Chainticks/labor-demand-index

Chainticks Labor Demand Index

An agent-friendly labor-demand panel for finding where organizations are still trying to hire humans for work that may be automatable.

This dataset intentionally publishes aggregates and official public-domain series only:

• official_labor_timeseries: BLS JOLTS monthly openings, hires, quits, layoffs/discharges, and separations by US sector (latest revision).
• jolts_sector_metrics: derived vacancy/hire, quit/hire, separation/hire, and YoY openings metrics from JOLTS.
• jolts_release_vintages: append-only snapshots of JOLTS values as observed on each collection date, partitioned by vintage_date, so revision history is preserved (first-release vs. latest).
• sampled_posting_aggregates: small, rate-limited JobSpy query aggregates by role family, geography, and snapshot date. Includes posting-age signals (new_postings_count, median_posting_age_days, stale_share_30d/60d/90d) derived from a private first-seen ledger, and salary/workflow/compliance texture.
• labor_pressure_index: the combined index — one auditable score per role family x geo x day that fuses automation fit, workflow digitization, posting persistence (can't-fill signal), and confidence-weighted JOLTS sector tightness. This is the join of the demand and tightness halves of the panel, materialized. Carries oews_annual_median_wage as a ground-truth labor-cost anchor.
• oews_wages: BLS OEWS national annual mean/median wages for the tracked SOC occupations (public domain; the ground-truth labor cost behind each role family).
• role_taxonomy: the tracked automatable role families, query terms, skill terms, and automation-fit labels.
• query_taxonomy: one row per tracked search query, so query panels can evolve without changing downstream schemas.
• occupation_bridge: candidate SOC, NAICS, and JOLTS-sector bridges for each tracked workflow.
• jolts_sector_taxonomy: the tracked JOLTS sector codes used by the collector.
• geo_taxonomy: the stable country/metro panel used for rotating sampled posting collection.

Raw job postings, job URLs, company names, emails, and descriptions are not included.

import pandas as pd

repo = "Chainticks/labor-demand-index"
date = "YYYY-MM-DD"
url = f"https://huggingface.co/datasets/{repo}/resolve/main/sampled_posting_aggregates/date={date}/part-0000.parquet"
df = pd.read_parquet(url)
print(df.head())

Use Cases

• Track sampled demand for automatable back-office roles, including how long requisitions stay open (stale_share_30d/60d/90d) — the clearest "can't hire a human" signal.
• Join live posting aggregates to official JOLTS sector tightness via the jolts_sector_candidates_json bridge.
• Nowcast monthly JOLTS from daily postings, and check whether postings led the revised number using jolts_release_vintages.
• Analyze posting-level workflow texture through aggregate software, workflow, and compliance term counts.
• Find role families where hiring demand, turnover, posting age, and automation fit are all high.
• Build daily research agents without storing raw scraped job-board content.

Limitations

sampled_posting_aggregates is a directional sample, not a job-posting census. Job boards rate-limit and change markup. Source coverage can shift by day, site, geography, and query wording. Treat counts as comparable only within the same query/source configuration and use BLS rows as the official macro anchor. A pinned core panel is sampled every day for stable per-cell time series; the long tail rotates.

Layout

official_labor_timeseries/date=YYYY-MM-DD/part-0000.parquet
jolts_sector_metrics/date=YYYY-MM-DD/part-0000.parquet
jolts_release_vintages/date=YYYY-MM-DD/part-0000.parquet
sampled_posting_aggregates/date=YYYY-MM-DD/part-0000.parquet
labor_pressure_index/date=YYYY-MM-DD/part-0000.parquet
oews_wages/date=YYYY-MM-DD/part-0000.parquet
role_taxonomy/date=YYYY-MM-DD/part-0000.parquet
query_taxonomy/date=YYYY-MM-DD/part-0000.parquet
occupation_bridge/date=YYYY-MM-DD/part-0000.parquet
jolts_sector_taxonomy/date=YYYY-MM-DD/part-0000.parquet
geo_taxonomy/date=YYYY-MM-DD/part-0000.parquet
_schema.json
_manifest.json
LATEST_DATE.txt

Provenance

Allowed source_kind values are ['derived', 'public_domain', 'sampled_public_web_aggregate']. Raw scraped postings are private collection artifacts and must not be uploaded to this public dataset.