datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
tenacious-bench-v0.1
Tenacious-Bench v0.1
A domain-specific evaluation benchmark for B2B outbound sales agents. Grades five failure modes that standard benchmarks (τ²-Bench retail) cannot measure.
300 tasks · 5 rubric dimensions · 4 authoring modes · CC-BY-4.0
Why This Benchmark Exists
τ²-Bench retail tests cooperative airline-service policy transactions. B2B outbound sales requires:
Dimension
Trigger Rate (Week 10 probes)
Commercial Risk
signal_grounding_fidelity
35%
CTO… See the full description on the dataset page: https://huggingface.co/datasets/rafiakedir/tenacious-bench-v0.1.tenacious-bench
Tenacious-Bench v0.1
Sales agent evaluation benchmark for B2B outbound honesty and tone compliance.
274 tasks (250 programmatic + 24 hand-authored). Machine-verifiable rubric.
Author: Eyobed Feleke
License: CC-BY-4.0
Model: eyobed7b/tenacious-bench-simpo-judge-v1
tenacious_bench_v0.1
Tenacious-Bench v0.1
A domain-specific evaluation benchmark for B2B sales AI agents in the technical staffing vertical. Tenacious-Bench tests whether an AI outreach agent is honest, proportionate, and professionally grounded — not just whether it completed a task.
Quickstart
git clone https://huggingface.co/datasets/Yohannesdn/tenacious_bench_v0.1
pip install jsonschema
python scoring_evaluator.py --schema schema.json --example 1
Run python scoring_evaluator.py --schema… See the full description on the dataset page: https://huggingface.co/datasets/Yohannesdn/tenacious_bench_v0.1.tenacious_bench_v0.1
Tenacious-Bench v0.1
A 266-task evaluation benchmark for B2B sales-outreach agents, grounded in the
Tenacious (B2B engineering-outsourcing) workflow. 10 failure dimensions, 5
authoring source modes, mechanically-gradable rubric (no human in the loop).
Held-out partition is sealed — see held_out/SEALED.md for details.
Dataset summary
Total tasks (train + dev): 191
Held-out (sealed): 75
Source modes: 5
Failure dimensions: 10
License: CC-BY-4.0
Quickstart
from… See the full description on the dataset page: https://huggingface.co/datasets/eyorata/tenacious_bench_v0.1.tenacious-bench
📊 Tenacious-Bench: B2B Sales Outreach Judge Preference Dataset
Version: v0.1Author: Bethelhem Abay · 10 Academy TRP1Date: 2026-05-02License: MIT
A curated preference dataset of 323 (chosen, rejected) pairs for training and evaluating a pre-send judge that blocks unsafe B2B sales outreach before it reaches the wrong people.
🔗 Quick Links
Resource
Link
📦 Dataset (this page)
bethelhem21/tenacious-bench
🤖 Trained Judge Model… See the full description on the dataset page: https://huggingface.co/datasets/bethelhem21/tenacious-bench.tenacious_bench_v0.1
Datasheet: Tenacious Bench v0.1
Quickstart
You can load this dataset directly in Python using the Hugging Face datasets library. The dataset contains three splits: train, dev, and held_out.
from datasets import load_dataset
# Load the dataset
dataset = load_dataset("sanoy24/tenacious_bench_v0.1")
# View a single training example
print(dataset["train"][0]["prompt"])
print(dataset["train"][0]["chosen"])
print(dataset["train"][0]["rejected"])
Baseline and Target… See the full description on the dataset page: https://huggingface.co/datasets/sanoy24/tenacious_bench_v0.1.tenacious-bench-v0.1
Tenacious-Bench v0.1
Tenacious-Bench v0.1 is a small, contamination-checked benchmark for Tenacious-style B2B sales outreach. It measures whether an agent can ground outreach in public hiring signals, stay within truthful capacity and pricing boundaries, preserve the Tenacious tone markers, and avoid brand-damaging failure modes that generic retail-style agent benchmarks do not measure.
What is in the dataset
train split: task authoring and training-pair construction
dev… See the full description on the dataset page: https://huggingface.co/datasets/abdulaziz0111/tenacious-bench-v0.1.tenacious-bench
Tenacious-Bench v0.2
A domain-specific evaluation benchmark for B2B sales outreach agents. Measures compliance with structured business rules that generic benchmarks (τ²-Bench retail, etc.) cannot grade: bench capacity constraints, ICP segment routing, signal confidence calibration, and tone compliance.
