phone+net+tics · phone-net · I.C.K.S. v0.7.6 · full analytics
Every day, side by side
One column per day from June 2025 to September 2026. Each strip is one signal, so a crisis day, a build day and a quiet day line up vertically. Darker = more. Counts only; no names or content. ← back to the summary page · sessions of therapy →
AI-agent incidents are the owner's own "CVE-FAUX" write-ups of mistakes made by AI coding assistants (27 dated). Letter = agent named in the report: G Grok, C Claude, W Cowork; hollow "?" = not named. Hover a dot for the one-line summary.
Crisis days = distinct days with an emergency response or a safety incident in the caregiver's own log. "Game sessions" has no data yet: it needs an export (platform playtime history or the game's replay-file dates) before it can be charted.
AI usage around the hard days
Local models on the owner's own GPUs vs cloud assistants, lined up with care notes, approved hours, emails sent, and "hard days" (a safety incident, an emergency, or contact with the care-plan or payroll company). March to September 2026.
Sources: session index (cloud assistants; output tokens only, and one chat tool doesn't record tokens), the model gateway's spend logs on two machines (local and cloud routes; spend recorded as $0 because cloud routes were free tiers or subscriptions), and the assistant's own usage log for recent weeks. The browser-history search tool has no token data. Gold marks on the token rows = hard days.
Notes per day vs hours those notes cover
Every day with at least two caregiving notes and a payroll record (Mar 2025 – Sep 2026). Across: how many notes that day. Up: hours from the first note to the last. Dashed gold: the paid cap in the payroll company's own exports.
"Hours covered" is the active window between the day's first and last note, not a timesheet: it shows when the caregiver was on duty and writing things down. The paid cap and the hours logged in the payroll app come from the payroll company's shift-status exports.