EdgeCalc is built, verified, documented and committed on main (commit 5c82a5f). The git bundle is 142 KB.

What was delivered

• Site: Vite 8 + React 19 + TypeScript single page with all four sections. Single bet computes edge (green/red with sign), EV per 1 USDC, full Kelly or "No bet", stake, shares, payout, profit and loss live. Section 2's fee raises break-even and flows into Section 1's figures. Arbitrage shows profit per 1 USDC of payout and the per-side allocation, or "No arbitrage". The position log supports add, inline edit, delete with undo, clear with undo, localStorage persistence, and CSV export/import with skipped-line reporting.
• Export: dist/ is committed with ./ relative asset URLs, alongside source, package.json and package-lock.json. No node_modules or archives are tracked.
• Docs: README.md covers install, preview, rebuild, publish and the validation record. DESIGN.md documents the oklch token tiers, type scale, spacing, container-query breakpoints, elevation, shapes and components from the final source. artifacts/validation.md holds the Better Interface review (all six domains, 10 findings with fixes, coverage and limitations); artifacts/ is excluded from git by the workspace and uploaded separately.

Checks on the final source, all passing:

Check	Result
npm run typecheck	clean
npm test	61 unit tests, 4 files
npm run build	index.html 0.98 kB, CSS 19.29 kB, JS 252.76 kB
npm run check:browser	6 viewport loads, 17 interactions, 56 contrast pairs, 0 failures

How the browser check ran: the MCP browser tool cannot open file:// and no managed preview existed, so scripts/browser-check.mjs starts its own ephemeral server, drives Chromium 154 through Playwright, inspects dist/ under a subpath at 320/390/768/1280 px in light and dark, and exits. Screenshots are in artifacts/screenshots/.

Notable findings fixed during review: clicking Edit immediately saved and left edit mode because React reused the button node and the browser's post-click activation submitted the form (fixed by remounting the edit row). The dark-mode primary button measured 4.10:1 (accent darkened to 4.87:1). Input focus rings, table action wrapping, and unlabeled totals cells were also corrected.

Limitations recorded: no screen-reader session, native 200% zoom, forced-colors rendering, RTL or physical-device testing; Chromium only. Section 1 assumes YES-anchored inputs (NO side uses 1 − price and 1 − P(YES)), stated in the UI and validation notes.