# Still Going

> Survival analysis for everyday belongings. List the umbrellas, mugs, phone chargers (or any one kind of
> thing) you have had: the month each arrived, and the month and way it went - lost, broke, wore out,
> borrowed and never returned, given away in working order - or that it is still going. The page draws a
> life table and a Kaplan-Meier survival curve (items still going or given away are censored), finds the
> median life, fits a Weibull shape and gives a bathtub-curve verdict. A metered run adds a deadpan
> actuary's note. Built on the SkillSafe skill @aj-geddes/survival-analysis.

https://still-going.skillsafe.ai/

## What is free
- The register (up to 12 items, .csv/.json drop), the lifelines card, the Kaplan-Meier curve, the median life, the share still going at 6, 12 and 24 months, the Weibull shape and the verdict all run in the browser.
- Verdicts, in check order: too few to tell (fewer than 4 endings), dies young (shape under 0.8), wears out (shape over 1.25), luck of the draw (otherwise).
- Four sample notes replay at no cost, one per verdict.
- Downloads: the card (SVG, PNG), the table (CSV), the full reading (Markdown, JSON).

## The metered task
One task, `report`, on gpt-terra, metered per run. The input is string fields computed by the page: `task`, `kind`, `as_of`, `items` (numbered lines), `counts`, `median`, `still_going_at`, `shape`, `verdict` and `leading_cause`. The reply is one JSON object: `verdict` (the page's, copied), `headline`, `reading`, `lives` (one {item, note} per item, in order), `advice`, `next_one` and `caveat`. The page re-checks the reply against its own numbers.

Medians and shapes from a handful of items are estimates, not laws.

## Links
- API tutorial: https://still-going.skillsafe.ai/api.html
- Source skill: https://skillsafe.ai/skill/@aj-geddes/survival-analysis (aj-geddes/useful-ai-prompts)
- Background: https://en.wikipedia.org/wiki/Kaplan%E2%80%93Meier_estimator and https://en.wikipedia.org/wiki/Bathtub_curve