Post-Workshop Survey: What People Want Improved

Stata Translation Workshop · Post-Workshop Survey

What people want improved

Drawn from the "concerns," "RSE role," and "other comments" free-text answers (9 respondents). Words are packed edge-to-edge into a circular cloud, upright or rotated a clean 90° only — no diagonal tilts. Concrete improvement asks are large and colorful; generic subject-matter words (testing, code, Stata…) stay small and gray — they're topic, not an ask.

Verifying Correctness Flagging Silent Changes Understanding Model Limitations Line-by-Line Review Human-Readable Test Scripts Reporting Discrepancies Verifying Self-Corrections Idiomatic Code Structure Standardized Translation Frameworks Testing Code Translation Increasing Rigour Drop-in Support Clinics Skill & Plugin Guidance Incremental Re-translation Generalize Beyond Claude Code User-Guided Pseudocode Scoping Maintaining Non-Human Code Environment & PATH Setup Stata Documentation Reliability Script Package Claude R
Bigger & more colorful = a stronger, more specific improvement point · small, gray & faint = general topic, not an ask

Biggest ask: a reliable way to verify translated code is actually correct — 5 of 9 respondents raised this independently (code that "looks right but isn't," wanting line-by-line checks, no clear way to check accuracy on large packages, an unverifiable self-correction claim from Claude). Everything else was raised by 1–2 respondents each; treat the smaller phrases as individual ideas worth watching, not consensus. Color here is decorative (one hue per phrase, cycled) rather than meaningful — it's there to make each phrase easy to pick out, not to group them.

View themes as table, with a representative quote
ThemeRespondentsRepresentative quote
Verifying correctness5"Code that appears to be correct but is not – over confidence."
Flagging silent changes2"My code got simplified by Claude and it just put a comment saying # simplified… report back if it does any simplifications"
Understanding model limitations2"Understanding the limitations, and… where it may need support."
Line-by-line review1"I would want to check it line-by-line and do extensive testing."
Human-readable test scripts1"the test script … wasn't very user-friendly (used testthat package)"
Reporting discrepancies1"discrepancies between my STATA documentation and STATA code… did not flag this issue"
Verifying self-corrections1"Claude claimed to find an error in its translation and correct it; I was unable to verify this"
Idiomatic code structure1"Lack of elegance (e.g. using optimal structures/organisation/logic for each coding language)"
Standardized translation frameworks1"Designing systems/frameworks to standardise code translation processes."
Increasing rigour1"Increasing the rigour of what we do"
Drop-in support clinics1"Drop-in clinic (e.g. every month) where we could ask for advice"
Skill & plugin guidance1"Guidance on writing skills / plugin"
Incremental re-translation1"find a way to translate… new feature… without re-translating everything"
Generalize beyond Claude Code1"Generalisation to beyond Claude Code would be great"
User-guided pseudocode scoping1"pseudo-code writing step should… engage with the user to see if there is functionality… not needed"
Maintaining non-human code1"How can packages be updated/maintained/checked if not human-written?"
Environment & PATH setup1"Need to make sure to set up PATH variables correctly."