neurodeveloper/real-workplace-occupational-telemetry

Real Workplace Occupational Telemetry («Del Diván al Dato»)

Real Workplace Occupational Telemetry is an open-access, clinically audited natural language processing dataset designed to benchmark, train, and evaluate occupational health telemetry, workplace friction, and psychosocial risk detection models.

Unlike synthetic or simulated datasets, this corpus is curated entirely from genuine real-world corporate communications extracted from the public record Annotated Enron Subject Line Corpus (AESLC / FERC), enriched with validated psychometric dimensions under the «Del Diván al Dato» paradigm engineered by Fabio Torres (Psychologist Specialist in Psychosocial Risk & M.Sc. in Data Engineering).

---

🚀 Quickstart (1 Line in Python)

You can load this dataset directly using the Hugging Face datasets library or pandas:

from datasets import load_dataset

# Load official dataset from Hugging Face Hub
ds = load_dataset("neurodeveloper/real-workplace-occupational-telemetry")

# Access the training split (1,000 real curated communications)
sample = ds["train"][0]
print(f"ID: {sample['id']}")
print(f"Subject: {sample['subject']}")
print(f"Burnout Exhaustion Score: {sample['maslach_exhaustion_score']}/100")
print(f"Psychological Safety (Amy Edmondson): {sample['edmondson_psychological_safety']}/100")
print(f"Karasek Quadrant: {sample['karasek_quadrant']}")
print(f"Resolution 2764 Domain: {sample['resolution_2764_domain']}")

Or using pandas and pyarrow:

import pandas as pd

df = pd.read_parquet("https://huggingface.co/datasets/neurodeveloper/real-workplace-occupational-telemetry/resolve/main/data/train.parquet")
print(df.head())

---

📊 Dataset Overview & Statistics

• Total Records: 1,000 genuine workplace email interactions.
• Languages: English (US corporate standard).
• Average Word Count: 136.4 words per interaction (focused, high-signal discursive units).
• PII Tokens Redacted: 1,669 sensitive entities masked (100% Zero-Leakage).
• Sampling Strategy: Balanced stratified sampling across all four Karasek Job Demand-Control quadrants (250 records each, 25.0% equal distribution).

Karasek Quadrant Distribution (Balanced)

Karasek Quadrant	Count	% of Dataset	Clinical Interpretation
high_strain	250	25.0%	High Demand + Low Control: Highest risk for psychosomatic symptoms and emotional exhaustion.
active	250	25.0%	High Demand + High Control: High challenge, learning opportunity, high decision latitude.
low_strain	250	25.0%	Low Demand + High Control: Low psychological strain, calm execution, high operational safety.
passive	250	25.0%	Low Demand + Low Control: Routine tasks, low initiative, risk of organizational disengagement.

Colombian Resolution 2764/2022 Intralaboral Domains

Psychosocial Domain	Count	Description
demandas_del_trabajo	476	Quantitative workload, accelerated pace, urgent deadlines, and time demands.
control_sobre_el_trabajo	286	Autonomy, initiative, decision authority, and participation in solutions.
recompensas_reconocimiento	150	Recognition, gratitude, feedback appreciation, and professional respect.
liderazgo_relaciones_sociales	88	Leadership behavior, directive commands, interpersonal friction, or social support.

Clinical Risk Tier Breakdown

Risk Tier	Count	Criteria
low	588	Stable indicators, positive psychological safety, and low friction.
moderate	285	Moderate urgency or mild interpersonal friction requiring monitoring.
high	105	Elevated exhaustion score (>= 50.0) or low psychological safety (<= 35.0).
critical	22	Acute burnout signals (>= 70.0) or severe interpersonal hostility (>= 65.0).

---

🔬 Clinical & Psychometric Framework

Every interaction has been scored using the deterministic psychometric engine from Workplace-Pulse-Telemetry:

1. Maslach Exhaustion Score (maslach_exhaustion_score):

   • Grounded in the Maslach Burnout Inventory (MBI) Emotional Exhaustion subscale.
   • Captures verbal expressions of cognitive depletion, chronic fatigue, cynicism, and pacing panic.
   • Continuous score: 0.0 (optimal stamina) to 100.0 (acute exhaustion).

