Historical one-off scripts kept here for reference. Use the replacements below for any new work.
| Legacy script | Replaced by | Why archived |
|---|---|---|
test-sample5.js |
scripts/batch-generate-queries.js (with --sample-n 5) |
First iteration, used the now-blocked mode: claude-code HTTP path against packy-cc gateway. Falls back to persona-fallback template; not useful for real LLM runs. |
test-sample5-cli.js |
scripts/batch-generate-queries.js (transport claude-cli) |
Working spike that proved subprocess-based CLI calling. Logic was generalized into scripts/lib/llm-batch.js; this file does only one shape (5 scenes × 3 complexities) without resume / errors.json / config.json. |
fill-missing.js |
scripts/batch-generate-queries.js (rerun with --output-dir <same>) |
The new batch script defaults to --resume: rerunning the same --output-dir automatically picks up missing rows. No separate fill step needed. |
build-comparison-html.js |
scripts/build-query-comparison.js |
Old version had hardcoded paths for the three sample5 runs. New script accepts arbitrary --run-dir / --runs and renders any number of datasets side-by-side. |
These files still run as-is, but they:
- duplicate utilities now centralized in
scripts/lib/llm-batch.jsandmvp/query_factory_v2.js, - don't share the standardized output layout (
plan.json/raw_queries.jsonl/errors.json/stats.json/config.json), - and are skipped by
npm run …scripts.
If you find yourself needing to revive any of them, prefer extending the lib module instead.