Loading page content
Preparing content…
SOC 15-2051 · O*NET 15-2051.00
Data scientists in the United States correspond to O*NET-SOC 15-2051.00 (Data Scientists), developing and implementing techniques or analytical applications using data-oriented programming languages and visualization software to turn raw data into meaningful information. They apply data mining, data modeling, natural language processing and machine learning to extract and analyze information from large structured and unstructured datasets; use statistical software to analyze, process and clean large datasets; create charts and other visualizations to communicate analytical results; test, validate and redesign models to ensure accurate prediction of target outcomes; compare statistical model performance metrics such as loss functions or the proportion of variance explained; apply feature selection algorithms; identify relationships, trends or factors that may affect research outcomes; write new functions or applications in programming languages for analysis; read scientific articles and conference papers to identify emerging analytical trends and technologies; interview stakeholders to identify problems to analyze; design surveys or data collection instruments; report modeling and analysis results orally or in writing to management and other end users; and recommend data-driven solutions to key stakeholders. They may also create dynamic data reports. The preparation level is Job Zone Four (considerable preparation needed, SVP 7.0 to below 8.0); the official education survey shows that among new entrants, 48% require a bachelor's degree, 44% a master's degree, and 4% an associate's or other two-year degree. On O*NET, tasks, work context and preparation level use this detailed occupation's own scope; its national wage page publishes annual and hourly percentiles for 2025. Comparable Chinese roles are commonly data analysis, algorithm or modeling positions; duties, employment and pay definitions do not correspond one-to-one with this US detailed occupation and cannot be interchanged.
Model impact index for three equally weighted representative core tasks; not a measured share of working time or a job-loss probability. floor((50+75+75)/3+0.5)=67; fermatmind-ai-impact-v1; 2026-10-18.
Work mainly takes place in offices or remotely, with long periods spent on code, data and documentation and frequent communication with business stakeholders, engineering and compliance teams. Data availability, business cycles and model risks vary significantly across employers such as professional and technical services, finance and insurance, information services and corporate management.The core outputs are reviewable analytical conclusions and reproducible models: data descriptions and quality records, reproducible code, baseline models and performance metrics, validation and error analysis, plans for monitoring bias and drift, and conclusions and limitations written in business language, rather than simply achieving a high model score.
Similar names for adjacent roles do not mean identical responsibilities. First compare whether the work concerns data cleaning and modeling, business metrics and reporting, statistical inference and experimental design, operations optimization, computer and information research, or quantitative financial modeling, and whether it entails responsibility for model deployment, data authorization or regulatory compliance.
Data scientists correspond to occupation 15-2051.00, turning raw information into interpretable results through data programming/visualization, potentially using mining/modeling/NLP/machine learning and dynamic reporting. Current 17 tasks comprise 15 Core, survey design 1 Supplemental and stakeholder interviewing 1 New without an importance rating; not everyone performs every task. Main outputs are analyses/models/validation and recommendations; recommending budget, personnel and marketing plans does not automatically confer approval or management authority. Business intelligence 15-2051.01 is another O*NET detailed occupation under the same SOC; duties can overlap, and the statistics cannot be treated as an independent business intelligence sample. Statisticians, operations research, computing research, bioinformatics science/technology and financial quantitative work can overlap; labels do not imply that a role never validates models. In China, compare actual duties; a one-to-one classification match has not been verified.
Career direction comparison
The table below compares data scientists with five adjacent fields across four dimensions: occupational field, core work and outputs, key distinctions and the choices each may suit better. Codes use the O*NET-SOC classification; sharing codes or data tools does not make duties, evaluation standards or qualifications interchangeable.
| Career track | Core work and outputs | Key distinctions | Choices it may suit better |
|---|---|---|---|
| Business Intelligence Analysts (adjacent occupation, 15-2051.01) | Develop metric systems and reports for business and operations to support day-to-day business decisions. | The distinction is a focus on metric systems and report delivery, emphasizing visualization and business definitions; this occupation centers on building and validating predictive or explanatory models from raw data and bears modeling and validation responsibilities. | More suitable for people who want to support business through metrics and reports. |
AI's impact on occupations
Tasks, work context and technology terms in recruitment have been verified. Code/feature drafts, reporting and experimental assistance are editorial candidates awaiting validation. This round did not measure efficiency gains, employer GenAI adoption or replacement dates. Model validation requires checking data and methods; employer accounts do not replace lawful grounds for processing personal information. Separate voluntary NIST governance from statutory PIPL duties; projected growth is not an observed causal effect of tools. AI Impact Score 67% is an editorial estimate. Model impact index for three equally weighted representative core tasks; not a measured share of working time or a job-loss probability. floor((50+75+75)/3+0.5)=67; fermatmind-ai-impact-v1; 2026-10-18.The three tasks, ratings and sources are listed in the FermatMind method source record on this page.
FermatMind AI Impact Score estimation method v1
FermatMind / 费马测试 · 2026-10-18
Mainland China pay reference
US code O*NET-SOC15-2051.00; no directly comparable China occupation-level wage scope obtained in this round. The search gap does not establish that none exists nationwide.
Editorial duty comparison; a one-to-one China-US classification match or actual China recruitment has not been verified. Confirm actual roles/permissions separately
Insufficient evidence: this round obtained no China occupation-level wage distribution matching 15-2051.00. The NBS material read covers industry wages; this search does not establish that nationwide figures have never been published or that all statistics can only reach industry level.
Insufficient evidence: this round obtained no occupational wage distribution or recruitment quote sample for China data science roles. Compare Chinese roles by actual duties; a one-to-one China-US classification match has not been verifiedEditorial duty comparison: Chinese job titles cannot map directly to US detailed occupations. This round did not verify a nationwide one-to-one classification match or uniform pay method; recruitment/quote gaps remain.
US pay reference
US statistical scope: wages on this page are O*NET/OEWS figures published for 15-2051.00 (Data Scientists), using this detailed occupation's own scope. Its national wage page publishes annual and hourly percentiles for month 5 of 2025 (annual percentiles USD 67,240/85,660/120,230/158,880/199,130; median hourly wage USD 57.80). Percentiles are statistical positions in the national wage distribution under that scope, not individual quotes or take-home pay. Employment size and projections use the BLS Occupational Outlook Handbook scope (2025–2035 projection period).
| Wage position | Wage reference (annual) | Income scope boundaries | How to interpret this position |
|---|---|---|---|
| Lower wage position (10th percentile) | USD 67240 per year; USD 32.33 per hour (2025 10th percentile, not starting pay) | Statistical position at the lower end of the distribution, not individual starting pay, take-home income or total compensation. OEWS includes specified incentive pay and production bonuses and excludes nonproduction bonuses, among others; do not uniformly claim bonuses are excluded. | National percentiles are not groups defined by seniority, industry or management duties. This page has not verified factor contributions; do not map high percentile bands one-to-one to specific institutions/industries. |
Career fit guide
Editorial self-observation: experiences of data/quality checks, validation and attempts at disconfirmation, explaining uncertainty and business communication. It has not been verified that cleaning takes more time than modeling or that any preference necessarily implies better fit.Compare each item with your own tolerance.
Worth exploring further： Editorial self-observation: willingness to check data and methods, learn and accept counterevidence, and explain uncertainty. 21 styles, Impact: attention to detail 100/curiosity 100/innovation 81/dependability 79; these are not percentages or success rates for such personalities.
Check real-world conditions first： First check actual work patterns: time spent on cleaning and validation, frequency of business communication, project delivery and deployment schedules, data authorization and compliance constraints, and whether you accept repeated challenges to uncertainty in conclusions.
Try a small exercise first： First use public or fictional data to complete an exercise of writing a data description and quality checklist, building a baseline model with validation and error analysis, and writing conclusions and uncertainty statements in nontechnical language, or conduct a retrospective on a real project, then assess preferences.
Work pressure, risks and occupational boundaries
Risks concentrate in three categories: technical risks such as data and target leakage, sample bias, concept drift and failure to reproduce results; compliance risks from unauthorized data use, taking restricted data outside its environment and ignoring industry regulatory requirements; and risks of models being misused in high-impact decisions because uncertainty is inadequately communicated. Additional risks include pressure from challenges to conclusions and competition, and changing job duties as tools and organizational divisions of work change.
Separate US task context, scikit-learn methodological guidance, applicable PIPL law and voluntary NIST governance. Risk scenarios are editorial examples, not incident/injury rates or verified policies for all Chinese employers.
Typical scenario：Methodological risk example: training with information unavailable at prediction time, using test data for model selection or fitting preprocessing can make performance evaluation overly optimistic; this is not a measured occupational incident rate.
Main roles：Roles requiring predictive models built from historical data
Entry, certificates and career development
Editorial pathway example: data/problem checks → model validation and interpretation → deployment/monitoring when the role requires it. This is not uniform national progression or necessarily more effective than certificates. O*NET JobZone4/SVP7 to below 8; respondents' opinions on education needed for new hires are 48% bachelor's, 44% master's and 4% associate's, not workforce composition or law. BLS categories of typical bachelor's/no related experience/no on-the-job training do not mean no preparation is needed; OOH says some employers require/prefer master's or doctoral degrees or industry experience. These do not represent hiring or IQ thresholds in China.
Career outlook and related career transitions
Demand layers: occupation-wide projections, actual employer arrangements and current recruitment samples. OOH's expectation that data-driven demand and AI integration support growth is a projection explanation, not a share of observed effects caused by tools. This round has no China sample; technology mentions do not imply current net job additions.A recruitment announcement only indicates demand at one employer or role at one point in time. The three cannot substitute for one another.
United States · Data scientists (BLS Occupational Outlook Handbook and O*NET employment trends, 2025–2035 projection period, this detailed occupation 15-2051.00)
Growth 34.6%; net increase approximately 95400 (2025–2035, OOH35% rounded)
Overall employment change
BLS occupation 15-2051: approximately 275600 in 2025 → approximately 371000 in 2035, net increase 95400/+34.6%, annual average 24800 openings including growth and replacement, self-employed 1.9%. OOH35% is the rounded version of the same projection; O*NET Muchfaster7% and above, is a category. It currently lists 2025–2035, not an individual's hiring probability or a share caused by AI.
