# Independent reproduction protocol KB-BIND-REPLICA-20260831

**Published:** 2026-08-31  
**Target run:** `KB-BIND-REPLICA-20260822`  
**Status:** Open protocol; no qualifying independent report received as of 2026-08-31

## Purpose

This protocol enables an unaffiliated analyst to test whether an explicit source-to-pack mapping prevents notes for one replica from being applied to another. Publishing it is not an independent reproduction.

## Materials

- Desk log `KB-BIND-REPLICA-20260822`
- Aggregate CSV `aggregate-KB-BIND-REPLICA-20260822.csv`
- Verification script `verify-KB-BIND-REPLICA-20260822.py`
- Two separately identified authorized or public datasets with overlapping schema names
- A versioned definition pack intended for only one dataset

The private sanitized composites behind the first-party run are not published.

## Procedure

1. Declare identity, affiliation, commercial conflicts, source grains, row counts, versions, retrieval dates, and hashes.
2. Publish the pack owner, version, hash, intended source, definition, and exclusions.
3. Run an unbound control; retain prompt, model version, SQL, passages, source identifiers, and outputs.
4. Apply the explicit bind to the intended source and rerun the unchanged goal.
5. Independently recalculate source row counts and verify which source supplied each passage.
6. Publish confirming and conflicting results, failed runs, timing, limitations, and non-sensitive artifacts.

## Acceptance criteria

A qualifying report must be authored outside InfiniSynapse, disclose conflicts, identify both source versions and the exact mapping, expose reproducible arithmetic, and state whether the bind prevented cross-source leakage. Running only the first-party CSV verification script does not qualify.

Send a public report URL or documented contradiction to [zhuhl@infinisynapse.com](mailto:zhuhl@infinisynapse.com). InfiniSynapse may link qualifying confirming or conflicting reports.