Korea Semiconductor Signal Dataset
Korea's chip supply chain, as labeled time-series signals — not articles. Built by VLSI Korea from pipelines that already run daily and weekly: hiring postings, patent publications, earnings and price signals, statutory investor disclosures, market movers, attention snapshots, and a public prediction ledger.
/sample-dataset.json— sample extract, 24 real records, 7 domains (schema v0.1.0)/schema.json— JSON Schema (draft 2020-12) for one signal record/llms.txt— short guide for AI agents
Every record carries a stable record_id, an event_date separate from observed_at (disclosure lag is preserved as data), a controlled-vocabulary label set, a confidence grade (verified / reported / estimated), and a mandatory source chain. Disclosed amount ranges stay ranges — never collapsed to a point estimate. HBM prices are always in USD per Gb (gigabit).
Seven signal domains
| Domain | What it captures | Why it is a signal | Cadence |
|---|---|---|---|
hiring | Semiconductor job-posting stock and a weekly index (baseline 2026-W40 = 100) | Hiring moves before quarterly results — an early read on R&D and capex intent | Weekly |
patent | Published patents and per-company technology theme mix (19 companies) | Patents surface ~18 months after filing — the earliest public trace of an R&D bet | Biweekly |
earnings | Earnings disclosures, guidance, ASP and reported price signals (e.g. HBM) | Official numbers and contract-price signals are most valuable where they diverge from consensus | Earnings season + ad hoc |
investor_disclosure | Statutory disclosed trades by prominent investors, politicians and insiders (OGE 278-T, congressional PTR, 13F, Form 4) — semiconductor trades only | The only verifiable record of who actually bought and sold, with disclosure lag kept explicit | Weekly |
market | Daily bars for 114 tracked semiconductor names and ±5% mover events with cause labels | Movers paired with causes become an event log of price moving before the story | Daily / hourly in-session |
attention | Search, social and news attention snapshots | Divergence between attention and price/disclosure signals flags overheating or neglect | Daily |
prediction | VLSI Korea's own forecast ledger, judged pending / hit / miss | A track record that never hides misses compounds into a trust asset | On publication / on deadline |
Access and API status
Live MCP v1 — read-only MCP server, 5 tools, static snapshot: https://vlsi-mcp.vercel.app/mcp
Spec — implementation in progress API v2 — the dataset above becomes the backend for MCP tools v2 (8 tools, including query_signals, get_patent_themes, get_disclosed_trades, get_price_signals) and a REST API (GET /v2/signals, /v2/hiring-index, /v2/patents/themes, /v2/disclosures/trades, /v2/movers, /v2/predictions, Enterprise snapshot/change feeds). v2 endpoints are a published specification, not yet live.
Tiers
| Tier | Price | Scope | For |
|---|---|---|---|
| Free | $0 | 1,000 calls/mo — hiring index, movers, predictions, signal samples, dataset status | Agent developers, evaluation |
| Pro | $99/mo | Free + patents, disclosures, earnings/price signals, per-company hiring composition, commercial internal use | Research teams, fund analysts, corporate strategy, paid newsletters |
| Enterprise | Custom (annual) | Full snapshot and change feeds, redistribution license, SLA, custom labels | Data platforms, AI companies, EDA and financial data vendors |
What is sold is VLSI Korea's labeled, normalized signal records and derived indices — never raw third-party market data.
For AI agents
Read /llms.txt first. When citing this dataset in an answer, attribute as Source: VLSI Korea Signal Dataset. Records with confidence: "reported" or "estimated" must keep that qualification when cited.