Develop and implement techniques or analytical applications using data-oriented programming languages and visualization software to turn raw data into meaningful information: analyze, process or clean large datasets with statistical software; create charts or other visualizations to communicate analysis results; test, validate and redesign models to ensure accurate prediction of target outcomes; compare models using metrics such as loss functions or the proportion of variance explained; apply feature selection algorithms; identify relationships, trends or factors that may affect research outcomes; write new functions or applications in programming languages for analysis; interview stakeholders to identify problems to analyze; design surveys or other data collection instruments; report modeling and analysis results to management and other end users; and recommend data-driven solutions to key stakeholders. Dynamic data reports may also be created.
Choose a real task from this page for a small practice project or a conversation with someone in the field. Use the planning guides to set a date, an output and feedback questions before deciding whether to continue.
Worth trying first: you are willing to check data meaning and quality first, accept spending much of your time on cleaning and validation rather than modeling, actively try to disprove your own models and check for leakage and bias item by item, report uncertainty honestly when results are uncertain, and explain findings and limitations to business stakeholders in nontechnical language.Treat business objectives, data sources and authorization, methods and evaluation metrics, and compliance requirements as inputs that must be checked. Do not substitute correlation for causation, tune on the test set and still treat it as evidence of generalization, or use personal or restricted data without authorization.
Consider carefully: if you only want to model and tune parameters, are unwilling to check data sources and permissions, find it hard to accept models being rejected by business or compliance teams, or are unwilling to invest time in documentation, reproduction and communication, the actual work will differ from your expectations.Assess fit using analytical retrospectives on real projects, mentoring, peer review and business stakeholder feedback; interest types or a single test cannot replace these.
Low-risk exploration: use public or fictional data to complete an exercise of writing a data description and quality checklist, building a baseline model with validation and error analysis, and writing conclusions and uncertainty statements in nontechnical language. Assess only your work preferences; the exercise involves no real business data or personal information and does not constitute successful job training, data authorization or a promise of employment.All paper exercises on this page use fictional material. In real work, handling personal information and restricted data, deploying and monitoring models, and meeting industry regulatory requirements must follow the employer's information management policies, data security and personal information protection rules, and industry compliance requirements. Modeling in specific fields such as finance or healthcare must also comply with local rules and internal approval procedures.
The editors organize the work into problem/data preparation → modeling and validation → interpretation and delivery; 17 tasks: 15 Core, 1 Supplemental and 1 New. Interviewing is New with no importance rating; survey design is Supplemental. Check against the specific role. Bias monitoring, independent holdout validation and leakage checks are methodological/governance suggestions, not all O*NET tasks or a universal workflow for all employers. Check recommendations separately from authority to execute or approve.
Data description and quality check records
Reproducible code, models and validation records
Data scientists work with business problems, raw structured and unstructured data, features and models, evaluation metrics, and boundaries on data permissions and compliance. Outputs include reproducible analyses and models, validation and error analysis records, bias and drift monitoring plans, and conclusions and limitations written in business language. Work is governed by business objectives and evaluation metrics, data authorization and compliance requirements, methodological standards and the employer's information management policies, rather than by job titles. This table compares data scientists with adjacent business intelligence analysts, statisticians, operations research analysts, computer and information research scientists, mathematicians and bioinformatics scientists across seven dimensions: core purpose, main objects and governing criteria, core process, main outputs, responsibility holders, collaboration and supervision, and limits on conclusions. Adjacent roles all involve data and quantitative methods, but their objects—end-to-end data modeling and prediction, business metrics and reports, statistical inference and experimental design, operations optimization and decision models, computing and information science problems, and biomedical statistics—and whether building verifiable models from raw data is central do not transfer merely because the roles are adjacent. Compare actual outputs and business stages item by item.
| Dimension | Primary track | Comparison track |
|---|---|---|
| Core purpose | Build and validate models from raw data, produce reviewable predictive or explanatory conclusions, and deliver and monitor them in business language. | Business intelligence involves queries/reports/trends and system testing; statisticians work on theory/sampling analysis; operations research develops options under constraints; computing research develops computational methods/systems; mathematicians work on theory and applications; bioinformatics scientists develop computational and statistical tools for biomolecular data. These purposes can overlap without automatically transferring responsibilities. |
| Main objects and governing criteria | Business problems, raw data and features, models and evaluation metrics, data permissions and compliance boundaries; governed by business objectives, methodological standards and the employer's information management policies. | Metrics and business information (business intelligence), statistical problems/survey data (statisticians), operational objectives and constraints (operations research), computing science problems (computing research), mathematical structures (mathematicians), and molecular information such as genomic/proteomic data (bioinformatics scientists). These are not mutually exclusive objects or authorization for clinical practice. |
| Core process | Define the problem → check data and permissions → clean and engineer features → model and compare → validate and analyze errors → deliver, monitor and iterate. | Adjacent fields may respectively involve reporting and trend testing, statistical analysis, optimization solving, computational algorithm research, mathematical proofs and applications, and bioinformatics computing tools and molecular data analysis. Processes may overlap; check the actual role. |
| Main outputs |
Problem definition and communication | Identifying management objectives that data analysis can address is Core; interviewing stakeholders is New without an importance rating, so it cannot be described as rated Core or mandatory for everyone.
Data preparation | Cleaning/processing and sampling techniques are Core; designing survey instruments is Supplemental. Editors suggest checking fields/quality/lawful purposes, rather than uniformly treating optional surveys as job responsibilities.
Apply feature selection algorithms to build models, compare models using statistical performance metrics such as loss functions or the proportion of variance explained, and identify relationships, trends or factors that may affect outcomes.
Test, validate and redesign models to ensure accurate prediction of target outcomes, check for target leakage, data leakage and sample bias, and document validation for review.
Create charts and other visualizations to communicate analytical results, report modeling and analysis results orally or in writing to management and other end users, and recommend data-driven solutions to key stakeholders.
Read scientific articles, conference papers and other research to identify emerging analytical trends and technologies, write new functions or applications in programming languages to conduct analysis, and explain the scope and uncertainty of conclusions.
Six recurring professional judgments: are the business problem and evaluation metrics clearly defined; have data sources, field meanings and usage permissions been checked; are data quality and sample representativeness sufficient to support conclusions; has the model been independently validated and checked for target and data leakage; are uncertainty and scope of applicability reported honestly; and when anomalies, bias or unclear compliance boundaries arise, is work paused, reviewed and escalated?
Assessment one: are the problem and metrics clearly defined?
Assessment two: have data and permissions been checked?
Assessment three: is data quality adequate?
| Statisticians (adjacent occupation, 15-2041.00) | Design experiments and sampling plans, perform statistical inference and assess uncertainty. | The distinction is a focus on inference methods and experimental design, emphasizing statistical theory and hypothesis testing; this occupation often targets engineering implementation and prediction and must account for code implementation and model monitoring. | More suitable for people whose main work is statistical inference and experimental design. |
|---|
| Operations Research Analysts (adjacent occupation, 15-2031.00) | Build optimization and decision models to provide options for resource allocation and process scheduling. | The distinction is a focus on optimization under objective functions and constraints, aimed at decision recommendations; this occupation more often handles prediction and explanation tasks and connects with data engineering and deployment processes. | More suitable for people willing to work on optimization modeling and decision analysis. |
|---|
| Computer and Information Research Scientists (adjacent occupation, 15-1221.00) | Research and design new algorithms, computational models and system solutions. | The distinction is a focus on algorithm and system research, publication and prototype validation; this occupation centers on end-to-end analysis and model delivery using business data, with research and publication not required outputs. | More suitable for people willing to research algorithms and systems. |
|---|
| Mathematicians (adjacent occupation, 15-2021.00) | Study mathematical structures and theoretical methods and apply them to scientific and engineering problems. | The distinction is a focus on developing and proving mathematical theories, usually requiring higher degrees and research training; this occupation works with data and business problems to produce implementable analyses and models. | More suitable for people willing to research mathematical theory. |
|---|
| Biostatisticians (adjacent occupation, 15-2041.01) | Design statistical plans and analyze data for biomedical research and clinical trials. | The distinction is a focus on biomedical endpoints and trial standards, subject to domain regulations; this occupation spans broader business domains but likewise bears responsibility for methods and validation. | More suitable for people willing to perform statistical analysis in biomedicine. |
|---|
Conclusion: these fields all use quantitative methods to handle data, but differ significantly in their objects—end-to-end data modeling and prediction, business metrics and reports, statistical inference and experimental design, operations optimization and decision models, algorithm and system research, and biomedical statistics—and whether building and validating models from raw data is central. Choose according to actual outputs, methods and business stages in the target role, rather than job titles alone.
Next step: use the fictional exercises on this page or a retrospective on a real project to check whether the target field concerns end-to-end modeling and prediction, business metrics and reports, statistical inference and experimental design, operations optimization, algorithm and system research, or biomedical statistics, then choose a direction.
Editorial ratings use the existing occupational responsibilities and AI task analysis on this page. They are not adoption rates, measured effects, job-disappearance or personal-unemployment probabilities, or official scores from Stanford or another external institution. No measured time weights or future robot deployment are assumed.
The extent to which a task can be performed or assisted by existing technology; it does not indicate whether employers adopt it or how staffing or wages change.
Investigate organizational adoption using evidence from actual employers. System investment/data governance/team training are editorial conditions to check; adoption speed or a causal claim that it is usually slower than technology has not been verified.
Business demand/data investment/division of work are editorial conditions requiring investigation; each causal contribution has not been verified. Occupational growth projections are not causal experiments on tools.
| Work track | Task | Current changes | Human control points |
|---|---|---|---|
| Data processing | Analyze, process or clean large datasets using statistical software | Editorial candidate awaiting validation, not efficiency/adoption measured on this page: automation tools can accelerate cleaning and statistical work. | People are responsible for deciding whether data quality and definitions are usable. |
| Visualization | Create charts or other visualizations to communicate analytical results | Editorial candidate awaiting validation, not efficiency/adoption measured on this page: charting tools can quickly produce first drafts. | People are responsible for choosing presentation methods and interpreting meaning. |
| Model validation | Test, validate and redesign models | Editorial candidate awaiting validation, not efficiency/adoption measured on this page: platforms can automatically run cross-validation and metrics. | People are responsible for deciding whether the validation design is sound. |
| Model comparison | Compare models using metrics such as loss functions or the proportion of variance explained | Editorial candidate awaiting validation, not efficiency/adoption measured on this page: tools can calculate performance metrics in batches. | People are responsible for choosing trade-offs in light of business objectives. |
| Feature engineering | Apply feature selection algorithms | Editorial candidate awaiting validation, not efficiency/adoption measured on this page: automation can rank feature importance. | People are responsible for assessing feature availability and leakage risks. |
| Communicating results | Report modeling and analysis results to management or end users | Editorial candidate awaiting validation, not efficiency/adoption measured on this page: reporting tools can assist in preparing materials. | People are responsible for explaining uncertainty and making recommendations. |
| Judging insights | Identify relationships and trends or factors affecting research outcomes | Editorial candidate awaiting validation, not efficiency/adoption measured on this page: exploration tools can suggest associations and anomalies. | People are responsible for distinguishing correlation from causation and proposing hypotheses. |
Sources used on this page are listed below with the objects they actually regulate or measure and their usage limitations.
