Dataset Viewer
Auto-converted to Parquet Duplicate
trace_id
stringlengths
36
36
timestamp
stringlengths
32
32
input
stringlengths
16
162
input_category
stringclasses
6 values
urgency
bool
2 classes
scam_tactics
stringclasses
7 values
result_summary
stringlengths
157
202
risk_label
stringclasses
5 values
reply_draft_policy
stringclasses
3 values
412911fd-6973-4f2a-adf8-2eb273adce5f
2026-06-07T07:06:03.455103+00:00
text: PAKISTAN POST: [NAME_OR_ENTITY] parcel [ADDRESS]. [NAME_OR_ENTITY] Rs. 85 today at [LINK] or the parcel will be destroyed.
courier
true
link, urgency, payment, courier
Likely scam: Courier-style content with link, urgency, payment, courier signals. Known courier pattern. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
913f15bb-bc1a-45bb-96ec-2cf711bb43a7
2026-06-07T07:06:45.233617+00:00
text: I hate you
unknown
false
none
Inappropriate: Unclassified content with no mapped signals. Unclassified pattern; this does not confirm a new scam type. The content was not suitable for a scam-risk assessment.
Inappropriate
suppressed
ea76c223-62f1-428e-bb52-bec267897050
2026-06-07T07:07:22.167475+00:00
text: FBR REFUND: [NAME_OR_ENTITY] your CNIC and bank card [REDACTED] at the link today to receive payment.
fbr
true
cnic, credentials, urgency, payment, refund_or_prize
Likely scam: FBR-style content with CNIC, credential, urgency, payment signals. Known FBR pattern. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
e3236504-f1fb-449c-87fe-d035b8ccf066
2026-06-07T13:19:30.610210+00:00
image: Unclassified content with no mapped signals
unknown
false
none
Likely scam: Unclassified content with no mapped signals. Unclassified pattern; this does not confirm a new scam type. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
1b66b7e7-1dd5-4ddd-a1df-1b6d48054a9b
2026-06-07T13:40:58.958579+00:00
image: Unclassified content with no mapped signals
unknown
false
none
Looks normal: Unclassified content with no mapped signals. Unclassified pattern; this does not confirm a new scam type. No strong scam indicators were found, but verify through an official channel.
Looks normal
suppressed
5473e89f-2839-4142-be7f-ab7a13562f33
2026-06-07T15:44:20.441519+00:00
image: Unclassified content with no mapped signals
unknown
false
none
Likely scam: Unclassified content with no mapped signals. Unclassified pattern; this does not confirm a new scam type. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
eee6c44b-09c8-458c-b778-e991def4386e
2026-06-07T15:45:37.207160+00:00
image: Traffic-challan-style content with link, urgency, payment, challan signals
traffic_challan
true
link, urgency, payment, challan
Likely scam: Traffic-challan-style content with link, urgency, payment, challan signals. Known traffic-challan pattern. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
3b61003c-42e7-4703-9096-5ecd2e1dddd4
2026-06-07T15:47:28.227702+00:00
image: Unclassified content with no mapped signals
unknown
false
none
Likely scam: Unclassified content with no mapped signals. Unclassified pattern; this does not confirm a new scam type. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
33fb2d0f-2cb4-474d-bfd3-d780ec09d818
2026-06-07T15:48:49.131381+00:00
image: Unclassified content with no mapped signals
unknown
false
none
Likely scam: Unclassified content with no mapped signals. Unclassified pattern; this does not confirm a new scam type. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
fcf6608b-d866-4ed2-85c3-914e9c3fb71d
2026-06-07T15:51:06.399305+00:00
image: Unclassified content with no mapped signals
unknown
false
none
Likely scam: Unclassified content with no mapped signals. Unclassified pattern; this does not confirm a new scam type. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
57ce6d34-4ead-43fc-87a8-92576cf6353e
2026-06-07T15:52:29.495745+00:00
image: Unclassified content with no mapped signals
unknown
false
none
Looks normal: Unclassified content with no mapped signals. Unclassified pattern; this does not confirm a new scam type. No strong scam indicators were found, but verify through an official channel.
Looks normal
suppressed
15e98cbc-f62c-4e41-8a9a-6fc46fa56b5b
2026-06-07T15:53:15.261370+00:00
text: [NAME_OR_ENTITY] me, you duffer
unknown
false
none
Inappropriate: Unclassified content with no mapped signals. Unclassified pattern; this does not confirm a new scam type. The content was not suitable for a scam-risk assessment.
Inappropriate
suppressed
df7bbd08-eae8-4ead-b71f-a63dedbb9256
2026-06-07T19:37:54.494734+00:00
image: Marketplace-style content with credential signals
marketplace
false
credentials
Likely scam: Marketplace-style content with credential signals. Known marketplace pattern. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
4e400902-dc3b-4136-9bc5-dc7c48e4b82a
2026-06-07T20:00:21.182508+00:00
image: Unclassified content with no mapped signals
unknown
false
none
none: Unclassified content with no mapped signals. Unclassified pattern; this does not confirm a new scam type. No completed assessment result was available.
none
not_applicable
0e083b7f-cd16-4908-977d-f09f7bf3fe23
2026-06-07T20:00:39.424146+00:00
image: Marketplace-style content with credential signals
marketplace
false
credentials
Likely scam: Marketplace-style content with credential signals. Known marketplace pattern. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
49d7a51c-ad65-422b-a258-895297a08e03
2026-06-07T20:00:46.830258+00:00
image: Unclassified content with no mapped signals
unknown
false
none
none: Unclassified content with no mapped signals. Unclassified pattern; this does not confirm a new scam type. No completed assessment result was available.
none
not_applicable
e2d13141-d232-4cb2-bef9-bfe8d6b5cf02
2026-06-07T20:16:52.510221+00:00
text: shut up, you bitch
unknown
false
none
Suspicious: Unclassified content with no mapped signals. Unclassified pattern; this does not confirm a new scam type. Multiple warning signs were found; use caution and verify independently.
Suspicious
allowed
6d85b5a6-f66e-4aa9-8478-37fa10145e92
2026-06-07T07:01:20.917037+00:00
text: PAKISTAN POST: [NAME_OR_ENTITY] parcel [ADDRESS]. [NAME_OR_ENTITY] Rs. 85 today at [LINK] or the parcel will be destroyed.
courier
true
link, urgency, payment, courier
Likely scam: Courier-style content with link, urgency, payment, courier signals. Known courier pattern. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
23bbdea8-824b-4b01-a797-de52813c94a3
2026-06-07T07:01:20.917210+00:00
text: FBR REFUND: [NAME_OR_ENTITY] are eligible for Rs 42,[NUMBER]. [NAME_OR_ENTITY] your CNIC and bank card [REDACTED] at the link today to receive payment.
fbr
true
cnic, credentials, urgency, payment, refund_or_prize
Likely scam: FBR-style content with CNIC, credential, urgency, payment signals. Known FBR pattern. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
17558327-848a-4746-a847-1a05b0b6d143
2026-06-07T07:01:20.917356+00:00
text: HBL [NAME_OR_ENTITY]: [NAME_OR_ENTITY] account [REDACTED] be suspended. [NAME_OR_ENTITY] the OTP [REDACTED] to your phone with our support team immediately.
bank
true
otp, urgency, account_threat
Likely scam: Bank-style content with OTP, urgency, account-threat signals. Known bank pattern. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
ce27c0cb-878a-44e7-bcc5-b8c3f39620e0
2026-06-07T07:01:20.917462+00:00
image: Courier-style content with link, urgency, courier signals
courier
true
link, urgency, courier
Likely scam: Courier-style content with link, urgency, courier signals. Known courier pattern. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
afc06c6e-a510-42e5-8a69-d8c0734a82cd
2026-06-07T07:01:20.917817+00:00
image: Marketplace-style content with credential signals
marketplace
false
credentials
Likely scam: Marketplace-style content with credential signals. Known marketplace pattern. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed
95dacd6e-3947-40d2-ae1f-6e5bc709081c
2026-06-07T07:01:20.917983+00:00
image: Traffic-challan-style content with link, urgency, payment, challan signals
traffic_challan
true
link, urgency, payment, challan
Likely scam: Traffic-challan-style content with link, urgency, payment, challan signals. Known traffic-challan pattern. Strong scam indicators were found; avoid payments, links, and sharing credentials.
Likely scam
suppressed

