Alice Technical Journal — 09/02

Project: Alice (Insights & Teacher Economics)

Today was about turning raw operational data into fair, explainable economic insight — and learning where naïve averages quietly break trust.

What I worked on

I spent the day evolving the Teachers Ranking & Bonuses system under Admin > Insights, pushing it from “interesting numbers” toward something that can actually support real decisions.

Key additions and refinements:

1. Teachers Ranking maturity
   • Added richer metrics beyond totals:
     • Average Estimated Salary
     • Average Paid
     • Average Hourly Rate (derived from worked hours)
   • Introduced a Top Penalty section to surface negative salary entries transparently instead of hiding them in aggregates.
   • Added support for longer analytical horizons (12 months) without breaking short-range views (7 / 30 / 90 days).
2. Bonus model design
   • Designed a bonus formula that balances:
     • Performance (E / A / M / Penalty)
     • Salary scale
     • Tenure realism
   • The key breakthrough was separating:
     • Performance quality from
     • Eligibility window
   • Final model:
     • Performance is normalized and capped
     • Bonus scales by work days since first attendance, capped at 365
     • Short ranges (e.g. 30 days) intentionally produce smaller bonuses than long ranges (365 days), without paradoxes
3. Tenure-aware averaging
   • Fixed a subtle but serious issue:
     • Teachers who joined recently were having their averages diluted by time they didn’t exist.
   • Introduced per-teacher effective averaging windows:
     • effective_days = min(work_days, selected_range, 365)
     • Averages are now fair for both new and long-tenured teachers.
   • This change cascaded correctly into:
     • Avg Salary
     • Avg Hourly Rate
     • Bonuses
4. Insights hygiene
   • Added the ability to exclude teachers from insights without touching:
     • Academic records
     • Finance history
     • Attendance truth
   • This keeps insights clean (no interns, test accounts, or transitional staff) while preserving data integrity.

---

What I learned

1. Averages are politicalAn average is never neutral.Dividing by the wrong window quietly punishes:
   • New teachers
   • Part-time contributors
   • Anyone who joined mid-cycle
   Fixing averages to be tenure-aware immediately made the numbers feel human again.
2. Bonuses must scale with time, not just performanceA “year-end bonus” that’s smaller than a “month-end bonus” isn’t a math bug — it’s a model bug.The insight:
   • Performance answers how well
   • Time answers how much opportunity existed
   You need both, explicitly.
3. Negative signals deserve first-class visibilityPenalties shouldn’t just reduce totals silently.Making them visible as a ranking:
   • Changes conversations
   • Reduces disputes
   • Encourages accountability without moralizing
4. One metric → many systemsChanging how Avg Salary is computed wasn’t isolated.It affected:
   • Rankings
   • Hourly rates
   • Bonuses
   • Perceived fairness
   This reinforced the importance of single-source-of-truth calculations and pushing logic server-side.

---

Where the system is now

• Teachers Ranking is no longer just “who’s biggest”
• It now reflects:
  • Value
  • Reliability
  • Discipline
  • Time
• Bonuses are:
  • Bounded
  • Explainable
  • Consistent across time ranges
• Most importantly:the numbers now align with intuition, which is the real test of a good system.

Tomorrow, this can easily evolve into:

• Bonus preview simulations
• Finance reconciliation overlays
• Or even teacher-facing transparency views

But today was about getting the foundation right.

And it finally feels right.