated dataset with the same refinement protocol.
Make the reasoning independently checkable. Compare model metrics on the shared evaluation set.
Test transfer by changing one condition while holding the others still. Change only the high-resolution boundary.
Keep the claim boundary visible. Paired refinement can still be affected by model setup, convergence and systematic error.
Build an audit receipt with input, unit, model, calculation, observation and limit. Use this independent check: Compare model metrics on the shared evaluation set. For transfer, Change only the high-resolution boundary. State why the conclusion moved or stayed.
The evidence receipt for “Use paired refinement as a downstream check” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Begin with the quantity rather than the convenient number. CC1/2 estimates can vary when the number of observations or unique reflections is small.
Work the teaching case in a visible order. Repeat the fictional split100 times and summarize the distribution.
Make the reasoning independently checkable. Save the split rule or seed.
Test transfer by changing one condition while holding the others still. Reduce the outer shell to twelve unique reflections.
Keep the claim boundary visible. A single correlation without sample size hides instability.
Ask a second reader to reproduce the decision from the stated evidence. They should be able to apply: Save the split rule or seed. Next, Reduce the outer shell to twelve unique reflections. Preserve any mismatch as evidence.
The evidence receipt for “Audit random split stability” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Keep observation, transformation and conclusion on separate lines. One radial resolution shell can combine strong directions with weak directions.
Work the teaching case in a visible order. Partition the teaching data by reciprocal-space direction before pooling.
Make the reasoning independently checkable. Compare directional completeness and correlation.
Test transfer by changing one condition while holding the others still. Rotate the crystal strategy while holding nominal radial cutoff fixed.
Keep the claim boundary visible. A spherical cutoff may not describe anisotropic information content.
Turn the claim into a falsifiable test. First, Compare directional completeness and correlation. Then change one consequential condition: Rotate the crystal strategy while holding nominal radial cutoff fixed. Name the observation that would count against the explanation.
The evidence receipt for “Audit anisotropy” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Use the short answer only after fixing the boundary and units. Half datasets can share systematic errors and therefore correlate even when both are biased.
Work the teaching case in a visible order. Add the same scale drift to both halves.
Make the reasoning independently checkable. Inspect independent controls, scaling residuals and geometry diagnostics.
Test transfer by changing one condition while holding the others still. Split by image range rather than random observations.
Keep the claim boundary visible. Correlation is not proof of accuracy.
Before accepting the result, predict what should change and what should remain fixed. Apply this check: Inspect independent controls, scaling residuals and geometry diagnostics. Then run the changed case: Split by image range rather than random observations. Keep both outcomes visible.
The evidence receipt for “Audit systematic error” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Trace the mechanism in the order another reader can reproduce. Removing weak or disagreeing observations can improve residuals while discarding useful signal.
Work the teaching case in a visible order. Calculate statistics before and after a stipulated rejection rule.
Make the reasoning independently checkable. Record every exclusion and its reason.
Test transfer by changing one condition while holding the others still. Tighten the rejection threshold.
Keep the claim boundary visible. Optimizing a summary statistic is not the same as improving the final model.
Build an audit receipt with input, unit, model, calculation, observation and limit. Use this independent check: Record every exclusion and its reason. For transfer, Tighten the rejection threshold. State why the conclusion moved or stayed.
The evidence receipt for “Audit outlier rejection” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Treat a plausible result as a prompt for a check rather than proof. A promising CC1/2 does not compensate for missing unique reflections, and high multiplicity does not guarantee directional coverage.
Work the teaching case in a visible order. Read CC1/2 beside completeness and multiplicity for the same shell.
Make the reasoning independently checkable. Align shell boundaries across the table.
Test transfer by changing one condition while holding the others still. Remove a wedge of reciprocal space.
Keep the claim boundary visible. Data-quality dimensions answer different questions.
Ask a second reader to reproduce the decision from the stated evidence. They should be able to apply: Align shell boundaries across the table. Next, Remove a wedge of reciprocal space. Preserve any mismatch as evidence.
The evidence receipt for “Audit completeness and multiplicity” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Begin with the quantity rather than the convenient number. Treat the cutoff as a documented decision using several indicators and downstream tests, not one magic number.