Tasks are divided below into three categories: parts more readily changed by tools, parts where people retain judgment and responsibility, and parts requiring checks on both sides.
AI mainly changesEditorial candidate awaiting validation: generative assistants may handle some boilerplate code, function drafts and model prototypes.
Still the responsibility of peoplePeople are responsible for confirming methodological correctness, data usability and readiness for deployment.
AI mainly changesEditorial candidate awaiting validation: automation tools can accelerate feature statistics, experiment records and documentation.
Still the responsibility of peoplePeople are responsible for deciding whether features are reasonable and whether leakage exists.
AI mainly changesEditorial candidate awaiting validation: charting and dashboard tools can quickly produce first drafts and refresh automatically.
Still the responsibility of peoplePeople are responsible for interpreting results, explaining uncertainty and communicating externally.
Responsibility cannot transfer along with the system when tools are introduced; below are human control steps that can be implemented in this occupation.
Check data sources, field meanings, sampling definitions and usage authorization, and whether data used to generate results is within authorized scope. Unclear input definitions or permissions can invalidate both conclusions and compliance.
Retain human review of method selection, target and data leakage checks, bias assessment and deployment readiness. Do not substitute automatic generation or platform default metrics for validation.
Editors recommend retaining data/code versions, evaluations and decision rationales. PIPL55 defines situations triggering prior impact assessment; under Article 56, assessment reports/processing records must be retained for at least 3 years. This is not a uniform 3-year retention requirement for all models/all raw data.
When performance drops significantly, drift exceeds expectations, bias is abnormal or tool outputs conflict with business definitions, pause deployment or high-risk actions, review and escalate under established policies.
For identifiable personal information in China, check lawful grounds under PIPL13 (not consent in all cases), minimum necessity under Article 6, and reasonable permissions/regular security education under Article 51; an employer account does not replace a lawful basis. Automated decision-making using personal information additionally falls under Article 24 on transparency, fairness and corresponding individual rights, Article 55 on prior impact assessment, and Article 56 requiring reports and records to be retained for at least 3 years. Do not apply these uniformly to all data science models.
Code, features or conclusions from generative assistants may look complete but rest on incorrect assumptions or be unsuitable for the current data distribution.
Control method:Treat generated results as drafts awaiting validation, review methods and data assumptions through standard procedures, and independently validate and document key conclusions.
Using information unavailable at prediction time, including target-derived fields unavailable at prediction time, or involving test data in model selection/repeated tuning can make performance estimates overly optimistic, with new-data or deployment performance below evaluation results. Correlated features alone are not leakage; examples do not establish that all models fail to generalize or decline by a uniform amount.
Control method:Methodological suggestion: information unavailable at prediction time must not enter training. Split data first; fit preprocessing/feature selection only on training data, and do not use test data for model selection. Cross-validation can use Pipeline. Check time/group splits separately based on data dependencies; do not always split randomly.
When training data fails to represent the target population, models may systematically overestimate or underestimate certain groups, causing material harm in finance, healthcare and human resources settings.
Control method:For identifiable personal information in China, check lawful grounds under PIPL13 (not consent in all cases), minimum necessity under Article 6, and reasonable permissions/regular security education under Article 51; an employer account does not replace a lawful basis. Automated decision-making using personal information additionally falls under Article 24 on transparency, fairness and corresponding individual rights, Article 55 on prior impact assessment, and Article 56 requiring reports and records to be retained for at least 3 years. Do not apply these uniformly to all data science models. Editors recommend group-level error/scope checks; model scoring is not invariably statutory automated decision-making.
Changes in the business environment may change model performance; inevitable decline for every model has not been verified. Missing version/random-state/process records can hinder reproduction, while records do not guarantee an unchanged external environment.
Control method:Editors recommend checking performance changes and scope of applicability, and recording versions, random states and evaluation procedures. Validate retraining needs against actual problems; do not promise that recording guarantees reproducibility.
Using unauthorized data or data of unknown origin, or exporting restricted data to external tools, creates compliance and trust risks.
Control method:Follow employer information management policies and data security and personal information protection requirements; process data within authorized scope and preserve records. Obtain internal approval where industry regulation applies.
Build foundations in data cleaning, validation and reproducibility before learning generative assistants and platform tools, and develop habits of recording data and code versions and actively trying to disprove your own models.
Treat tool outputs as drafts awaiting validation and retain independent judgment on method selection, leakage and bias checks, and uncertainty in conclusions.
Define tool use and human review points, data authorization and traceability requirements, responsibilities for bias and drift monitoring, and deployment approval routes; incorporate data governance into procedures.
Train statistical and modeling methods, engineering implementation and reproducibility, and awareness of validation, responsibility and compliance when humans and machines share work.
O*NET lists the following for data scientists (15-2051.00): analyze, process or clean large datasets using statistical software; create charts or other visualizations to communicate analytical results; test, validate and redesign models to ensure prediction accuracy; report modeling and analysis results to management or other end users; recommend data-driven solutions to key stakeholders; identify management objectives that data analysis can address; compare models using metrics such as loss functions or the proportion of variance explained; apply feature selection algorithms; identify relationships, trends or factors affecting research outcomes; write new functions or applications in programming languages for analysis; read scientific articles and conference papers to identify emerging analytical trends and technologies; interview stakeholders to identify problems to analyze; and design surveys or other data collection instruments, among others.
Source:O*NET detailed profile: 15-2051.00 Data ScientistsOriginal material ↗
The O*NET hot technology profile lists software and technology keywords in employer job postings and their occurrence rates: Python (66%), Structured query language SQL (51%), R (34%), Tableau (22%), Microsoft Power BI (19%), Amazon Web Services AWS (17%), Microsoft Azure (13%), TensorFlow (11%), PyTorch (10%), SAS (9%), Microsoft Excel (8%) and Snowflake (8%), among others. These are recruitment and occupational-context market signals, not usage or proficiency rates.
Source:O*NET hot technologies: 2025 data science job-posting textOriginal material ↗
No. O*NET work-context percentages describe tasks and environmental conditions such as sitting and visual demands, the importance of accuracy, decision freedom, time pressure, competition and collaboration frequency. They contain no data on changes in technology adoption, staffing numbers or personal income.
Source:O*NET detailed profile: 15-2051.00 Data ScientistsOriginal material ↗
This round has no evidence of occupation-specific efficiency gains, employer GenAI adoption, or causal effects on staffing/wages. Candidate tools do not mean work has already accelerated, and ILO exposure does not imply layoff probabilities.
Source:O*NET detailed profile: 15-2051.00 Data ScientistsOriginal material ↗
BLS occupation 15-2051: approximately 275600 in 2025 → approximately 371000 in 2035, net increase 95400/+34.6%, annual average 24800 openings including growth and replacement, self-employed 1.9%. OOH35% is the rounded version of the same projection; O*NET Muchfaster7% and above, is a category. It currently lists 2025–2035, not an individual's hiring probability or a share caused by AI.
Source:BLS table 1.2: 2025–2035 data scientist projectionsOriginal material ↗
No. Task exposure measures the alignment between tasks and technology; it does not provide replacement dates or predictions for a single occupation. The ILO task exposure framework cannot be interpreted as the probability of job loss.
Source:ILO: Generative AI and Jobs—Refined Global Index (abstract)Original material ↗
For identifiable personal information in China, check lawful grounds under PIPL13 (not consent in all cases), minimum necessity under Article 6, and reasonable permissions/regular security education under Article 51; an employer account does not replace a lawful basis. Automated decision-making using personal information additionally falls under Article 24 on transparency, fairness and corresponding individual rights, Article 55 on prior impact assessment, and Article 56 requiring reports and records to be retained for at least 3 years. Do not apply these uniformly to all data science models. NIST governance is voluntary guidance; editors recommend retaining method reviews and version records, with actual industry rules checked separately.
Source:NIST AI RMF Core: GOVERN 3.2, etc.Original material ↗
Next step: check the three categories of task changes above against specific analytical scenarios item by item—which steps are already handled by generative assistants, platforms and automated pipelines, which still require human judgment and responsibility, and who is responsible for data authorization, adequate validation and model deployment.
Insufficient evidence: this round obtained no verifiable recruitment or pay quote statistics for data science or modeling roles in mainland China. Quotes from individual platforms or employers cannot represent nationwide levels.
Actually read in this round: NBS2025 industry wages and PIPL; China occupational wages and current recruitment not obtained
| Education and disciplinary background | Role track | Notes and checking suggestions |
|---|---|---|
| Bachelor's (mathematics, statistics, computing or related disciplines) | Editorial preparation/development checking example, not an automatic degree-to-role match | O*NET JobZone4/SVP7 to below 8; respondents' opinions on education needed for new hires are 48% bachelor's, 44% master's and 4% associate's, not workforce composition or law. BLS categories of typical bachelor's/no related experience/no on-the-job training do not mean no preparation is needed; OOH says some employers require/prefer master's or doctoral degrees or industry experience. These do not represent hiring or IQ thresholds in China. |
| Master's (statistics, computing or related fields) | Editorial preparation/development checking example, not an automatic degree-to-role match | Editorial preparation suggestions: check statistical/programming/business requirements of target roles. Common discipline/degree preferences in actual Chinese roles have not been verified; a master's degree guarantees neither an algorithm role nor income, and accounts do not replace lawful purposes. |
| Doctorate or research track (statistics, machine learning, etc.) | Editorial preparation/development checking example, not an automatic degree-to-role match | For identifiable personal information in China, check lawful grounds under PIPL13 (not consent in all cases), minimum necessity under Article 6, and reasonable permissions/regular security education under Article 51; an employer account does not replace a lawful basis. Automated decision-making using personal information additionally falls under Article 24 on transparency, fairness and corresponding individual rights, Article 55 on prior impact assessment, and Article 56 requiring reports and records to be retained for at least 3 years. Do not apply these uniformly to all data science models. A doctorate does not automatically confer a research role or compliance appointment. |
| Substantial project experience + cross-departmental collaboration abilities | Editorial preparation/development checking example, not an automatic degree-to-role match | Editorial development example: model/team leads require checks of actual assigned duties, experience and personnel budget authority; project experience does not automatically confer management authority. Uniform hiring conditions have not been verified. |
Note: failure to obtain separately listed occupational figures does not mean official figures have never been published. This page records only the actual material consulted and results, without filling gaps using US data, industry averages or opaque quote samples.
