meeting-load-balancer

Meeting Load Balancer

Overview

Knowledge workers lose 10–20+ hours a week to meetings, and almost nobody measures it. Calendars fill by accretion: each recurring meeting was individually justifiable in isolation and collectively catastrophic. The damage isn't only the summed hours — it's fragmentation: a day with 4 hours of meetings scattered into 45-minute gaps has zero deep-work blocks left, and a team's most expensive people are often its most double-booked.

This skill turns an exported calendar (.ics) into an evidence-based meeting audit:

1. Import — parses standard ICS files (Google/Outlook/Apple all export it), expands recurring events over a date window, dedupes cancelled occurrences.
2. Measure — hours/week total, per weekday, per meeting series, per attendee; double-booked overlaps; meeting-free gaps and the longest focus block per day.
3. Fragmentation score — counts usable focus blocks (≥90 uninterrupted minutes) destroyed by meeting placement, per day and per week. 4 hours of meetings can cost 8 hours of focus depending on placement.
4. True-cost ranking — each series scored by hours × attendees × frequency with placement damage, producing the "this standing meeting costs the team a work-week per month" list.
5. Prescriptions — explicit rules flag meetings to KILL (low attendance, no decision record), SHORTEN (25/50-min default), DE-DENSITY (weekly→biweekly), BATCH (scattered→clustered), or MAKE-ASYNC (status-shaped), plus a simulate mode that shows hours and focus blocks recovered if you applied the changes.

Everything runs offline on calendar files you control; nothing is uploaded.

When to Use

• Personal calendar audit — "where did my week go?" → stats, focus
• Before proposing a new recurring meeting — check the team's existing load first
• Team retro / effectiveness review — aggregate several people's exports → team
• Defending focus time — find the batchable/blocked windows → focus --advice
• Killing a standing meeting with evidence — rank then simulate the removal
• Don't use for: room/equipment scheduling (different problem), or individual-event time zone archaeology (the parser handles standard ICS but exotic recurrence rules may be simplified — always sanity-check counts against your calendar app).

Commands

# Export your calendar first (see references/ics-export.md), then:
python3 scripts/meeting_load.py stats --ics calendar.ics --weeks 4
python3 scripts/meeting_load.py stats --ics calendar.ics --weeks 4 --person "you@corp.com"

# Focus-time analysis (fragmentation, longest blocks, batchable windows)
python3 scripts/meeting_load.py focus --ics calendar.ics --weeks 4

# Rank every recurring series by true cost
python3 scripts/meeting_load.py rank --ics calendar.ics --weeks 4

# Rule-based prescriptions: kill / shorten / de-density / batch / async
python3 scripts/meeting_load.py prescribe --ics calendar.ics --weeks 4

# Simulate applying the prescriptions (before arguing about it)
python3 scripts/meeting_load.py simulate --ics calendar.ics --weeks 4

# Multi-person: combine exports (cat them or pass --ics twice)
python3 scripts/meeting_load.py team --ics alice.ics --ics bob.ics --weeks 4

# Worked example without any calendar
python3 scripts/meeting_load.py demo

Metrics Definitions

Working hours default 09:00–17:00 local, Mon–Fri (--day-start 9 --day-end 17 to change). Multi-day and all-day events are excluded from meeting metrics by default.

What Prescriptions Look Like

PRESCRIPTIONS (12 meetings analyzed, 4 series):

KILL      'Weekly Status Sync'        60min × 8 ppl × 1.0/wk — status-shaped,
          avg 53% accepted, no decisions logged → replace with written update
SHORTEN   'Design Review'             60→45min — 62% of occurrences end early
          (median actual use 40min)
DE-DENSITY '1:1 Manager'              weekly→biweekly — 4/4 weeks had
          cancellation/reschedule
BATCH     'sprint-planning' Mon 14:00 + 'retro' Thu 11:00 + 'estimation' Tue 10:00
          → same day cluster frees 2 full days/week of ≥3h blocks
ASYNC     'Incident Readout'          12 occurrences, avg 6 min used of 30,
          1 speaker — record/video instead

simulated effect: −5.6 meeting hrs/week, +2.1 focus blocks/week for the team

Common Pitfalls

1. Exporting too narrow a window. Recurring meetings reveal their sins over 4+ weeks (cancellations, attendance decay). Use --weeks 4 minimum; 8 is better.
2. Counting declined events. The tool excludes events you declined where the ICS records it; re-check surprises with list --day before believing a number.
3. Treating hours as the metric. Two calendars with identical meeting hours can differ 2× in usable focus time. Always read focus alongside stats — placement, not just volume, is the enemy.
4. Forgetting attendee cost. A 30-min meeting for 10 people is 5 person-hours; the rank output exists so the "quick sync" gets priced honestly.
5. Shortening everything reflexively. Some 60s earn their hour (decisions with real debate). The rules flag evidence of waste (early endings, decay, status-shape), not length per se.
6. All-day/multi-day events polluting stats. They're excluded by default; if your org encodes real meetings as all-days (some do), --include-allday.

Verification Checklist

• ICS export covers ≥4 weeks and includes recurring expansions
• stats --person matches what your calendar app reports for a sample week
• focus reviewed — fragmentation is usually the hidden finding
• simulate run before proposing changes; numbers shown to attendees
• Changes applied incrementally (kill one, measure again next month)

One-Shot Recipes

Personal "where did March go" audit

python3 scripts/meeting_load.py stats --ics march.ics --weeks 4
python3 scripts/meeting_load.py focus --ics march.ics --weeks 4

The 'we should add a weekly sync' moment

python3 scripts/meeting_load.py team --ics a.ics --ics b.ics --ics c.ics --weeks 4
# if anyone is >12 hrs/wk or <1 focus block/day → propose async instead

Annual meeting purge

python3 scripts/meeting_load.py rank --ics q3.ics --weeks 8
python3 scripts/meeting_load.py simulate --ics q3.ics --weeks 8