Detect spam split across messages #3

Merged

backmeupplz merged 1 commit from fix/contextual-spam-detection into main

2026-09-19 01:25:06 +00:00

backmeupplz commented

2026-09-19 01:17:01 +00:00

(Migrated from github.com)

Summary

keep a bounded 10-minute per-chat/per-sender message history with physical expiry
classify current messages with recent sender context and a dedicated multi-message spam signal
after the 0.90 aggregate gate, delete only the contiguous recent suffix whose per-message linkage reaches 0.75
treat edited messages as replacements and reclassify them with context

Validation

bun run check: 13 passed, 5 live tests skipped without credentials
live credentialed suite: 5 passed, including the reported Russian split-spam sequence, a legitimate multi-message control, and mixed legitimate+spam history
git diff --check

Deployment

Redeploy the existing Easypanel service bots/jev-antispam-bot; no config or secret changes.

## Summary - keep a bounded 10-minute per-chat/per-sender message history with physical expiry - classify current messages with recent sender context and a dedicated multi-message spam signal - after the 0.90 aggregate gate, delete only the contiguous recent suffix whose per-message linkage reaches 0.75 - treat edited messages as replacements and reclassify them with context ## Validation - `bun run check`: 13 passed, 5 live tests skipped without credentials - live credentialed suite: 5 passed, including the reported Russian split-spam sequence, a legitimate multi-message control, and mixed legitimate+spam history - `git diff --check` ## Deployment Redeploy the existing Easypanel service `bots/jev-antispam-bot`; no config or secret changes.

backmeupplz commented

2026-09-19 01:24:55 +00:00

(Migrated from github.com)

Independent GPT-5.6 Sol review — approved exact head 97c50112695f09e3d678c46072a5bb560e417d22; no blockers.

Confirmed: aggregate contextual-spam gate remains 0.90; only the contiguous recent suffix linked at >=0.75 is deleted; unrelated-message boundary, physical TTL pruning, per-chat/per-sender isolation, 10-message cap, edit replacement, fail-open behavior, and privacy-safe logging are correct. Validation: bun run check 13 passed, credentialed live suite 5 passed, git diff --check passed.

Independent GPT-5.6 Sol review — **approved** exact head `97c50112695f09e3d678c46072a5bb560e417d22`; no blockers. Confirmed: aggregate contextual-spam gate remains 0.90; only the contiguous recent suffix linked at >=0.75 is deleted; unrelated-message boundary, physical TTL pruning, per-chat/per-sender isolation, 10-message cap, edit replacement, fail-open behavior, and privacy-safe logging are correct. Validation: `bun run check` 13 passed, credentialed live suite 5 passed, `git diff --check` passed.