Evaluation reportGenerated by the project's CI evaluation on held-out data.About the projectCode on GitHub

Meeting operations: evaluation

Scheduling across six time zones, and action items from transcripts. Settings tuned on separate organisations and phrasing.

75.6%fewer night and weekend meeting hours (17 vs 68 a week)
1680 of 1680requests placed in held-out organisations
0.72owner-and-task F1 on unseen phrasing (rules: 0.00)
96.4%of action items extracted or put in front of a reviewer

Scheduling

OrganisationsPolicyOut-of-hours h/weekNight h/weekWorst-off personFocus lostOptional can come
Held-out organisationsFirst free slot (organiser's hours)108.067.94.24 h2.37 h32.9%
Held-out organisationsFirst slot in everyone's hours103.766.14.06 h2.53 h34.5%
Held-out organisationsBalanced (this platform)93.016.63.85 h2.37 h39.6%
Other time zones, fuller calendarsFirst free slot (organiser's hours)122.075.94.28 h1.87 h34.9%
Other time zones, fuller calendarsFirst slot in everyone's hours118.774.44.28 h2.02 h35.2%
Other time zones, fuller calendarsBalanced (this platform)111.320.44.13 h1.73 h36.9%

Action items: owner-and-task F1 (with due date)

PhrasingKeyword searchHand-written rulesLearned readerSurfaced
Familiar phrasing (rules built on)0.173 (0.173)1.000 (1.000)1.000 (1.000)100.0%
Familiar phrasing, new meetings0.170 (0.170)1.000 (1.000)1.000 (1.000)100.0%
New phrasing A (validation)0.128 (0.095)0.000 (0.000)0.752 (0.549)92.6%
New phrasing B (test)0.112 (0.056)0.000 (0.000)0.722 (0.335)96.4%

Gate (regression limits)

CheckValueLimitResult
heldout_night_vs_first_slot_max0.2440.3pass
fresh_night_vs_first_slot_max0.2690.33pass
heldout_out_of_hours_vs_first_slot_max0.8610.92pass
heldout_placed_share_min1.00.99pass
heldout_learned_f1_min1.00.99pass
fresh2_learned_owner_task_f1_min0.7220.68pass
fresh2_surfaced_min0.9640.93pass