This dataset contains compact, deterministic metadata about NoticeCheck message-review requests. It does not contain hidden model reasoning or autonomous-agent trajectories.
The hosted application uses MiniCPM5-1B through Transformers on Hugging Face ZeroGPU, with NVIDIA Nemotron-Parse v1.2 for supported screenshots. The same pipeline can run locally on an NVIDIA GPU with Docker Compose. Creating a trace never makes an additional AI model call. Traces only observe the existing request path and convert it into allow-listed categories, booleans, and fixed descriptions. For image submissions, the existing assessment's explanation and red flags are inspected transiently for this mapping, but their text is not stored. The trace mapper predicts the privacy-safe image category and tactics directly from that result summary using deterministic English/Urdu evidence rules.
input: aggressively redacted, length-limited text for text submissions, or
a fixed-template image: ... descriptioninput_category: deterministic category such as courier, bank, or FBRurgency: deterministic boolean urgency signalresult_summary: deterministic summary of the mapped pattern, whether it is
known or unclassified, and why the risk label mattersscam_tactics: readable comma-separated tacticsrisk_label and reply_draft_policyEvery 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 model runtime metadata.
The dataset never stores:
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, OCR text, model explanations, and model red flags are not stored. Users see a checked trace disclosure in the app and may opt out before each request.
When image analysis fails before an assessment is available, new records use
image: Assessment unavailable. This keeps service failures separate from
successful assessments whose content is genuinely unclassified.
The default configuration contains privacy-safe runtime traces and may include
successful, rejected, or failed requests. The examples configuration contains
the six reviewed public examples bundled with NoticeCheck.
Trace generation itself does not invoke the model.
The seed rows are illustrative examples, not an evaluation split. All six
currently have the Likely scam label, so they must not be used to estimate
class balance, accuracy, recall, or real-world scam prevalence.
Default runtime rows intentionally preserve repeated requests. Consequently,
repeated examples, unclassified image descriptions, and incomplete none
assessments may be common. For training or evaluation, curate the default
configuration by:
risk_label: noneinput and result columnsThe source repository includes traces/scripts/analyze_trace_dataset.py for
schema validation and a reproducible summary of these quality indicators.
CC BY 4.0.
This dataset contains compact, deterministic metadata about NoticeCheck message-review requests. It does not contain hidden model reasoning or autonomous-agent trajectories.
The hosted application uses MiniCPM5-1B through Transformers on Hugging Face ZeroGPU, with NVIDIA Nemotron-Parse v1.2 for supported screenshots. The same pipeline can run locally on an NVIDIA GPU with Docker Compose. Creating a trace never makes an additional AI model call. Traces only observe the existing request path and convert it into allow-listed categories, booleans, and fixed descriptions. For image submissions, the existing assessment's explanation and red flags are inspected transiently for this mapping, but their text is not stored. The trace mapper predicts the privacy-safe image category and tactics directly from that result summary using deterministic English/Urdu evidence rules.
input: aggressively redacted, length-limited text for text submissions, or
a fixed-template image: ... descriptioninput_category: deterministic category such as courier, bank, or FBRurgency: deterministic boolean urgency signalresult_summary: deterministic summary of the mapped pattern, whether it is
known or unclassified, and why the risk label mattersscam_tactics: readable comma-separated tacticsrisk_label and reply_draft_policyEvery 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 model runtime metadata.
The dataset never stores:
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, OCR text, model explanations, and model red flags are not stored. Users see a checked trace disclosure in the app and may opt out before each request.
When image analysis fails before an assessment is available, new records use
image: Assessment unavailable. This keeps service failures separate from
successful assessments whose content is genuinely unclassified.
The default configuration contains privacy-safe runtime traces and may include
successful, rejected, or failed requests. The examples configuration contains
the six reviewed public examples bundled with NoticeCheck.
Trace generation itself does not invoke the model.
The seed rows are illustrative examples, not an evaluation split. All six
currently have the Likely scam label, so they must not be used to estimate
class balance, accuracy, recall, or real-world scam prevalence.
Default runtime rows intentionally preserve repeated requests. Consequently,
repeated examples, unclassified image descriptions, and incomplete none
assessments may be common. For training or evaluation, curate the default
configuration by:
risk_label: noneinput and result columnsThe source repository includes traces/scripts/analyze_trace_dataset.py for
schema validation and a reproducible summary of these quality indicators.
CC BY 4.0.