Work the teaching case in a visible order. The fictional shell at2.1–2.0 Å is retained provisionally because CC1/2 is positive and paired refinement improves, despite high Rmerge.
Make the reasoning independently checkable. State the policy before seeing the preferred model.
Test transfer by changing one condition while holding the others still. Repeat after anisotropic truncation.
Keep the claim boundary visible. No threshold from this teaching case is universal.
Turn the claim into a falsifiable test. First, State the policy before seeing the preferred model. Then change one consequential condition: Repeat after anisotropic truncation. Name the observation that would count against the explanation.
The evidence receipt for “Transfer to a high-resolution cutoff” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Keep observation, transformation and conclusion on separate lines. Combining crystals introduces scale, non-isomorphism and batch effects beyond ordinary repeated measurements.
Work the teaching case in a visible order. Compare within-crystal and between-crystal half-set schemes.
Make the reasoning independently checkable. Keep crystal identity in the observation ledger.
Test transfer by changing one condition while holding the others still. Remove one non-isomorphous crystal.
Keep the claim boundary visible. A better pooled correlation can hide a dominant crystal.
Before accepting the result, predict what should change and what should remain fixed. Apply this check: Keep crystal identity in the observation ledger. Then run the changed case: Remove one non-isomorphous crystal. Keep both outcomes visible.
The evidence receipt for “Transfer to merging crystals” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Use the short answer only after fixing the boundary and units. Small differences may require statistics designed for anomalous pairs and careful half-set construction.
Work the teaching case in a visible order. Keep ordinary CC1/2 separate from anomalous correlation.
Make the reasoning independently checkable. Check the software’s exact statistic label.
Test transfer by changing one condition while holding the others still. Merge Friedel mates and observe which signal disappears.
Keep the claim boundary visible. This article does not replace anomalous-signal diagnostics.
Build an audit receipt with input, unit, model, calculation, observation and limit. Use this independent check: Check the software’s exact statistic label. For transfer, Merge Friedel mates and observe which signal disappears. State why the conclusion moved or stayed.
The evidence receipt for “Transfer to weak anomalous signal” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Trace the mechanism in the order another reader can reproduce. Large numbers of partial observations change multiplicity, scaling and split design.
Work the teaching case in a visible order. Split the fictional serial observations by image or crystal identity.
Make the reasoning independently checkable. Prevent leakage of correlated frames across halves.
Test transfer by changing one condition while holding the others still. Randomize at the wrong hierarchy and detect inflated correlation.
Keep the claim boundary visible. The correct independence unit depends on acquisition design.
Ask a second reader to reproduce the decision from the stated evidence. They should be able to apply: Prevent leakage of correlated frames across halves. Next, Randomize at the wrong hierarchy and detect inflated correlation. Preserve any mismatch as evidence.
The evidence receipt for “Transfer to serial data” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Treat a plausible result as a prompt for a check rather than proof. Record shell bounds, counts, completeness, multiplicity, Rmerge-family definition, CC1/2 split method, uncertainty and paired-refinement result.
Work the teaching case in a visible order. Another reader can reproduce why the teaching shell was retained.
Make the reasoning independently checkable. Delete one field and identify the unsupported inference.
Test transfer by changing one condition while holding the others still. Version the receipt after reprocessing.
Keep the claim boundary visible. Documentation does not eliminate model dependence.
Turn the claim into a falsifiable test. First, Delete one field and identify the unsupported inference. Then change one consequential condition: Version the receipt after reprocessing. Name the observation that would count against the explanation.
The evidence receipt for “Build a cutoff decision receipt” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Begin with the quantity rather than the convenient number. Residuals can fall when useful weak data are removed or multiplicity changes.
Work the teaching case in a visible order. Compare the full and truncated teaching datasets.
Make the reasoning independently checkable. Ask whether the downstream model improved on common reflections.
Test transfer by changing one condition while holding the others still. Add precise repeated measurements and watch multiplicity effects.
Keep the claim boundary visible. A cosmetically smaller statistic is not a standalone objective.
