RQ3 training-data to generated-code quality Sankey diagram
Study-family-source-target path assignments connect twelve observed RQ2 training-data quality leaves through twelve mapping families to eight RQ1 generated-code quality dimensions.
Training-data quality and generated-code quality relationships
Band widths and percentages show literature coverage; one study may support multiple paths
RQ2 DATA-QUALITY LEAF
MAPPING FAMILY
RQ1 GENERATED-CODE QUALITY
Correctness → Defect-Pattern Carryover Share of path assignments: 13.9% Distinct studies: 14 Distinct evidence rows: 16 Studies: S001; S010; S013; S014; S021; S047; S048; S050; S063; S086; S087; S089; S094; S100 Evidence: A-E015; A-E031; A-E032; A-E034; A-E046; B021; B027; C002; C010; C015; C024; C029; C030; C033; C042; C046
Security → Vulnerability-Pattern Carryover Share of path assignments: 9.9% Distinct studies: 10 Distinct evidence rows: 10 Studies: S007; S017; S018; S019; S037; S082; S087; S089; S112; S114 Evidence: A-E001; A-E003; A-E041; A-E045; A-E047; B008; B032; B034; C008; C048
Security → Restricted/Sensitive-Content Reproduction Risks Share of path assignments: 3.0% Distinct studies: 3 Distinct evidence rows: 3 Studies: S030; S075; S082 Evidence: A-E037; A-E042; B007
Compliance → Restricted/Sensitive-Content Reproduction Risks Share of path assignments: 3.0% Distinct studies: 3 Distinct evidence rows: 3 Studies: S041; S075; S087 Evidence: A-E008; A-E036; A-E048
Compliance and Security Risks (Textual) → Restricted/Sensitive-Content Reproduction Risks Share of path assignments: 2.0% Distinct studies: 2 Distinct evidence rows: 2 Studies: S082; S091 Evidence: A-E027; A-E050
Robustness → Robustness-Pattern Associations Share of path assignments: 2.0% Distinct studies: 2 Distinct evidence rows: 2 Studies: S002; S106 Evidence: A-E049; B041
Maintainability → Structural-Quality Associations Share of path assignments: 1.0% Distinct studies: 1 Distinct evidence rows: 1 Studies: S065 Evidence: A-E039
Understandability → Clarity-Pattern Associations Share of path assignments: 2.0% Distinct studies: 2 Distinct evidence rows: 2 Studies: S086; S103 Evidence: C018; C028
Efficiency → Efficiency-Pattern Carryover Share of path assignments: 4.0% Distinct studies: 4 Distinct evidence rows: 4 Studies: S011; S012; S089; S103 Evidence: B011; B037; C009; C019
Distribution Imbalance Issues → Prevalence-Associated Quality Skew Share of path assignments: 17.8% Distinct studies: 17 Distinct evidence rows: 18 Studies: S001; S006; S011; S024; S029; S042; S059; S063; S064; S069; S073; S082; S083; S090; S094; S104; S114 Evidence: A-E002; A-E007; A-E009; A-E012; A-E019; A-E023; A-E030; A-E040; B002; B012; B036; B039; B040; B042; C022; C036; C037; C044
Redundancy Issues → Redundancy-Associated Reinforcement and Degeneration Share of path assignments: 3.0% Distinct studies: 3 Distinct evidence rows: 3 Studies: S010; S059; S108 Evidence: A-E022; A-E035; B001
Inadequate Diversity → Underexposure-Associated Quality Gaps Share of path assignments: 23.8% Distinct studies: 22 Distinct evidence rows: 27 Studies: S003; S004; S009; S010; S014; S015; S023; S031; S037; S039; S047; S049; S057; S070; S080; S081; S091; S100; S104; S105; S110; S112 Evidence: A-E004; A-E005; A-E006; A-E018; A-E025; A-E028; A-E029; A-E033; B009; B018; B020; B022; B033; B038; C004; C005; C011; C012; C013; C016; C020; C021; C031; C032; C038; C040; C045
Data Contamination Risks → Contamination-Associated Familiarity Differences Share of path assignments: 3.0% Distinct studies: 3 Distinct evidence rows: 3 Studies: S073; S088; S097 Evidence: A-E011; B025; C034
Low-Value Data → Supervision-Signal Quality Associations Share of path assignments: 11.9% Distinct studies: 11 Distinct evidence rows: 12 Studies: S002; S016; S046; S059; S060; S061; S071; S072; S090; S091; S092 Evidence: A-E010; A-E014; A-E016; A-E017; A-E024; B001; B005; B010; B028; B030; B031; B035
Defect-Pattern Carryover → Correctness Share of path assignments: 13.9% Distinct studies: 14 Distinct evidence rows: 16 Studies: S001; S010; S013; S014; S021; S047; S048; S050; S063; S086; S087; S089; S094; S100 Evidence: A-E015; A-E031; A-E032; A-E034; A-E046; B021; B027; C002; C010; C015; C024; C029; C030; C033; C042; C046
Vulnerability-Pattern Carryover → Security Share of path assignments: 9.9% Distinct studies: 10 Distinct evidence rows: 10 Studies: S007; S017; S018; S019; S037; S082; S087; S089; S112; S114 Evidence: A-E001; A-E003; A-E041; A-E045; A-E047; B008; B032; B034; C008; C048
