[finom]

Five audits (ES5 build, router and form parser, events, Matreshka.Array, docs and skill); every fix has a spec that fails on fec0a1c (test/spec/regressions/sweep5_*_spec.js, docs/test/code-parity.test.js).

Events

• DOM listeners (click::x) follow listener owners: a second path or the object's own on keeps them
• off('click::x') no longer leaves _unbind:x entries behind
• undelegating from one object no longer cancels a debounced call pending from another
• off cancels a pending DOM onDebounce call
• once fires once when an earlier handler triggers the same event again

Router

• history mode adds no entry when the path doesn't change; another spelling of the same route is rewritten in place
• a property set to null nulls the following ones even when the URL doesn't change
• emptying a hash route leaves no bare #
• Router.hash and Router.history work on first use, without a subscriber
• Router methods aren't enumerable in dist/

ES5 build

• <col> renders without <template>
• text, html, output binders and {{x}} show String(value) in dist/ as in src/
• Array.from fallback keeps holes (Matreshka.Array.from({length: 3}, fn))
• a __proto__ data key no longer sets the prototype (fromEntries)
• a Class chain over an ES class works in src/ and .min.mjs, as in dist/
• assign and forOwn fallbacks take primitives

Matreshka.Array

• splice that adds items works after the container is rebound
• restore() with a Matreshka.Array Model gives empty collections
• a method that throws on render (push of a rendered item, fill, copyWithin) leaves the array unchanged
• flat and flatMap spread nested Matreshka.Array instances, as concat does

Docs and skill

• broken samples fixed (faq, restore, instantiate, onItemRender, trackBy, progress, style, types/matreshka, toJSON)
• ua and ru code blocks are the same code as en; a test checks it (it also found 'change: a' with a space in ua)
• SKILL.md: matreshka.min.mjs is a little smaller than the script-tag file

Also: the legacy scenario waits through timers, so it doesn't flake on a busy machine.

Not changed: a route without * ('/x/') still drops deeper URL parts on write; the documented pattern for classes sharing a URL pads with *.

Gates: lint, typecheck, 1050 tests with 100% coverage, build, check:dist (1002 tests on dist, 16 legacy cases), build:docs. Checked in Chrome on the built bundle against the 3.0.1 bundle.