Measurement

What is measured about maxstack, how, and — as importantly — what the numbers do not say.

Every figure here is non-comparative and self-checkable. Nothing on this page is a claim about being faster or better than another tool: no head-to-head has been run with a control arm, so no such claim would be supportable. What follows are properties of this system that you can reproduce yourself.

Cold start #

What is measured. Wall-clock from npx maxstack start "<description>" on a machine that has never seen maxstack, to the moment the app answers on its port.

Why it is fussy. "Cold" is the whole measurement, and it is easy to fake by accident. A run that reuses a warm npm cache measures almost nothing. The method therefore forces a fresh HOME, a fresh npm cache and a fresh install prefix, and refuses to call itself cold if it fetched less than a megabyte — because a run that downloaded nothing was not a cold run, however long it took.

What dominates it. Dependency weight, not code. The bytes fetched over the network are the number that moves the total; that is why the runtime tarball's contents are treated as a budget rather than an afterthought, and why test files are excluded from the vendored runtime snapshot.

How to check it yourself. Time it on a machine that has never installed maxstack, or in a fresh container:

shell

docker run --rm -it node:22 sh -c \
  'time npx --yes maxstack@latest start "a bug tracker" --no-dev'

Caveat, stated plainly. The last recorded figure was taken against the init entry on an older release, before the runtime snapshot was pruned. It has not been re-measured against start on the current version, so treat any specific number you find in older notes as unverified for today's entry point. Measure it yourself; that is the point of publishing the method rather than the number.

Regeneration safety #

What it means. Regenerating an app from its spec must never destroy work you own, and must never lose a manually added item.

How it is enforced. Not as a target — as a gate. Every regeneration in the test suite is bracketed by a safety check, and a run that is not 100% safe fails rather than reporting a percentage. Three invariants carry it:

regeneration never deletes manually added items;
generation grounds only on accepted items;
eject copies with a banner and never overwrites an existing file.

How to check it yourself. maxstack validate in any project reports spec validity, manifest integrity and regeneration safety. Eject a route, edit it, regenerate, and confirm your edit survives.

Combination safety #

What it means. Feature bundles compose. Any subset of the catalog that a user could install must produce a valid spec and a working app.

How it is checked. By sweeping bundle subsets rather than sampling the ones that seemed likely — order-dependence between bundles is exactly the class of bug that a hand-picked sample misses. The sweep is expensive, so it runs in the maintainer's own CI rather than on every pull request here.

What it does not prove. That every combination of user specs works — only that catalog bundles compose. Your own entities and pages are outside its reach.

Upgrade safety #

What it means. Upgrading a project to newer framework generators must not silently change or drop things the user added by hand.

How it is checked. Against pinned fixtures captured at an older bundle version — including hand-added spec items an upgrade must not touch — then regenerated forward and compared.

What is deliberately not measured here #

Any comparison to another framework or stack. A comparison needs a control arm run under a pre-registered protocol. That has not happened, so there is no number, and an unqualified "faster than" would be unsupported.
Leverage ratios (application produced per line of declaration). The measurement apparatus for this exists but its results depend heavily on how the denominator is defined, and a ratio does not license a claim in units.
Time-to-Nth-safe-change over a long-lived project. Interesting, and instrumented, but it takes months of real use to say anything honest.

If you want to argue with a number here, the method is the argument surface. Open an issue with what you measured and how.