The Filter reads line by line
Type or paste an email and a naive-Bayes spam filter grades it live, line by line — every trigger word highlighted, a deliverability score from 0–100 updating on each keystroke, and a predicted inbox / promotions / spam placement. Words like free, winner, and guaranteed push the score up; meeting, invoice, and regards pull it down. Type your worst subject line on purpose and watch the score crater. Runs entirely in your browser (0 uploads, works offline).
How to use it
- Type or paste your email subject and body into the editor.
- Watch the spam score and predicted placement update on every keystroke.
- Read the per-line grades below the editor — each triggering word is chipped with its weight (red = spammy, green = hammy).
- Load worst subject line or clean work email to see the two extremes, then open the reveal to see the exact rules.
What this clears up (the fundamentals)
- A spam filter is not magic — it's a lexicon and a sum. A naive-Bayes classifier gives every word a learned log-odds weight, adds up the weights of the words that fired, and squashes the total through a sigmoid to a probability. That's the whole mechanism.
- The score is per-word, so it's per-line. Each line contributes its own signed score; one greedy subject line can outweigh a whole paragraph of legitimate copy.
- Trigger words are real, but context wins. "Free" alone won't sink you — it's the sum. Balance a promotional word against genuine, specific content and the total comes back down.
- You can out-write the filter because you can see it. The normally invisible rule engine is fully exposed here, so you edit against the math instead of guessing after a bounce report.
Where it's used
A free public spam-score checker — the fastest way to feel why your campaign landed in Promotions before you ever hit send. It's a miatz build-lab concept playable: learn the naive-Bayes mechanic here, then build the deliverability scorer yourself in the lab.
FAQ
What is a naive-Bayes spam filter?
A classifier that treats an email as a bag of words. Each word carries a log-odds weight learned from labelled spam and ham; the filter sums the weights of the words present and converts the total to a spam probability with the logistic (sigmoid) function. "Naive" because it assumes the words are independent.
Why is my legitimate email flagged as spam?
Because the sum of your word weights is high, not because any single word is banned. A subject stuffed with "free", "winner", and "urgent" can outweigh a body full of specific, useful content. Rebalance toward concrete detail and the total drops.
Do real filters only use trigger words?
No — production filters also weigh sender reputation, SPF/DKIM/DMARC authentication, and link-to-text ratio. This playground isolates the word-scoring core so the mechanic is visible; the same log-odds-and-sigmoid math sits underneath the rest.
Is anything uploaded?
No. Your email text is tokenized and scored entirely in your browser against a fixed lexicon — nothing is transmitted, logged, or stored. Turn off your Wi-Fi and it still works.
Limits
A teaching model with a fixed 16-word lexicon and hand-set weights — real filters train on millions of messages and weigh authentication, reputation, and links too. The mechanism (per-word log-odds, summed, through a sigmoid) is exactly the real practice; the specific numbers are illustrative.
Related
Part of the Demystify Playgrounds. Explore the rest from the Playgrounds home.
Bookmark this page (Ctrl+D, or ⌘D on Mac) — it works offline the next time you need it.