Why this benchmark exists
The Week 10 Tenacious agent failed every bench-capacity probe (40/40, 100%) and misrouted 54% of ICP classification tasks — while producing… See the full description on the dataset page: https://huggingface.co/datasets/Natnaela/tenacious-bench.tenacious-bench-v0.1
Tenacious-Bench v0.1
260-task, machine-verifiable evaluation benchmark for the Tenacious Conversion Engine AI outbound sales agent.
Purpose-built to measure the failure modes that τ²-Bench cannot grade: confidence-calibrated phrasing, staleness disclosure, abstention routing, and thread isolation.
Split
Tasks
Use
train
143
LoRA fine-tuning corpus
dev
55
Validation / prompt iteration
held_out
62
Released post-training for independent Delta B verification… See the full description on the dataset page: https://huggingface.co/datasets/kirutew17654321/tenacious-bench-v0.1.tenacious-bench-v1
Datasheet for Tenacious-Bench v0.1
1. Motivation
Tenacious-Bench v0.1 is a domain-specific evaluation dataset for B2B sales-agent behavior in the Tenacious Conversion Engine workflow. Existing public agent benchmarks evaluate generic tool use, transaction completion, and sequential reasoning, but they do not sufficiently measure Tenacious-specific behavior: signal-grounded outreach, low-hype sales tone, qualification before pitching, objection handling, pricing safety… See the full description on the dataset page: https://huggingface.co/datasets/AddisuT/tenacious-bench-v1.tenacious-bench-v0.1
Datasheet for Tenacious-Bench v0.1
Following Gebru et al. (2021) "Datasheets for Datasets" framework, extended with Pushkarna et al. (FAccT 2022) layered documentation structure.
Executive Summary (Telescopic View)
Pushkarna et al. (FAccT 2022) — Layer 1: one-paragraph overview for any reader.
Tenacious-Bench v0.1 is a 300-task evaluation benchmark for B2B sales outreach agents in the Tenacious Consulting & Outsourcing conversion engine. It was built because… See the full description on the dataset page: https://huggingface.co/datasets/Mella123/tenacious-bench-v0.1.tenacious_bench_v0.1
TenaciousBench v0.1
A 220-task evaluation benchmark for B2B outbound sales agents.
TenaciousBench v0.1 evaluates LLM-based B2B outreach systems across ten failure dimensions: ICP targeting precision, confidence-aware phrasing, signal grounding fidelity, tone safety, hallucination avoidance, CTA behavior, competitor gap reasoning, pricing discipline, multi-turn objection handling, and thread continuation coherence.
Why This Benchmark Exists
Existing LLM benchmarks… See the full description on the dataset page: https://huggingface.co/datasets/eyorg/tenacious_bench_v0.1.tenacious_bench_v0.1
Datasheet — Tenacious-Bench v0.1
Following Gebru et al., Datasheets for Datasets (2021) and Pushkarna
et al., Data Cards: Purposeful and Transparent Dataset Documentation
(FAccT 2022). Pushkarna's telescopic/periscopic/microscopic layered
detail is shown inline at the start of each section.
Telescopic (one line): A 237-task, machine-verifiable evaluation
bench for B2B engineering-talent sales agents — segment classification,
signal grounding, bench honesty, tone preservation, gap… See the full description on the dataset page: https://huggingface.co/datasets/kgutd/tenacious_bench_v0.1.tenacious_bench_v0.1
License
This dataset is released under the CC-BY-4.0 license. Users are free to use, modify, and distribute the dataset with appropriate attribution.
Tenacious Bench v0.1
Motivation
This dataset evaluates B2B outreach quality and failure modes in Tenacious-style sales workflows.
Composition
230 tasks across four source modes:
trace-derived
programmatic
multi-LLM synthesis
hand-authored
Partition
Train: 115
Dev: 68
Held-out: 47… See the full description on the dataset page: https://huggingface.co/datasets/beti4/tenacious_bench_v0.1.tenacious-bench-path-b-preference
Tenacious Bench Path B Preference
1. Motivation
This dataset exists because generic assistant benchmarks do not reliably measure the failure modes that matter in Tenacious-style B2B outbound work. The goal here is not broad conversational quality; it is grounded business behavior under uncertainty.
