Single-answer verifiable. Benchmarked against frontier models. License-clean.

Frontier-grade data for the models that need it most

The catalog

Physical AI Factory

Egocentric manipulation from real production floors

egocentric-manipulation

First-person video of skilled manual work captured on real factory and workshop floors, each clip densely annotated with 21-keypoint hand pose for both hands, hand-visibility state, per-hand detection confidence, and frame-accurate contact and release events. Built for the embodied models that learn manipulation from how people actually use their hands.

Vision-language-action training, manipulation policy learning, and embodied world models.
Volume
8,000 clips
Fields
6 columns
data_idenvironmentvideohand_posecontact_eventstask_label

First example

01FAC-01Abrasive finishing

taskBimanual disc finishing — one hand holds a metal disc against the fixture while the other drives an abrasive tool across its face, including the brief release moment where both hands leave the workpiece.

labelsPer-frame 21-keypoint hand pose for both hands, a hand-visibility state ('None / Left only / Right only / Both'), and tool-contact transitions with 'hands off' release windows timed to a tenth of a second.

Voice & Speech

Expressive speech across tone, accent, and identity

expressive-speech-safety

Recorded and synthetic speech annotated along the dimensions that matter for audio models: emotional tone and prosody, regional and non-native accents, speaker identity with paired real and cloned voices for impersonation detection, scenario intent including fraud and social-engineering red-team content, code-switching, and acoustic environment. Every clip ships with a verified transcript and per-dimension labels.

Speech safety classifiers, deepfake and spoof detection, TTS and voice-clone evaluation, and ASR robustness.
Volume
25,000 clips
Fields
8 columns
data_idaudiotranscripttoneaccentimpersonationprovenanceintent

First example

01VOX-01Impersonation red-team

scenarioVoice-clone impersonation of a public figure delivering a cryptocurrency giveaway pitch — a negative example for training and evaluating audio deepfake and scam-call detectors.

Impersonation
Public figure — Barack Obama (cloned voice)
Provenance
Synthetic — TTS voice clone
Tone
Measured, authoritative, persuasive
Accent
General American
Intent
Fraud — crypto giveaway scam
Audio
Mono, 44.1 kHz, 14 s

STEM Reasoning Traces

Step-verified derivations for scientific reasoning

verified-reasoning-traces

Graduate-level physics and STEM problems paired with complete, step-checked reasoning traces: explicit givens, a numbered derivation with every intermediate value verified, a unique machine-checkable answer, and an independent sanity check. Items that reduce to one-line arithmetic, leak the answer in the prompt, or whose derivation doesn't land on the stated answer are filtered out in QA.

Process-reward models, verifier training, and chain-of-thought SFT for scientific reasoning.
Volume
10,000 tasks
Fields
6 columns
problemgivenstrace_stepsfinal_answersanity_checkarea

First example

01Quantum Field TheoryQCD coupling

With b₀ = 11 − (2/3)n_f and leading-order running 1/α_s(Q) = 1/α_s(M_Z) + (b₀/2π)·ln(Q/M_Z), evaluate α_s at Q = 500 GeV. Report to two decimal places.

given
α_s(M_Z) = 0.118    M_Z = 91.2 GeV    Q = 500 GeV    n_f = 5
Show reasoning trace (6 steps)

1 · Beta-function coefficient: b₀ = 11 − (2/3)(5) = 23/3 ≈ 7.667 2 · Invert the coupling: 1/α_s(M_Z) = 1/0.118 = 8.475 3 · Running coefficient: b₀/(2π) = 7.667 / 6.2832 = 1.220 4 · Energy logarithm: ln(Q/M_Z) = ln(500/91.2) = ln(5.482) = 1.702 5 · Evolve the inverse coupling: 1/α_s(Q) = 8.475 + (1.220)(1.702) = 10.551 6 · Invert back: α_s(500 GeV) = 1/10.551 = 0.0948, rounded to two decimals.

verified answer0.09

sanity checkCoupling fell from 0.118 → 0.095 as energy rose — the correct direction for asymptotic freedom, confirming the sign convention was applied right.

Medical AI Data Infrastructure

Longitudinal EHR and clinical conversations at population scale

longitudinal-ehr

Longitudinal electronic health records spanning visits, diagnoses, medications, and lab series, shipped as verified reasoning tasks over the raw records: multi-hop temporal QA, clinical entity extraction, and longitudinal aggregation, each with a step-by-step evidence-grounded trace, a programmatically checked gold answer, and label-quality flags. Alongside the records: 6,000+ hours of doctor-patient conversations with transcripts available on demand. All datasets can be structured, filtered, de-identified, annotated, and quality-reviewed to the partner's use case, specialty, language, modality, and compliance requirements.

Longitudinal clinical reasoning, disease progression and risk prediction, patient history summarisation, care-gap identification, medical scribing and SOAP note generation, and healthcare copilot evaluation.
6M+patients >1yr history
27M+patients <1yr history
6,000+hrs conversations
Volume
33M+ patients
Fields
7 columns
patient_idcontextquestionreasoning_tracefinal_answerverificationquality_flags

First example

01EHR-TRACE-001Multi-hop temporal QA

qRosuvastatin was newly initiated for this patient at one visit (medicine_status = 'entry'). What was the change in LDL cholesterol from the last reading before that visit to the first reading after it? Report the initiation date, both readings, and the absolute change in mg/dL.

