# Entity Match Quality Review — Round 4 (2026-06-09)

## Context

Fourth round of quality review for the WIMS entity-matching pipeline. Rounds 1–3
removed 437 false positives and led to five code fixes. Two additional rows (Orion
Industries Ltd, Orion Industries Ltd variants) and eight Middleton person-name
variants were deleted between rounds as extensions of already-confirmed Round 3
verdicts.

This round focuses on four newly identified patterns:

- **Group A**: A systematic *single-initial person-name* pattern — patent inventors
  listed as "Surname X" or "Surname X Y" (e.g. "Atkins D", "Garrison John M")
  are being fuzzy-matched to companies whose name contains that surname. The
  existing `isPersonNameCommaFormat` guard catches the "Lastname, Firstname"
  format but misses initial-only suffixes and multi-word no-comma variants.
  Confirming the pattern will justify a code fix covering all ~97 instances at once.
- **Group B**: Chinese Shenzhen tech/semiconductor companies fuzzy-matched to
  *multiple different* unrelated targets simultaneously — a clear sign the match is
  driven by shared generic tokens ("semiconductor", "intelligent", "tech") rather
  than entity identity. 'Semiconductor' and 'shenzhen' were added to
  `GENERIC_CORPORATE_TOKENS` in Round 3; these rows appear to have been created
  before that fix and were not cleared.
- **Group C**: Cross-sector and cross-geography mismatches — unrelated companies
  sharing a single distinctive word (chemistry, medical, garden) or abbreviation.
- **Group D**: Disambiguation and legitimacy questions — three cases requiring
  domain knowledge to adjudicate.

## How to respond

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

For Group A, a single verdict covering the entire pattern is acceptable — the code fix
will batch-delete all instances matching the pattern, not just the sampled rows.

---

## Group A — Single-initial person-name → company (fuzzy direct, conf 0.75)

Context: These source names consist of a surname plus one or two initials (e.g.
"Atkins D", "Gilbert V"), which is the standard EPO/USPTO format for listing a
patent inventor's last name and first initial when the applicant is an individual.
The fuzzy matcher scores them highly against companies whose names begin with the
same surname. The `isPersonNameCommaFormat` guard (added in Round 2) only catches
the "Lastname, Firstname" comma format — it does not catch these initial-only variants.

### A1: Atkins → ATKINS NUTRITIONALS, INC.

Context: Atkins Nutritionals is the diet/nutrition company founded by Dr. Robert Atkins.
"Atkins D", "Atkins J", "Atkins L", "Atkins S" are single-initial patent inventor entries.

| em.id | Source name | Target company |
|---|---|---|
| 3008290 | Atkins D | ATKINS NUTRITIONALS, INC. |
| 3012810 | Atkins J | ATKINS NUTRITIONALS, INC. |
| 3012811 | Atkins L | ATKINS NUTRITIONALS, INC. |
| 3012645 | Atkins S | ATKINS NUTRITIONALS, INC. |

---

### A2: Nelson → A Nelson & Co Limited

Context: A. Nelson & Co. is a UK homeopathic medicine manufacturer (founded 1860, now
Nelson's Homeopathics). "Nelson J" and "Nelson N" are single-initial inventor entries.

| em.id | Source name | Target company |
|---|---|---|
| 3017163 | Nelson J | A Nelson & Co Limited |
| 3014499 | Nelson N | A Nelson & Co Limited |

---

### A3: Bergstrom / Chiaro / Gilbert / Kennedy → eponymous companies

Context: Each company is named after its founder, causing inventor entries with that
surname to score highly. Same pattern as A1/A2.

| em.id | Source name | Target company |
|---|---|---|
| 3012592 | Bergstrom A | Bergstrom Innovations, LLC |
| 3015277 | Chiaro J | Chiaro Technology Limited |
| 3015278 | Chiaro S | Chiaro Technology Limited |
| 3013267 | Gilbert B | Gilbert Technologies B.V. |
| 3014771 | Gilbert V | Gilbert Technologies B.V. |
| 3013768 | Kennedy J | Kennedy Enterprises, Inc |
| 3011328 | Kennedy M | Kennedy Enterprises, Inc |

---

### A4: Garrison cluster → John Garrison Capital LLC

Context: "John Garrison Capital LLC" is a small US holding company named after its
founder John Garrison. Multiple name formats — single initial, full name + middle
initial, comma-format with extra initials — all match to it. The comma-format variants
(Harris, John, G; Harrison, John, M) have 2 commas and were not caught by the existing
guard (which requires exactly one comma).

| em.id | Source name | Target company |
|---|---|---|
| 3012122 | Garrison J | John Garrison Capital LLC |
| 3010508 | Garrison John M | John Garrison Capital LLC |
| 3016853 | Harris, John, G, | John Garrison Capital LLC |
| 3015391 | Harrison John M | John Garrison Capital LLC |
| 3015811 | Harrison, John, M | John Garrison Capital LLC |

