# Patent substantivity Phase 0 probe

This is a retrieval/span feasibility artifact only. It does not merge into Poucher, train a model, push to Hugging Face, or bump a version.

## Query Gate

- `("substantivity" OR "tenacity" OR "longevity") fragrance (C11B9 OR A61Q13) (hours OR score OR panel)`
- `("fragrance intensity" OR "odor intensity") (skin OR fabric OR hair) (hours OR "after application") (C11B9 OR A61Q13)`
- `("retention" OR "persistence") perfume compound (hours OR table) (C11B9 OR C11D)`

## Counts

- Sample patents fetched: 10
- Retrieval status: ok
- Patents with numeric keyword spans: 5
- Numeric keyword spans: 41
- High-attention manual-review spans: 2
- Raw numeric span yield per 100 patents: 410.0
- High-attention review span yield per 100 patents: 20.0

## Construct Inventory

- `substantivity`: 13
- `longevity`: 8
- `fragrance_intensity`: 8
- `unknown_numeric_measure`: 3
- `tenacity`: 3
- `headspace_concentration_or_intensity`: 2
- `odor_intensity`: 2
- `persistence`: 1
- `duration`: 1

## Substrate Inventory

- `unspecified`: 35
- `skin`: 5
- `headspace`: 2
- `hair`: 2
- `fabric`: 1

## Resolution

- Spans with printed CAS references: 1/41
- Spans with a printed compound reference or CAS: 3/41

## Poucher Alignability

Not established by this automated span probe. The constructs found here are method-specific patent constructs; merge eligibility still requires resolved overlapping molecules and an empirical correlation against `data/poucher_substantivity.jsonl`.

## GO/NO-GO

NO-GO for full build until manual/LLM review converts spans into true single-molecule measured tuples and confirms construct overlap; this probe validates query retrieval and estimates raw span yield only