# Entity Match Quality Review — Round 3 (2026-06-09)

## Context

This is the third round of third-party quality review for the WIMS entity-matching
pipeline. Previous rounds (v1, v2) corrected 347 false positives across the patent and
trademark sides and led to two code fixes (LEI `hasSharedDistinctiveToken` guard,
`&`→`amp` FUZZY_STOPWORDS entry).

This round focuses on:
- **Groups A1–A3**: Three *known* Wikidata false-positive clusters that re-appear on
  every fresh re-scan because the underlying `wikidata_relationships` records were never
  deleted. Confirm they are false positives so we can permanently remove the source records.
- **Groups B1–B7**: Newly identified suspicious Wikidata subsidiary matches — candidate
  false positives discovered after the re-scan, including two large clusters.
- **Group C**: Patent LEI subsidiary entries linking South-Africa-named companies to Nestlé,
  and one apparent person name.
- **Group D**: Patent fuzzy-direct matches at conf 0.75 that look suspicious.

## How to respond

For each pair, provide a verdict:
- **CORRECT** — the source entity is genuinely a subsidiary / acquired company of the target
- **FALSE_POSITIVE** — the match is wrong
- **UNCERTAIN** — not enough public information to decide

For FALSE_POSITIVE wikidata rows (Groups A–B), also indicate whether the
`wikidata_relationships` record itself should be **deleted** (to prevent re-appearing on
the next re-scan) or whether it may be a stale/edge-case claim worth keeping for reference.

---

## Group A1 — Eltek SpA/Srl → Delta Electronics, Inc. [wikidata subsidiary]

Context: Eltek SpA (IT) is an Italian automotive circuit board manufacturer.
Eltek AS (NO) is a Norwegian power supply company acquired by Delta Electronics.
Wikidata apparently conflates these two distinct "Eltek" entities via Delta Electronics.

| em.id | Source name | Target company |
|---|---|---|
| 3010232 | Eltek, Srl | Delta Electronics, Inc. |
| 3011896 | Eltek, Spa | Delta Electronics, Inc. |
| 3012981 | Eltek Srl | Delta Electronics, Inc. |
| 3013058 | Eltek Spa | Delta Electronics, Inc. |

---

## Group A2 — Merck Frosst Canada → Merck KGaA [wikidata subsidiary]

Context: Merck Frosst Canada is a subsidiary of *Merck & Co.* (US Merck / MSD), which
is a *completely separate* company from *Merck KGaA* (German Merck). The two Mercks have
operated independently since 1917 and are competitors in many markets.

| em.id | Source name | Target company |
|---|---|---|
| 3011875 | Merck Frosst Canada | Merck KGaA |
| 3011690 | Merck Frosst Canada Incorporated | Merck KGaA |
| 3013232 | Merck Frosst Canada Inc | Merck KGaA |
| 3016462 | Merck Frosst Canada Ltd | Merck KGaA |
| 3017067 | Merck Frosst, Canada Inc | Merck KGaA |

---

## Group A3 — Wilkinson Sword → Swedish Match AB [wikidata subsidiary]

Context: Wilkinson Sword was briefly owned by Swedish Match in the 1980s but has been
part of Gillette (now P&G) since 1993 and currently under Edgewell Personal Care. The
Wikidata claim is significantly outdated.

| em.id | Source name | Target company |
|---|---|---|
| 3009675 | Wilkinson Sword Co Ltd | Swedish Match AB |
| 3009676 | The Wilkinson Sword Company Ltd | Swedish Match AB |
| 3011232 | Wilkinson Sword Gmbh | Swedish Match AB |
| 3012956 | Wilkinson Sword Ltd | Swedish Match AB |
| 3011917 | Wilkinson Sword Lvd | Swedish Match AB |

(12 rows total; 5 representative rows shown — verdict applies to all 12.)

---

## Group B1 — Taiwan Semiconductor Manufacturing (TSMC) → ON Semiconductor Corporation [wikidata subsidiary]

Context: TSMC is the world's largest independent semiconductor foundry, headquartered in
Hsinchu, Taiwan. It is a publicly traded, independent company (NYSE: TSM). ON Semiconductor
(onsemi) is a separate US-based semiconductor company. These two companies are
*not* in a parent-subsidiary relationship of any kind.