---

### A5: Forrest, Kevin, M → Kevin Forest Kft.

Context: "Kevin Forest Kft." is a Hungarian LLC named after a person (Kevin Forest).
"Forrest, Kevin, M," is a comma-format patent inventor entry with a middle initial and
a different spelling ("Forrest" vs. "Forest"). Is this the company owner filing a patent
under his own name, or a different person named Kevin Forrest?

| em.id | Source name | Target company |
|---|---|---|
| 3014241 | Forrest, Kevin, M, | Kevin Forest Kft. |
| 3014240 | Forrest, Kevin, M, | Kevin Forest, Kft. |

---

## Group B — Shenzhen tech companies matching multiple unrelated targets (fuzzy direct, conf 0.75)

Context: Each source entity below matches 3–4 completely different target companies
simultaneously. A source that legitimately corresponds to a target would produce one
strong match; multiple conflicting matches at the same confidence indicate the match
is driven by shared generic tokens ("semiconductor", "technology", "intelligent") not
entity identity. Note: "semiconductor" and "shenzhen" were added to
`GENERIC_CORPORATE_TOKENS` in Round 3 — these rows pre-date that fix.

### B1: Shenzhen Dongke Semiconductor Co Ltd (3 targets)

| em.id | Source name | Target company |
|---|---|---|
| 3006808 | Shenzhen Dongke Semiconductor Co Ltd | ON Semiconductor Corporation |
| 3006809 | Shenzhen Dongke Semiconductor Co Ltd | Diotec Semiconductor AG |
| 3006810 | Shenzhen Dongke Semiconductor Co Ltd | Dialog Semiconductor PLC |

---

### B2: Shenzhen Kia Semiconductor Tech Co Ltd (4 targets, two name variants)

| em.id | Source name | Target company |
|---|---|---|
| 3010111 | Shenzhen Kia Semiconductor Tech Co Ltd | ON Semiconductor Corporation |
| 3010112 | Shenzhen Kia Semiconductor Tech Co Ltd | Dialog Semiconductor PLC |
| 3010113 | Shenzhen Kia Semiconductor Tech Co Ltd | ITM Semiconductor Co., Ltd. |
| 3010114 | Shenzhen Kia Semiconductor Tech Co Ltd | Alpha & Omega Semiconductor Inc. |
| 3008615 | Shenzhen Kia Semiconductor Technology Co Ltd | ON Semiconductor Corporation |
| 3008616 | Shenzhen Kia Semiconductor Technology Co Ltd | Dialog Semiconductor PLC |
| 3008617 | Shenzhen Kia Semiconductor Technology Co Ltd | ITM Semiconductor Co., Ltd. |
| 3008618 | Shenzhen Kia Semiconductor Technology Co Ltd | Alpha & Omega Semiconductor Inc. |

---

### B3: Shenzhen Tete Semiconductor Co Ltd (4 targets)

| em.id | Source name | Target company |
|---|---|---|
| 3006838 | Shenzhen Tete Semiconductor Co Ltd | Torex Semiconductor Ltd. |
| 3006839 | Shenzhen Tete Semiconductor Co Ltd | Semtech Semiconductor Holdings Limited |
| 3006840 | Shenzhen Tete Semiconductor Co Ltd | Semtech Semiconductor (Shanghai) Co., Ltd. |
| 3006841 | Shenzhen Tete Semiconductor Co Ltd | Semtech Semiconductor (Shenzhen) Co., Ltd. |

---

## Group C — Cross-sector and cross-geography mismatches (fuzzy direct, conf 0.75)

### C1: LEI path anomaly — Chicago Board Of Trade → Colin Mear Engineering Ltd.

Context: Chicago Board of Trade (now part of CME Group) is one of the world's largest
futures and options exchanges, based in Chicago. Colin Mear Engineering Ltd. is a UK
mechanical engineering company. This match came through the LEI parent-walk path, not
fuzzy — likely a GLEIF data collision. These two organizations have no conceivable
connection.

| em.id | Source name | Target company | Method |
|---|---|---|---|
| 3010016 | Chicago Board Of Trade | Colin Mear Engineering Ltd. | lei |

---

### C2: Biotech geography mismatch

| em.id | Source name | Target company | Notes |
|---|---|---|---|
| 3006939 | Oisin Biotechnologies Inc | Shenzhen Oplus Biotechnologies Ltd. | US longevity biotech vs. CN optics company |

---

### C3: Medical device geography mismatch

Context: AO Medical Products AB is a Swedish orthopedic company (AO = Arbeitsgemeinschaft
für Osteosynthesefragen). KOR Medical LLC is a US sports medicine company. "se" suffix
= Sweden country code appended to source name in the patent database.

| em.id | Source name | Target company |
|---|---|---|
| 3006945 | Ao Medical Products Ab | KOR Medical, LLC |
| 3006943 | Ao Medical Products Ab se | KOR Medical, LLC |