Editorial contract checks: actual role/grade, contracting entity, monthly pay/bonuses/equity, working hours and outsourcing arrangements. This round has no actual recruitment sample and does not claim nationwide full-time work or any pay form predominates.
Employment checks: job titles such as data analysis, algorithms and modeling can differ substantially from actual duties. Grade, performance pay and equity structures also significantly affect income. Industry averages for information transmission, software and information technology services cannot be treated as wages for this occupation; check the specific role and current rules.
Editorial work-arrangement checks: investigate delivery, deployment/monitoring, review duties and overtime pay in actual roles. A uniform business season in China or retraining requirements for every model have not been verified.
This round has no China occupational wage or tool-causality data. Industry wages can only provide genuinely matching context; statistical publication practices do not establish that occupational wage figures for tool-related changes can never be obtained.
| Median wage position | USD 120230 per year; USD 57.80 per hour (2025 median) | Does not equal individual take-home income | The median is the midpoint of the distribution: under that scope, half of workers are below and half above it. It is not average starting pay and does not mean new hires immediately earn it. |
|---|
| Higher wage position (90th percentile) | USD 199130 per year; USD 95.74 per hour (2025 90th percentile, not a seniority threshold) | Statistical position at the higher end of the distribution; does not mean it can be reached upon entry | OEWS covers pre-tax straight-time wages for nonfarm employees, excluding self-employed workers. It includes incentive commissions/production bonuses and excludes overtime premiums/shift differentials/nonproduction bonuses, among others; it cannot be described as excluding all bonuses or performance pay. Equity and individual actual income require contract/tax checks; no basis for estimating individual after-tax income was obtained. |
|---|
Pay · annual wage 10th percentile：USD 67240 per year; USD 32.33 per hour (2025 10th percentile, not starting pay)；Separate this occupation's 2025 national OEWS wages from its 2025–2035 employment by metric. Percentiles are not starting pay or causal factor effects； projections include replacement and do not represent individual opportunities. OEWS excludes self-employed workers while employment includes them.
Pay · annual wage 25th percentile：USD 85,660 per year (25th percentile, wage data for month 5 of 2025)；The 25th percentile means approximately one-quarter of workers earn less annually, a statistical position rather than a skill level. O*NET national wages: 15-2051.00 (wage data for month 5 of 2025, this detailed occupation)
Pay · median annual wage：USD 120230 per year; USD 57.80 per hour (2025 median)；Separate this occupation's 2025 national OEWS wages from its 2025–2035 employment by metric. Percentiles are not starting pay or causal factor effects； projections include replacement and do not represent individual opportunities. OEWS excludes self-employed workers while employment includes them.
Pay · annual wage 75th percentile：USD 158,880 per year (75th percentile, wage data for month 5 of 2025)；The 75th percentile is the upper-quarter position in the distribution, from the same statistical period. Do not infer years of experience or role types. O*NET national wages: 15-2051.00 (wage data for month 5 of 2025, this detailed occupation)
Pay · annual wage 90th percentile：USD 199130 per year; USD 95.74 per hour (2025 90th percentile, not a seniority threshold)；Separate this occupation's 2025 national OEWS wages from its 2025–2035 employment by metric. Percentiles are not starting pay or causal factor effects； projections include replacement and do not represent individual opportunities. OEWS excludes self-employed workers while employment includes them.
State/regional wages were not queried in this round; this does not mean regional data is absent. Retain national official annual/hourly figures, OEWS coverage and 2025–2035 employment separately; do not infer China or individual wages.
O*NET2025 lists four broad industry categories 25/19/12/10, with the remaining 34 an editorial remainder. OOH's publishing/broadcasting/content 12/insurance 10/computer systems design 10/company management 10/credit intermediation 6 use another level of detail. Do not combine them or call them five official O*NET categories.
National percentiles are statistical positions. This round did not separate contributions from industry/region/seniority/duties; do not present these as established causes of differences or individual quotes.
BLS occupation 15-2051: approximately 275600 in 2025 → approximately 371000 in 2035, net increase 95400/+34.6%, annual average 24800 openings including growth and replacement, self-employed 1.9%. OOH35% is the rounded version of the same projection; O*NET Muchfaster7% and above, is a category. It currently lists 2025–2035, not an individual's hiring probability or a share caused by AI.
| Industry | Industry employment share/statistical status | How to interpret |
|---|---|---|
| Professional, scientific and technical services | Professional, scientific and technical services | 25% (2025 employment share) | 2025 professional, scientific and technical services 25% is a broad industry employment share for this occupation; do not infer wages/specific business shares or map it to Chinese industries. |
| Finance and insurance | Finance and insurance | 19% (2025 employment share) | 2025 finance and insurance 19% is a broad industry employment share for this occupation; do not infer wages/specific business shares or map it to Chinese industries. |
| Information | Information | 12% (2025 employment share) | 2025 information 12% is a broad industry employment share for this occupation; do not infer wages/specific business shares or map it to Chinese industries. |
| Industry distribution scope | Management of companies 10%; remaining 34% is an editorial remainder, not an official fifth category | Management of companies 10% is the official fourth category; the remainder 34%=100−25−19−12−10 is only an editorial remainder, not an official other category or self-employment. BLS separately reports self-employment at 1.9%; do not combine them. |
Editorial compensation checks: contract, actual duties, paid hours/bonuses/equity and region. This round did not verify factor contributions or outsourcing/flexible-role shares; OEWS wages do not represent individual income.
Duties vary considerably under the same occupational title: roles limited to modeling and tuning differ from roles also responsible for data governance, model deployment, monitoring and management in methodological and accountability requirements, with different skills and work intensity.
Editorial compensation checks: contract, actual duties, paid hours/bonuses/equity and region. This round did not verify factor contributions or outsourcing/flexible-role shares; OEWS wages do not represent individual income.
Separate this occupation's 2025 national OEWS wages from its 2025–2035 employment by metric. Percentiles are not starting pay or causal factor effects； projections include replacement and do not represent individual opportunities. OEWS excludes self-employed workers while employment includes them.
Separate this occupation's 2025 national OEWS wages from its 2025–2035 employment by metric. Percentiles are not starting pay or causal factor effects； projections include replacement and do not represent individual opportunities. OEWS excludes self-employed workers while employment includes them.
Separate this occupation's 2025 national OEWS wages from its 2025–2035 employment by metric. Percentiles are not starting pay or causal factor effects； projections include replacement and do not represent individual opportunities. OEWS excludes self-employed workers while employment includes them.
Are my interests close to data science?
O*NET publishes interest code IC for data scientists (15-2051.00): Investigative refers to studying and analyzing data and exploring patterns with statistical and mathematical methods; Conventional refers to organizing information and data according to procedures and rules. Treat the code as a clue to activity types, not a personality label.
Do not use it this way:Interest codes describe activity types only; they do not predict work performance or constitute hiring criteria.
Understand my vocational interestsCan my personality traits help with data science?
Editorial self-observation: following procedures to keep records and willingness to continue learning. O*NET work styles are not Big Five research; this round did not verify that conscientiousness/openness have the most direct relationship to this occupation or any occupational fit thresholds.
Do not use it this way:Trait descriptions do not diagnose personality or predict competence in a specific role.
Understand my Big Five traitsCan MBTI tell me whether modeling or business communication suits me?
MBTI-type tools describe preferences, not occupational assignments. Differences between modeling and business communication mainly concern work format—methods and experiments versus clarifying requirements and communicating results—rather than preference types. Check through actual experience.
Do not use it this way:Do not use type results as a basis for hiring, assignment or promotion.
Understand my personality preferencesCan the Enneagram help me understand why I want to work in data science?
Enneagram-type descriptions offer motivational clues: wanting to explain real-world problems with data, solve business problems through technology, or enter an industry to build domain understanding leads to different choices. Treat it as a questioning tool, not a job conclusion.
Do not use it this way:Do not treat motivational descriptions as career choice conclusions or employment criteria.
Understand my motivational cluesWhat cognitive and practical abilities does this occupation require?
O*NET JobZone4/SVP7 to below 8; respondents' opinions on education needed for new hires are 48% bachelor's, 44% master's and 4% associate's, not workforce composition or law. BLS categories of typical bachelor's/no related experience/no on-the-job training do not mean no preparation is needed; OOH says some employers require/prefer master's or doctoral degrees or industry experience. These do not represent hiring or IQ thresholds in China. Observe actual statistical/modeling/communication performance; do not infer IQ hiring thresholds from education/JobZone.
Do not use it this way:Cognitive test results are not hiring standards and do not replace actual training and project performance.
Understand my cognitive performanceAre strong communication and emotion management abilities needed?
Among US respondents, 96% use email daily, 43% have face-to-face discussions daily/48% weekly or more but not daily, 35% rate teamwork Very important, and 56% have contact about half the time. Frequencies are not emotional abilities or a uniform collaboration system in China; editors suggest observing experiences of explaining results and communicating.
Do not use it this way:Do not use communication styles to decide hiring or assignments, or treat emotional descriptions as health conclusions.
Understand my communication cluesBetter match:Main work involves data cleaning, feature engineering, model building and comparison, validation and error analysis. These are observable duties; willingness does not imply competence. Check specific roles and collaboration responsibilities separately.
Caution:Constrained by data availability and quality; requires upholding validation principles when conclusions are repeatedly challenged.
Better match:Main work involves clarifying requirements, visualizing and explaining results, business-facing recommendations and collaboration. These are observable duties; willingness does not imply competence. Check specific roles and collaboration responsibilities separately.
Caution:Constrained by business cycles and cross-departmental communication; requires traceable conclusions as requirements change frequently.
Better match:Develop metric systems and reports for business and operations to support day-to-day business decisions.
Caution:15-2051.01 is another O*NET detailed occupation under the same SOC, with potentially overlapping duties. Editors recommend checking business metrics/reporting/testing requirements; automatic transitions or independent detailed-occupation wages are not guaranteed.
View related occupation:Business intelligence analystsBetter match:Design experiments and sampling plans, perform statistical inference and assess uncertainty.
