BEHAVIORAL LAYER
Mirror
The layer that tunes the team to you.
How you use Eidros — what you click, how long you dwell, what you dismiss, when you show up. Behavioral metadata only, never the contents of your work.
Why Mirror exists
Every other agent produces signals. Mirror decides which of them actually matter to you. Without Mirror, the feed is an equal-weighted river — a user who lives in Pulse and never opens Ledger drowns in Ledger noise. Mirror closes that loop.
Routines
What Mirror does every cycle.
Behavioral capture
Events flow from the Chrome extension + server-side logging. app_switch, dwell, finding_click, finding_dismiss, finding_act, friction.
Weight profile
Per-user row: peak_decision_hours, domain_weights, agent_engagement, friction_map, alert_response_time, digest_preferences, confidence_score.
autoDream consolidation
Idle-time event rollup. Can't churn the profile per-event; consolidates periodically.
Re-ranking
Every other agent's output is re-ranked against the profile before landing in your feed.
Confidence gating
Requires ≥7 days of data before confidence_score exceeds 0.6. Free tier exists for this calibration.
DATA SOURCES
- →Chrome extension (mirror-extension/)
- →Server-side dashboard interaction logs
- →mirror_events table
MODEL USE
No Claude calls — pure data-driven rollup
WHAT IT EMITS
- →weight_profiles row updates
- →Re-ranked feed order
- →Friction findings (workflow areas slowing you down)
The rest of the team