---

### C4: Chemistry/pharma name collisions

| em.id | Source name | Target company | Notes |
|---|---|---|---|
| 3006786 | Wuhan Phoenix Chem Co Ltd | E-Phoenix SÀRL | Wuhan chemistry company vs. Swiss holding |
| 3006958 | Baf Aktiengesellschaft | Hanf Aktiengesellschaft | German AG names sharing "aktiengesellschaft" + one token |
| 3006952 | Mc Medical Kk | CBD Medical Solutions, Inc. | Japanese medical KK vs. US CBD company |

---

### C5: Geography / business type mismatch

| em.id | Source name | Target company | Notes |
|---|---|---|---|
| 3006951 | Green Garden Inc | Shenzhen Green Green Technology Co., Ltd. | US garden retailer vs. CN tech |
| 3006956 | Winchester Dev Associates | Winchester MD Ltd. | Real estate developer vs. medical device company |
| 3006968 | Spink Benjamin, | Spa Benjamin LLC | Person name (Benjamin Spink) vs. small LLC |

---

## Group D — Disambiguation and legitimacy questions

### D1: Panasonic IP Management Corporation → Panasonic Corporation [fuzzy direct, conf 0.75]

Context: Panasonic IP Management Corporation is a real Panasonic subsidiary created
specifically to hold and manage Panasonic's patent portfolio. It files many patent
applications in its own name. The match method is `fuzzy direct` — should it be
CORRECT (and then the link_type would be `subsidiary`, not `direct`)?

| em.id | Source name | Target company |
|---|---|---|
| 3006899 | Panasonic Ip Management Corporation | Panasonic Corporation |
| 3006900 | Panasonic Ip Management Corporation | Panasonic Corporation |

Note: same match appears twice (two patent records). If CORRECT, both are valid.

---

### D2: Merck Canada Inc → Merck KGaA [wikidata subsidiary, conf 0.80]

Context: There are *two* "Merck Canada" entities:
1. **Merck Canada Inc.** (subsidiary of Merck & Co. / US Merck / MSD) — the prescription
   drug business in Canada, headquartered in Kirkland, Québec.
2. **EMD Inc.** / previously "Merck Frosst Canada" — historically the German Merck KGaA
   presence in Canada.

Wikidata has "Merck (Canada)" (Q30338225) linked to Merck KGaA (Q156959). If the patent
applicants named "Merck Canada Inc" are actually US Merck entities, this is a false
positive — the same Merck disambiguation issue as Round 3 A2. If they are German Merck
KGaA's Canadian entity, this could be CORRECT.

| em.id | Source name | Target company |
|---|---|---|
| 3017233 | Merck Canada Inc ca | Merck KGaA |
| 3017234 | Merck Canada Inc, | Merck KGaA |
| 3017235 | Merck Canada Inc | Merck KGaA |

---

### D3: Jiangsu Highstar Battery Mfg Co Ltd → Jiangsu Highstar Battery Manufacturing Co., Ltd. [fuzzy direct, conf 0.75]

Context: "Mfg" is a standard abbreviation for "Manufacturing". Source and target appear
to be the same company — this looks like a CORRECT match where the source name uses
an abbreviation. If so, confidence 0.75 seems low for what is effectively an exact
name match.

| em.id | Source name | Target company |
|---|---|---|
| 3006870 | Jiangsu Highstar Battery Mfg Co Ltd | Jiangsu Highstar Battery Manufacturing Co., Ltd. |

---

## Summary table

| Group | Pairs | Key question |
|---|---|---|
| A1 Atkins → Atkins Nutritionals | 4 | Person initials or company? |
| A2 Nelson → A Nelson & Co | 2 | Person initials or company? |
| A3 Bergstrom/Chiaro/Gilbert/Kennedy | 7 | Person initials or company? |
| A4 Garrison cluster | 5 | All person names? |
| A5 Forrest Kevin M → Kevin Forest Kft | 2 | Same person as owner, or different? |
| B1 Shenzhen Dongke → 3 targets | 3 | All FP (multi-match)? |
| B2 Shenzhen Kia → 4 targets ×2 | 8 | All FP (multi-match)? |
| B3 Shenzhen Tete → 4 targets | 4 | All FP (multi-match)? |
| C1 Chicago Board of Trade → Colin Mear | 1 | FP via LEI path? |
| C2 Oisin Biotech → Shenzhen Oplus | 1 | FP cross-geography? |
| C3 AO Medical → KOR Medical | 2 | FP cross-sector? |
| C4 Chemistry/pharma collisions | 3 | FP? |
| C5 Geography/business mismatches | 3 | FP? |
| D1 Panasonic IP Mgmt → Panasonic | 2 | CORRECT subsidiary or FP? |
| D2 Merck Canada Inc → Merck KGaA | 3 | Disambiguation: which Merck? |
| D3 Jiangsu Highstar Mfg → Manufacturing | 1 | CORRECT same-company? |
| **Total** | **51** | |