[Fyfar]

Summary

Statistics panel entries (Average/Min/Max/Peak-to-Peak/Std Dev) and the realtime chart's y-axis now render with SI unit prefixes (e.g. 11.13 MΩ) instead of raw base-unit numbers (e.g. 11130000.00 Ω). Display only — no change to stored values, resolution, CSV export, or Data Log rows.

• Each Statistics panel entry picks its own SI prefix from its own magnitude; the chart's y-axis picks one shared prefix per redraw from the visible range (lib/si.ts: siPrefixOf, formatSiValue).
• Decimals come from the existing measurement-resolution tracking (binWidth/recordedResolution), never a new precision source.
• The SI prefix is chosen from a value's true, unrounded magnitude and never re-picked after rounding, so a value can't be promoted into the next prefix by rounding (999999.6 Ω at 1 Ω resolution reads 1000.000 kΩ, never 1.00 MΩ).
• Capacitance's internal base unit is nF, not physical F — corrected via toPhysicalUnit wherever a capacitance number reaches the formatter.
• siPrefixOf relocated from lib/passfail.ts to a new lib/si.ts, now shared by the Dashboard, Pass/Fail, and the chart.

Hardware testing

Tested against a ZOYI ZT703s (28/29 cases passing on first run); three real issues found and fixed:

• A genuine zero (e.g. Peak-to-Peak before a 2nd sample) briefly rendered as a long decimal string (0.00000000000000F) instead of 0.
• Pass/Fail's Measured Spread panel showed a flat 3 decimals against the meter's real resolution once SI-scaled next to the Dashboard's resolution-accurate numbers. VerdictRow now carries the capturing reading's resolution; Pass/Fail tracks the coarsest resolution across the batch's captured rows, same as the Dashboard.
• Clear Batch could re-trigger a capture (and verdict beep) for a part still sitting on the probes, from any view, because it redundantly reset the same-part capture guard that already re-arms correctly on a genuine probe lift.

Three unrelated findings from the same test pass (unbounded manual y-range inputs, unscaled histogram bin labels, a bad x-axis offset after reload-before-reconnect) are recorded as known issues, not fixed here.

Verification

• npx tsc --noEmit — clean
• npx eslint — clean
• npm run check — all 8 self-checks pass, including new check-si.mts (19/19)
• npm run build — production build succeeds
• Two independent cold code reviews of the full diff (must-fix: none)
• Manual hardware verification against a real ZT703s

Version bumped 1.7.2 → 1.8.0 (minor, non-breaking addition).

🤖 Generated with Claude Code