To search this category programmatically: GET https://api.databazaar.io/datasets?category=scientific
Full API docs: https://api.databazaar.io/llms.txt
Browse / scientific
Scientific & Research Datasets
48 listingsResearch-grade data is hard to find outside institutional silos — this category surfaces lab results, clinical measurements, and structured research observations that sellers are licensed to distribute. Provenance and collection methodology matter most here, and both are documented on each listing. Samples are free, so you can assess rigor before spending budget.
Warnings That Proved Right: 88 Researched Cases of Technology and Product Risk, Scored
Updated Oct 20, 2026
88 historical cases where a new technology or product met a warning, researched and scored: the claim, who warned, who dismissed it, whether the incumbent had a stake, the verdict (fear right or mixed), years to verdict, cost of the error and the lesson, with 258 mappings to today's analogs (AI, GLP-1, self-driving, crypto, microplastics ...) and source links. One CSV, 88 rows; column meanings in the schema. No third-party clippings or archive material; no verbatim quotations. Our own summaries only.
- Records
- 88
- Format
- CSV
- Downloads
- 0
ERP Analytics Preview — Python Demo
9 protected rows; relative 28-day periods; no calendar dates · Updated Oct 14, 2026
Sample evaluation preview with nine aggregate rows across three cohorts, field dictionary, and runnable Python demonstration using standard library only. Shows last-value and two-point-mean prediction comparison against subsequent observations. Data is normalized, aggregated, and perturbed from one ERP source; no raw transactions, original amounts, entity identities, or calendar dates included. Integration illustration only, not a forecasting validation or learned-agent benchmark.
- Format
- ZIP
- Downloads
- 1
Protected ERP Analytics Pilot: Real-Derived Signals + 30 Analyst Tasks
Relative 28-day periods; calendar dates intentionally withheld · Updated Oct 14, 2026
Evaluate operational analytics and analyst agents with a compact, documented real-derived pilot. Includes 188 protected observations across three relative 28-day operational series, six aggregate first-receipt timing profiles, and 30 authored, machine-checkable analytic tasks. CSV and Parquet data, answer-free JSONL inputs, reference evaluation cases, four local forecasting baselines, a standard-library agent evaluator, data dictionary, transformation/privacy reports and reproducible integrity checks are included. Examples run locally without a paid AI API. The target is normalized recorded outflow workload, not physical inventory, market demand, revenue or cash flow. The reference agent example is scripted, not evidence of learned-agent performance. Internal exploratory baseline error reductions were 0%, 25.98% and 13.89% across the three cohorts; the smaller improvement is statistically uncertain. There is no external performance guarantee. One source and small coverage. Payment profiles are conditional on six retained, marked-paid groups and describe first linked receipts, not full settlement. Roughly 15% source receipt-link coverage limits interpretation. Original identities, exact dates, amounts, products and raw transactions are excluded. Aggregation, perturbation and content scans do not certify anonymity; organization-pattern linkage remains unquantified. USD 250 one-time license for this fixed edition and one purchasing entity. Internal commercial analytics, research, ML training and evaluation allowed. Non-reconstructive derived models may be commercialized. No dataset resale, redistribution, raw publication, sublicensing or re-identification. Future editions, custom consulting and support services are not included. See custom reuse terms. An evaluation preview is prepared and available on request. Review the scope and limitations before purchasing. CUSTOM COMMERCIAL LICENSE — FULL TERMS Protected ERP Analytics Pilot v0.2 — Commercial Internal-Use License Licensor: Alessandro Barrios. This text is proposed for the DataBazaar listing; it takes effect for a purchaser only upon an accepted paid order identifying the purchaser, this edition and these terms. No rights arise from this local draft. The paid license is non-exclusive, worldwide and non-transferable, for use by one purchasing individual or legal entity and its authorized personnel. It permits internal commercial analytics, research, software development, machine learning training and evaluation. There