# Query fanout repetition protocol

Blank research protocol. This template neither runs a provider request nor grants permission to publish outputs.

## Question and permissions

- Specific question being investigated:
- Why a new run is necessary instead of reusing approved data:
- Provider, endpoint, exact model/version and date:
- Evidence mode (native / modelled); do not mix them:
- Topic, market and response language:
- Tool configuration and endpoint/version constraints:
- Planned number of attempts and provider allocation:
- Hard monetary cap and stop conditions:
- Permission to retain each field:
- Permission to share/publish each field:
- Private-input handling and retention deadline:

## Define comparisons before running

- Compare exact exposed query strings, normalised strings or domains? Define each transformation:
- Keep query/action/run source scope separate:
- Completed zero-query runs: retain as zero, not failure:
- Failed, blocked, cancelled or timed-out attempts: count separately:
- Missing model/version/usage fields: unknown, not inferred:
- Matching locale/input conditions:
- Confounders and deviations to record:
- What this experiment cannot establish:

## Attempt ledger

| Attempt | UTC time | Conditions reference | Completion state | Exposed query count or unknown | Source scope | Usage/cost or unknown | Retained artifact permitted? | Deviation |
|---|---|---|---|---|---|---|---|---|
| | | | | | | | | |

## Review and disposal

- Denominator includes which completed attempts:
- Failed attempts reported separately:
- Finding supported only within these conditions:
- Alternative explanations and missing information:
- Public evidence approval or private-only:
- Disposal date and location:

On ai-fanout.com, local history is optional, limited to 20 saves and 30 days, and removes Gemini grounded-result/search-suggestion payloads. Do not use it as an unrestricted archive. Consumer searches, hidden reasoning, ranking and demand are not measured by this protocol.