# Entity Match Quality Review — WIMS Pipeline (Round 2)
**Date:** 2026-06-09  
**Reviewer:** third-party AI (web interface)

---

## Context

A pipeline matched trademark holders and **patent applicants** (from patent filings) to
known company records in a tobacco/IP database. Each match has a `link_type`:

- **direct** — the applicant IS the company (same legal entity, name variant, abbreviation, or rebrand)
- **subsidiary** — the applicant is a subsidiary, brand, or division OF the company (parent relationship)

Pairs are grouped by the matching method used. For each pair, decide:

- **CORRECT** — the link is accurate and the link_type is right
- **INCORRECT** — wrong company, wrong link_type, or no real relationship
- **UNCERTAIN** — plausible but unverifiable without deeper research

Provide a one-line reason for any INCORRECT or UNCERTAIN verdict.

This is a second review batch. The focus this time is on the **patent applicant side**
(never reviewed before) and on verifying several recently-applied fixes to the TM side.

---

## GROUP A — Patent exact name match (deterministic, calibration)

These were matched by normalized name equality. Confidence 0.90 = cross-country match
(penalty applied when applicant country ≠ company country). Expected to be mostly correct.

| # | Patent Applicant | → Company | Link | Countries |
|---|---|---|---|---|
| A1 | Te Connectivity Corp | TE Connectivity Ltd. | direct | US→CH |
| A2 | Philip Morris Ab | Philip Morris International, Inc. | direct | SE→US |
| A3 | Sks Co Ltd | SKS Co., Ltd. | direct | (unknown)→CN |
| A4 | A & F Corp | A&F Corporation | direct | (unknown)→KR |

---

## GROUP B — Patent fuzzy match, high confidence (conf 0.90–0.95)

Matched by fuzzy string similarity. Confidence ≥0.90 = near-exact ratio (≥95 fuzzy score)
or same-country bonus applied. Lower false-positive risk than the 0.75 tier.

| # | Patent Applicant | → Company | Link | Countries |
|---|---|---|---|---|
| B1 | British American Tobacco (Investments) Limited | British American Tobacco Group | direct | GB→GB |
| B2 | Shenzhen L-Rider Tech Co Ltd | Shenzhen L-Rider Technology Co., Ltd. | direct | CN→CN |
| B3 | Buzzelet Development And Tech Ltd | BUZZELET DEVELOPMENT AND TECHNOLOGIES LTD. | direct | IL→IL |
| B4 | Chongqing China Tobacco Industry Co Ltd | China Tobacco Yunnan Industrial Co., Ltd. | direct | CN→CN |
| B5 | Morris, Phillip | Philip Morris International, Inc. | direct | (unknown)→US |
| B6 | Shenzhen Woody Vapes Tech Co Ltd | Dongguan Woody Vapes Technology Co., Ltd. | direct | CN→CN |

---

## GROUP C — Patent fuzzy match, medium confidence (conf 0.75)

The base fuzzy tier. Highest false-positive risk — matched on token overlap without strong
identity signal. Includes both clear correct and likely incorrect examples.

| # | Patent Applicant | → Company | Link | Countries |
|---|---|---|---|---|
| C1 | The Procter & Gamble Co Ltd | The Procter & Gamble Company | direct | US→US |
| C2 | Concept Matrix Solutions us | Concept Matrix Solutions, Inc. | direct | US→US |
| C3 | Samaung Sdi Co Ltd | Samsung SDI Co., Ltd. | direct | KR→KR |
| C4 | Plasma Processing Corporation | Advanced Plant Processing, LLC | direct | US→US |
| C5 | Heyi Intelligent Technology (Shenzhen) Co Ltd | HQ INTELLIGENT HEALTH PTY LTD | direct | CN→AU |
| C6 | Nmb Semiconductor Co Ltd | ITM Semiconductor Co., Ltd. | direct | KR→KR |
| C7 | Sarkar Pte Ltd | Starker International Pte. Ltd. | direct | SG→SG |

---

## GROUP D — Patent LEI subsidiary (GLEIF parent-walk, conf 0.88)

Matched via the GLEIF global registry: the applicant name fuzzy-matched (≥65% similarity)
a GLEIF-registered legal entity, which was then walked up the ownership chain to a known
company. All are `subsidiary` link_type.

| # | Patent Applicant | → Company | Link | Countries |
|---|---|---|---|---|
| D1 | Silicon Laboratories Finland Oy | Silicon Laboratories Inc. | subsidiary | FI→US |
| D2 | Infineon Technologies North American Corp | Infineon Technologies AG | subsidiary | US→DE |
| D3 | Rothmans, Benson & Hedges Inc Et Al | Philip Morris International, Inc. | subsidiary | CA→US |
| D4 | Deka Research & Development Corp | Nestlé, S.A. | subsidiary | US→CH |
| D5 | Yeda Research & Development Co Ltd | Nestlé, S.A. | subsidiary | IL→CH |

**Note:** D4 and D5 represent a cluster — there are 231 total patent applicants matched to
Nestlé via this method, many containing "Research & Development" in their name (e.g.
"Johnson Research & Development", "LG Electronic China Research & Development",
"International Research & Development Co Ltd"). Please assess whether the Nestlé match is
valid for D4 and D5 specifically, and whether the pattern looks like a systematic error.

