News / All News / Article

Data Governance Is Not About Patching Things Up, but Rethinking the Approach: Weiming Penguin’s StarFlow Module Redefines Real-Time Data Governance

Article images may contain Chinese text.





Anyone who has worked with data for 15 years knows this: even after writing more than a thousand cleansing rules, new anomalies keep appearing; even after round after round of manual checking, accuracy still fails to improve.


The problem is not the rules, but the paradigm.



1. The Old Paradigm: 3 Fatal Flaws in Rule-Based Systems




Fatal Flaw
Symptom
Root Cause
Rules cannot keep up with variations
New anomalies appear every day; rules cannot be written fast enough
Rules are reactive; anomalies keep emerging
Semantic understanding depends on people
Are “amoxicillin capsules” and “amoxicillin capsules 0.5g*24s” the same product? People know; rules do not
Rules only “match” characters; they do not “understand” data
One-dimensional validation
Checks whether the format is correct, not whether the logic makes sense
Each rule covers only one dimension, leaving contradictions across dimensions unchecked

Rule-based systems address “known errors” but are powerless against “unknown variations.”




2. The New Paradigm: Cognitive Intelligence × Multidimensional Decision-Making




StarFlow does not patch up the old system. It uses cognitive intelligence + multidimensional decision-making to rebuild the entire data governance paradigm.


🔹 Layer 1: Deep Semantic Understanding

Enable the system to “understand” data rather than “match” characters.

Engine
Capability
Result
Knowledge graph
A network linking drugs, companies, and indications
Automatically link “amoxicillin capsules” to all strengths, pack sizes, and dosage forms
Natural language understanding
Parse nonstandard fields and free text
“People’s Hospital of an Unnamed Province” = “XX Provincial People’s Hospital,” with no manual mapping required
Entity recognition
Extract drug names, company names, dosages, and dosage forms
Automatically split a single-line address into province, city, district, and street

Semantic understanding is the “first gate” in data cleansing. If it is not handled properly, even the most sophisticated downstream work has no solid foundation.


🔹 Layer 2: Multidimensional Cross-Validation

You cannot determine whether a data record is correct by looking at just one dimension.

Raw data → Quality inspection rules library + Integrated industry databases + Scenario application library → Consistency checks → Optimal result

Validation Dimension
Description
Example
Quality inspection rules
Format + logic + business requirements
Numeric ranges, field completeness
Integrated industry databases
15 years of pharmaceutical data accumulated by Weiming Penguin
Historical price benchmarks for a drug in a given province
Scenario applications
Validate in the context of the use case
Reasonableness check for “purchases > sales” in purchase, sales, and inventory data

Single-dimensional validation can produce results that “look correct but are logically wrong.” Only multidimensional cross-validation can ensure results are “both correct and reasonable.”


🔹 Layer 3: Support for Every Data Type

Data Type
Typical Sources
Governance Challenges
Purchase, sales, inventory, and distribution flow data
NKA/LKA distributors and chain headquarters
Inconsistent formats and missing fields
Consumer sales
Pyramid selling / retail outlets
Inconsistent granularity and nonuniform codes
New media
Douyin, Xiaohongshu, Kuaishou, Meituan, e-commerce, O2O
Different platform definitions and many nonstandard fields
Compliance
Regulatory platforms and tendering/procurement systems
Rigid formats and substantial differences in definitions

Different data types, one set of governance logic—that is what a “paradigm” means.




3. Tiangong Engine: A Continuously Evolving Technology Foundation




Dimension
Traditional Rule-Based Systems
StarFlow · Tiangong Engine
Cleansing method
Manually written rules + batch processing
Cognitive intelligence + real-time governance
Semantic understanding
Character matching
Knowledge graph + NLP + entity recognition
Validation dimensions
Single dimension
Multidimensional cross-validation
Anomaly handling
After-the-fact remediation
Proactive prevention + real-time interception
Supported data types
A separate rule for each type
One framework for all types
How it evolves
People write new rules
Engine self-learning + continuous upgrades



4. A Commitment




Faster—real-time  governance cleanses data as it arrives, without waiting for T+1


More  accurate—cognitive intelligence + multidimensional validation improve both the cleansing rate and accuracy


Broader coverage—purchase,  sales, and inventory; consumer; new media; and compliance data, covering all types


Less manual  work—from manually written rules to intelligent, autonomous governance




Final Thoughts




Data governance is not about patching things up, but rethinking the approach.


The ceiling of a rule-based system is not a shortage of rules, but the limits of the paradigm itself.


Cognitive intelligence + multidimensional decision-making do not patch up the old system—they replace the operating system.