is no limit on internal experiments. Rights continue unless terminated for material breach, subject to mandatory law. The USD 250 price covers this fixed edition only; future editions, consulting, customization and support services are not included. Ownership is not transferred. No resale, redistribution, sublicensing of the dataset, raw publication, public hosting, or sharing with another legal entity. No attempted re-identification, linkage to the original organization or persons, credential recovery or unlawful surveillance. Preserve provenance notices. Do not present these aggregates as raw transactions, physical inventory, company-wide financial records or verified fraud labels. Contractors may access only on the purchasing entity's behalf, under equivalent restrictions; that entity remains responsible for their use. Non-reconstructive derived models and summary statistics may be commercialized. They must not expose source identities, permit substantial dataset reconstruction, or function as a dataset redistribution service. Do not publish substantial portions of the licensed dataset as examples or model outputs. Data is aggregated and perturbed from one real operational source. Direct identifiers, calendar dates, original amounts and raw records are excluded. Temporal organization linkage remains unquantified. This is not formal differential privacy, anonymity certification or an external utility guarantee. It is a small evaluation pilot: 188 observations in three relative 28-day series, six retained timing profiles and 30 authored analytic tasks. Timing profiles describe first linked receipts, not full settlement or business-wide behavior. Provided as-is to the extent permitted by applicable law. No promise of fitness, external prediction accuracy, industry representativeness or privacy certification. Nothing limits mandatory, non-excludable legal rights or overrides applicable marketplace dispute/refund procedures. Source database ownership is not transferred. For material breach, the Licensor may terminate the license by notice; cease dataset use and delete copies subject to mandatory retention. Compliant, non-reconstructive models/statistics may be retained unless mandatory law requires otherwise. No separate governing law or arbitration clause is imposed here.
- Format
- ZIP
- Downloads
- 0
Tyvek Manufacturing & Process Data
Updated Oct 11, 2026
Structured technical dataset of processing parameters and material measurements for Tyvek® applications, including process variables and measurement outputs. Intended for materials research, manufacturing analytics, process optimization, quality control, and predictive modeling.
- Format
- JPG
- Downloads
- 0
Clinical Trial Outcomes
2020–2026
Studies from ClinicalTrials.gov with posted results, normalized across phase, condition, endpoint type, and status. Includes rule-based endpoint classification (OS, PFS, SAFETY_AE, QOL_PRO, etc.) derived from registered measure names.
- Records
- 14,132
- Format
- CSV
- Downloads
- 0
Humanoid Robot Preliminary FMEA
FMEA covering 42 generic humanoid-robot failure modes with effects, causes, controls, severity/occurrence/detection scores, RPN, recommended actions, and verification methods. Preliminary engineering aid; not certification or substitute for architecture-specific safety analysis.
- Records
- 42
- Format
- CSV
- Downloads
- 0
Clinical Trial Outcomes Normalized Snapshot
2020 onward
Normalized dataset of 38,638 ClinicalTrials.gov studies with results posted from 2020 onward, including 326,638 endpoint rows and 2.6 million group/category measurement rows across four CSV tables. Includes trial metadata, statistical analyses, data dictionary, and validation report with SHA-256 hashes for reproducibility.
- Records
- 38,638
- Format
- ZIP
- Downloads
- 0
Iris Flower Classification
Ronald Fisher's canonical classification benchmark with iris samples across three species (setosa, versicolor, virginica), measured by sepal and petal length and width.
- Records
- 150
- Format
- PARQUET
- Downloads
- 5
Pen-Based Digit Recognition — Synthetic BNG
Synthetic expansion of the Pen Digits dataset generated via Bayesian Network, containing 16 pen-trajectory features per sample for handwritten-digit classification.
- Records
- 1,000,000
- Format
- PARQUET
- Downloads
- 5
Glass Type Identification — Synthetic BNG
Synthetic expansion of the UCI Glass Identification dataset generated via Bayesian Network, classifying glass type from refractive index and oxide composition.