Pakistan Notice Helper Privacy-Safe Traces

Purpose

This dataset contains compact, deterministic metadata about Pakistan Notice Helper scam-check requests. It does not contain hidden model reasoning or autonomous-agent trajectories.

The application uses a Modal-hosted Qwen model for normal assessments. Creating a trace never calls an AI model. Traces only observe the existing request path and convert it into allow-listed categories, booleans, buckets, and counts.

Fields

  • Trace identity: random trace ID and UTC timestamp
  • input: aggressively redacted, length-limited text for text submissions, or a fixed-template image: ... description
  • input_category: deterministic category such as courier, bank, or FBR
  • urgency: deterministic boolean urgency signal
  • result_summary: deterministic summary of the mapped pattern, whether it is known or unclassified, and why the risk label matters
  • scam_tactics: readable comma-separated tactics
  • Flat result columns such as risk_label and reply_draft_policy

Every dataset cell is a scalar string, number, or boolean. No column contains a dictionary or nested object, which keeps the Hugging Face table easy to read. trace_id is the first serialized column, and all detected boolean signals are combined into the single scam_tactics category column.

Records do not contain pipeline steps, cache fields, app commits, failure details, request source, schema versions, size buckets, language hints, or Modal metadata.

Privacy

The dataset never stores:

  • Raw message text
  • Screenshots, image bytes, or base64
  • URLs, phone numbers, CNICs, names, addresses, account/card numbers, or tracking numbers
  • Model explanations, red flags, safe-step text, reply drafts, or raw model output
  • Exceptions, credentials, tokens, or endpoint headers

Text traces store an aggressively redacted form capped at 500 characters. Regexes remove common URLs, emails, phones, CNICs, card/account numbers, credentials, addresses, tracking IDs, long numbers, and title-case names or entities. Regex redaction cannot guarantee removal of every possible identifier, so users should opt out when submitting sensitive content.

Images store only fixed descriptions; screenshots and OCR text are not stored. Users see a checked trace disclosure in the app and may opt out before each request.

Provenance

Seed traces represent the six public examples bundled with Pakistan Notice Helper. Runtime traces may represent successful, rejected, or failed requests. Trace generation itself does not invoke the model.

Limitations

  • Regex signals and category detection are approximate.
  • Regex redaction may miss unusual personal or confidential information.
  • Novelty is not researched against external threat-intelligence sources.
  • The dataset cannot reproduce original messages or screenshots.
  • A risk label is safety guidance, not official verification.

Links

License

CC BY 4.0.

Downloads last month
-

Space using build-small-hackathon/pakistan-notice-helper-traces 1