The preference pairs focus on:
grounded language
weak-confidence handling
over-claiming avoidance
pricing handoff safety
qualification correctness
channel routing… See the full description on the dataset page: https://huggingface.co/datasets/ephorata/tenacious-bench-path-b-preference.tenacious-bench-v0.1
Tenacious-Bench v0.1 — Sales Agent Evaluation Bench
Domain-specific evaluation benchmark for Tenacious's B2B sales agent. Built because τ²-Bench retail (pass@1=0.7267) completely missed five categories of production failure — the agent was using banned phrases in 38% of runs while scoring 0.82 on τ²-Bench.
Headline result: ORPO fine-tuning on 381 preference pairs lifted held-out mean from 4.008 → 4.462 (+11.3%), Δ=+0.454, p=0.001, beating a prompt-engineered baseline by Δ=+0.290… See the full description on the dataset page: https://huggingface.co/datasets/Chalie-lijalem/tenacious-bench-v0.1.tenacious_bench_v0.1tenacious_bench_v0.1
Tenacious-Bench v0.1
Tenacious-Bench v0.1 is a domain benchmark for evaluating sales-agent outputs under Tenacious-specific business and policy constraints.
What This Benchmark Measures
This dataset evaluates whether an agent (or critic/judge) correctly handles:
unsupported pricing or scope claims
overclaimed signal or maturity claims
generic/ungrounded outreach
incorrect CRM/HubSpot/calendar next actions
reply escalation or objection-handling failures
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/Nurye/tenacious_bench_v0.1.tenacious-bench-v0.1
Tenacious-Bench v0.1
A 200-task evaluation benchmark for B2B sales outreach agents. Measures five failure dimensions that general-purpose benchmarks (τ²-Bench retail) cannot detect.
Why this benchmark exists
The Tenacious AI sales agent achieves 38.7% pass@1 on τ²-Bench retail — but the failures are invisible to that rubric. τ²-Bench has no bench inventory model, no style guide, and no signal confidence grading. This benchmark was built specifically to catch what τ²-Bench… See the full description on the dataset page: https://huggingface.co/datasets/ketewodros41/tenacious-bench-v0.1.tenacious-bench-v0-1
Datasheet: Tenacious-Bench v0.1
Gebru et al. (2021) 7-section format with Pushkarna et al. (2022) layered detail.
1. Motivation
Why was this dataset created?
Tenacious-Bench v0.1 was created to fill a critical evaluation gap for the Tenacious Intelligence Corporation sales agent. The τ²-Bench retail dataset (cancel-order, exchange-items, track-shipment tasks) shares no domain vocabulary, decision logic, or grounding constraints with Tenacious's B2B engineering staff… See the full description on the dataset page: https://huggingface.co/datasets/MajorKemeriya/tenacious-bench-v0-1.tenacious-bench-v0.1
Datasheet: Tenacious-Bench v0.1
Summary Layer
Tenacious-Bench v0.1 is a 200-task benchmark for evaluating B2B sales-agent outreach under Tenacious-specific constraints: public-signal grounding, weak-signal humility, capacity safety, pricing boundaries, one-ask structure, and the Tenacious style guide. It follows Gebru et al.'s Datasheets sections and Pushkarna et al.'s layered data-card pattern: summary, structured overview, and schema-level notes.
The core object is not… See the full description on the dataset page: https://huggingface.co/datasets/ruthasol/tenacious-bench-v0.1.tenacious-bench-v0.1
Datasheet for Tenacious-Bench v0.1
Following Gebru et al. (2021) seven-section format, supplemented by Pushkarna et al. (2022) Data Cards layered detail.
Version: 0.1 | Release date: 2026-04-29 | Author: richh-s (10Academy TRP1)Dataset size: 200 tasks (97 train / 60 dev / 43 held_out)
1. Motivation
Why was this dataset created?
τ²-Bench retail (the Week 10 evaluation benchmark for conversational agents) scores binary pass/fail on task resolution in a generic shopping… See the full description on the dataset page: https://huggingface.co/datasets/rich6/tenacious-bench-v0.1.tenacious-bench-v0.1tenacious-bench-v0.1tenacious-bench-v01