Implementation TODO

​

Current state

• ManaRepository and DriftManaRepository assemble ManaDayInputs
• application actions and Riverpod providers exist for input and compute paths
• engine-facing models are defined
• the engine sub-services and test harness exist

​

Engine implementation

• Implement AxisFactorCalculator behavior from the check-in rules
• Implement DailyPoolCalculator with cold start, conservative bias, EWMA updates, and explicit clamping
• Implement TaskCostCalculator so per-task costs are derived deterministically from task properties and the current axis factor
• Implement TaskSurfacingService so essentials, timing constraints, low-capacity rules, and actionability filtering follow spec
• Implement ReturnRecoveryService for the neutral return-after-absence behavior
• Implement PemDetector using the documented thresholds and confidence rules
• Implement RegimeShiftDetector for sustained baseline changes
• Implement ManaTransparencyBuilder so explainers match actual deterministic behavior
• Wire the above into ComposedManaEngine.compute so ComputeManaDay and WatchComputedManaDay return real ManaDayResult values

​

Event and learning semantics

• Decide which parts of learning depend only on current state and which must be event-replay-driven
• Start emitting and consuming dayTypeSet with opportunityWeight where end-of-day learning uses it
• Start emitting and consuming taskCostCalibrated when the task-cost override flow exists
• Start emitting and consuming taskRecoveryWeightUpdated if recovery weights become learned state rather than static fields
• Start emitting and consuming pemDetected / pemResolved once PEM detection becomes real
• Ensure any learned state can still be reconstructed deterministically from persisted inputs and events

​

Determinism and algorithm runtime

• Keep the mana engine pure for a fixed ManaDayInputs bundle
• Add deterministic maths helpers for EWMA, clamping, bounded updates, and safe rounding where needed
• Implement algorithm versioning/runtime migration support as described in Versioning, Migration, Rollback
• Ensure no double-counting between pool changes and cost changes from the same signal

​

Test work

• Turn the intentionally red engine tests under app/test/domain/mana/engine/ green
• Add focused subsystem tests for:
  • axis factor
  • pool EWMA and conservative bias
  • task costing
  • task surfacing
  • return recovery
  • PEM detection
  • regime shift detection
  • transparency output
• Add replay/integration tests that exercise DriftManaRepository -> ManaDayInputs -> ManaEngine -> ManaDayResult
• Add property-style tests for boundedness, determinism, and no-feedback-loop invariants
• Add scenario regression tests matching the worked examples in Example Scenarios

​

Explicitly out of scope here

• Mana screen and widget implementation
• Visual explainers and UI affordances
• Non-Mana task UI flows