Tuning the Scoring Engine: Weighting Each Layer

Spamify’s decision is only as good as the score behind it — and the score is a weighted sum of what each layer found. The weights setting lets you decide how much each signal matters, so the final number reflects your risk tolerance.

Why weighting matters

Consider two submissions:

• One from an address whose domain has no mail server at all — a near-certain dead end.
• One that merely came in a little faster than your timing threshold — suspicious, but not proof.

These shouldn’t contribute equally to the score. Weighting lets the high-confidence signal push hard toward blocking while the soft signal only nudges the submission toward review.

How weights feed the thresholds

Each layer produces a result; its weight scales that result’s contribution to the total. The total is then compared against your flag and block thresholds. So weights and thresholds are two halves of one dial:

• Weights shape how the score is built from signals.
• Thresholds decide what the finished score triggers.

A sensible weighting strategy

1. Give high confidence to definitive signals — a confirmed-dead mailbox or a resolutely invalid domain.
2. Give moderate weight to structural signals — syntax problems that strongly suggest fakery.
3. Give low weight to soft heuristics — a marginally fast fill time, a borderline content flag governed by content_block_threshold.

Change one weight at a time, then watch the dashboard to see how the distribution of blocked vs. flagged shifts before adjusting the next. Small, observed changes beat a big guess.