Enrichment
Fill the gaps you can prove, show the ones you can’t.
Enrichment adds missing specifications from appropriate sources — with the document, location and method recorded for every value, and conflicts kept for review.
Delivered in scoped assessments
The problem
Why “enriched” data
is often worse
Filling every blank is easy. Filling blanks with values someone can defend is the actual work.
- 01
Values without sources
If you cannot see where a number came from, you cannot correct it, defend it to a customer or update it when the source changes.
- 02
Conflicts resolved by accident
Two sources disagree; whichever was processed last wins. The disagreement — often the most useful signal — disappears.
- 03
Scope confusion
A marketplace seller’s country of origin gets written onto the manufacturer product, where it does not belong.
Worked example
One attribute,
fully accounted for
An attribute assertion from the synthetic fixture: the value, its scope, its sources and its review state.
| Field | Assertion | Source | Location | Method | Outcome |
|---|---|---|---|---|---|
| bore_diameter | 25 mm | Illustrative datasheet | Row 1, “Bore” | source_extraction | populated |
| dynamic_load_rating | — vs — | Datasheet rev. A / rev. B | Table 2 / Table 3 | source_extraction | conflicting · 2 assertions |
| total_width | — | No source located | — | — | unknown |
| country_of_origin | — | Not a product-level fact | — | — | lot-scoped · not researched |
The conflicting load rating keeps both assertions. A review policy selects an accepted value and records why; the other remains in history.
Method
Source hierarchy
per attribute
One appropriate manufacturer source can be enough. More cross-checking is used when context, conflict or impact demand it.
Agree sources and rights
Which documents may be used, and for what — internal use, redistribution or neither.
Target applicable fields
Only research attributes that apply to the family; mark the rest not applicable.
Extract with location
Record document, revision, page or row, original label and unit for each value.
Keep conflicts
Disagreeing supported values are preserved as separate assertions.
Report gaps
Unknowns stay unknown, with what was searched — not filled with estimates.
Outcome and limits
What you get — and what you don’t
Missing attributes filled where evidence supports them, every value traceable, conflicts and gaps explicitly reported.
| Detail | |
|---|---|
| Included | Per-attribute source, location and method |
| Included | Conflicting values preserved for review |
| Included | Explicit unknown / not-researched states |
| Included | Scope separation: product, listing, lot |
| Not included | Values estimated from similar products |
| Not included | Guaranteed completeness of every attribute |
| Not included | Redistribution of sources without permission |
Delivery options
How it is delivered today
Related
Keep reading
What does “confidence” mean in your outputs?
A method-specific estimate of how well a value is supported. It is not a guaranteed probability of correctness; calibration against expert-reviewed samples is separate work.
Do you require two sources for every value?
No. Source requirements are attribute-specific. A single appropriate manufacturer source is often sufficient; we cross-check more when there is conflict or high impact.
Which attributes are missing?
Let’s see what can be sourced.
Tell us the category, the attributes you need and the sources you trust. We will scope what can be supported with evidence.