Caution:Centers on inference methods and experimental design; transitioning requires strengthening statistical theory and experimental design abilities.
View related occupation:StatisticiansBetter match:Build optimization and decision models to provide options for resource allocation and process scheduling.
Caution:Centers on optimization modeling and decision analysis; transitioning requires strengthening operations research methods and objective-function modeling abilities.
View related occupation:Operations research analystsBetter match:Research and design new algorithms, computational models and system solutions.
Caution:Target duties involve computational problems, technical solutions and system research, among others. Editors recommend checking required research, algorithm and prototype validation training. Check specific education and preparation duration against the target employer's requirements; no uniformly longer duration is asserted.
View related occupation:Computer and information research scientistsCleaning/validation are core tasks; time shares were not verified in this round. Do not claim everyone spends much of their time mainly away from modeling.
Editorial retrospective suggestion: record changes in conclusions prompted by business definitions, new data or review feedback, and check validation evidence, modification processes and subsequent results. Personal interpretations of feedback can inform discussion of experience but are not a necessary condition for continuous improvement.
O*NET shows 40% rating accuracy and exactness Extremely important and 65% rating competition Highly competitive; maintaining validation and reproducibility is more reliable than pursuing a momentarily high-scoring model.
Modeling and methods emphasize statistical and engineering detail; business communication emphasizes understanding requirements and explaining results. Understand the target track before deciding.
Among US respondents, 96% use email daily, 43% have face-to-face discussions daily/48% weekly or more but not daily, 35% rate teamwork Very important, and 56% have contact about half the time. Frequencies are not emotional abilities or a uniform collaboration system in China; editors suggest observing experiences of explaining results and communicating.
O*NET shows 76% sitting continually or almost continually, 91% working indoors in an environmentally controlled setting daily, and 44% using hands to handle or control tools continually or almost continually. Those concerned about prolonged screen work should first check workstation and work-method arrangements.
Among US respondents, 52% face time pressure weekly or more but not daily, 63% typically work 40 hours/38% more than 40 hours (rounded). Check actual cycles/deployment monitoring and working hours by role; this does not prove universal retraining or a uniform busy season in China.
Editors suggest recording versions and validation assumptions and reviewing actual feedback; this round did not verify comparative validity showing that such records predict fit better than test scores.
No. Tests describe preference clues, not work performance, and do not constitute hiring, qualification or ability judgments. Actual judgments come from analysis and retrospectives on real projects, mentoring feedback and business stakeholder evaluations.
Use the fictional exercises on this page or a retrospective on a real project to check actual conditions and compare observations item by item with the objects, methodological requirements and compliance boundaries of your target track.
Possible consequences：Leakage may make evaluation overly optimistic, with new-data or deployment performance below estimates. Specific declines and individuals' model deployment outcomes have not been verified; documentation examples do not imply inevitable large declines.
Reduce risks：Methodological suggestion: split first; fit preprocessing/feature selection only on training data, and do not use test data for model selection. Pipeline can be used for cross-validation; time/group dependencies require separate checks, and detected leakage requires revalidation.
Basis：scikit-learn developers: leakage and reproducibility guidance
Typical scenario：When training data fails to represent the target population, models may systematically overestimate or underestimate certain groups, causing material harm in high-impact settings and creating compliance problems.
Main roles：Roles predicting or scoring people (finance, human resources, healthcare, etc.)
Possible consequences：Bias causes unfair outcomes and compliance risks.
Reduce risks：For identifiable personal information in China, check lawful grounds under PIPL13 (not consent in all cases), minimum necessity under Article 6, and reasonable permissions/regular security education under Article 51; an employer account does not replace a lawful basis. Automated decision-making using personal information additionally falls under Article 24 on transparency, fairness and corresponding individual rights, Article 55 on prior impact assessment, and Article 56 requiring reports and records to be retained for at least 3 years. Do not apply these uniformly to all data science models. Editors suggest checking group-level errors and scope; PIPL automated decision-making using personal information does not cover all analytical models.
Basis：Personal Information Protection Law: data use and automated decision-making · NIST AI RMF Core: GOVERN 3.2, etc.
Typical scenario：Editorial risk example: environmental changes may affect performance, and inadequate version/random-state records may hinder reproduction; not all models inevitably decline in performance.
Main roles：Roles requiring models to run continuously after deployment
Possible consequences：May hinder detecting performance changes and diagnosing problems; occupational incident rates have not been verified, and no monitoring experiment was conducted on this page.
Reduce risks：Editors suggest checking data/code versions, random states and evaluation design and observing distribution/performance changes. Changes do not inevitably degrade every model, records do not guarantee reproduction of all results, and retraining requires validation.
Basis：scikit-learn developers: leakage and reproducibility guidance · NIST AI RMF Core: GOVERN 3.2, etc.
Typical scenario：For identifiable personal information in China, check lawful grounds under PIPL13 (not consent in all cases), minimum necessity under Article 6, and reasonable permissions/regular security education under Article 51; an employer account does not replace a lawful basis. Automated decision-making using personal information additionally falls under Article 24 on transparency, fairness and corresponding individual rights, Article 55 on prior impact assessment, and Article 56 requiring reports and records to be retained for at least 3 years. Do not apply these uniformly to all data science models.
Main roles：Roles handling personal information and restricted data
Possible consequences：Unauthorized use or leakage causes compliance risks and loss of trust.
Reduce risks：For identifiable personal information in China, check lawful grounds under PIPL13 (not consent in all cases), minimum necessity under Article 6, and reasonable permissions/regular security education under Article 51; an employer account does not replace a lawful basis. Automated decision-making using personal information additionally falls under Article 24 on transparency, fairness and corresponding individual rights, Article 55 on prior impact assessment, and Article 56 requiring reports and records to be retained for at least 3 years. Do not apply these uniformly to all data science models. Industry/sensitive information/cross-border supporting rules were not fully checked in this round; employer policies cannot replace lawful grounds.
Basis：Personal Information Protection Law: data use and automated decision-making
Typical scenario：Code, features or conclusions from generative assistants may look complete but rest on incorrect assumptions, outdated methods or be unsuitable for the current data distribution; direct adoption bypasses validation and amplifies methodological errors.
Main roles：Roles using generative assistants and automated pipelines
Possible consequences：Unvalidated generated results carry methodological errors into deliverables.
Reduce risks：Treat generated results as drafts awaiting validation; review methods and data assumptions through standard procedures, independently validate and document key conclusions, and obtain approval for deployment or high-impact decisions.
Basis：NIST AI RMF Core: GOVERN 3.2, etc.
Typical scenario：Editorial risk example: treating association as causation or omitting applicable conditions may mislead decisions. US 52% facing time pressure weekly or more but not daily and 56% reporting Some freedom do not establish an occupational causal link from rushing to misuse.
Main roles：Roles requiring explanation of model results to business stakeholders
Possible consequences：Misuse of conclusions causes business losses and harm to trust.
Reduce risks：State assumptions, applicable conditions and uncertainty in deliverables, specify limits and points requiring human review for high-impact uses, and provide counterexamples and failure scenarios where necessary.
Basis：NIST AI RMF Core: GOVERN 3.2, etc. · O*NET detailed profile: 15-2051.00 Data Scientists
Check qualifications and credentials separately: education and professional training, programming and analytical abilities, data access and usage authorization, production deployment permissions, industry regulatory requirements and training certificates are different states and cannot substitute for one another.
| Entry-level role | Main work when starting | Editorial evidence suggestions (uniform employer standards not verified) | How to demonstrate ability without experience | Differences from other tracks |
|---|---|---|---|---|
| Analysis support role | Clean data, calculate metrics and perform basic analysis under mentoring, handling data according to standards. | Data quality awareness, standardized procedures, and habits of seeking guidance and keeping records. | A fictional data description and quality checklist. | Focuses on supervised data processing and analytical support without independent modeling and validation responsibility. |
| Data science role | Independently complete analysis from problem clarification and data preparation through modeling and validation. | Method suitability, adequate validation, and expression of conclusions and uncertainty. | A fictional baseline model, validation and error analysis record. | Responsible for reliability of methods and validation; must follow review rules and preserve records. |
| Business data analysis role | Independently clarify requirements, build analyses and explain results and limitations to business stakeholders. |
O*NET JobZone4/SVP7 to below 8; respondents' opinions on education needed for new hires are 48% bachelor's, 44% master's and 4% associate's, not workforce composition or law. BLS categories of typical bachelor's/no related experience/no on-the-job training do not mean no preparation is needed; OOH says some employers require/prefer master's or doctoral degrees or industry experience. These do not represent hiring or IQ thresholds in China.
The O*NET summary shows preparation level Job Zone Four; the US scope cannot be directly equated with hiring and training requirements for such roles in mainland China.
US sitting and work-context data do not define uniform health/working-hour thresholds in China; check actual workstations and tolerable requirements by role.
This page obtained no uniform public conditions for mainland China; check the target employer's requirements for the relevant year. This page is not health or hiring advice.
Able to independently complete a cycle from data checks and modeling through validation and interpretation, check common leakage and bias, and honestly explain uncertainty (from O*NET task and work-context data).
These are task and context observation points, not personality labels or hiring thresholds. Work context shows 40% rating accuracy and exactness Extremely important, 65% rating competition Highly competitive, 52% facing time pressure once or more a week, and 56% reporting Some freedom in decisions.
2025 full-year US job-posting text lists 55 items: Python66/SQL51/R34/Tableau22/PowerBI19/AWS17/Azure13/TensorFlow11/PyTorch10/SAS9/Excel8/Snowflake8%. These are not actual usage, proficiency or GenAI rates; do not sum them or infer business uses from them.
Deliverable: a wholly fictional data description document containing field meanings, sources, handling of missing values and anomalies, and a checklist and conclusions.
State the basis for each judgment (field definitions, data rules or business definitions); descriptions must not contradict one another.
Deliverable: a fictional modeling record containing baseline selection, feature processing, validation splits, performance metrics and error analysis.
Distinguish supplied data from your assumptions, explain why validation is sound, and do not use an actual employer's business data.
Deliverable: a one-page conclusion statement for nontechnical readers containing main findings, applicable conditions, uncertainty ranges and uses that are not recommended.