There are 30 rows matching various TSMC name variants to ON Semiconductor. Representative sample:

| em.id | Source name | Target company |
|---|---|---|
| 3009261 | Taiwan Semiconductor Manufacturing Company Ltd | ON Semiconductor Corporation |
| 3010706 | Taiwan Semiconductor Manufacturing Co Ltd | ON Semiconductor Corporation |
| 3012780 | Taiwan Semiconductor Manufacturing Co | ON Semiconductor Corporation |

---

## Group B2 — Renesas Northern Japan Semiconductor Inc → ON Semiconductor Corporation [wikidata subsidiary]

Context: Renesas Electronics is a major Japanese semiconductor manufacturer formed from
the merger of Hitachi Semiconductor, Mitsubishi Electric Semiconductor, and NEC
Electronics. It is an independent, publicly traded company — *not* an ON Semiconductor
subsidiary.

| em.id | Source name | Target company |
|---|---|---|
| 3015691 | Renesas Northern Japan Semiconductor Inc | ON Semiconductor Corporation |

---

## Group B3 — Orion Corp / Orion Industries → Nestlé, S.A. [wikidata subsidiary]

Context: "Orion Corporation" is a Finnish pharmaceutical company (Helsinki Stock Exchange:
ORNBV), best known for manufacturing pharmaceuticals and diagnostics reagents. "Orion
Industries" is a US aerospace/defense contract manufacturing company. Neither has any
known corporate connection to Nestlé.

| em.id | Source name | Target company |
|---|---|---|
| 3008748 | Orion Corporation Ltd | Nestlé, S.A. |
| 3014925 | Orion Corporation | Nestlé, S.A. |
| 3014924 | Orion Corp | Nestlé, S.A. |
| 3017169 | Orion Industries | Nestlé, S.A. |
| 3016221 | Orion Industries Inc | Nestlé, S.A. |

---

## Group B4 — ETA SA → HP Tronic [wikidata subsidiary]

Context: ETA SA Manufacture Horlogère Suisse is the Swiss watch movement manufacturer
owned by the Swatch Group — one of the most important movement suppliers in the world.
HP Tronic is a Czech electronics company (retail + own-brand consumer electronics).
These two companies appear completely unrelated.

| em.id | Source name | Target company |
|---|---|---|
| 3013552 | Eta Sa | HP Tronic |
| 3015805 | Eta草图制造公司 | HP Tronic |

---

## Group B5 — Middleton → Altria Group, Inc. [wikidata subsidiary]

Context: "John Middleton Inc." (also "John Middleton Co.") is a real Altria subsidiary —
a US cigar company acquired by Altria in 2007. However, the patent database contains
several entries with "Middleton" that appear to be *person names* (patent inventors):
- "Middleton, John L." — comma-firstname format suggests inventor; however John L. Middleton
  was an Altria/Philip Morris executive, so this could go either way.
- "Middleton, Nigel John" / "Nigel John Middleton" — British-style full name, likely an inventor.

Please assess whether each represents the *John Middleton Inc.* company or a *person*.

| em.id | Source name | Target company |
|---|---|---|
| 3015054 | Middleton John C us | Altria Group, Inc. |
| 3010452 | Middleton John L us | Altria Group, Inc. |
| 3014319 | Middleton Nigel John gb | Altria Group, Inc. |

---

## Group B6 — John Knight / Knight, John → Unilever PLC [wikidata subsidiary]

Context: "John Knight Ltd" was a Victorian-era British soap manufacturer that merged with
Lever Brothers in 1897 and became part of the Unilever lineage — a legitimate historical
relationship. However, several entries look like *person names* filed as patent inventors
(e.g., "Knight, John B, Jr"; "Knight John gb"; "John V Knight").

| em.id | Source name | Target company |
|---|---|---|
| 3011538 | John Knight | Unilever PLC |
| 3011257 | Knight John gb | Unilever PLC |
| 3013103 | Knight, John B, Jr | Unilever PLC |

(8 rows total; 3 representative rows shown. Full ID list in appendix.)

Please assess for each: is this the *John Knight Ltd* soap company (CORRECT) or a person
inventor (FALSE_POSITIVE)?

---

## Group B7 — Alcon Inc / Alcon Ltd → Nestlé, S.A. [wikidata subsidiary]

Context: Alcon was acquired by Nestlé in 1977 and was a Nestlé subsidiary through the
1990s–2000s. Nestlé sold Alcon to Novartis in a deal completed in 2010–2011. Alcon
was subsequently spun off from Novartis as an independent company in 2019 (NYSE: ALC).
The match is historically accurate but reflects an ownership relationship that ended ~15
years ago. Should we treat this as CORRECT (historical record) or FALSE_POSITIVE (current
state)?

| em.id | Source name | Target company |
|---|---|---|
| 3008798 | Alcon Inc | Nestlé, S.A. |
| 3016775 | Alcon Ltd Inc | Nestlé, S.A. |

