Threat intelligence / live workspace

Detect. Explain. Protect.

Analyze message signals with an explainable spam classifier built for fast, confident decisions.

Ready for analysis0 scans today

01 / INPUT

Email analysis

Clear

Sender email Subject Email body Optional headersOptional

Analyze email→ Load sample ↑Upload .EML

Paste from clipboard⌘ V

02 / DECISION

Detection result

LR · v1.4

0.0%spam probability

Awaiting analysis

Submit an email to reveal the model decision and its supporting evidence.

Why this result?

Model confidence—

Threshold50%

ModelLogistic regression

Model score—

03 / SIGNALS

Extracted features

Suspicious words—

—

Links detected—

—

Suspicious links—

—

Email length—

—

Word count—

—

Urgency indicators—

—

⌁Signals will appear here after analysis.

04 / EXPLANATION

Detection evidence

Why the model decided

◎No evidence yetRun an analysis to see the decision broken into clear signals.

05 / INSPECT

Message highlights

Annotated

Suspicious wordsRisky linksUrgency

Your analyzed message will be annotated here.

06 / INTERPRETATION

What influenced the prediction?

Local baseline + contextual weights, not guaranteed causal explanations

Analyze an email to compare feature influence.

07 / MULTIMODAL

Risk breakdown

Enhanced signals

Analyze an email to map text, URL, structure, and metadata risk.

08 / CONTENT SIGNALS

AI-assisted phishing indicators

Heuristic

—AI-content indicator

—

Indicators associated with automated or AI-assisted writing are not definitive proof of AI authorship.

Phishing risk—

Analyze an email to see separate phishing and AI-content signals.

Operational record

Analysis history

Review recent classifications and reopen any message for its evidence trail.

Clear history

AllSpamNot spamHigh risk

0 analyses

Time	Sender	Subject	Probability	Result	Open

◌No analyses match this filterYour completed scans will appear here.

Explainable machine learning

Model insights

A transparent look at how feature signals become a probability.

LOGISTIC REGRESSION · V1.4

MODEL FUNCTION

Sigmoid decision curve

σ(z) = 1 / (1 + e⁻ᶻ)

Model score—

Sigmoid output—

Spam probability—

Classification—

FEATURE VECTOR

Input features

X1 · Word frequency—

X2 · Link risk—

X3 · Email length—

w1 · vocabulary4.20

w2 · link risk3.10

w3 · length signal0.95

b · intercept−2.10

Weighted score→Sigmoid→Probability→Class

MODEL STORY

Baseline → enhancement

Baseline model3 features

Word frequency, link risk, and email length feed the existing sigmoid model.

→

Enhanced assessment10+ signals

Text, URLs, structure, metadata, explanations, and content indicators add context without replacing the baseline.

EVALUATION

Model metrics

Dataset required

Connect a labeled test set to calculate baseline-versus-enhanced precision, recall, F1, ROC-AUC, and PR-AUC. No performance numbers are fabricated in this demo.

Investigation canvas

Signal analysis

A focused view of the latest message and its extracted indicators.

Run latest analysis

⌁No message analyzed yetYour latest feature map will appear here after a scan.

MailGuard AI Spam Detector