Datasetpublic

huyxdang/adaption-market-analysis-final

This dataset is a remastered version prepared using Adaption's Adaptive Data platform. adaption-market-analysis-final This dataset contains prompt/completion pairs for updating dated market-research notes using only supplied evidence. Each entry includes specific financial data points, such as historical actuals or analyst forecasts, and a corresponding response that strictly separates observed results from forward-looking claims. The completions emphasize causal limitations… See the full description on the dataset page: https://huggingface.co/datasets/huyxdang/adaption-market-analysis-final.

sourceHugging Faceupdated 2mo agoView on Hugging Face

0likes26downloads

Dataset Card

This dataset is a remastered version prepared using Adaption's Adaptive Data platform.

adaption-market-analysis-final

This dataset contains prompt/completion pairs for updating dated market-research notes using only supplied evidence. Each entry includes specific financial data points, such as historical actuals or analyst forecasts, and a corresponding response that strictly separates observed results from forward-looking claims. The completions emphasize causal limitations, preserve counterevidence, and avoid providing executable trade instructions or personalized advice.

Dataset size

There are 45,758 data points in this dataset. This is an instruction tuning dataset.

Quality of Remastered Dataset

The final quality is A, with a relative quality improvement of 17.5% (original text 8.0, adaptive text 9.4). Percentile improved from 15.8 to 49.5.

Domain

—Market-analysis (100%)

Language

—English (100%)

Composition

Built with the Adaptive Data Combine feature from three generations over the same source corpus:

Source	Rows	Recipe
adaption-market-analysis-1	15,252	completion-only enhancement
adaption-market-analysis-2	15,253	prompt + completion, response constraints
adaption-market-analysis-3	15,197	prompt + completion, response constraints
Combined	45,758	45,702 from the three inputs, plus 56 rows

Schema

Combine normalizes its inputs to four columns. The task, asset, relationship, and leakage-group metadata carried by the three source datasets is not retained here; use the linked source datasets for per-task-family analysis or grouped splits.

Field	Description
`original_prompt`	Source instruction with system framing, as-of date, and evidence
`original_completion`	Source analyst-style response
`enhanced_prompt`	Adaptive Data rewrite of the prompt (null in 15,253 rows, 33.3%)
`enhanced_completion`	Adaptive Data rewrite of the completion (null in 56 rows)

Notes

—56 rows have no enhanced_completion; filter them before training on that column.
—The three source generations cover the same underlying items, so the combined set contains roughly 15,250 unique source items at three enhancement variants each.