Restricted/Sensitive-Content Reproduction Risks → Compliance Share of path assignments: 7.9% Distinct studies: 6 Distinct evidence rows: 8 Studies: S030; S041; S075; S082; S087; S091 Evidence: A-E008; A-E027; A-E036; A-E037; A-E042; A-E048; A-E050; B007
Robustness-Pattern Associations → Robustness Share of path assignments: 1.0% Distinct studies: 1 Distinct evidence rows: 1 Studies: S106 Evidence: A-E049
Robustness-Pattern Associations → Correctness Share of path assignments: 1.0% Distinct studies: 1 Distinct evidence rows: 1 Studies: S002 Evidence: B041
Structural-Quality Associations → Maintainability Share of path assignments: 1.0% Distinct studies: 1 Distinct evidence rows: 1 Studies: S065 Evidence: A-E039
Clarity-Pattern Associations → Understandability Share of path assignments: 2.0% Distinct studies: 2 Distinct evidence rows: 2 Studies: S086; S103 Evidence: C018; C028
Efficiency-Pattern Carryover → Efficiency Share of path assignments: 4.0% Distinct studies: 4 Distinct evidence rows: 4 Studies: S011; S012; S089; S103 Evidence: B011; B037; C009; C019
Prevalence-Associated Quality Skew → Maintainability Share of path assignments: 1.0% Distinct studies: 1 Distinct evidence rows: 1 Studies: S042 Evidence: C036
Prevalence-Associated Quality Skew → Efficiency Share of path assignments: 2.0% Distinct studies: 2 Distinct evidence rows: 2 Studies: S011; S042 Evidence: B012; C037
Prevalence-Associated Quality Skew → Correctness Share of path assignments: 13.9% Distinct studies: 14 Distinct evidence rows: 14 Studies: S001; S006; S024; S029; S059; S063; S064; S069; S082; S083; S090; S094; S104; S114 Evidence: A-E002; A-E007; A-E009; A-E019; A-E023; A-E030; A-E040; B002; B036; B039; B040; B042; C022; C044
Prevalence-Associated Quality Skew → Generation Derailment Share of path assignments: 1.0% Distinct studies: 1 Distinct evidence rows: 1 Studies: S073 Evidence: A-E012
Redundancy-Associated Reinforcement and Degeneration → Correctness Share of path assignments: 2.0% Distinct studies: 2 Distinct evidence rows: 2 Studies: S010; S059 Evidence: A-E035; B001
Redundancy-Associated Reinforcement and Degeneration → Generation Derailment Share of path assignments: 1.0% Distinct studies: 1 Distinct evidence rows: 1 Studies: S108 Evidence: A-E022
Underexposure-Associated Quality Gaps → Security Share of path assignments: 2.0% Distinct studies: 2 Distinct evidence rows: 2 Studies: S039; S049 Evidence: C016; C021
Underexposure-Associated Quality Gaps → Efficiency Share of path assignments: 1.0% Distinct studies: 1 Distinct evidence rows: 1 Studies: S010 Evidence: A-E033
Underexposure-Associated Quality Gaps → Correctness Share of path assignments: 17.8% Distinct studies: 18 Distinct evidence rows: 21 Studies: S003; S004; S009; S014; S015; S023; S031; S039; S047; S057; S070; S080; S081; S091; S100; S104; S110; S112 Evidence: A-E004; A-E005; A-E006; A-E018; A-E025; A-E028; B018; B020; B022; B033; C004; C005; C011; C012; C013; C020; C031; C032; C038; C040; C045
Underexposure-Associated Quality Gaps → Generation Derailment Share of path assignments: 3.0% Distinct studies: 3 Distinct evidence rows: 3 Studies: S037; S081; S105 Evidence: A-E029; B009; B038
Contamination-Associated Familiarity Differences → Correctness Share of path assignments: 2.0% Distinct studies: 2 Distinct evidence rows: 2 Studies: S088; S097 Evidence: B025; C034
Contamination-Associated Familiarity Differences → Generation Derailment Share of path assignments: 1.0% Distinct studies: 1 Distinct evidence rows: 1 Studies: S073 Evidence: A-E011
Supervision-Signal Quality Associations → Security Share of path assignments: 1.0% Distinct studies: 1 Distinct evidence rows: 1 Studies: S016 Evidence: B028
Supervision-Signal Quality Associations → Understandability Share of path assignments: 1.0% Distinct studies: 1 Distinct evidence rows: 1 Studies: S046 Evidence: A-E016
Supervision-Signal Quality Associations → Correctness Share of path assignments: 9.9% Distinct studies: 10 Distinct evidence rows: 11 Studies: S002; S016; S059; S060; S061; S071; S072; S090; S091; S092 Evidence: A-E010; A-E014; A-E017; A-E024; B001; B005; B010; B028; B030; B031; B035
RQ2 training-data quality leaf: Correctness Share of path assignments: 13.9% Distinct studies: 14 Training code samples contain syntax logical functionality API or dependency defects.
