S?Scam Ba ’To?

Technical transparency

WHAT'S UNDER
THE VERDICT.

A compact technical view of the current Scam Ba ’To? intelligence stack, data pipeline, registry coverage, and validation corpus.

This page separates recognition from authentication, controlled tests from real incidents, and verified evidence from candidate data.

01 · Architecture

Decision pipeline.

01Input + privacy handlingmessage validation and sensitive-value handling
02Normalization + semantic interpretationlanguage, intent, event, and requested-action context
03Identity + destination analysisclaimed organization, URLs, domains, and destination integrity
04Transaction + Action RiskOTP, payment, transfer, account, enrollment, delivery, government-service context
05Verified Channel Intelligenceorganization policies, known practices, destinations, and source-backed facts
06Evidence fusion + safety gateRED / YELLOW / GREEN with uncertainty preserved

02 · Data lineage

Verified data is earned, not assumed.

Automatically collected records do not become trusted links simply because they came from a directory.

AUTHORITATIVE SOURCE↓BRONZE / RAW↓SILVER / CANDIDATE↓CROSS-CHECK + REVIEW↓GOLD VERIFIED↓RUNTIME
Silver can help recognize an organization. Only verified Gold intelligence can supply an official destination to the user.

03 · Current data snapshot

Registry and corpus.

77Gold verified entities
648Silver recognition entities
716organization intelligence profiles
186verified destinations
211verified facts
94source-grounded cases

04 · Coverage

Largest intelligence sectors.

Financial176
Education144
Government112
Utilities67
Health51
Recruitment44
NGO / Charity41
Courier15
Telecom14

05 · Benchmark anatomy

Raw volume is not diversity.

48,000controlled regression executions
→
16,447exact-unique messages
→
6,144behavioral representatives

These figures are intentionally shown separately. Controlled variants are useful for regression testing, but they must not be presented as independent real-world scam reports.

06 · Validation

What the engine is challenged with.

RegressionKnown behavior must remain correct after changes.

Hard negativesLegitimate messages with alarming words, OTPs, payments, or security notices.

AdversarialMessages deliberately written to evade or confuse simple rules.

MetamorphicBehavior preserved while wording, language, noise, or presentation changes.

Source-groundedRepresentative cases tied to documented organization guidance.

07 · Decision standard

Three verdicts. No “100% safe.”

RED

Material evidence of dangerous or deceptive behavior.

YELLOW

A critical fact cannot be resolved safely from the message alone.

GREEN

No material danger observed from available evidence. Not proof of legitimacy.

08 · Limits

What this system cannot know.

It cannot know whether you personally initiated a transaction unless that context is supplied. It cannot authenticate an SMS sender from message text alone, and an official-looking URL does not automatically make the entire message legitimate.

Organization intelligence is evidence, not endorsement. Scam tactics and official communication practices can change, so verified information requires continued review.

Back to the public story

Technical enough?

The About page explains the same system without requiring the reader to think like an engineer.

Back to About →