@gverni suggested in #69 that the manual test harness (scripts/manual-test-plan.js) could share fixtures with the existing Jest tests, so the same inputs get validated both locally and against live Twilio. This would give us stronger confidence that local tests match real Twilio behavior.
Current state:
- Jest: 671 tests covering comprehensive encoding paths, edge cases, grapheme splits
- Manual harness: 16 cases targeting the 4 areas that changed since v1.2.0 (CRLF, combining marks, Smart Encoding, UCS-2 display)
Challenge:
Running all 671 tests against live Twilio would be expensive (time + cost per segment). We'd need a smaller "golden set" that both systems consume.
Question for @gverni:
What should be in that golden set? Should it:
- Match the current 16 manual cases (focused discriminators for recent changes)?
- Expand to ~20-30 cases covering broader encoding boundaries and edge cases?
- Be a different set entirely based on what's most likely to diverge between the calculator and Twilio?
Would love your thoughts on scope and which test cases are the highest signal.
@gverni suggested in #69 that the manual test harness (
scripts/manual-test-plan.js) could share fixtures with the existing Jest tests, so the same inputs get validated both locally and against live Twilio. This would give us stronger confidence that local tests match real Twilio behavior.Current state:
Challenge:
Running all 671 tests against live Twilio would be expensive (time + cost per segment). We'd need a smaller "golden set" that both systems consume.
Question for @gverni:
What should be in that golden set? Should it:
Would love your thoughts on scope and which test cases are the highest signal.