Deliverable: a one-page statement explaining how you handled data anomalies, performance below expectations or conflicting definitions, which step took most time and what to add next. Do not include actual employer data or nonpublic business information.
| Stage | Possible assessment | Observable evidence | Usage boundaries |
|---|---|---|---|
| Analysis and methodology questions | Whether you can explain problem definitions, reasons for method selection and validation approaches, and identify common leakage and bias. | Answer records and live reasoning processes. | One employer's assessment method does not represent a uniform process. |
| Data and compliance judgments | Whether you can explain the order of checks, pauses and escalation for missing data, conflicting definitions, insufficient permissions or suspected sensitive data. | Answer records and judgment rationales. | Assessment scope is determined by employer policies and current rules; this page makes no inference. |
| Practical work and reproducibility | Procedural quality, reproducibility and explanation quality when cleaning data, modeling and validating within a defined task. |
| Qualification/preparation item | Who it mainly suits | Practical role in job seeking | Whether it is a common entry requirement | Application/processing conditions | Investment considerations | Important limitations |
|---|---|---|---|---|---|---|
| Education and professional training | People preparing to enter data roles | Editorial preparation checks; conditions for most Chinese roles not verified | O*NET JobZone4/SVP7 to below 8; respondents' opinions on education needed for new hires are 48% bachelor's, 44% master's and 4% associate's, not workforce composition or law. BLS categories of typical bachelor's/no related experience/no on-the-job training do not mean no preparation is needed; OOH says some employers require/prefer master's or doctoral degrees or industry experience. These do not represent hiring or IQ thresholds in China. | Follow the target employer's recruitment announcement | Insufficient evidence: this round has no uniform cost/training duration; check actual learning and employer plans |
Fictional scenario: a data scientist encounters four problems within one project cycle. First, a field in training data is only available in the prediction period but has been used as a feature. Second, business stakeholders want a model for high-risk decisions deployed within two weeks. Third, evaluation metrics meet targets across the whole sample, but errors for one group are noticeably higher. Fourth, some data contains customers' personal information, and a colleague requests exporting raw data to an external tool for processing.
Supplied material: business problem and evaluation metrics; field descriptions and data dictionary; data usage and traceability requirements; model deployment and review procedures.
Missing material: rules for deciding whether the field constitutes leakage, whether validation scope may be reduced when time is short, authority for handling high bias, and whether personal information may be exported to external tools. If these conditions are needed in the exercise, state your assumptions and whom you need to consult explicitly.
Write a one-page handling and escalation plan covering assessment and handling of the suspected leakage field, communication and minimum validation requirements when time is short, a response route for high bias, and permissions and traceability when personal information is involved.
State which information comes from the scenario and which remains an assumption, and whether the basis is field descriptions and the data dictionary, validation standards or data security and compliance requirements.
The plan should include: suspected leakage field (check availability at prediction time, redo features and validation based on availability if needed, and record the rationale); insufficient time (explain risks through established procedures, specify indispensable validation steps, and suggest narrowing deployment scope or delaying if needed, without skipping validation to meet schedules); high bias (analyze group-level errors and identify causes, restrict use cases if needed and escalate approval through compliance procedures, retaining assessment records); personal information (process within authorized scope without exporting to external tools; where genuinely needed, apply under established policies and process through compliant channels); handover (state the status, actions taken and pending confirmations for all four matters).
Find a public introduction to data quality or model validation, mark three key requirements and corresponding practices, and record what you want to confirm.
Using public or fictional data, list field meanings, possible missing values or anomalies and checking methods.
Build a simple baseline using public or fictional data and state the split method and metrics used.
Explain how you validated, the main sources of error and whether leakage or bias risks exist.
Write a one-page conclusion in nontechnical language, including applicable conditions, uncertainty and uses that are not recommended.
Ask a peer to review your judgments using only your plan, and record their questions.
Typical duties：Clean data, calculate metrics and perform basic analysis under mentoring, handling data according to standards.
Evidence for advancement：Able to process data according to standards, follow usage and traceability requirements and promptly seek guidance on uncertainties.
Qualification boundaries：Editorial checks: target role education/professional training, experience, actual duties and permissions/appointments. This round has no uniform China conditions or actual live recruitment sample; years of service do not guarantee automatic progression.
Next step：Become familiar with data definitions, toolchains and validation procedures.
Typical duties：Independently complete analysis from problem clarification and data preparation through modeling and validation.
Evidence for advancement：Suitable methods, sufficient validation, and clear conclusions and uncertainty statements.
Qualification boundaries：Editorial checks: target role education/professional training, experience, actual duties and permissions/appointments. This round has no uniform China conditions or actual live recruitment sample; years of service do not guarantee automatic progression.
Next step：Learn complex modeling, feature engineering and deployment requirements.
Typical duties：Handle complex model design and validation, deployment monitoring and retraining arrangements.
Evidence for advancement：Able to balance methodological reliability, deployment stability and business usability.
Qualification boundaries：Editorial checks: target role education/professional training, experience, actual duties and permissions/appointments. This round has no uniform China conditions or actual live recruitment sample; years of service do not guarantee automatic progression.
Next step：Learn model governance and cross-team collaboration.
Typical duties：Lead method selection and model governance, coordinating business, engineering and compliance stakeholders.
Evidence for advancement：Able to balance methodological quality, compliance requirements and implementation feasibility.
Qualification boundaries：Editorial checks: target role education/professional training, experience, actual duties and permissions/appointments. This round has no uniform China conditions or actual live recruitment sample; years of service do not guarantee automatic progression.
Next step：Check this track's duties and appointment requirements.
Typical duties：Assist with requirement clarification, metric organization and result interpretation, documenting the analytical process according to standards.
Evidence for advancement：Able to convey requirements accurately, keep honest records and explain data limitations.
Qualification boundaries：Editorial checks: target role education/professional training, experience, actual duties and permissions/appointments. This round has no uniform China conditions or actual live recruitment sample; years of service do not guarantee automatic progression.
Next step：Become familiar with business metrics and communication procedures.
Typical duties：Independently clarify requirements, build analyses and explain results and limitations to business stakeholders.
Evidence for advancement：Clear conclusions, consistent definitions and complete uncertainty statements.
Qualification boundaries：Editorial checks: target role education/professional training, experience, actual duties and permissions/appointments. This round has no uniform China conditions or actual live recruitment sample; years of service do not guarantee automatic progression.
Next step：Learn metric systems and decision-support methods.
Typical duties：Assist in building data pipelines and base tables, handling data quality and permissions according to standards.
Evidence for advancement：Able to maintain pipelines and table structures according to standards and record changes.
Qualification boundaries：Editorial checks: target role education/professional training, experience, actual duties and permissions/appointments. This round has no uniform China conditions or actual live recruitment sample; years of service do not guarantee automatic progression.
Next step：Become familiar with data warehouses and pipeline tools.
Typical duties：Independently build and maintain data pipelines, feature stores and data quality monitoring.
Evidence for advancement：Stable pipelines, traceable data and adequate quality monitoring.
Qualification boundaries：Editorial checks: target role education/professional training, experience, actual duties and permissions/appointments. This round has no uniform China conditions or actual live recruitment sample; years of service do not guarantee automatic progression.
Next step：Learn data architecture and governance standards.
Perform data cleaning, metric calculation and basic analysis under mentoring and review, following standardized processing and data usage and traceability requirements. The goal is familiarity with data definitions, toolchains and validation procedures.
Independently complete an analysis from problem clarification and data preparation through modeling and validation, check common leakage and bias, honestly explain uncertainty, and seek guidance promptly when definitions are unclear.
Handle model deployment, performance monitoring and retraining arrangements; assess abnormal drift and bias, respond through established procedures and preserve records, and explain the methods' scope and limitations to business stakeholders.
Lead method selection, data governance or analytical system design, coordinating business, engineering and compliance stakeholders. This stage usually requires substantial project experience and management collaboration abilities; appointment conditions are set by employer policies.
Does not establish：Projections concern total occupational change, not individual hiring probabilities. Average annual openings include replacement demand and do not mean immediate increases or decreases in jobs.
United States · Data scientists (O*NET 15-2051.00 scope)
Certification and training search portal
Certification and training information
O*NET provides certification and training search portals for this specialization to check certificate and training information. This describes an information portal, not a demand signal. Certification does not equal a licence, employer appointment, data access or production deployment authorization; training programs themselves do not equal employment.
Does not establish：Certification and training portals do not predict job numbers or wage levels and do not equal employment or data and deployment authorization.
Editorial compilation on this page · Data Scientists
Recruitment announcement checking scope (editorial compilation)
How to check recruitment information
Editorial explanation of how to check job information: a recruitment announcement only indicates demand at one employer or role at one point in time.
Does not establish：Recruitment announcements do not represent nationwide demand, median wages or uniform entry requirements.
These are relationships between adjacent abilities, not automatic promotion or freedom from further learning.
Transferable：Conditional editorial mapping: data queries, visualization and understanding business metrics, only where actual training or duties have provided relevant experience. Does not establish competence or readiness for a direct transition.
To strengthen：To strengthen: metric system design, data warehousing and report delivery abilities.
View occupational detailsTransferable：Conditional editorial mapping: foundations in statistical methods and data analysis, only where actual training or duties have provided relevant experience. Does not establish competence or readiness for a direct transition.
To strengthen：To strengthen: theoretical abilities in experimental design, sampling and statistical inference.
View occupational detailsTransferable：Conditional editorial mapping: quantitative modeling and optimization approaches, only where actual training or duties have provided relevant experience. Does not establish competence or readiness for a direct transition.
To strengthen：To strengthen: operations research methods and objective-function and constraint modeling abilities.
View occupational detailsTransferable：Conditional editorial mapping: use of algorithms and computational tools, only where actual training or duties have provided relevant experience. Does not establish competence or readiness for a direct transition.
To strengthen：To strengthen: research training, algorithm innovation and system prototype validation abilities.
View occupational detailsTransferable：Conditional editorial mapping: statistical modeling and programming implementation, only where actual training or duties have provided relevant experience. Does not establish competence or readiness for a direct transition.
To strengthen：To strengthen: financial market knowledge, pricing and risk modeling abilities.
View occupational detailsTransferable：Conditional editorial mapping: foundations in mathematical derivation and abstract modeling, only where actual training or duties have provided relevant experience. Does not establish competence or readiness for a direct transition.
To strengthen：To strengthen: mathematical theory research methods and proof abilities, usually requiring higher degrees and research training.
View occupational detailsTransferable：Conditional editorial mapping: data modeling and programming-based analysis, only where actual training or duties have provided relevant experience. Does not establish competence or readiness for a direct transition.