---

## Group C — Patent LEI subsidiary: South Africa cluster → Nestlé + person name

### C1: South Africa companies → Nestlé, S.A. [lei subsidiary]

Context: The LEI parent-walk matched these South-Africa-registered companies to Nestlé
via a GLEIF subsidiary. The `hasSharedDistinctiveToken` guard passed because "south" and
"africa" are shared geographic tokens not yet in `GENERIC_CORPORATE_TOKENS`. Companies
like Plessey (UK/SA electronics) and Behr (German automotive thermal management) have
no known Nestlé connection.

| em.id | Source name | Target company |
|---|---|---|
| 3011655 | Behr South Africa Pty Ltd | Nestlé, S.A. |
| 3016918 | Plessey South Africa Ltd | Nestlé, S.A. |
| 3016818 | Leer South Africa Ltd | Nestlé, S.A. |
| 3012544 | Gunson Seeds South Africa Pty | Nestlé, S.A. |
| 3009666 | Noel Lab South Africa Pty | Nestlé, S.A. |

Note: If these are FALSE_POSITIVE, it implies "south" and "africa" need to be added to
`GENERIC_CORPORATE_TOKENS` (same fix as "amp", "development" etc.).

### C2: Person name → Kenvue [lei subsidiary]

Context: McNeil Consumer Healthcare (now Kenvue) makes Tylenol, Motrin etc. — a
legitimate entity. "Mcneil Tom," appears to be a person name (Tom McNeil), not the
McNeil company.

| em.id | Source name | Target company |
|---|---|---|
| 3014598 | Mcneil Tom, | Kenvue, Inc. |

---

## Group D — Patent fuzzy direct: suspicious low-confidence matches (conf 0.75)

| em.id | Source name | Target company | Notes |
|---|---|---|---|
| 3014873 | Stout Medical Group Inc | Sotsu Medical Co., Ltd. | US spinal implants vs JP toy/media licensing |
| 3010753 | Fisher Scientific Company Llc | Inter Scientific Ltd. | Thermo Fisher sub vs UK testing lab |
| 3012872 | Kirin Machinery Kk | Shanghai Kuixing Packaging Machinery Co., Ltd. | JP vs CN, different products |
| 3009954 | Rush, Benjamin, M | Dr. Benjamin Rush | Person name vs company named after Founding Father |
| 3009953 | Rush, Benjamin, M, | Dr. Benjamin Rush | Same as above (variant with trailing comma) |
| 3011038 | Sensor Diagnostics Incorporated | Siemens Healthcare Diagnostics Inc. | Small startup vs Siemens division |
| 3016562 | Fuchs Dan, | Shenzhen Fuchs science and Technology Co., Ltd. | Reversed person name "Dan Fuchs" vs CN company |
| 3017005 | Hong Jiang | Chang Jiang Electronics Co., Ltd. | Person name or wrong geographic match |
| 3013971 | Hk Medical Tech Inc | KOR Medical, LLC | HK = Hong Kong; KOR = US sports medicine |
| 3009299 | Nitta Belting Co Ltd | Nitta Gelatin Inc. | Both Nitta-brand JP companies — are they related? |

---

## Knight group IDs (for Group B6)

Full list of Knight → Unilever entries for reference:

| em.id | Source name |
|---|---|
| 3011538 | John Knight |
| 3012567 | John R. Knight |
| 3014755 | John V Knight |
| 3011257 | Knight John gb |
| 3011537 | Knight John us |
| 3011258 | Knight, John |
| 3013103 | Knight, John B, Jr |
| 3013067 | Knight, John R |

---

## Summary table for verdicts

| Group | Pairs | Key question |
|---|---|---|
| A1 Eltek→Delta | 4 | FP — delete wikidata_relationship? |
| A2 Merck Frosst→Merck KGaA | 5 | FP — delete wikidata_relationship? |
| A3 Wilkinson→Swedish Match | 5 (of 12) | FP/stale — delete wikidata_relationship? |
| B1 TSMC→ON Semi | 3 (of 30) | FP — delete wikidata_relationship? |
| B2 Renesas→ON Semi | 1 | FP — delete wikidata_relationship? |
| B3 Orion→Nestlé | 5 | FP? — which "Orion" did Wikidata claim? |
| B4 ETA SA→HP Tronic | 2 | FP? — Swatch Group vs Czech electronics |
| B5 Middleton→Altria | 3 | Company or person name? |
| B6 Knight→Unilever | 3 | Historical company or person? |
| B7 Alcon→Nestlé | 2 | Historical CORRECT or current FP? |
| C1 South Africa→Nestlé | 5 | FP? — geographic token guard bypass? |
| C2 McNeil Tom→Kenvue | 1 | Person name FP |
| D fuzzy suspicious | 10 | Mix of FP and uncertain |
| **Total** | **49** | |