# Sand Left

> A rescue order for irreplaceable photos, writing, art, messages and recordings. Type one keepsake per line - what it is | where it lives | how many years that place has existed | copies elsewhere | signals - and see a shelf of hourglasses: the sand left in each is the rule-of-thumb odds that the place it lives on is still there in 5 years. A lifeboat seats the riskiest keepsakes first. One metered run writes the rescue plan: a verdict, and for every keepsake why it sits where it does, the first step and what to keep it as, plus a weekend's steps and what to watch out for.

URL: https://sand-left.skillsafe.ai/
API tutorial: https://sand-left.skillsafe.ai/api.html

## Free, in the browser
- Reading the shelf: years as 12, a start year (2014) or ? (unknown, assumed 3); copies as 0, 1, 2 or none; signals as comma-separated words - closing notice, no export, export tool, lapsed plan, an ageing device (old phone, drive, disc), app-only format - or your own words. A .txt or .csv file can be dropped in.
- The sand: the Lindy rule of thumb - a place that has lasted A years has about A/(A+5) odds of lasting 5 more. A closing notice drops the odds to 0%, a lapsed plan halves them, an ageing device caps them at 50%. Risk = (1 - odds) x copy factor (no copy 1, one 0.4, two or more 0.15), + 0.1 with no export. Red 0.6 or more, amber 0.25 or more, green below. These are a rough rule of thumb, an estimate, not a forecast for any real place.
- Actions: red - rescue first; not red with an app-only format - convert it; amber - add a copy; green - leave it be.
- The lifeboat: red and amber keepsakes seated in order of risk; green ones stay ashore.
- The verdict rule, checked in order: rescue now (a keepsake under a closing notice with no copy, or at least half the shelf red), steady (nothing red), copy soon (everything else).
- Exports: a rescue list (.csv), Markdown and JSON; kept shelves (in the browser, or in your account when signed in).
- The loop: "+1 copy made" on any keepsake updates its copies and re-colours the sand; the weekend steps tick off; a kept shelf remembers its colours and shows what moved since (red 3 -> 1).

## Metered (gpt-terra), one task: rescue
The model gets every figure from the page and does no arithmetic of its own. It returns the verdict by the rule above, one plan per keepsake in lifeboat order (the page's action, why - the place's age by the rule of thumb, then what overrides it - the first step and what to keep it as), 2-5 weekend steps and 0-3 watch-outs. A first step never relies on an export the keepsake does not have. It names no product, site or brand and adds no person, date, deadline or backstory the input does not state. The page re-checks every reply: the verdict against its own rule, a plan for every keepsake with the page's action, and no number, capitalised name, he/she, field name or brand the input does not hold.

## Sources
Built on the SkillSafe skill @tjboudreaux/thinking-lindy-effect - https://skillsafe.ai/skill/@tjboudreaux/thinking-lindy-effect - (treat survival duration as a remaining-life prior, then check domain drift before favouring the proven), narrowed to one question: which of my keepsakes do I copy out first.