TechAPI dataset roadmap and status: all categories (1989-2026)

[Seungpyo1007]

Purpose

Roadmap and status board for the TechAPI dataset: every category from 1989 to 2026, published as a static JSON API at https://gettechapi.github.io/TechAPI/.

This issue stays open. The body is the current status, updated each release. Comments are one summary per release. Day-to-day work is tracked in the workstream issues below and on Project #2.

How work is tracked (since 2026-09-27)

Workstream	Issue	Which PRs link there
Roadmap & releases	#1 (this issue)	develop → main release PRs and the batched dump PRs
New records	#295	feat(data…): new CPUs/GPUs/SoCs/devices, benchmark backfill, new sources
Data accuracy	#296	fix(data…): corrections, duplicate merges, dates, placeholder cleanup
Engine, bot & automation	#297	ci/app PRs in TechAPI; TechEngine PRs reference GetTechAPI/TechAPI#297
Site	#19	Homepage and site PRs

pr-metadata adds the matching Closes #N line automatically. Auto-close is disabled, so trackers stay open. Project #2 has a Workstream field with the same split.

Current status (release #315, 2026-09-27)

Category	Records	Verified	Verified %	Notes
smartphone	93,407	17,581	18.8%	39,622 models; Wikipedia/Wikidata matching confirms only ~1.5% of Tier-3-only models
software	42,493	358	0.8%	Domain scoring since TechEngine #103; most records lack dates/developers even on Wikidata
website	40,084	7,635	19.0%	Domain scoring + Wikidata liveness since TechEngine #103
cpu	4,507	4,211	93.4%
tablet	3,459	288	8.3%
soc	2,117	280	13.2%	996 red = placeholder imports with no failing check (weak source only)
gpu	2,076	1,493	71.9%
laptop	1,995	105	5.3%	Domain scoring since TechEngine #102
monitor	882	0	0.0%	Domain scoring since TechEngine #102; no T1/T2 sources yet
watch	447	94	21.0%
brand	207	123	59.4%
pda	140	74	52.9%
devices (excl. software, website)	109,237	24,249	22.2%	Comparable to the pre-09-27 figure (20.8%)
all 12 categories	191,814	32,242	16.8%

Two figures are shown because software and website (82,577 records) entered the denominator on 2026-09-27. Their verification is a same-source check: the cited Wikidata item still exists and the record passes domain rules. Most software records are too sparse for green, even after filling from Wikidata. Verified records have only grown.

Goals

1. Verification: 30% of records in the scored categories (smartphone, tablet, watch, pda, cpu, gpu, soc, brand); laptop and monitor scoring is done (TechEngine data: add laptop category — 970 laptops (first non-mobile class) #102).
2. Coverage: keep new releases (2025-2026 phones, SoCs, CPUs, GPUs, laptops, watches) within one release of launch, when a citable source exists.
3. Accuracy: no known impossible values (storage-as-RAM, placeholder dates, estimate-only benchmarks); duplicate-free slugs.
4. Automation: every PR is checked on all 12 categories; weekly refresh and ingest run green.

Recent releases

Release	Date	Main content
#315	2026-09-27	8,839 verified promotions (first software), Wikidata-filled software/website fields, check-urls fixes
#308	2026-09-27	Dump with 312 accuracy fixes (storage, battery, monitor fields, SoC years), watch SoC links
#304	2026-09-27	Wearable SoCs, Galaxy Watch9 / Ultra 2, Pixel Watch 4, laptop/monitor scoring, 1,306 promotions
#292	2026-09-26	Galaxy Tab S11 / iPad Pro M5, 13 Surface laptops, benchmark backfill (~1,900 records), 182 SoC dates, GPU duplicate merge
#284	2026-09-25	2025-26 watches and phones, 83 Xeon thread counts, RAM and duplicate fixes
#274	2026-09-25	470 new CPUs/GPUs/SoCs/MacBooks, Intel and GPU cross-check fixes
#252	2026-09-25	SoC dedup, Commons images, tablet citations, 146 verified promotions
#249	2026-09-25	TechPicks image fixes, base_model_slug fill

Data rules (summary)

• Every record cites source_urls. Nothing is invented; if no source states a value, it stays null.
• Bulk imports land as verified: false. The verify layer promotes a record only when a live T1/T2 source agrees.
• Sources: Wikipedia, Wikidata, manufacturer pages and CC datasets are fine. GSMArena and PhoneDB may only be used to correct existing values (ADR-013). Namuwiki is banned. DeviceSpecifications is paused.
• PRs target develop. Before each release, develop gets one batched dump (site/public/v1); main only moves through release PRs.

History

This issue began on 2026-05-28 as the "Massive dataset rebuild" tracker for 8 categories (51,783 records at PR #36). The original description is preserved in the comment that announced this restructure.

[Seungpyo1007]

Rolled out in 11 commits on main

#	Commit	Subject
1	4dc4b57	chore(scripts): widen validator bounds and refresh seed loader
2	9708ee3	feat(app): expand GPU model and schema for richer hardware specs
3	7a9df9e	feat(data/brand): restructure into country folders and expand catalog
4	de9b32c	feat(data/cpu): expand AMD lineup from K5 to Zen 6 with segment split
5	b08e2cf	feat(data/cpu): expand Intel lineup from i486 to Panther Lake with segment split
6	9781670	feat(data/cpu): add IBM POWER, z mainframe, PowerPC, Cell, Motorola G3/G4
7	5cd4b5b	feat(data/cpu): add historic and niche vendors across 11 manufacturers
8	2a434bf	feat(data/gpu): rebuild discrete GPU catalog across 10 vendors with segment split
9	fc7bcd1	feat(data/smartphone): rebuild smartphone catalog across major brands
10	54306c8	feat(data/soc): rebuild SoC catalog covering Apple, Qualcomm, MediaTek, Samsung, Google, HiSilicon
11	f64a74a	docs: update README and data/README for expanded dataset and new layout

(The site homepage update was already on main as ef15e00 feat(site): add interactive API playground and redesign homepage, so the rebase deduplicated it.)

Result

• ✅ python -m scripts.validate PASS across brand / cpu / gpu / smartphone / soc
• 464 CPU records, 330 GPU records, ~366 smartphone records, ~100+ SoC records, 129 brand records
• New layouts: brand/<country>/<slug>.json + cpu/<mfr>/<year>/<consumer|enterprise>/<slug>.json
• Tracked in Project: TechAPI Dataset Rebuild