RQ2 training-data quality leaf: Security Share of path assignments: 12.9% Distinct studies: 12 Training code samples contain vulnerabilities credentials or high-risk implementation patterns.
RQ2 training-data quality leaf: Compliance Share of path assignments: 3.0% Distinct studies: 3 Training code samples violate legal ethical or regulatory requirements.
RQ2 training-data quality leaf: Robustness Share of path assignments: 2.0% Distinct studies: 2 Training-code fault-tolerance, input-validation, exception-handling, or boundary-checking patterns; path-level direction and qualifiers distinguish stronger practices from deficits.
RQ2 training-data quality leaf: Maintainability Share of path assignments: 1.0% Distinct studies: 1 Training code samples have structures that are difficult to modify extend or reuse.
RQ2 training-data quality leaf: Understandability Share of path assignments: 2.0% Distinct studies: 2 Training code samples express logic or intent unclearly.
RQ2 training-data quality leaf: Efficiency Share of path assignments: 4.0% Distinct studies: 4 Training code samples contain resource-wasting or algorithmically suboptimal implementations.
RQ2 training-data quality leaf: Compliance and Security Risks (Textual) Share of path assignments: 2.0% Distinct studies: 2 Non-code textual training content is harmful copyright-infringing or privacy-leaking.
RQ2 training-data quality leaf: Distribution Imbalance Issues Share of path assignments: 17.8% Distinct studies: 17 The corpus is uneven across languages domains resources data types tasks or difficulty.
RQ2 training-data quality leaf: Redundancy Issues Share of path assignments: 3.0% Distinct studies: 3 The corpus contains duplicate near-duplicate or excessively repetitive samples.
RQ2 training-data quality leaf: Inadequate Diversity Share of path assignments: 23.8% Distinct studies: 22 The corpus insufficiently covers scenarios domains edge cases environments or solution forms.
RQ2 training-data quality leaf: Data Contamination Risks Share of path assignments: 3.0% Distinct studies: 3 Training data overlap with evaluation benchmarks or otherwise leak test solutions.
RQ2 training-data quality leaf: Low-Value Data Share of path assignments: 11.9% Distinct studies: 11 Training data contributes little useful learning signal or introduces noise and error.
RQ3 mapping family: Defect-Pattern Carryover Share of path assignments: 13.9% Distinct studies: 14 Groups atomically coded relations whose final RQ2 source is a Correctness defect within code, while retaining each observed final RQ1 target branch.
RQ3 mapping family: Vulnerability-Pattern Carryover Share of path assignments: 9.9% Distinct studies: 10 Groups atomically coded relations from Security defects within code to downstream non-Compliance generated-code quality problems.
RQ3 mapping family: Restricted/Sensitive-Content Reproduction Risks Share of path assignments: 7.9% Distinct studies: 6 Groups atomically coded governance, privacy, licensing, harmful-content, and sensitive-content relations whose final generated outcome is Compliance.
RQ3 mapping family: Robustness-Pattern Associations Share of path assignments: 2.0% Distinct studies: 2 Groups atomically coded relations from fault-tolerance, input-validation, exception-handling, or boundary-checking patterns in training code to generated-code Robustness or task-specific Correctness outcomes.
