llms.txt
Compare with other sites' llms.txt files →# life. scored.
> A transparency tool that exposes how real-world systems — credit bureaus, actuarial life tables, lenders, and audit-study research — already turn a person into a number. It rebuilds those scores in the open: every rule cited to public data or flagged as a guess, every weight shown and editable. The goal is transparency, not judgment.
## How to use this as a skill
life. scored. is "just math with personal information." You do not need to send anything to our servers to use it — and you shouldn't. The complete rulebook and the exact formulas are published as machine-readable data:
- [Rulebook + math (rules.json)](https://lifescored.com/rules.json): every rule's weight, bounds, evidence, citation, and formula; the input schema (what to ask the user); the engine constants (medians, the household-income equivalence, the net-worth age curve, the credit-score mapping); and the methodology prose.
- [Full reference (llms-full.txt)](https://lifescored.com/llms-full.txt): the entire rulebook + methodology as one flat markdown document — every rule's logic, evidence, cited source, and weight rationale in prose. The readable twin of rules.json.
- [Ready-made skill (SKILL.md)](https://github.com/royashbrook/lifescored/blob/main/skills/lifescored/SKILL.md): a drop-in agent skill that does everything below. Copy the `skills/lifescored/` folder into your agent's skills directory (e.g. `~/.claude/skills/`), or just follow the steps here.
To compute a life score for someone:
1. Read `inputs` in rules.json. Ask the user only for fields you don't already know — anything missing falls back to its `default`.
2. Get each rule's `position` exactly: a finite-input rule ships a `positions` table — look up the user's value under `positions.byValue`; a numeric-input rule has its formula in `constants` (keyed by rule id). Then `points = round(clamp(position, bounds[0], bounds[1]) × weight)`.
3. Sum points over enabled rules (the `core` pack is always on; other packs are opt-in) for the composite — but the breakdown is the point, not the total.
Their answers stay with you. Nothing about a person should be sent to lifescored.com to get a score.
## MCP endpoint
If you're an MCP-capable agent, the same rules are available as a tool server (Streamable HTTP):
- Endpoint: `https://lifescored.com/mcp`
- Stateless, no auth, and it accepts no personal data — it computes nothing. Tools: `get_rulebook`, `get_input_schema`, `get_methodology`, `how_to_give_feedback`. Pull the rulebook, ask the user what's missing, and compute on your own side.
## Privacy
Inputs never leave the user's device. The website computes everything client-side; any API or MCP we offer serves only the rules, never receives answers. No tracking, no accounts, no data sales, no affiliate links — by design.
## How it works
- [Why this exists / methodology](https://lifescored.com/about): what's measured, what's deliberately left out (protected characteristics, luck), and how the scoring works.
- [The rulebook](https://lifescored.com/rulebook): every rule with its plain-English logic, evidence tag, public source, and a tunable weight.
- [Free ways to improve a score](https://lifescored.com/improve): the most basic free next step per area, plus public resources — no affiliate links.
## Contributing / feedback
The weights are editorial and the rules are arguable — that's the design. Improvements from many people are aggregated in the open.
- [Open an issue](https://github.com/royashbrook/lifescored/issues): name the rule id and what you'd change (weight, source, bounds, or a new rule), with a public citation if you have one.
- [Contributing guide](https://github.com/royashbrook/lifescored/blob/main/CONTRIBUTING.md)
- [Source code](https://github.com/royashbrook/lifescored)