You need to agree to share your contact information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
This dataset is public but manually gated. It is deprecated for model training and ordinary property-prediction benchmarking. Access is intended for research on dataset curation, auditing, provenance tracing, and agent evaluation.
Log in or Sign Up to review the conditions and access this dataset content.
AM11 DED316L Texture Hardness — Agent Only Deprecated
This dataset contains 45 tabular records derived from DED 316L thin-wall texture/hardness experiments. It is uploaded under this separate name as an agent-only, deprecated version.
Do not use this dataset as a foundation-model training set or as a normal process-to-property benchmark. The converted data are too small for training, and the current property-prediction framing has ambiguous same-input/different-GT cases. This repo is intended only for agent tasks such as dataset audit, provenance checking, source-table lookup, and conversion-quality review.
Why Deprecated
- Only 45 records are available in the converted HF dataset.
- The source contains many thermal/EBSD files, but they are not reliably aligned to the scalar hardness/texture targets per record.
- Five process-height records have the same visible input but different GT because Wall A1 and Wall A2 are replicate walls with identical process parameters (
300 W,2750) and different measured hardness. - Single-value scoring is inappropriate unless the task explicitly includes provenance/wall identity or asks the agent to inspect the source CSV.
Records
- 25 records: process parameters + wall height -> average Vickers hardness HV (
Fig3.csv). - 7 records: cooling rate -> texture anisotropy Ahkl + average Vickers hardness HV (
Fig10.csv). - 13 records: cooling rate -> average Vickers hardness HV only, because Ahkl is missing in the source CSV.
All rows use task T-AM1; image and reasoning are null.
Known Ambiguity
The following rows have the same visible input but different answers. The wall value exists only in metadata, not in the rendered query/input.
| laser_power_W | scan_speed | height_within_wall_mm | Wall A1 annot | Wall A2 annot |
|---|---|---|---|---|
| 300 | 2750 | 1.0 | {"avg_hardness_HV": 203.33} |
{"avg_hardness_HV": 207.5} |
| 300 | 2750 | 3.0 | {"avg_hardness_HV": 205.83} |
{"avg_hardness_HV": 208.33} |
| 300 | 2750 | 5.0 | {"avg_hardness_HV": 208.33} |
{"avg_hardness_HV": 207.17} |
| 300 | 2750 | 7.0 | {"avg_hardness_HV": 211.83} |
{"avg_hardness_HV": 206.83} |
| 300 | 2750 | 9.0 | {"avg_hardness_HV": 209.67} |
{"avg_hardness_HV": 209.0} |
Suggested Agent-Only Uses
- Given source CSV rows, ask an agent to retrieve HV/Ahkl and cite the source row.
- Ask an agent to detect the A1/A2 repeated-input conflict and explain why the conversion is ambiguous.
- Ask an agent to judge whether the dataset is suitable for SFT, numeric regression evaluation, or only curation/audit.
Schema
| field | type | meaning |
|---|---|---|
query |
string | task text plus structured input values |
annot |
string | JSON target string |
reasoning |
null | no native chain-of-thought |
cate |
string | nearest category label |
task |
string | unified task ID (T-AM1) |
metadata |
string | JSON metadata, including source CSV, units, wall, and raw input |
image |
null | no input image |
Provenance and License
Underlying local source: 11_DED_316L_texture_hardness, associated with the DED 316L texture/hardness study. The package has no shipped LICENSE file; upstream terms must be verified before redistribution or downstream use. Access is therefore manual gated.
See stats.json, sample.jsonl, and audits/AM11_DED316L_texture_pre_audit.md for conversion/audit context.
- Downloads last month
- 10