RQ3 mapping family: Structural-Quality Associations Share of path assignments: 1.0% Distinct studies: 1 Groups atomically coded relations whose final RQ2 source is a Maintainability defect within code.
RQ3 mapping family: Clarity-Pattern Associations Share of path assignments: 2.0% Distinct studies: 2 Groups atomically coded relations whose final RQ2 source is an Understandability defect within code.
RQ3 mapping family: Efficiency-Pattern Carryover Share of path assignments: 4.0% Distinct studies: 4 Groups atomically coded relations whose final RQ2 source is an Efficiency defect within code.
RQ3 mapping family: Prevalence-Associated Quality Skew Share of path assignments: 17.8% Distinct studies: 17 Groups atomically coded relations from relative prevalence, representation, allocation, or distribution mismatch in the training data.
RQ3 mapping family: Redundancy-Associated Reinforcement and Degeneration Share of path assignments: 3.0% Distinct studies: 3 Groups atomically coded relations from duplicate, cloned, or highly similar samples within the training corpus.
RQ3 mapping family: Underexposure-Associated Quality Gaps Share of path assignments: 23.8% Distinct studies: 22 Groups atomically coded relations from missing or insufficient task, domain, language, API, temporal, security, performance, or format coverage.
RQ3 mapping family: Contamination-Associated Familiarity Differences Share of path assignments: 3.0% Distinct studies: 3 Groups atomically coded relations from verified or explicitly qualified potential overlap between training data and evaluation material.
RQ3 mapping family: Supervision-Signal Quality Associations Share of path assignments: 11.9% Distinct studies: 11 Groups atomically coded relations from noisy, erroneous, low-information, irrelevant, malformed, or misaligned non-code supervision.
RQ1 generated-code quality dimension: Security Share of path assignments: 12.9% Distinct studies: 13 Generated code contains weaknesses permitting unauthorized access disclosure alteration or unsafe execution.
RQ1 generated-code quality dimension: Compliance Share of path assignments: 7.9% Distinct studies: 6 Generated code violates legal ethical or regulatory constraints.
RQ1 generated-code quality dimension: Robustness Share of path assignments: 1.0% Distinct studies: 1 Generated code cannot remain stable under abnormal inputs boundary conditions or non-malicious runtime disruptions.
RQ1 generated-code quality dimension: Maintainability Share of path assignments: 2.0% Distinct studies: 2 Generated code is difficult to modify extend or reuse without introducing risk.
RQ1 generated-code quality dimension: Understandability Share of path assignments: 3.0% Distinct studies: 3 Humans have difficulty comprehending the generated code's logic intent or structure.
RQ1 generated-code quality dimension: Efficiency Share of path assignments: 6.9% Distinct studies: 6 Generated code unnecessarily consumes time or memory resources.
RQ1 generated-code quality dimension: Correctness Share of path assignments: 60.4% Distinct studies: 48 Generated code fails to implement intended semantics or to execute validly.
RQ1 generated-code quality dimension: Generation Derailment Share of path assignments: 5.9% Distinct studies: 5 The generated response departs from the task-directed course needed to provide a usable code deliverable.
Correctness (13.9%)
Security (12.9%)
Compliance (3.0%)
Robustness (2.0%)
Maintainability (1.0%)
Understandability (2.0%)
Efficiency (4.0%)
Compliance and Security Risks (Textual) (2.0%)
Distribution imbalance (17.8%)
Redundancy (3.0%)
Inadequate diversity (23.8%)
Contamination (3.0%)
Low-value data (11.9%)
Defect Carryover (13.9%)
Vulnerability Carryover (9.9%)
Content Reproduction Risks (7.9%)
Robustness Patterns (2.0%)
Structural Quality (1.0%)
Clarity Patterns (2.0%)
Efficiency Carryover (4.0%)
Prevalence Quality Skew (17.8%)
Redundancy Reinforcement (3.0%)
Underexposure Gaps (23.8%)
Contamination Familiarity (3.0%)
Supervision Signal Quality (11.9%)
Security (12.9%)
Compliance (7.9%)
Robustness (1.0%)
Maintainability (2.0%)
Understandability (3.0%)
Efficiency (6.9%)
Correctness (60.4%)
Generation derailment (5.9%)
RQ2 code-attribute leaf
RQ2 non-code/corpus leaf
Band color identifies mapping family
RQ1 generated-code dimension
Band widths and percentages use path assignments; one study may support multiple paths.
Hover a band or node to inspect counts, study coverage, evidence coverage, and definitions. Hovering also highlights the corresponding mapping family. This file is self-contained and makes no network requests.