Classify Wikidata brand references as Tier 2 sources

[Seungpyo1007]

Summary

Wikidata entity pages were unclassified (host tier 0), even though Wikidata is a secondary structured knowledge base. Add wikidata.org to Tier 2 and cover its host classification and effect on a fully populated brand record. This changes source classification only; the completeness, consistency, provenance, score threshold, and live-source promotion gates stay intact.

Wikidata describes itself as a collaborative secondary knowledge base: https://www.wikidata.org/wiki/Wikidata:Introduction/simple

Near-miss analysis

Regenerated 83,847 unverified Tier-0 scores using TECHAPI_DATA_DIR=C:\Users\29\Desktop\TechAPI\data and python -m app.verify score --unverified-only. The requested pool (yellow, best_host_tier <= 2, 65 <= score < 75) contains 2,757 records, 2,698 without flags. The <= 2 filter includes tier 0, which means unclassified hosts.

Category (count)	Completeness: points × records	Consistency	Host	Provenance
smartphone (2,057)	3.1×12, 6.2×13, 9.4×13, 12.5×1,964, 15.6×50, 18.8×4, 25×1	23.3×3, 26.2×3, 28×49, 35×2,002	3×1, 18×2,040, 26×16	7×2,051, 8×6
soc (438)	0×239, 4.2×195, 8.3×4	35×438	18×112, 26×221, 30×105	7×25, 8×413
cpu (113)	10.7×62, 14.3×51	35×113	18×112, 22×1	7×113
brand (101)	25×101	35×101	3×101	7×101
tablet (44)	0×1, 10×43	35×44	18×43, 26×1	7×44
watch (3)	8.3×1, 16.7×2	23.3×2, 26.2×1	18×2, 26×1	7×2, 8×1
gpu (1)	17.9×1	23.3×1	26×1	7×1

The dominant 1,963-smartphone joint cohort scores 12.5 + 35 + 18 + 7 = 72.5. All 1,963 lack cameras, wired charging power, and OS version; 1,962 also lack storage options. The records have OS family in os, but that does not supply the absent version, and the missing camera/storage/charging data are real gaps in these records. I left those deductions unchanged. The 101-brand cohort has both richness fields and full consistency but gets only 3 host points because its cited hosts are unclassified. Sixty-eight of those records cite Wikidata item pages. Other low completeness cohorts contain actual missing normalized fields (for example SoC CPU configuration and CPU socket, TDP, or L3 cache), so they were not changed.

Measured effect

Re-scoring brand records against the pre-change scores cache yields 68 requested near-miss records changing yellow → green (70.0 → 85.0). Thirty more brands outside that pool change yellow → green (73.0 → 85.0) because they cite Wikidata alongside a Tier-3 host. Total band changes: 98 yellow → green; no other brand band changes. These are offline green candidates; promotion still requires the existing live-source or cross-reference decision.

Checks

• python -m pytest tests/verify --basetemp C:\Users\29\AppData\Local\Temp\techengine-verify-all-pytest — 145 passed
• python -m mypy app — passed
• python -m ruff check app tests — passed
• git diff --check — passed

No TechAPI data JSON was edited. The pre-existing local TechAPI submodule change was left untouched.