K-12 AI Governance · Tool Intelligence

The First Watch

Drift Timeline · Audits 001 → 004
Audit 001 · April 6, 2026 · 7 tools · 8-week rotation
Audit 002 · May 6, 2026 · 8 tools · 4-week rotation
Audit 003 · June 6, 2026 · 9 tools · 4-week rotation
Audit 004 · July 6, 2026 · 10 tools · 4-week rotation
Cross-references verified via vendor primary sources
Drift Category No Drift Watch Flag
Forensic Pattern Convergence Asymmetric Divergence
01View · Tool by Tool

How each tool moved across four audits

Each tool's findings from Audit 001 (April 6), Audit 002 (May 6), Audit 003 (June 6), and Audit 004 (July 6), read side by side. The Audit 004 column flags whether the forensic pattern shifted, held, or is new this cycle. Two tools crossed categories this cycle (Khanmigo to Watch / Asymmetric; Claude for Education to No Drift / Convergence), and MagicSchool enters as a baseline. No tool's decision label changed in Audit 004.

02View · Chronological Events

Every dated movement, February → July 2026

Each documented event from all four audits plotted on a single timeline. Hover any dot for details. Four vertical rust lines mark each audit's data-collection cutoff (April 6, May 6, June 5, July 6).

03View · Pattern Shifts

What four cycles on record now show

With four audits on file, patterns read across cycles, not just within one. Tools that held a pattern for four consecutive cycles now constitute a genuine track record. Tools whose pattern shifted this cycle (Khanmigo moving to asymmetric after three convergence cycles; Claude for Education settling to convergence) show that even a stable decision label can sit over moving parts.

Pattern Shifted · This Cycle
2
Khanmigo → Asymmetric (after convergence ×3); Claude for Education → Convergence. Both under unchanged decision labels.
Four-Cycle Track Records
3
ChatGPT Free (divergence ×4); ChatGPT for Teachers and Microsoft 365 Copilot (asymmetric ×4).
Held Asymmetric · 003 → 004
4
ChatGPT for Teachers, NotebookLM, Microsoft 365 Copilot, Claude Enterprise: asymmetric again.
New Baseline
1
MagicSchool enters at convergence; external check corroborates. First true drift finding carried to Audit 005.
Tool
Forensic pattern · 001 → 002 → 003 → 004
Longitudinal read
003 → 004