HowToLiveBetter: an evidence-graded life guide sorted by cost-effectiveness

What HowToLiveBetter solves, and who it is written for

Most life advice arrives without a number attached. HowToLiveBetter is an attempt to attach one. The README states the project covers longevity and disease prevention, accidents and first aid, saving and personal finance, fraud and legal limits, unemployment support, starting a business, platform compliance, dating and marriage, going abroad, and skills, across 498 entries. Each entry is supposed to answer two questions: what you spend (money, time, energy, willpower) and what you get back (change in all-cause mortality, a drop in a specific cause of death, time and energy saved, money saved, or protection and personal freedom).

The intended reader is someone deciding what to do next rather than someone browsing for reassurance. The table of contents is framed as questions a person actually asks: how much smoking, drinking, sitting and late nights cost in years; which supplements and health check packages can be skipped; what to claim after losing a job or being owed wages; whose money a bride price or a premarital property legally is; what to do first when someone collapses without breathing. Entries are ordered by cost-effectiveness inside each section, not by topic, so near-zero-cost items with large payoffs come first. That ordering is the project's main editorial claim, and it is worth knowing that it is a claim: the README says the cost-effectiveness tier is the author's judgement, essentially C-grade, and orthogonal to the evidence grade.

Evidence grades, cost tiers and the plain-language line

Every entry carries an evidence grade. A means quantifiable evidence from a meta-analysis, a large cohort or an RCT, with a concrete figure such as an HR, RR or percentage drop. B means research support that is hard to quantify, or evidence from a small or single study. C means the author's experience or general consensus with no direct literature. Of the 498 entries, 323 are A, 126 are B and 49 are C; 45 entries are marked as disputed and 39 places are marked TODO pending verification.

The README is explicit that the grade answers whether a number can be trusted, not whether an action is worth taking. A separate payoff magnitude (large, medium, small) and a unit (life, money, time and energy, personal freedom) feed a cost-effectiveness tier computed from three cost dimensions: whether it costs money, how much time it takes, and whether it needs willpower. The README states the tiers are not comparable across units: a 12 percent drop in all-cause mortality and 500 yuan saved per year are not on the same ruler. Of the 498 entries, 88 are rated extremely high cost-effectiveness, 248 high and 162 ordinary. The README says the middle band is deliberately thick because the underlying payoff magnitude has only three levels, and cutting finer would fake precision. That is an unusually honest admission and it also means the tier is a rough sort, not a ranking.

Each entry also has a plain-language line that restates the payoff column in everyday terms, for example turning a hazard ratio into a statement about how much lower the probability of dying in the same period was. The README says this line uses only facts already present in the entry and introduces no new numbers.

The data flow: book/ is the source, index.html is the view

The repository is small and flat. Top-level entries are .gitignore, .nojekyll, CLAUDE.md, LICENSE, README.md, book/, docs/, index.html, og.png, robots.txt, sitemap.xml and tools/. The README says the online search page reads its data directly from the Markdown files under book/, so editing the prose changes the search page. That is the whole architecture: content in Markdown, one HTML page in front of it, no database and no build service described in the README.

The search page supports filtering by keyword, by chapter, by evidence grade, and by three cost dimensions (whether it costs money, how much time, whether it needs willpower). The README gives two worked filters. Selecting evidence grade A leaves the 323 entries with concrete numbers from meta-analyses or large trials. Selecting cost-effectiveness "extremely high" yields 85 entries that cost nothing, take no time and need no willpower, with the payoff in the largest band; the README notes that stacking a payoff unit on top gives a priority list for that unit. Note the small inconsistency: the README's badge and tier paragraph say 88 entries are extremely high, while the reading instructions say 85. I cannot tell from the README which count is current, and that is the kind of discrepancy you should expect from a hand-maintained dataset.

A worked entry in the README shows the shape: swapping household salt for low-sodium (potassium) salt, costing a few yuan more per bag, with a stated 14 percent reduction in stroke, 13 percent in cardiovascular events and 12 percent in total mortality, graded A, sourced to a 2021 NEJM paper with a DOI, and marked disputed with the counter-evidence and its own citation listed in the note.