2. Amy Edmondson Psychological Safety Index (edmondson_psychological_safety):

   • Grounded in Dr. Amy Edmondson's organizational research (Harvard Business School).
   • Measures interpersonal climate: openness to admit mistakes, asking questions, offering alternative solutions, vs fear of punitive backlash.
   • Continuous score: 0.0 (toxic/fear-based) to 100.0 (high psychological safety).

3. Interpersonal Friction Score (interpersonal_friction_score):

   • Identifies hostile language, defensive communication, passive-aggressive phrasing ("as per my last email", "not my job", "read carefully"), and finger-pointing.
   • Continuous score: 0.0 (harmonious) to 100.0 (destructive friction).

4. Karasek Job Demand-Control Model (karasek_quadrant):

   • Evaluates psychological demands (karasek_demand_score) against decision authority (karasek_autonomy_score).
   • Categorizes each interaction into high_strain, active, low_strain, or passive.

5. Resolution 2764/2022 Intralaboral Domains (resolution_2764_domain):

   • Aligned with Colombia's official psychosocial risk assessment battery (Ministerio del Trabajo).
   • Maps natural language interactions into one of the four regulated intralaboral macro-domains.

---

🛡️ Zero-Leakage Anonymization Protocol

To comply with international data privacy standards (GDPR, HIPAA, and Colombian Statutory Law 1581/2012):

• Corporate Email Addresses $\rightarrow$ [EMAIL_REDACTED]
• Telephone Numbers & Direct Extensions $\rightarrow$ [PHONE_REDACTED]
• Individual Names in Salutations & Sign-offs $\rightarrow$ [NAME_REDACTED]
• Financial Deals & Monetary Figures $\rightarrow$ [FINANCIAL_REDACTED]
• Internal Server Paths, IP Addresses & URLs $\rightarrow$ [URL_REDACTED] / [IP_REDACTED]

All anonymization was verified through automated negative regex sweeps to guarantee zero leakage of personal data.

---

📁 Dataset Schema & Field Descriptions

Column	Type	Description
id	string	Unique record identifier (enron_telemetry_0001 to enron_telemetry_1000).
source_corpus	string	Attribution to original open dataset (Annotated Enron Subject Line Corpus - FERC / Yale LILY).
subject	string	Anonymized email subject line.
body	string	Anonymized real workplace email body.
word_count	int64	Word count of the cleaned message body.
maslach_exhaustion_score	float64	Maslach emotional exhaustion intensity (0.0 to 100.0).
edmondson_psychological_safety	float64	Psychological safety index (0.0 to 100.0).
interpersonal_friction_score	float64	Interpersonal friction and passive aggression score (0.0 to 100.0).
karasek_demand_score	float64	Karasek quantitative and emotional demand score (0.0 to 100.0).
karasek_autonomy_score	float64	Karasek decision authority and autonomy score (0.0 to 100.0).
karasek_quadrant	string	Karasek quadrant: high_strain, active, low_strain, or passive.
resolution_2764_domain	string	Colombian psychosocial domain under Resolution 2764/2022.
clinical_risk_tier	string	Overall occupational health risk level: low, moderate, high, or critical.
detected_markers	list[string]	Clinical telemetry markers identified in the text.
pii_masked_count	int64	Total count of PII entities redacted in the interaction.

---

📚 Source Attribution & Academic Citation

The underlying textual corpus originates from the Annotated Enron Subject Line Corpus (AESLC), published by the Yale LILY Lab and derived from the public FERC Enron email collection.

@inproceedings{zhang2019email,
  title={This Email Could Have Been an Email: Subject Line Generation in Corporate Communications},
  author={Zhang, Rui and Tetreault, Joel},
  booktitle={Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL)},
  year={2019}
}

@dataset{torres2026realtelemetry,
  author       = {Fabio Torres},
  title        = {Real Workplace Occupational Telemetry (Del Diván al Dato)},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/neurodeveloper/real-workplace-occupational-telemetry}}
}

---

👨‍💻 Author & Scientific Stewardship

Fabio Torres (@neurodeveloper / neurodeveloper11)
Psychologist Specialist in Psychosocial Risk & M.Sc. in Data Engineering
Creator of the «Del Diván al Dato» Architectural Framework

• Hugging Face: @neurodeveloper
• GitHub: neurodeveloper11
• LinkedIn: Fabio Torres