LinkedIn Behavioral Detection : Account Restricted

What Happened

LinkedIn restricted the account on 2026-03-25 for “use of software that automates activity.” Despite using CDP (undetectable at the DOM/network level), the behavioral pattern was inhuman: 62+ application attempts in one session, sub-second delays, repeated click retries, Tab-spamming through forms.

Root Cause

See definitions/root-cause-analysis for the analytical framework. Specific cause: technical undetectability (CDP pierce, keyboard nav) is necessary but not sufficient. LinkedIn’s detection is behavioral: it models human typing/clicking patterns and flags sessions that are statistically impossible for a human.

How to Avoid

10 rules : Gaussian distributions everywhere, never uniform Math.random(). Human behavior has specific statistical signatures that uniform randomness does not replicate.

1. Max 10-12 applications per day (cap is ~20, stay well under)
2. 5-15 min between submissions (Gaussian mean 8min, stddev 3min)
3. 1 retry max per job : skip permanently if click fails twice
4. Scroll + read description before applying (15-30s pause)
5. Limit search keywords: 5-8 per session, not all 17
6. Session: 2-3 hours max, then 2hr break
7. 10+ min between mining cycles
8. 300ms mean between keystrokes (Gaussian), 200-400ms between Tab presses
9. Stop ALL automation on auth_expired, 403, unusual redirects, captcha spike
10. Mouse movement before every click: Bezier curves, hover 300-600ms

Related

• projects/jobs-apply/_index : parent project
• ideas/2026-03-31-gaussian-timing-for-all-automation : idea that generalizes this pattern
• skills/behavioral-anti-detection : skill developed from this pitfall
• jobs-apply : project context
• Full plan: docs/LINKEDIN_REMEDIATION_PLAN.md