Show reasoning trace (4 steps)

1 · locate_medication_event — scan medication records for rosuvastatin with status = 'entry'. Initiation visit = V00214, dated 2025-04-02. 2 · partition_lab_series — sort the LDL series and split around 2025-04-02. Last before: 120 mg/dL (2025-03-31). First after: 65 mg/dL (2025-05-17). 3 · compute_delta — 65 − 120 = −55 mg/dL (−45.8% relative change). 4 · sanity_check — a high-intensity statin (rosuvastatin 40 mg) initiation is typically followed by a large LDL drop within 4–6 weeks; the 46-day gap fits. Answer accepted.

Initiation date
2025-04-02
Absolute change
−55 mg/dL
Last LDL before
120 mg/dL · 2025-03-31
First LDL after
65 mg/dL · 2025-05-17

verificationProgrammatic exact match — recompute from source JSON: filter meds by (generic~rosuvastatin, status=entry); partition sorted LDL readings around visit_date; assert delta == −55.0.

quality flagSame visit V00214 maps raw text 'kco ald with old pancreatitis' to diagnosis 'adrenoleukodystrophy'; clinical context (pancreatitis, Indian shorthand) strongly suggests 'alcoholic liver disease'. Ontology-mapping defect — excluded from this task's context but relevant for dataset QA.

MATH Core

Verifiable reasoning at scale

Non-trivial undergraduate and early-graduate problems with single, machine-verifiable answers. Broad coverage across algebra, combinatorics, geometry, topology, number theory, probability, optimization, graph theory, and analysis. The workhorse corpus for reasoning RL and supervised fine-tuning.

High-volume training data, enterprise evaluations, and general reasoning coverage.
Volume
20,000 tasks
Fields
4 columns
problemfinal_answertopicdifficulty

First example

01GeometryCore Breaker

Let SS be the set consisting of 00 together with all 5151 complex roots of z51=1z^{51} = 1. How many monic degree-88 complex polynomials p(z)p(z) have eight distinct roots, all lying in SS, with the additional property that every root is a corner point of the convex hull of the eight roots in the complex plane?

answer645,795,150645\\,795\\,150

MATH Undergrad Breakers

Curated to break frontier models

undergrad-frontier-breaking

Undergraduate-level tasks specifically engineered to defeat frontier models. Every item is benchmarked against strong models and kept only if they fail to solve it, giving you dense, high-signal training and eval material without crossing into graduate research.

Buyers who want harder math tasks without moving fully into graduate research material.
Volume
20,000 tasks
Fields
5 columns
problemfinal_answertopicdifficultybroke_model

First example

01Graph Theory & NetworksUndergrad Breaker

The complete graph K10K_{10} on 1010 labeled vertices has 10810^8 spanning trees by Cayley's formula. Determine the number of spanning trees in which, for every edge, the sum of the degrees (in the tree) of its endpoints is at most 55.

answer35,078,40035\\,078\\,400

MATH Graduate Breakers

Frontier-grade difficulty

graduate-frontier-breaking

Graduate and research-style tasks designed to challenge frontier models. The hardest tier we ship, sourced from advanced coursework and research-adjacent problems, then benchmarked against the strongest available models to confirm genuine frontier difficulty.

Frontier-lab evaluation, difficult training data, and private math benchmarks.
Volume
10,000 tasks
Fields
5 columns
problemfinal_answertopicdifficultybroke_model

First example

01Finite Symplectic GeometryGraduate Breaker

Work over F37F_{37} with basis e1,e2,f1,f2e_1, e_2, f_1, f_2 and alternating form langlex,yrangle=x1y3+x2y4−x3y1−x4y2\\langle x, y \\rangle = x_1 y_3 + x_2 y_4 - x_3 y_1 - x_4 y_2. Let XX be the set of incident pairs ([v],L)([v], L), where [v][v] is a 1-dimensional subspace of F374F_{37}^4 and LL is a 2-dimensional subspace containing [v][v] on which this form is identically zero. Let WW be the F2F_2-vector space of all labelings c:Xto0,1c : X \\to \\{0,1\\} such that, for every [v][v], the sum of c([v],L)c([v], L) over all LL containing [v][v] is 0 in F2F_2, and for every LL, the sum of c([v],L)c([v], L) over all [v][v] contained in LL is 0 in F2F_2. With EijE_{ij} denoting the matrix in row ii and column jj, define A+(t)=I+t(E12−E43)A_+(t) = I + t(E_{12} - E_{43}), A−(t)=I+t(E21−E34)A_-(t) = I + t(E_{21} - E_{34}), B+(t)=I+tE13B_+(t) = I + t E_{13}, and B−(t)=I+tE31B_-(t) = I + t E_{31} for tt in F37F_{37}. Let mAm_A be the F2F_2-dimension of the part of WW fixed by every A+(t)A_+(t) and A−(t)A_-(t), and define mBm_B similarly using B+(t)B_+(t) and B−(t)B_-(t). Let rr be the index, among projective linear transformations preserving the incidence relation on XX, of those induced by matrices preserving the alternating form exactly. The reported score is mA+mBm_A + m_B if mA+mBm_A + m_B is congruent to rr modulo 3 and the two numbers mA,mBm_A, m_B have opposite parity; otherwise the reported score is 0. What is the reported score?

