Subspace continuously observes company career boards and hiring infrastructure over time, turning public records into verifiable hiring-behavior data — ghost-job rates, stuck roles, hiring posture — with methodology and confidence labels, not AI guesses. Every metric on a Subspace company trace is labeled observed (read directly from a primary source), inferred (computed from observed inputs by deterministic logic), or estimated (model output carrying explicit uncertainty — reserved for future use; no metric is estimated as of this writing). The table below lists every metric Subspace publishes, its exact source, and its label.
For the snapshot-reproducibility contract that lets you cite a Trace URL in an IC memo and have it resolve to the same JSON forever, see Methodology · Verification.
Number of active job postings on the company's ATS-powered career page.
Source: ATS API where available (Greenhouse, Lever, Ashby, Workday Career Sites, Oracle HCM); HTML scrape fallback for boards without an API. Sampled slice capped per scan; ATS-reported total preferred when surfaced.
Which applicant-tracking system the company uses for its public career page.
Source: Fingerprint regex set matching board HTML, JS bundles, and canonical-URL patterns. Multi-source evidence reconciliation with confidence weighting.
The customer-identifier path component in the ATS URL (e.g. `cloudflare` in `boards.greenhouse.io/cloudflare`).
Source: Direct parse of the canonical career-page URL. Drift between slug and apex domain is itself a moat-relevant signal — flags acquisitions, rebrands, international variants.
Share of postings exhibiting ghost-job patterns: long-open with no engagement, reposted identical JDs, vague evergreen titles, mismatched job-vs-company signals.
Source: Composite score over per-job indicators. Each job carries its own ghost score; the index is the corpus rollup.
Net direction of open-role count over rolling windows; rendered as a qualitative story in v0 (e.g. "rapid expansion", "selective hiring").
Source: Time-series of `job_signal_snapshots` weekly counts. Bucketed by week, slope-fitted, qualitative label assigned by threshold bands.
Whether we're confident the company is actively hiring (`confident_jobs`), confirmed not hiring (`confident_zero`), only present in talent-pool sources (`talent_pool_only`), or uncertain (`unconfident`).
Source: Engine rollup over ATS-board status, job counts, scrape outcome history, and enrichment-source coverage.
Press mentions of the company within the most recent 90 days. Funding events flagged separately.
Source: RSS ingest from TechCrunch Venture, Crunchbase News, Tech.eu, and other reputable feeds. Entity-resolved against company domain via `mentioned_companies` JSONB.
Disclosed funding rounds, M&A announcements, IPO filings.
Source: News articles with `is_funding_event=true` flag (RSS-based detection) + SEC EDGAR RSS for public-company filings + SEC Form 4 for insider transactions.
Legal entity status from the Delaware Division of Corporations (most US-headquartered private companies incorporate here).
Source: Direct query against the public Delaware corporations registry. Captures entity name, file number, active/forfeited/dissolved status.
Worker-adjustment-and-retraining-notification filings — state-level mass-layoff disclosures required by US federal law for >100-person events.
Source: Per-state Department of Labor public WARN portals. Indexed back 2 years.
Multi-year arc of company size, rendered as a line chart.
Source: Two complementary sources: live ATS API per-job posting dates (Greenhouse `first_published`, Lever `createdAt`, Ashby `publishedAt`) bucketed by week — survivorship-biased but precise on shape; and Wayback Machine archive scrapes of the careers page. Wayback is broken on JS-rendered SPAs (Stripe, Coinbase, Google) — never published as an absolute number.
Corporate ownership relationships — which parent owns this subsidiary, which subsidiaries belong to this parent.
Source: Hiring.cafe `companyInfo.parent_company` and `subsidiaries[]` payload. Names resolved to canonical domains where possible.