Dataset record

Support Ticket Corpus

Zendesk export, managed connector

Observed at 24 Sep 2026, 06:15 UTCds-support-ticketseu-west-1sensitivity confidential

Summary

Owner: Priya Raman
Residency: eu-west-1
Retention policy: 36 months from ticket closeexpires 24 Sep 2029 · no legal hold
Lineage: IncompleteAttachment text extracted before 2025-11 has no derivation record. Downstream links from those chunks are shown as unknown, not as absent.

Prompt logs derived from this corpus expire at 90 days.

Detector, confidence and the share of content the detector could actually read.

Scroll table sideways for more columns →

Label	Detector	Confidence	Coverage
Customer contact details	pii.contact v7	96%	Pass
Payment card fragments	pci.pan v4	88%	Partial
Account credentials in pasted logs	secret.generic v3	61%	Partial
Health references in accessibility tickets	phi.mention v2	44%	Unknown

Scanned share is measured per version; unsupported formats stay visible.

Scroll table sideways for more columns →

Version	Created	Records	Classification	Scanned	Unsupported formats	Note
v2026.09.3	24 Sep 2026, 02:00 UTC	4.1M tickets	Partial	94%	.heic, password-protected .zip	6% of attachments skipped. Skipped content is not counted as scanned.
v2026.08.1	4 Aug 2026, 02:00 UTC	3.9M tickets	Pass	99%	none	Superseded. Still referenced by index support-eu-v11.

Each use mode is decided separately. An absent decision is unknown, never an allow.

retrieval
Support purpose is inside the approved scope and the EU region restriction is satisfied.
dup-2026.07 · reviewer Legal Ops — D. Okoye · expires 31 Jan 2027
Allow
training
Reading for retrieval is not training permission. No customer consent record covers model training on ticket content, so deny takes precedence.
dup-2026.07 · reviewer Legal Ops — D. Okoye · no expiry recorded
Deny
evaluation
Evaluation permitted on de-identified samples only.
dup-2026.07 · reviewer Legal Ops — D. Okoye · expires 31 Dec 2026
Allow
export
No reviewed decision exists for export. Unknown does not become allow.
dup-2026.07 · reviewer — · no expiry recorded
Unknown

Granted access and what has actually been observed using it.

Scroll table sideways for more columns →

Principal	Kind	Grant	Observed use
grp-support-agents	group	read (source ACL)	Observed: 812 readers in 30 days
svc-index-builder	workload	read	Observed: nightly full read
uc-support-copilot	agent	read via delegated user	Observed: 41k retrievals in 30 days
grp-data-science	group	read	No observed access in 90 days

Constraints must be inherited by copies, embeddings and extracts.

Chunked retrieval corpus
der-chunks-eu · Split 800 tokens, overlap 120
ACLs and purpose limits carried to chunk level.
Constraints inherited
Legacy OCR attachment text
der-ocr-2024 · OCR, pre-2025 pipeline
Inheritance unproven. Treated as restricted until reviewed.
Constraints not inherited
De-identified evaluation sample
der-eval-sample · Tokenized identifiers, 20k rows
Transformation cannot lower sensitivity without review.
Constraints inherited

Training runs, retrieval indexes and evaluations that consume this dataset.

Scroll table sideways for more columns →

Kind	Consumer	Status
RAG index	support-eu-v12idx-support-eu	Fail
evaluation	Support QA suite Q3eval-support-q3	Pass

Scan coverage receipt v2026.09.3
ev-scan-77 · scan · producer collector-eu-03 · collected 24 Sep 2026, 02:00 UTC (4 h ago)
SignedValid to 24 Oct 2026, 02:00 UTC
Effective access snapshot restricted preview
ev-acl-19 · attestation · producer collector-eu-03 · collected 23 Sep 2026, 02:00 UTC (28 h ago)
SignedValid to 7 Oct 2026, 02:00 UTC