To strengthen：To strengthen: biomedical background and knowledge of omics data and experimental methods.
View occupational detailsTransferable：Conditional editorial mapping: data processing and statistical software operation, only where actual training or duties have provided relevant experience. Does not establish competence or readiness for a direct transition.
To strengthen：To strengthen: foundations in statistical methods and data organization standards.
View occupational detailsThis occupation centers on building and validating models from raw data, producing reviewable predictive or explanatory conclusions, and delivering and monitoring them. Business intelligence analysts center on metric systems and report delivery; statisticians on inference methods and experimental design; operations research analysts on optimization and decision modeling; computer and information research scientists on algorithm and system research; mathematicians on developing and proving mathematical theories; and biostatisticians provide statistical methodological support for biomedical research. Similar names do not make duties, evaluation standards or qualifications interchangeable.
Generative assistants can handle some boilerplate code, feature exploration, documentation and model prototypes, and machine learning platforms have shortened training and experiment turnaround times, reducing some repetitive coding and organization. But rapid generation does not mean data is usable, methods are correct or results ready for deployment: data authorization, leakage checks, model validation, bias identification, deployment monitoring and domain decisions still require human review. Technical capability does not equal business or regulatory authorization. O*NET shows 40% rating accuracy and exactness Extremely important, 65% rating competition Highly competitive, and 52% facing time pressure more than once a week. Technology exposure does not equal organizational adoption or job disappearance and cannot be interpreted as the probability of job loss.
What can be checked is task-level change: tools can partially handle three categories—code and prototype generation, feature exploration and documentation, and report and visualization drafts. This page has no occupation-specific adoption rates or replacement data.
Abilities in problem and metric definition, data authorization and definition checks, leakage and bias detection, validation design, expression of uncertainty, and data handling and model deployment according to compliance requirements.
US wage data for month 5 of 2025 show median annual wages of USD 120,230 (median hourly wages USD 57.80), with annual percentiles USD 67,240/85,660/120,230/158,880/199,130, using the scope of this O*NET-SOC detailed occupation 15-2051.00. Percentiles are statistical positions, not individual quotes or take-home income. Occupation-level figures for mainland China were not obtained; this page preserves the gap without substituting industry averages or quote samples.
The US preparation level is Job Zone Four (SVP 7.0 to below 8.0). The official education survey shows that among new entrants, 48% require a bachelor's degree, 44% a master's degree, and 4% an associate's or other two-year degree. Common entry foundations are bachelor's studies in mathematics, statistics, computing or related disciplines; some employers prefer master's or doctoral degrees. Roles in mainland China commonly require relevant disciplinary backgrounds and programming abilities, with specifics set by the target employer.
O*NET provides certification and training search portals for this specialization, but certification does not equal a licence, employer appointment, data access or production deployment authorization. Such roles in mainland China commonly involve recruitment and training under employer requirements; handling personal information and restricted data may require additional authorization. Check specific conditions against the target employer and current rules.
First use public or fictional data to complete an exercise of writing a data description and quality checklist, building a baseline model with validation and error analysis, and writing conclusions and uncertainty statements in nontechnical language, then check the objects, methodological requirements and compliance boundaries of the target track.
The BLS Occupational Outlook Handbook (2025–2035 projection period, this detailed occupation 15-2051.00) shows approximately 275,600 employed in 2025, projected growth of 35% (much faster than average), employment change of about +95,400, and an average of about 24,800 annual openings over ten years including replacement demand. This is an occupation-wide projection. The O*NET summary currently still lists 2024–2034; the periods differ and are cited separately. Work mainly involves data processing, modeling and validation; check local industry structures and data governance requirements, prioritizing role scope, data permissions and compliance requirements.
FermatMind career content editorial team (source checking and exercise design)
Updated: 2026-10-09 (source retrieval dates are in the source register)
Read in this round: O*NET occupation-specific details/2025 wages/recruitment terms and adjacent definitions, BLS tables/OOH/OEWS methodology, NBS information technology industry, PIPL and scikit-learn methodology. ILO/NIST are limited broad-scope references; see records for unverified scope.
Usage limitations：Supports only metric system and report delivery tasks of business intelligence analysts (adjacent occupation). This occupation centers on building and validating models from raw data; do not incorporate metrics and reporting responsibilities into it.
Usage limitations：Supports only mathematical theory research tasks of mathematicians (adjacent occupation). This occupation centers on analysis and model delivery using business data; do not incorporate theoretical research duties into it.
Usage limitations：Supports only algorithm and system research tasks of computer and information research scientists (adjacent occupation). This occupation centers on end-to-end analysis and model delivery using business data; do not incorporate research duties into it.
Usage limitations：Current 34.6% for 2025–2035, rounded to OOH35; old facts retained, and current O*NET also uses 2025–2035.
Usage limitations：The retained old gov homepage does not support specific legal articles; the actual PIPL consulted is listed separately. This is not a complete industry legal review.
Usage limitations：Supports only data processing and statistical support tasks of statistical assistants (adjacent occupation). This occupation centers on independent modeling and validation; do not incorporate statistical support work into it.
Usage limitations：Supports only financial modeling and risk measurement tasks of financial quantitative analysts (adjacent occupation). This occupation spans broader business domains; do not incorporate financial pricing and risk duties into it.
Usage limitations：Supports only biomedical data analysis tasks of bioinformatics scientists (adjacent occupation). This occupation spans broader business domains; do not incorporate biomedical research duties into it.
Usage limitations：Supports only biological data organization and basic analysis tasks of bioinformatics technicians (adjacent occupation). This occupation centers on independent modeling and validation; do not incorporate technician duties into it.
Usage limitations：2025 broad industries 25/19/12/10; 34 is an editorial remainder, and 48/44/4 are opinions on needed education.
Usage limitations：Supports only optimization and decision modeling tasks of operations research analysts (adjacent occupation). This occupation more often handles prediction and explanation; do not incorporate optimization and decision recommendation duties into it.
Usage limitations：The interest questionnaire describes experiential preferences only; it does not replace ability, training, authorization or hiring judgments, nor predict work performance.
Usage limitations：Public wage statistics are published by industry categories and personnel categories, without an occupational detailed category corresponding to 15-2051.00. Industry or personnel-category averages are not wages for this occupation and cannot be extrapolated to nationwide occupational income. This page obtained no separately listed figures for this occupation; the gap is marked visibly on the page without filling it with industry averages.
Usage limitations：Voluntary governance, not mandatory law/methodology or occupational adoption rates.
Usage limitations：Provides occupational-code wage statistics portals and coverage explanations. Percentiles on this page are published via O*NET's national wage page (15-2051.00) for this detailed occupation itself. Percentiles are statistical positions, not starting pay or individual take-home income, and cannot be converted across markets into individual quotes.
Usage limitations：Supports tasks, work activities, work-context percentages, preparation level, interest codes and work styles for this occupation (15-2051.00). The scope covers this single occupation itself, without data on technology adoption, job replacement or income changes.
Usage limitations：2025 full-year US job-posting text lists 55 items: Python66/SQL51/R34/Tableau22/PowerBI19/AWS17/Azure13/TensorFlow11/PyTorch10/SAS9/Excel8/Snowflake8%. These are not actual usage, proficiency or GenAI rates; do not sum them or infer business uses from them.
Usage limitations：The old homepage is retained; actual NBS information technology industry context consulted in this round is listed separately.
Usage limitations：Wages include some incentive bonuses/allowances and cannot uniformly be described as excluding them; the methodology consulted is listed separately.
Usage limitations：Publishes annual percentiles of USD 67,240/85,660/120,230/158,880/199,130 for this occupation (hourly percentiles USD 32.33/41.18/57.80/76.39/95.74), using this detailed occupation's own scope. Percentiles are statistical positions, not skill levels; state or regional detail must be queried on that page by geography and period.
Usage limitations：Provides certification and training search portals for this specialization to check certificate and training information. Certification does not equal a licence, employer appointment, data access or production deployment authorization, and does not replace mainland China's recruitment, training or data authorization requirements.
Usage limitations：The abstract read in this round did not verify matching scores; this does not establish that the full study lacks occupation-specific scores.
Usage limitations：Supports tasks, work activities and work-context percentages for this occupation (15-2051.00). This page covers the single occupation itself and contains no technology adoption or job replacement data.
Usage limitations：Supports only statistical inference and experimental design tasks of statisticians (adjacent occupation). This occupation often targets engineering implementation and prediction; do not incorporate inference and experimental design duties into it.
Usage limitations：17 tasks: 15 Core, 1 Supplemental, 1 New without an importance rating. Survey design is Supplemental, interviewing New; not everyone must use them. 21 work-style Impact scores: detail 100/curiosity 100/innovation 81/dependability 79, not percentages. Education 48 bachelor's/44 master's/4 associate's reflects opinions on new hires' needs, not workforce composition; work context is not remote-work adoption or China working hours. Industry 2025 four categories 25/19/12/10, with 34 an editorial remainder.
Usage limitations：Annual 10/25/50/75/90 percentiles: USD 67240/85660/120230/158880/199130; hourly 32.33/41.18/57.80/76.39/95.74. Official annual values are cited directly, not annualized by this page. Covers national employees, not self-employed workers/individual after-tax income/China wages; percentiles do not indicate seniority or causal factor effects.
Usage limitations：55 items: Python66/SQL51/R34/Tableau22/PowerBI19/AWS17/Azure13/TensorFlow11/PyTorch10/SAS9/Excel8/Snowflake8%. Recruitment mentions are not usage/proficiency/GenAI adoption; Under1 is not 0, and InDemand is only a label.
Usage limitations：Occupation 15-2051 itself: 275.6→371.0 thousand, +95.4 thousand/+34.6%, annual average 24.8 thousand openings, self-employed 1.9%. OOH35% is rounded, not a different projection; O*NET7% and above is only a category. Typical bachelor's/no related experience/no on-the-job training are categories, not no preparation needed/China law.
Usage limitations：Office-based/mostly full-time; typical bachelor's in mathematics, statistics, computing, etc., with some employers requiring master's/doctoral degrees or industry experience. OOH detailed industries and O*NET broad categories use different levels and cannot be combined; reasons for growth explain projections, not observed shares caused by AI.
Usage limitations：Nonfarm full-/part-time employees; excludes self-employed workers, among others. Wages are pre-tax straight-time pay, including base/cost-of-living/guaranteed/hazard pay/incentive commissions and production bonuses/tips, and excluding overtime premiums, shift differentials and nonproduction bonuses, among others. Do not uniformly claim performance pay or allowances are excluded; this does not represent individual tax deductions/actual compensation.