---

## GROUP E — Patent Wikidata subsidiary (Wikidata P749/P355 corporate graph, conf 0.80)

Matched by walking the Wikidata corporate-ownership graph (P749 "parent organization" /
P355 "subsidiary"). These are 313 new matches not previously reviewed.

| # | Patent Applicant | → Company | Link | Countries |
|---|---|---|---|---|
| E1 | Sandoz Sa | Novartis AG | subsidiary | (unknown)→CH |
| E2 | Fairchild Semiconductor Suzhou Co Ltd | ON Semiconductor Corporation | subsidiary | CN→US |
| E3 | Amylin Pharmaceuticals Inc | AstraZeneca | subsidiary | US→GB |
| E4 | Eltek Spa | Delta Electronics, Inc. | subsidiary | IT→TW |
| E5 | Merck Frosst, Canada Inc | Merck KGaA | subsidiary | CA→DE |
| E6 | Wilkinson Sword Gmbh | Swedish Match AB | subsidiary | DE→SE |

---

## GROUP F — Patent DBA / alias match (conf 0.95)

Matched via "doing business as" or alternative trade names registered in the database.
High confidence — requires a registered alias hit.

| # | Patent Applicant | → Company | Link | Countries |
|---|---|---|---|---|
| F1 | Pepsi Co Inc | Pepsico Inc | direct | US→US |
| F2 | Vapor 4 Life Llc | Vapor4Life Inc. | direct | US→US |
| F3 | Jt 国际公司 | Japan Tobacco Inc. | direct | JP→JP |
| F4 | Shenzhen Zun Yi Pin Technology Co Ltd | Shenzhen KingZone (Tesla) Technology Co., Ltd. | direct | CN→CN |

---

## GROUP G — TM wikidata matches, link_type corrected to subsidiary

These trademark-holder rows originally had `link_type=direct` but were corrected to
`subsidiary` after a code fix (the pipeline was allowing the AI to override the corporate
graph signal). Please verify: (a) is the match itself correct, and (b) is `subsidiary` the
right link_type (i.e., the TM holder is a subsidiary/brand OF the company, not the same
legal entity)?

| # | Trademark Holder | → Company | Link | Countries |
|---|---|---|---|---|
| G1 | Asia Pacific Breweries Ltd | Heineken N.V. | subsidiary | SG→NL |
| G2 | Sandoz Ag. | Novartis AG | subsidiary | CH→CH |
| G3 | Heineken Ceska republika, a.s. | Heineken N.V. | subsidiary | CZ→NL |
| G4 | Meiji Seika Pharma Co., Ltd. | Meiji Holdings Co., Ltd. | subsidiary | JP→JP |
| G5 | John Middleton Co.(US) | Altria Group, Inc. | subsidiary | US→US |

---

## GROUP H — TM fuzzy survivors (ai_confirmed from pre-code-fix run)

These TM matches were confirmed by AI during an earlier pipeline run, before recent
false-positive guards were added (single-token stricter threshold, Indonesian "PT" suffix).
The new code would have suppressed some of these; they remain in the database from the
prior run. Please assess each — is the AI confirmation correct?

| # | Trademark Holder | → Company | Link | Countries |
|---|---|---|---|---|
| H1 | GRE Grand River Enterprises Deutschland GmbH | Grand River Enterprises (Deutschland) GmbH | direct | DE→DE |
| H2 | Ceed Vitality Group Ltd | Vitality CBD, Ltd. | direct | GB→GB |
| H3 | DNA Nutraceuticals | Nutraceuticals International LLC | direct | US→US |
| H4 | Lin Luo | Dongguan Lei Luo Technology Co., Ltd. | direct | CN→CN |
| H5 | PT. ZUNA GROUP INDONESIA | Pt You Indonesia, Ltd. | direct | ID→ID |

---

## Instructions summary

For each pair, return a table with columns: **#**, **Verdict** (CORRECT / INCORRECT /
UNCERTAIN), **Reason** (one line, only if not CORRECT).

Additionally, for **GROUP D**: after reviewing D4 and D5, please comment on whether the
231-match Nestlé cluster looks like a systematic false-positive caused by shared generic
tokens ("Research & Development") in the GLEIF registry.

Flag any other systematic patterns you notice across a group.