- Records
- 137,781
- Format
- PARQUET
- Downloads
- 7
Cleveland Heart Disease
Clinical dataset of patients with 13 attributes including age, sex, chest-pain type, blood pressure, cholesterol, and ECG results, annotated with heart-disease diagnosis labels. A foundational UCI benchmark for classification.
- Records
- 303
- Format
- PARQUET
- Downloads
- 4
Hepatitis Prognosis — Synthetic BNG
Synthetic expansion of the UCI Hepatitis dataset via Bayesian Network generation, predicting patient survival from clinical and laboratory attributes.
- Records
- 1,000,000
- Format
- PARQUET
- Downloads
- 3
Space Shuttle Autolanding Control
NASA decision dataset specifying conditions for automatic versus manual Space Shuttle landing, including stability, error, sign, wind, magnitude, and visibility. A classic UCI benchmark.
- Records
- 15
- Format
- PARQUET
- Downloads
- 3
Page Blocks Classification — Synthetic BNG
Synthetic expansion of the classic Page Blocks dataset via Bayesian Network generation, classifying document layout blocks by geometric and pixel-density features.
- Records
- 295,245
- Format
- PARQUET
- Downloads
- 3
Heart Disease Statlog — Synthetic BNG
Synthetic expansion of the Statlog Heart dataset generated via Bayesian Network, predicting heart-disease presence from 13 clinical attributes.
- Records
- 1,000,000
- Format
- PARQUET
- Downloads
- 4
Heart Disease Cholesterol Prediction
Clinical dataset with serum cholesterol as the prediction target alongside 13 clinical attributes from patients. Used as a numeric regression benchmark.
- Records
- 303
- Format
- PARQUET
- Downloads
- 4
E. coli Promoter Gene Sequences
DNA sequences from E. coli labeled as promoter or non-promoter, each described by 57 sequential nucleotide positions. A classic molecular-biology classification benchmark for sequence analysis.
- Records
- 106
- Format
- PARQUET
- Downloads
- 4
SEA Concept-Drift Stream
Synthetic data stream from the classic SEA generator with three numeric attributes and binary class label, designed as a standard benchmark for concept-drift detection and streaming classifier evaluation.
- Records
- 1,000,000
- Format
- PARQUET
- Downloads
- 4
Statlog Heart Disease
Clinical dataset with 13 attributes (age, sex, chest-pain type, blood pressure, cholesterol, ECG results, and related measurements) and binary heart-disease presence labels. A standard UCI classification benchmark.
- Records
- 270
- Format
- PARQUET
- Downloads
- 3
Robotics and Humanoid FMEA Public Source Dataset
Curated index of 18 public-source FMEA references for robotics, humanoids, collaborative, and autonomous platforms. Includes URLs, source type, relevance scores, and manual review labels. Compiled metadata and links only; does not redistribute original papers or proprietary documents.
- Records
- 18
- Format
- CSV
- Downloads
- 0
Magneton Protein Representation Learning Dataset
SwissProt proteins annotated with DSSP secondary structure and InterPro 103.0 substructure annotations for training and evaluating protein representation models.
- Records
- 530,601
- Format
- PARQUET
- Downloads
- 3
Open Schematics Electronic Circuit Designs
Electronic schematics from hardware projects with visual representations, component metadata, and KiCad source files. Suitable for training on circuit design, component recognition, and hardware engineering tasks.
- Records
- 84,470
- Format
- PARQUET
- Downloads
- 5
Physiotherapy Evidence QA Bilingual
Expert-curated Turkish and English question-answer pairs covering evidence-based physiotherapy, musculoskeletal rehabilitation, outcome measures, and clinical research methodology.
- Records
- 143,711
- Format
- PARQUET
- Downloads
- 4
ChemBench Chemistry & Materials LLM Evaluation Benchmark
Expert-curated question-answer and multiple-choice benchmark for evaluating large language model performance on chemistry and materials science tasks.
- Records
- 2,785
- Format
- PARQUET
- Downloads
- 4