Humanly Not Understandable Tables and Columns
Description
This recipe identifies column and tables names from the schemas across your data ecosystem that are not easily understandable. It identifies columns and tables with very short names, numeric-only names, absence of vowels, technical names and numeric/underscore-heavy patterns and also explains the reason for the identification
Details
Cryptic names (short tokens, hashes, numeric-only fields) reduce discoverability, cause analysts to choose wrong objects, lengthen onboarding, and increase stewardship effort. This recipe flags unreadable metadata using heuristics, entropy scoring, and AI-assisted judgement to generate governance-ready remediation outputs.
Business value
- Improves data discoverability and reduces analyst time lost to cryptic objects.
- Enables automated naming standardisation and prioritised remediation.
- Reduces manual review hours and speeds up governance and migration projects.
- Increases trust in the catalog by separating meaningful names from junk.
- Targets engineering fixes to ingestion sources producing unreadable metadata.
Who benefits
- Data Governance Teams
- Data Stewards & Catalog Managers
- Data Architects
- Platform Engineering Teams
- BI & Analytics Owners
What you receive
How the recipe works (guided flow)
Step 1 — Read: Ingest catalog metadata
Read tables from oetable, columns from oecolumn, join with oeschema and connectioninfo to gather full context.
Step 2 — Compute: Rule-based flagging
Apply rules (too short, no vowels, numeric-only, high-entropy tokens, repeated hashes) to identify candidate unreadable names.
Step 3 — Compute: Whitelist & false-positive reduction
Use curated whitelist and pattern exceptions to filter valid short business tokens and reduce noise.
Step 4 — Compute: AI-assisted readability judgement
Run AI checks to confirm whether flagged names convey business meaning or are junk; attach a Readability verdict and confidence score.
Step 5 — Compute: Classify reasons and severity
Assign explicit reason codes (auto-generated, numeric-hash, too-short, acronyms) and severity levels for remediation prioritisation.
Step 6 — Output: Produce governance-ready exports
Generate humanly_not_understandable_columns, humanly_not_understandable_tables, and a prioritized remediation dashboard for stewards.
Sample insights (illustrative)
| Insight Category | What the recipe discovered | Business Impact |
|---|---|---|
| Unreadable column share | 12% of catalog columns flagged as too short or vowel-less. | Prioritise cleanup to reduce analyst selection errors and improve search results. |
| Top violating schemas | Three schemas contain the majority of underscore-heavy and numeric names—likely ingestion pipelines to review. | Focus engineering fixes on those ingestion sources to reduce future junk creation. |
| AI correction suggestion rate | AI judged ~30% of short names as valid business acronyms; remaining flagged for rename. | Reduces false-positive workload for stewards and speeds up remediation cadence. |
Make sure the following ingredients are available in your workspace:
- Connected datasets are crawled and profiled
- Columns have distinct counts and top values generated