Skip to content

VERIFICATION.md

Latest commit

 

History

History
47 lines (40 loc) · 3.04 KB

VERIFICATION.md

File metadata and controls

47 lines (40 loc) · 3.04 KB
Raw

Verification — 0.1.0

Local verification completed September 8, 2026. This is bounded engineering evidence, not a claim about paid demand or statistical performance comparisons.

  • 30 engine tests pass with Bun 1.3.14: 28 independently authored tests plus two unaltered MIT Speedscope upstream examples. The independent review caught and verified a fix for non-string units being accepted through property coercion.
  • 15 real-browser interaction checks pass in Firefox 155.0 on Linux. They exercise totals, flame widths/zoom, search, share deltas, new/removed filters, inclusive inspection, swap, malformed replacement, native file input and active profile selection, dimension rejection, competing asynchronous imports, JSON provenance, escaped snapshot export/reopening, pagination, and clearing.
  • Responsive checks pass at a 560-pixel browser window: inspector visible, no horizontal page overflow. Desktop and narrow screenshots were inspected.
  • Synthetic scale check: two identical 100,000-sample files, each 1,045,371 decoded UTF-8 bytes, 1,001 function identities and total weight 300,000 ms. Import, comparison, table paging and search completed in one measured 683 ms interval on the test laptop. This is one synthetic smoke measurement, not a representative profiler benchmark or a latency guarantee near resource caps.
  • python3 build.py --check confirms the committed standalone HTML exactly matches source, styles and examples. Runtime package dependencies: zero.

The first browser run passed 13/14 checks. The remaining assertion demanded JavaScript-double precision from a CSS percentage string, which Firefox serializes at lower precision. The check now allows 0.001 percentage points of serialization rounding; the graph formula did not change. The final suite adds a scale/pagination case and passes 15/15. Early screenshot-only automation captured a page before its asynchronous tests finished; the committed runner waits for the completion object.

The GitHub Actions workflow runs the 30 engine tests under Node 22 and Node 24 and checks the generated build. Its public run is separate evidence from the local Bun execution; consult the Actions tab at the commit you use.

Not established: Safari/Chromium behavior, a full accessibility audit, every third-party exporter, independently supplied real workload comparisons, statistical significance, paid use, or adoption. The upstream examples are small public format fixtures. The checkout demo and scale data are explicitly synthetic.

The app has no fetch, XHR, WebSocket, telemetry, server, storage or runtime imports. Its CSP blocks outbound connections. Imported strings use text nodes; snapshot JSON escapes HTML-significant characters. The browser test includes a hostile closing-script/image name and verifies it stays inert after reopening. This is specific regression coverage, not a complete security audit of the browser.

Reproduce with the commands in the README. Browser test outputs go to the requested temporary output directory and are not part of the application or repository.