CHINZY / FDE and robotics acceptance fieldbook / 09
Task success versus dataset yield
A failed task can produce a complete sample. Check required fields, timing and labels; use separate numerators for data validity and task success.
Test setup
Freeze the workcell, fixture and capture settings using samples or replay logs. Judge motion completion, correct labels and reusable data packages separately. Retain exceptions by batch.
Contrasting examples
Figures and IDs are illustrative fixtures, not product results or contract terms. Save inputs, versions and conditions before repeating an unexpected result.
| Example input | Expected observation |
|---|---|
attempt=A; task=failed; fields=complete; sync=ok; label=present | task_success=0/1; valid_data=1/1 |
attempt=B; task=passed; fields=missing; sync=ok; label=present | task_success=1/1; valid_data=0/1 |
Recording the result
Keep a separate row for each example; one pass cannot stand in for the other. Enter the actual output, evidence location, environment version and review decision. Leave unresolved results open.
Download test worksheet (CSV) ↓