Trace column-level lineage
The outcome: traces every number at column level from source to report, through views and manual steps.
Builds on: Describe the catalogue entries.
Which AI model is best at trace column-level lineage?
| Model | Provider | Tier | Test score | Knobits mastered (expected) | Cost per test run |
|---|---|---|---|---|---|
| gemini-4-argon | Flagship | – | 89% | $0.021 | |
| gemini-3.5-flash-lite | Low cost | – | 73% | $0.003 | |
| gpt-6-luna | OpenAI | Low cost | – | 81% | $0.002 |
| claude-haiku-5-5 | Anthropic | Low cost | – | 88% | $0.006 |
Test score: mean rubric score on the test tasks for this activity. Knobits mastered: expected share of the activity's knobits the model masters. Cost: mean API cost of one test run at list prices, half price for answers sent as a batch. Method and caveats.
The 11 knobits of this activity
0 that AI does well, 0 it does partly, and 11 where people still lead.
| ID | Knobit | Type | AI exposure (models mastering) |
|---|---|---|---|
| M-501 | Recognises a broken lineage hop | Skill | People still lead 0/4 |
| M-502 | Recognises lineage ending at a view | Skill | People still lead 0/4 |
| M-503 | Recognises a manual step missing from a lineage path | Skill | People still lead 0/4 |
| M-504 | Recognises a lineage edge with placeholder logic | Skill | People still lead 0/4 |
| M-505 | Recognises an integration lineage edge without its mapping | Skill | People still lead 0/4 |
| M-506 | Recognises a table-level lineage edge where column level is needed | Skill | People still lead 0/4 |
| M-551 | Explains data lineage as a graph of derivations | Knowledge | People still lead 0/4 |
| M-552 | Explains lineage granularity | Knowledge | People still lead 0/4 |
| M-553 | Explains view resolution in lineage | Knowledge | People still lead 0/4 |
| M-554 | Explains BCBS 239 requirements on lineage and manual steps | Knowledge | People still lead 0/4 |
| M-555 | Explains source-to-target mappings as lineage edges | Knowledge | People still lead 0/4 |