answer149149

LEAN Formalization

Informal to formal, fully proved

Lean 4 data drawn from the undergraduate and graduate difficulty bands. Each record pairs a natural-language statement with its Lean formal statement and a complete, type-checked Lean proof, exactly what you need to train and evaluate autoformalization and proof-synthesis models.

Training and evaluating models that translate informal mathematics into correct Lean.
Volume
1,000 tasks
Fields
3 columns
nl_statementlean_statementlean_proof

First example

01Number Theory

The sum of two even integers is even.

lean statement
theorem even_add_even (m n : ℤ) (hm : Even m) (hn : Even n) : Even (m + n)
proof
theorem even_add_even (m n : ℤ) (hm : Even m) (hn : Even n) :
    Even (m + n) := by
  obtain ⟨a, ha⟩ := hm
  obtain ⟨b, hb⟩ := hn
  exact ⟨a + b, by rw [ha, hb]; ring⟩

Chart Understanding

Multimodal quantitative reasoning

Real-world charts across finance, healthcare, government, education, and scientific research, each paired with a grounded, verifiable question, a gold answer, and a full step-by-step chain of thought. Questions demand reading exact values, multi-step computation, and cross-panel reasoning, built for multimodal reasoning models.

Vision-language training, multimodal evaluation, and document/analytics reasoning.
Volume
15,000 tasks
Fields
6 columns
data_iddomainimagequestiongold_answerCOT

First example

01FIN-01Financial services
Chart for sample FIN-01

qWhat is the difference between Unilever's and Colgate-Palmolive's market share in Apparel Care, and which brand's delta suggests it is closing the gap faster?

a11.9 percentage points; Unilever
Show chain of thought

Unilever holds 21.3% share in Apparel Care (delta -0.7pp, share erosion flagged) versus Colgate-Palmolive's 9.4% (delta +0.3pp). The gap is 21.3 - 9.4 = 11.9 percentage points. With Unilever eroding 0.7pp and Colgate-Palmolive gaining 0.3pp, the gap narrows at a combined 1.0pp per quarter, and Unilever's erosion is the dominant factor closing it.

GIS Spatial Reasoning

Reason over maps and geospatial layers

Real cartographic maps spanning land cover, hydrology and flood risk, terrain elevation, and urban transportation, each paired with a grounded, verifiable question, a gold answer, and a full chain of thought. Questions require reading legends, scale bars, contour spacing, and choropleth gradients, then performing multi-layer spatial inference, built for multimodal models that need to reason over geospatial data.

Vision-language training, geospatial analytics, and remote-sensing or mapping evaluation.
Volume
12,000 tasks
Fields
6 columns
data_iddomainimagequestiongold_answerCOT

First example

01GIS-01Land cover & remote sensing
Map for sample GIS-01

qUsing the land-cover classification, which class occupies the largest contiguous area along the coastline, and what is the dominant adjacency relationship between urban (gray) and water (blue) classes?

aForest (green) dominates the inland coastline; urban built-up areas are directly adjacent to water along the estuary, indicating a port/harbor settlement pattern.
Show chain of thought

Reading the legend, green = forest, gray = urban, blue = water. Tracing contiguous polygons along the coast, the green forest class forms the largest unbroken band inland of the shore. The gray urban polygons consistently border the blue water class at the estuary mouth rather than being embedded inland, which is the classic signature of a port settlement that grew around a harbor. Cropland (yellow) sits behind the urban fringe, away from the immediate shoreline.

Why buyers choose us

Data you can actually train and evaluate on.

Quality is the product. Each corpus is built to be hard, clean, and machine-gradable from day one.

Sourced & authored

Problems are drawn and authored across nine mathematical domains, spanning undergraduate coursework to research-adjacent material, never scraped boilerplate.

Benchmarked against frontier models

Hard tiers are tested against the strongest available models first. Any problem they solve gets dropped, so what's left is the dense, high-signal material that actually moves evals.

Verifiable by construction

Math ships with single checkable answers, Lean ships with type-checked proofs, and chart QA is grounded in the underlying data, so grading is unambiguous.

Built by people who understand the math

Our core team comes from mathematics and computer science at the University of Pennsylvania, with access to a broader network of PhDs, postdocs, and faculty across technical fields. The data is written and reviewed by people who understand the mathematics, not scraped from public problem banks or lightly rewritten from competitions.

Get a sample pack scoped to your eval

Tell us the domains, difficulty band, and volume you need. We'll send representative samples and a quote. Bulk and exclusive licensing available.

Which data are you interested in?