Analysis
The fairness report
No score appears here without its arithmetic. Every factor carries its weight, its raw sub-score and the sentence that produced it, so you can argue with the method instead of the number.
Our Home · last 28 days
81.1
spread across the household: 7.3 points · 2 of 3 exactly on their share
Engine
griha-fairness/2026.10.1
computed 2026-10-03 07:48:07 UTC
The single next action
"Reseal the bathroom window" is 6 day(s) past due; Ishita is 11 points under their capacity share and should take the next one.
Who carries what
Per-member breakdown
Household score
Four weighted factors
Weights are hand-chosen and documented, not fitted — with a few dozen completions, fitting weights would be theatre. Every factor shows the raw sub-score it started from.
- Share parityweight 0.40 · raw 0.854 · 34.16 pts
Mean absolute deviation is 7.3 points across 3 member(s).
- Overdue pressureweight 0.25 · raw 0.571 · 14.29 pts
4 overdue chore(s), worst 6 day(s) late.
- Momentumweight 0.20 · raw 1.000 · 20.00 pts
14 completion(s) in 28 days against a target of 12.
- Effort honestyweight 0.15 · raw 0.845 · 12.67 pts
Recorded minutes differ from chore estimates by 7.8% on average.
Weather-shaped work
Outdoor suitability
- suitability 0.50workable
No forecast for this date; outdoor work is scored at the neutral 0.50.
- suitability 0.87good
20% rain, 3.1 m/s wind, 29.4°C mean → suitability 0.87
- suitability 0.94good
10% rain, 2.4 m/s wind, 30.2°C mean → suitability 0.935
- suitability 0.97good
5% rain, 2 m/s wind, 31°C mean → suitability 0.968
Both feeds are running on Griha's sealed offline sample. Nothing here is a live reading, and the suitability figures above are the fallback forecast, not today's weather.
Backlog
60 overdue minutes
Overdue work is weighted at 0.25 of the household score. It is the factor families feel fastest.
Head seal 27afa0d14bde3a21ef78c69a…
Open weights · on-device · TabPFN (Prior Labs)
Will it actually get done?
The fair-share engine above is deterministic: it knows who is carrying what. This asks a different question — given this chore, on this date, at this effort, assigned to this person at this point in their load, how likely is it to actually get done? It fits TabPFN v2 on 14 completed and 6 missed chore(s) from your own board. The weights are downloaded once, the training runs in this tab, and nothing about your household is sent anywhere.
Backend
not loaded
Precision
—
Training rows
20
Built with PriorLabs-TabPFN. Model weights are covered by the Prior Labs licence; the WebTabPFN runtime is Apache-2.0. Inference happens in your browser, so no completion data leaves this device.
The model has not been loaded yet, so no probabilities are shown. Nothing here is estimated or pre-filled.
The eight features TabPFN sees
- 1.
capacity - 2.
categoryShare - 3.
effortMinutes - 4.
weekday - 5.
daysSinceLastOfCategory - 6.
assigneeLoadShare - 7.
isPublicHoliday - 8.
outdoor