investor-behavior-dataset

Self-produced structured data from the investor_behavior_analysis research (Finalsystems, part 2 of the startup full-lifecycle decision toolchain). It profiles global investors' pre- and post-investment behavior around semiconductor (chip) startups, reconstructed from free public sources and organized into entity / event / chain / company tables.

Contents

Config Files Grain Row est.
entity investors_entity_v0.3 + investors_provenance_v0.2 1 investor per row ~1,915
events funding / exposure / interface (clean_human v0.2, identity_patch v1) 1 event per row funding ~1.4K / exposure ~3.3K / interface ~3.2K
chains chains_chip_v0.3 (clean_human v0.2, identity_patch v1) 1 funding-level outcome chain per row ~1,342
companies companies_entity_v0.2 + category_v0.2 + edges_long_v0.1 + associated_* 1 company per row ~43,474

Data dictionary & machine contracts

Each table ships a human-readable schema spec (*.spec.md) and, where a programmatic contract exists, a machine JSON Schema (*.schema.json) under schemas/. These are authoritative for column semantics, keys, enums and null policy.

  • investors_entity_v0.3 โ€” 1,915 investors (frozen v0.2 989-row base + 926 interface-recovery entities), 22 columns, stable IDs (wikidata:<QID> / interface_recovery:<hex>)
  • chains_chip_v0.3 โ€” one temporal chain per funding-level company outcome; schema_version v0.3.0; full Arrow schema incl. exposure/interface event ids in-window

Scope

  • Universe: semiconductor / chip startups (2,278 Crunchbase seed universe) + global investors across 9 types (IB / PE / AM / HF / VC / Pension / CVC / SWF / Angel).
  • Time: coverage through 2026 (static snapshot; see provenance for collection dates per source).

Data quality & limitations

  • Self-produced analysis fields from public sources (SEC EDGAR, Wikidata CC0, company/investor sites, news via Tavily, AKShare, etc.).
  • Contains no redistributable third-party raw rows โ€” only aggregate / derived / self-produced fields.
  • Known coverage ceilings per investor type (e.g. SWF ~50%, interface ~70% paid-source); per_vc_amount_usd ~16% populated; descriptive, not causal or predictive.

Source

Reconstructed from free public data. See the accompanying research report (investor_behavior_analysis/_REPORT.md) for full methodology and source provenance.

Related

  • Companion skills repo: startup-decision-skills (event-claim-resolution, event-chain-builder, entity-dedup, ...).
  • Companion paper (arXiv/SSRN) covering both parts of the toolchain.
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support