Public sample is ready.
Specification-driven public demonstration
This POC converts a public basic-skirt Tech Pack into a different public Excel template. It exposes every mapping decision, preserves where each value came from, and sends uncertainty to review.
Try the conversion
The adapter deliberately supports one published source specification. Uploading an unrelated file will fail clearly rather than produce a plausible-looking result.
Public sample is ready.
Waiting to process.
Conversion evidence
| Source evidence | Destination field | Decision | Why |
|---|---|---|---|
| WAIST14 in | Waist width · S | High · mapped | Direct semantic and unit match. |
| CB LENGTH14 in | Center back length · S | High · mapped | Known alias: CB means center back. |
| SWEEP18 in | Hem width · S | Medium · review | Likely equivalent, but the destination’s half-width convention must be confirmed. |
| HIP at 7 in17 inLOW HIP at 9 in17.5 in | Hip width at 8 in · S | Low · not mapped | The reference position differs. The POC does not interpolate a production measurement. |
| FABRIC A52% poly / 48% acetate, black | BOM · Shell fabric | High · mapped | Material name normalized; missing quantity and supplier remain blank. |
| Pages 2–3Visual construction content | Construction notes | Manual review | Drawing callouts and image-only annotations cannot be recovered from the text layer. |
Before and after
The comparison is about the handling of evidence, not a claim that every conversion becomes fully automatic.
The conversion specification
This POC is deterministic. The adapter declares what it accepts, what it can map, and when it must stop.
Expected metadata labels, measurement rows, three-page structure and file fingerprint.
Source terminology is normalized without losing the original label or value.
Cells and fields are populated only where the destination meaning is compatible.
Unmapped, ambiguous and visual-only content appears in the conversion report.
What the next POC must prove
The next step is to run the same pipeline on two anonymized incoming Tech Packs and one real downstream template from a buying house. That is where we test coverage, exception rates, review time and operational trust.