Matched
7
Data engineering ยท QA reconciliation
plan-vs-actual-recon
A reconciliation is trustworthy only when missing, extra, borderline, and ambiguous records stay visible instead of being forced into matches.
Public · synthetic demoThe problem
Picking the nearest record can hide the exact deviation a reviewer needs.
Plans and inspection logs rarely line up perfectly. A safe reconciliation converts units, applies asymmetric tolerances, and refuses to guess when more than one actual row matches.
Input
A structured plan and structured actual log.
Plan items: 9 Actual records: 10 Match keys: item_id first location second Inputs: JSON or CSV records No drawings or images
The money shot
Seven matches do not erase the rows that need a person.
Missing
1
Extra
1
Ambiguous
1
P-002 measured 2415 mm against 2400 mm with tolerance 10 mm, so its delta 15 is out of tolerance. P-003 used 7.5 mm of an 8 mm tolerance and stayed visible as borderline.
How it's verified
The report names every outcome instead of collapsing them into pass/fail.
| Status | Count | Example |
|---|---|---|
| within tolerance | 3 | P-001, P-007, P-008 |
| material match | 1 | P-006 |
| material mismatch | 1 | P-005 |
| out of tolerance | 1 | P-002 |
| borderline | 1 | P-003 |
Honest limitations
This checks recorded measurements, not the physical work.
- The tool does not read CAD, drawings, PDFs, or photos.
- It trusts every value in the actual log.
- Unknown or incompatible units stop comparison rather than being coerced.
- Multiple candidates stay ambiguous until a human resolves them.
Structured evidence is reconciled; physical correctness remains outside scope. jigonyoo.com · Back to hub