Before accepting the result, predict what should change and what should remain fixed. Apply this check: Ask whether the downstream model improved on common reflections. Then run the changed case: Add precise repeated measurements and watch multiplicity effects. Keep both outcomes visible.
The evidence receipt for “Repair lower Rmerge always means better” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
CHAPTER 22 OF 24 · Report and continue
22. Repair positive CC1/2 proves correctness
Keep observation, transformation and conclusion on separate lines. Shared bias, small samples or leverage can produce a positive correlation.
Work the teaching case in a visible order. Add a common offset pattern to both halves.
Make the reasoning independently checkable. Inspect plots, counts and independent diagnostics.
Test transfer by changing one condition while holding the others still. Change the split hierarchy.
Keep the claim boundary visible. Correlation measures reproducibility of variation, not truth.
Build an audit receipt with input, unit, model, calculation, observation and limit. Use this independent check: Inspect plots, counts and independent diagnostics. For transfer, Change the split hierarchy. State why the conclusion moved or stayed.
The evidence receipt for “Repair positive CC1/2 proves correctness” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Use the short answer only after fixing the boundary and units. The method supports a transparent, multi-indicator resolution boundary and shows why residual and correlation statistics cannot be substituted.
Work the teaching case in a visible order. It can flag a cutoff chosen only to beautify Rmerge.
Make the reasoning independently checkable. Write the criterion and validation path before final refinement.
Test transfer by changing one condition while holding the others still. Move from teaching data to a deposited dataset.
Keep the claim boundary visible. Use current community practice, processing documentation and expert crystallographic review.
Ask a second reader to reproduce the decision from the stated evidence. They should be able to apply: Write the criterion and validation path before final refinement. Next, Move from teaching data to a deposited dataset. Preserve any mismatch as evidence.
The evidence receipt for “State the supported decision” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
CHAPTER 24 OF 24 · Report and continue
24. Continue through verified crystallography routes
Trace the mechanism in the order another reader can reproduce. This distinction sits beside detector coverage, completeness, redundancy, saturation and rocking-width lessons.
Work the teaching case in a visible order. Choose the sibling that owns the next failure mode.
Make the reasoning independently checkable. Keep geometric reach, observation coverage, repetition and signal reproducibility separate.
Test transfer by changing one condition while holding the others still. Move to detector coverage without treating it as a statistical cutoff.
Keep the claim boundary visible. This article owns Rmerge-versus-CC1/2 reasoning for a declared shell and split method.
Turn the claim into a falsifiable test. First, Keep geometric reach, observation coverage, repetition and signal reproducibility separate. Then change one consequential condition: Move to detector coverage without treating it as a statistical cutoff. Name the observation that would count against the explanation.
The evidence receipt for “Continue through verified crystallography routes” records the source, input, transformation, changed-context prediction and limit on separate lines. Keep shell limits, counts, multiplicity, exact statistic definitions, split method and downstream validation together.
Contents · Previous chapter · Return to X-ray Crystallography
Continue through the verified reading route
Resolution cutoff and detector coverage
Resolution cutoff and detector coverage. Use this when geometric reach is confused with an analysis boundary.
Reflection completeness and redundancy
Reflection completeness and redundancy. Use this when unique coverage is confused with repeated observations.
Detector saturation and integrated strength
Detector saturation and integrated strength. Use this when pixel nonlinearity affects intensity integration.
Frequently asked questions
Does a high Rmerge prove an outer shell is useless?
No. Multiplicity and weak intensities affect it; inspect CC1/2, completeness, counts and downstream tests.
Does positive CC1/2 prove the data are accurate?
No. Shared systematic errors, leverage and small samples can still mislead.
Is there one universal CC1/2 cutoff?
No. Report the shell, uncertainty, processing and validation strategy rather than treating one value as universal.
Authoritative sources and further reading
IUCr: Better models by discarding data?. Primary paper comparing CC1/2, merging R values and paired-refinement evidence.
IUCr: How good are my data and what is the resolution?. Primary discussion of CC1/2, information content and resolution choice.
IUCr: Data-quality indicators in macromolecular crystallography. Primary review of merging statistics and correlation-based indicators.
Return to the X-ray Crystallography subject guide, The Gold Standard of Education, or the How X Works hub.