Count the markers.
Verify the behavior.

Normal builds now summarize final route HTML with bounded counts and paths. A literal marker scan is an inventory, not proof of script freedom, visible content, or accessibility.

Read the build reference Inspect the release

64maximum concurrent route-file reads

0 Bbrowser JavaScript delta

0new runtime concepts or dependencies

0.16.28

BOUNDED VERIFICATION

Separate artifact facts.
Observe real behavior.

OUTPUT

Final route HTML inventory

The scan runs after public copying and afterBuild, then prints only after successful output promotion. It counts files with and without script/modulepreload text markers and unreadable files, showing up to five bounded, sorted paths per category. Quiet and JSON builds omit it.

LIMITS

Markers are not zero-JavaScript proof

Comments, JSON-LD and inert content can match. Inline event attributes and extra public HTML are not checked. Compiler semantics, artifact schemas, runtime behavior and deploy bytes remain unchanged.

REPOSITORY

Browser tools stay repository-only

Trusted-local Linux Chrome smoke tooling observes rendered text and accessible names, uses native input, and fails on observed browser or resource errors. Bounded exact-target diagnostics and copied-tool integrity checks are covered by regressions. These tools are not shipped in npm, a framework API, a sandbox, or full accessibility certification.

EVIDENCE

No demonstrated AI-token advantage

Historical full-suite r8 remains Kudzu 23/25 versus React + Vite 24/25. Focused r9 records 3/5 versus 5/5; separate local summary r10 and equal-browser-tools r11 each record 4/5 versus 5/5. Failures remain included and experiments are not rescored. Browser adoption does not establish cost superiority; 1.0.0 remains blocked.

UPGRADE

Same semantics. Bounded output.

$npm install @kudzujs/core@^0.16.28

Count the markers.Verify the behavior.