| --- |
| license: apache-2.0 |
| language: |
| - en |
| tags: |
| - hr |
| - layoffs |
| - workforce |
| - employment |
| - global |
| - autoscientist |
| base_model: mistralai/Mixtral-8x7B-Instruct-v0.1 |
| --- |
| |
| # Global Layoffs & Workforce Interpreter |
|
|
| **Author:** Hussein Adeiza (mabera) |
| **Role:** Licensed Environmental Health Officer, Abuja Nigeria |
| **Base Model:** Mixtral 8x7B |
| **Fine-tuned with:** AutoScientist by Adaption Labs |
|
|
| ## Model Description |
| A LoRA adapter fine-tuned to interpret real global tech layoffs data, |
| producing structured analytical reasoning grounded entirely in |
| directly-downloaded statistics from a real tracked dataset (layoffs.fyi). |
| This submission completes all 7 categories of the AutoScientist |
| Challenge 2026 Part 2. |
|
|
| ## A Note on Subject Matter |
| This dataset addresses real workforce reductions affecting real |
| people's livelihoods. All interpretation stays measured and analytical, |
| including honest disclosure of data limitations (missing figures, |
| likely reporting bias) rather than overstating confidence in the |
| underlying figures. |
|
|
| ## Training Data |
| - Source: layoffs.fyi tracked data, accessed via a properly-cited |
| GitHub repository, downloaded directly, 2,361 real company entries |
| (2020-2023) |
| - Dataset: 5 original prompt-completion pairs, every number computed |
| via pandas directly from the raw source file, expanded via Adaptive |
| Data (no recipe modifications, full 20K+ datapoint expansion) |
| - Kaggle: https://www.kaggle.com/datasets/yunusahusseinadeiza/global-layoffs-and-workforce-interpreter |
|
|
| ## Training Metrics |
| - **Win rate (on dataset): 70% adapted vs 30% base model** |
| - Base model: mistralai/Mixtral-8x7B-Instruct-v0.1 |
| - Method: LoRA, no recipe modifications |
| - Dataset quality: 8.0 → 8.6 (+7.5% relative improvement, Grade B) |
| - Percentile: 31.5 |
| - Domain classification: Data-analysis-visualization (60%) / Market- |
| analysis (20%) / Corporate-business (20%), not a clean HR match; |
| disclosed here as this submission was intended for the HR category |
|
|
| ## Key Cited Findings (from the raw downloaded source only) |
| - Global tracked layoffs followed a non-monotonic trend: 81,068 in |
| 2020 (COVID shock), falling to 15,823 in 2021 (recovery), then |
| rising to 161,711 in 2022 and 127,277 in the first quarter of 2023 |
| alone |
| - Funding raised shows almost no correlation with layoff scale |
| (r = 0.077 across 1,491 companies with complete data), challenging |
| the intuitive assumption that better-funded companies cut more |
| - Post-IPO (mature, publicly-traded) companies account for 53.1% of |
| total tracked layoffs, likely reflecting company scale rather than |
| instability |
| - Roughly a third of entries (31.3% missing headcount, 33.2% missing |
| percentage) lack key figures, meaning aggregate totals are |
| conservative lower bounds, not precise counts |
| - The dataset's heavy US concentration (66% of the geographic total) |
| most likely reflects reporting bias in the underlying tracker toward |
| English-language tech media, not a true global distribution |
|
|
| ## Credits |
| Powered by Adaptive Data — Adaption Labs |
| AutoScientist Challenge 2026, Part 2 — HR Category |
