messages listlengths 2 2 | metadata dict | quality_score float64 7.5 8.5 | decontamination_flags listlengths 0 0 | decontamination_evidence listlengths 0 0 |
|---|---|---|---|---|
[
{
"role": "user",
"content": "I’m seeing my order marked “fulfilled” in your portal, but I never received a shipping confirmation and there’s no tracking number. I also didn’t get any email (checked inbox). I need this shipped to a forwarder and if it’s not out today I have to cancel. Can you tell me what “... | {
"quality_rule_issues": [
"assistant_too_long"
],
"quality_issues": [
"Does not directly answer why no tracking exists yet beyond plausible causes; could more explicitly state the most common reason (label created but not scanned / email failure) and what the user can do immediately in the portal.",
... | 7.5 | [] | [] |
[
{
"role": "user",
"content": "We’re on a Business plan and placed a hardware add-on order through the admin console for 50 units. The dashboard says “processing,” but it’s been 6 days with no shipping confirmation. Support articles mention 1–2 business days. We also have a credit-card preauth still pending.... | {
"quality_rule_issues": [
"assistant_too_long"
],
"quality_issues": [
"Implies the assistant can directly access internal order timelines, payment state, and escalate to fulfillment; this may be unrealistic without clarifying limitations.",
"Requests potentially sensitive account-linking details (adm... | 7.5 | [] | [] |
[
{
"role": "user",
"content": "I placed order #A19-77421 six days ago and still haven’t received a shipping confirmation. My dashboard says “Processing,” but my card was captured (not pending) and your checkout page promised 1–2 business day dispatch. I also didn’t get an order confirmation email—nothing in ... | {
"quality_rule_issues": [
"assistant_too_long"
],
"quality_issues": [
"Assumes the assistant can 'take ownership' and directly verify internal order/payment/warehouse systems; in many contexts the assistant should clarify limitations or route to support.",
"Some timeline claims (e.g., customs 1–5 bus... | 8.5 | [] | [] |
[
{
"role": "user",
"content": "I placed order #A-104883 two days ago (paid via Apple Pay), and my dashboard still shows “Confirmed” with no shipping email. I need it to arrive before a conference next week. Can you verify whether it actually left the warehouse and, if it hasn’t, what the fastest reroute/upgr... | {
"quality_rule_issues": [],
"quality_issues": [
"Implies the assistant can directly access internal fulfillment events/warehouse status without clarifying limitations or that it needs system access; could be misleading in a generic assistant context.",
"Mentions 'temporary duplicate authorization depending... | 8 | [] | [] |
expanded_E-commerce_and_SaaS_customer_support_(orders,_shipping,_billing,_subscriptions,_app_troubleshooting)
Synthetic customer-support dataset generated with synth-dataset-kit.
What Is Included
train.jsonl: training split in JSONL chat formateval_summary.json: minimal evaluation and generation summaryexpanded_E-commerce_and_SaaS_customer_support_(orders,_shipping,_billing,_subscriptions,_app_troubleshooting)_quality_report.html: visual quality reportexpanded_E-commerce_and_SaaS_customer_support_(orders,_shipping,_billing,_subscriptions,_app_troubleshooting)_quality_report.json: machine-readable quality report
Summary
- examples: 4
- generator: synth-dataset-kit
- avg quality: 7.88
- passed: 4
- contamination hits: 0
- distribution divergence: 0.5000
- rebalancing rounds: 0
Data Generation Process
This dataset was created from a small customer-support seed set. Generation used staged candidate creation, quality filtering, decontamination, and distribution-aware rebalancing.
Quality Signals
- lexical diversity: 0.4176
- self-BLEU proxy: 0.0159
- embedding diversity: None
Distribution Alignment
- divergence: 0.5000
- distribution match score: 0.00/100
- semantic coverage score: 0.00%
- graph coverage score: 0.00%
- underrepresented clusters: none
- semantic coverage gaps: none
- graph frontier clusters: none
Contamination Audit
- contamination hits: 0
- verdicts: clean=4
Recommended Use
Use this dataset for instruction tuning or support-assistant fine-tuning experiments. Validate on your own holdout set before production use.
Reproducibility
This bundle was generated by synth-dataset-kit from a small seed set.
Use eval_summary.json plus the full HTML/JSON report to reproduce the run context.
Baseline Comparison
- baseline dataset: customer_support_seeds
- example delta: -2
- avg quality delta: +0.8750
- pass rate delta: +33.33%
- lexical diversity delta: -0.1072
- distribution divergence delta: +0.0000
- contamination hit delta: +0
Reference Dataset Comparison
- reference dataset: customer_support_seeds
- avg user length delta: +53.0833
- avg assistant length delta: +398.7500
- diversity delta: -0.0138
- lexical diversity delta: -0.1072
- style distribution distance: 0.5000
- cluster coverage distance: 0.0000
- difficulty profile distance: 1.0000
- topic overlap ratio: 100.00%
- topic novelty ratio: 0.00%
- exact overlap ratio: 0.00%
- near overlap ratio: 0.00%
- semantic overlap ratio: n/a
- reference alignment score: 70.50/100
- Downloads last month
- 16