Keep this worksheet inside the data-owner environment. Freeze it before viewing results. Export only the schema-approved hashes and counts produced through the evidence request.
- Operator contract:
codeprobe.zero-code-access.operator.v1 - Selection method:
predeclared_explicit/predeclared_stratified/predeclared_window - Window start (
YYYY-MM-DD): - Window end (
YYYY-MM-DD): - Mining goal:
quality/navigation/mcp/general - Candidate count, maximum 20:
- Minimum paired distinct scorable tasks:
10 - Repeats per task and configuration:
3
- Configuration A local description:
- Configuration B local description:
- The one dimension that differs:
- All controlled fields held equal:
- Both arms use the identical frozen task set.
- Neither arm receives information produced by the other arm.
Local descriptions may contain sensitive detail and must not enter the evidence request. Record only configuration SHA-256 digests there.
Assign stable anonymous categories before results:
| Category | Selection rule | Selected count | Paired scorable count |
|---|---|---|---|
category_01 |
Local-only description | ||
category_02 |
Local-only description |
Allowed exclusion codes are predeclared, duplicate, out_of_window, and
unsupported_task_type. Record counts, not task identities, in the evidence
request.
For every selected task, record locally:
| Anonymous row | Task SHA-256 | Verifier SHA-256 | Category |
|---|---|---|---|
| 01 | category_01 |
- Task digests are unique.
- Category selected counts equal the number of hashed task rows.
- Attrition equals selected minus paired scorable counts.
- At least ten paired rows remain.
- The sample did not change after results were visible.
- The data-owner technical owner attests the frozen sample represents the declared workload.
If any final checkbox is false, use insufficient_evidence.