# Independent reproduction protocol KB-MARGIN-20260831

**Published:** 2026-08-31  
**Target run:** `KB-MARGIN-20260822`  
**Status:** Open protocol; no independent reproduction received as of 2026-08-31

## Purpose

This protocol lets an unaffiliated analyst test whether a bound document pack changes an ambiguous metric answer in an inspectable way. It is not itself an independent report.

## Materials

- Published desk log `KB-MARGIN-20260822`
- Aggregate CSV `aggregate-KB-MARGIN-20260822.csv`
- Verification script `verify-KB-MARGIN-20260822.py`
- A separately obtained authorized dataset or public extract
- Two or more disclosed definitions for one metric

Do not request or publish the private sanitized composite behind the first-party desk run.

## Procedure

1. Declare identity, affiliation, conflicts, source version, row grain, file hashes, and retrieval date.
2. Publish the competing metric formulas, document versions, owners, and exclusions.
3. Run an unbound control and preserve the query, SQL, retrieved passages, and outputs.
4. Bind the selected document pack, rerun the same goal, and preserve the same artifacts.
5. Recalculate each reported metric independently of the generated answer.
6. Publish confirming and conflicting results, failed runs, limitations, and non-sensitive evidence.

## Acceptance criteria

An independent report must be authored outside InfiniSynapse, disclose conflicts, identify source and document versions, reproduce its arithmetic, and state whether retrieval and labels agree with or contradict the page. Running only the first-party CSV verification script does not count.

Send a public report URL or documented contradiction to [zhuhl@infinisynapse.com](mailto:zhuhl@infinisynapse.com). InfiniSynapse may link qualifying reports whether they confirm or contradict the first-party run.