Reading it without installing anything

There is no install step. The project is a static site plus Markdown files, and the README points readers to the hosted search page at eternity4719.github.io/HowToLiveBetter/ and to the book/ directory for sequential reading. If you want the whole thing locally, the only step is cloning the repository; the README does not document a build step, a package manager or a server command, so treat anything beyond cloning as unverified.

git clone https://github.com/eternity4719/HowToLiveBetter.git
cd HowToLiveBetter
ls book/

After that you should see the chapter files that the README's table of contents links to, such as 01-不要早死.md and 13-紧急情况.md. Opening index.html directly in a browser is the local equivalent of the hosted search page, based on the repository layout; the README does not describe how the page loads the Markdown, so whether it works from file:// depends on the tooling under tools/, which the README does not explain.

A first real use is filtering rather than reading. Open the hosted page, set the evidence grade filter to A, and set the cost-effectiveness filter to extremely high. You should be left with a short list of entries that cost nothing, take no time, need no willpower and carry a large payoff. Pick one, open its cited source, and check the number in the entry against the abstract. That check is the intended workflow, and it is the reason the sources are DOIs and official documents rather than secondary summaries.

Where the guide is the wrong tool

The evidence grade is self-reported and the README concedes that 39 places are marked TODO pending verification. Nothing in the repository enforces that a cited paper says what the entry claims; the only check is you opening the DOI. For anything with a legal consequence, that matters more than for anything else, because the entries on legal limits, criminal exposure for programmers, labour injury recognition and platform licensing summarise rules that change. The README does not document a verification cadence or a review date per entry, so you cannot tell from the page alone how old a given legal statement is.

The unit separation is also a real limitation rather than a feature. Because the four payoff units are not comparable, the guide cannot tell you whether to spend an hour on exercise or an hour on a tax claim. It sorts within units and leaves the cross-unit trade-off to you. The cost-effectiveness tier is explicitly the author's judgement and the README says so; if you disagree with the thresholds (a relative reduction of 20 percent or more counts as large for life outcomes), the ordering changes.

Finally, the content is in Chinese. The repository is HTML and Markdown with no translation layer described, so a reader without Chinese cannot use the entries at all, and the search page's value is entirely inside that language.

How it differs from a wiki, a checklist app or a curated reading list

The closest familiar alternative is a community wiki or a curated list of links, and the difference is the unit of content. A wiki page accumulates claims; a link list defers the judgement to whatever is linked. HowToLiveBetter's unit is a single recommendation with four fixed fields (cost, payoff, evidence grade, source) plus a plain-language line, and the README states that sources are limited to primary literature with DOIs or PubMed links, or official documents from bodies such as WHO, CDC or the national statistics bureau, with no secondary retellings. A link list cannot enforce that because it has no field for it.

The second difference is the explicit refusal to compare unlike things. A general advice site will happily put "sleep more" next to "refinance your mortgage" and rank them. This project separates life, money, time-and-energy and personal-freedom units and says it does not convert between them. That is a constraint on usefulness, and it is also the reason the ordering means something within a section. If you want a single ranked list across every domain of life, this project deliberately will not give you one.

Licence, maintenance and what upgrades cost you

The licence is Unlicense, which the README's badge repeats. Unlicense places the work in the public domain, so copying, modifying and redistributing the text and the site does not require attribution under that licence. That has a practical consequence worth stating plainly: if you fork it and change a number, nothing in the licence or the repository stops you from republishing the result under the same name, and no upstream review exists. I am not giving legal advice; read the LICENSE file for the actual terms.

On maintenance, the last push to the default branch was on 2026-09-11, five days before this writing, and the repository is not archived. The README does not describe a release process, and no releases were retrieved, so versioning is by commit. Upgrading means pulling the branch and re-reading the entries you rely on, because the README says the search page reads book/ directly: a prose edit changes what the page shows with no separate data file to diff. The 39 TODO markers and 45 disputed markers are the places where a pull is most likely to change an answer, and they are the places to re-check first.

Editorial conclusion

Adopt HowToLiveBetter if you read Chinese and want the citation attached to each recommendation, or if you want to filter 498 entries down to the 323 graded A before acting. Do not adopt it as a medical or legal authority: the README itself flags 45 entries as disputed and 39 as TODO, and a C-grade entry is the author's own judgement. Before relying on anything, open the linked DOI or official document for that entry and check the number still says what the entry claims.

Frequently asked questions

Official sources

1. eternity4719/HowToLiveBetter on GitHub
2. Issues
3. License: Unlicense
4. Project website
5. README