The Big Picture
Canthus uses a single metaphor: mana. It represents both the energy a person has today and the energy a task will require today. The system has three core jobs:- Estimate today’s budget (mana pool)
- Cost each task (mana cost)
- Show only what’s achievable (with a narrow essential-task exception)
| Signal | What it captures | Mechanism | Changes how often |
|---|---|---|---|
| Condition severity | How ill the person is overall | Personal coefficient | Over weeks/months |
| Daily variation | ”Today I feel worse/better than usual” | Axis factor | Daily |
| Spending pattern | What the user actually accomplishes | Pool EWMA | Over days/weeks |
Core concepts
Mana
An abstract unit of energy. The absolute number is less important than the ratio between pool and task costs.
Mana pool
Daily budget, typically between 8 and 130 depending on severity and history.
Mana cost
The cost of a specific task for this person on this day.
Check-in
Daily body, mind, mood ratings that supply context for today’s costs.
Relative cost
Effort of an activity relative to personal baseline.
Personal coefficient
Long-term severity multiplier that scales all task costs.
Axis factor
Daily multiplier from check-in that modulates costs for today.
Essential tasks
User-defined tasks that remain visible even when over pool.
System map
- User checks in (full, quick, momentum, or none)
- Engine derives axis factor from body/mind state and task weights
- Engine computes task costs from relative cost, duration, coefficient, and axis factor
- Engine computes daily pool from spending EWMA and confidence bias
- Engine surfaces only tasks that fit remaining pool, except essentials and temporal horizon exemptions
- Detection systems monitor PEM and longer pattern shifts
- Calibration updates slowly to preserve stability and trust
Where to go deeper
Onboarding
Calibration startup and cold start behavior.
Task costing and budget
Formulas, coefficients, axis factor, pool confidence model.
Temporal tasks and surfacing
Deadlines, scheduled tasks, recurrence, horizon windows.
Detection systems
PEM, regime shift, cumulative load, under-rating detection.
Example scenarios
Worked examples with concrete numbers.
Data model
Tables, key fields, storage model, and constraints.
Versioning, migration and rollback
Safe algorithm evolution rules.
Design principles and user contract
Why the system is framed this way.