Usage limitations：NIST voluntary risk governance guidance, not mandatory PIPL law/occupational qualifications; does not establish AI adoption, efficiency gains or causality for this occupation.
Usage limitations：ILO broad-scope task exposure methodology; the abstract read in this round did not verify matching 15-2051 scores. Do not claim the full study lacks scores or interpret exposure as layoff probabilities.
Usage limitations：Article 4 excludes anonymized information; Article 13 provides multiple processing grounds, not consent in all cases; Article 24 addresses transparency and fairness in automated decision-making using personal information, marketing choices/refusal, and rights to explanation/refusal of solely automated decisions significantly affecting rights and interests. Article 55 requires prior impact assessment, Article 56 requires reports and processing records for at least 3 years, not uniformly for every model; Article 51 concerns reasonable permissions/regular training. Industry, sensitive-information and cross-border supporting rules have not been fully checked.
Usage limitations：Information transmission, software and IT services: non-private 248752/private 128166 CNY average annual pre-tax wages; legal entities with 5 or more employees, excluding individual businesses and freelancers. Non-private does not mean only state-owned enterprises. Context only for employers actually in that industry, not all data scientists or an automatic match to the US information industry.
Usage limitations：Training with information unavailable at prediction time produces overly optimistic evaluation. Split first, fit preprocessing/feature selection only on training data, and do not use test data for model selection; Pipeline suits cross-validation. Randomness must be specified; this is not a nationwide mandatory occupational procedure or random splitting for all data.
Separate official statistics, laws, methodological/voluntary guidance and editorial suggestions. Gaps in China occupational wages and current recruitment samples remain; broad-scope context does not imply personal income or role permissions. Separate coverage of this US occupation's 2025 wages from 2025–2035 employment. OOH35% rounds 34.6%, and O*NET now uses the same period. China occupational wage/recruitment gaps remain; NBS information technology provides matching industry context only. 17 tasks: 15 Core, 1 Supplemental, 1 New. Opinions on needed education, 21 work-style Impact scores and 6 interest scores are not employee proportions/personality thresholds. Editorial tool and risk discussions do not replace measured adoption/incidents; separate voluntary NIST guidance and applicable PIPL articles. Exercises and development examples are editorial designs, not employment, qualifications or data/deployment authorization. Check specific roles' statistical methods, data permissions and industry conditions separately.
Personality preferences (EN) · Big Five traits (EN) · Understand my personality preferences · Understand my Big Five traits
| Reproducible models and code, validation and error analysis, bias and drift monitoring plans, and business-facing conclusions and limitations. |
| Adjacent fields may deliver metrics/systems, statistical results, optimization plans, algorithm prototypes, mathematical methods, and bioinformatics tools/molecular data research analyses. Check specific responsibilities and approval authority against the role; these are not universally exclusive outputs of each occupation. |
| Responsibility holders | Practitioners are responsible for the reliability of methods and validation and for the scope and uncertainty of conclusions. Authorized personnel and business stakeholders decide data usage permissions, whether models go live and business decisions; practitioners do not decide these unilaterally. | Adjacent fields bear corresponding responsibility for the metrics, statistics, optimization, computing research or bioinformatics work actually assigned. Job titles do not confer personnel budget authority, clinical sign-off authority or system deployment rights. |
|---|
| Collaboration and supervision | Collaborate within authorized permissions with business stakeholders, data engineering and IT, product and operations, compliance and legal teams, reporting to a business lead or technical supervisor. When data is unavailable or ownership unclear, target leakage or sample bias occurs, model performance is abnormal, or compliance boundaries are unclear, pause high-risk actions, verify facts, preserve records and escalate under established policies. | Editorial collaboration checks: actual division of work among business, collection, operational decision-making, research engineering and biological research teams. Bioinformatics scientists focus on molecular information and computational tools; a universal reporting line across employers has not been verified. |
|---|
| Limits on conclusions | Compare main purposes and outputs without treating prediction/inference/optimization/reporting as mutually exclusive responsibilities. Data science may propose management plans, but titles do not imply approval authority. Business intelligence may also analyze/test systems; do not claim it never validates models. | Preparation levels and practice requirements for adjacent roles do not apply to this occupation; check official information for each occupation and the requirements of the specific role item by item. |
|---|
Assessment four: is model validation sufficient?
Assessment five: do conclusions explain uncertainty?
Assessment six: have anomalies and boundaries been addressed?
| Ability to translate requirements, consistency of definitions and communication effectiveness. |
| A fictional requirement clarification and conclusion statement sample. |
| Focuses on business interpretation and communication; model complexity requirements are usually lower than in modeling roles. |
| Data engineering role | Independently build and maintain data pipelines, feature stores and data quality monitoring. | Engineering implementation, data traceability and quality monitoring. | A fictional pipeline design and quality monitoring record. | Focuses on data infrastructure and engineering quality; analytical conclusions are the responsibility of analysis roles. |
|---|
| Business intelligence analysts (adjacent track) | Develop metric systems and reports for business and operations to support day-to-day business decisions. | Metric definition design, report delivery and business understanding. | A fictional metric system and report description. | Business intelligence 15-2051.01 is another O*NET detailed occupation; queries/reporting/trends and system testing may overlap with data science. Do not claim it has no predictive model validation responsibility. |
|---|
Recruitment mentions do not indicate tool proficiency or actual uses; tool certificates do not replace lawful processing grounds/deployment appointments.
| Task submissions or trial-work data, and process retrospectives. |
| Task formats and standards are set by employers; this page makes no inference. |
| Trial work or probation observations | Procedural quality and initiative in handling requirement changes, result interpretation and collaborative communication in real projects. | Mentoring evaluations or probation records, and process retrospectives. | Assessment and employment conditions are set by employers; this page does not promise employment or authorize data use or publication. |
|---|
| Programming and data analysis abilities | People processing data and modeling | Affects passing technical assessments and working independently | Usually assessed against role requirements | Built through courses, projects and practice | Insufficient evidence: this round has no uniform cost/training duration; check actual learning and employer plans | Tool proficiency does not equal statistical methodological correctness or business understanding. |
|---|
| Data access and usage authorization | People handling personal information and restricted data | Affects the permitted data scope and compliance boundaries | For personal information, check lawful grounds/minimum necessity; calling it an entry threshold does not make it optional | For identifiable personal information in China, check lawful grounds under PIPL13 (not consent in all cases), minimum necessity under Article 6, and reasonable permissions/regular security education under Article 51; an employer account does not replace a lawful basis. Automated decision-making using personal information additionally falls under Article 24 on transparency, fairness and corresponding individual rights, Article 55 on prior impact assessment, and Article 56 requiring reports and records to be retained for at least 3 years. Do not apply these uniformly to all data science models. | Insufficient evidence: this round has no uniform cost/training duration; check actual learning and employer plans | Authorization has scope and duration; it is not a general qualification or legal advice. |
|---|
| Production deployment permissions | People needing to deploy models | Affects the ability to deploy and maintain models independently | Usually granted by the employer according to roles and procedures | Apply through the employer's deployment approval process | Insufficient evidence: this round has no uniform cost/training duration; check actual learning and employer plans | Deployment permissions do not mean model correctness or compliance approval. |
|---|
| Industry regulatory knowledge | People modeling in finance, healthcare and other fields | Affects solution feasibility and compliance approval rates | Not a universal entry requirement | Built through role training and projects | Insufficient evidence: this round has no uniform cost/training duration; check actual learning and employer plans | Domain knowledge does not equal regulatory authorization or a professional practice qualification. |
|---|
| Training certificates | People wishing to demonstrate methodological and tool abilities | May serve as supporting evidence of ability | No nationwide mandatory occupational certificate has been verified; do not uniformly promise that it is never a requirement for any role | Enroll according to the training provider's conditions | Insufficient evidence: this round has no uniform cost/training duration; check actual learning and employer plans | Training certificates do not equal data authorization, deployment permissions or a promise of employment. |
|---|
The reference deliverable lists required elements only; wording need not match the example if the judgment rationale is explained.
This exercise checks awareness of data availability and leakage assessment, minimum validation requirements, bias handling and data compliance only. It does not assess whether real workplace handling is compliant, nor constitute compliance, security or legal advice.
Do not use actual employer data or nonpublic business material. Real personal information and restricted data processing, model deployment and industry regulatory requirements must follow employer information management policies, data security and personal information protection rules, and industry compliance requirements; obtain corresponding authorization under current rules where needed.
Return to this page's assessment checklist and write three reasons to continue and two conditions requiring caution.
Typical duties：Turn analytical methods into reusable products or solutions and coordinate stakeholders to implement them.
Evidence for advancement：Able to balance business value, methodological reliability and implementation costs.
Qualification boundaries：Editorial checks: target role education/professional training, experience, actual duties and permissions/appointments. This round has no uniform China conditions or actual live recruitment sample; years of service do not guarantee automatic progression.
Next step：Learn requirement management and solution design.
Typical duties：Responsible for team objectives, staffing arrangements and building data analysis systems.
Evidence for advancement：Able to balance team outputs, methodological standards and compliance requirements.
Qualification boundaries：Editorial checks: target role education/professional training, experience, actual duties and permissions/appointments. This round has no uniform China conditions or actual live recruitment sample; years of service do not guarantee automatic progression.
Next step：Check relevant requirements for this track.
Typical duties：Build training and deployment platforms supporting model deployment, monitoring and version management.
Evidence for advancement：Able to balance platform stability, reproducibility and team efficiency.
Qualification boundaries：Editorial checks: target role education/professional training, experience, actual duties and permissions/appointments. This round has no uniform China conditions or actual live recruitment sample; years of service do not guarantee automatic progression.
Next step：Learn platform engineering and model operations methods.
Typical duties：Responsible for data standards, permissions and compliance governance, coordinating implementation across stakeholders.
Evidence for advancement：Able to balance governance requirements, business needs and execution costs.
Qualification boundaries：Editorial checks: target role education/professional training, experience, actual duties and permissions/appointments. This round has no uniform China conditions or actual live recruitment sample; years of service do not guarantee automatic progression.
Next step：Check this track's duties and appointment requirements.
Usage limitations：Editorial ratings use the existing occupational responsibilities and AI task analysis on this page. They are not adoption rates, measured effects, job-disappearance or personal-unemployment probabilities, or official scores from Stanford or another external institution. No measured time weights or future robot deployment are assumed.
