# Desk log KB-AIKB-FAQ-20260822

**Status:** First-party sanitized composite; not a customer extract, independent benchmark, third-party dataset, certification, media report, or endorsement  
**Page:** https://infinisynapse.com/en/blog/ai-knowledge-base-for-data-analysis  
**Run ID:** `KB-AIKB-FAQ-20260822`  
**Date:** 2026-08-22  
**Evidence package checked:** 2026-08-31
**Operator:** InfiniSynapse Data Team  
**Attestor:** William Zhu, InfiniSynapse cofounder ([GitHub @allwefantasy](https://github.com/allwefantasy))  
**Contact for contradictions:** zhuhl@infinisynapse.com

## What this file is

A downloadable record of FAQ-margin versus finance-margin retrieval. It documents the six-step method, input boundary, labeled percents, and artifact inventory. The companion [aggregate CSV](https://infinisynapse.com/blog-media/ai-knowledge-base-for-data-analysis/downloads/aggregate-KB-AIKB-FAQ-20260822.csv) contains only the published aggregate rows—not 12,800 raw rows.

## Verification materials

- [Verify the aggregate CSV](https://infinisynapse.com/blog-media/ai-knowledge-base-for-data-analysis/downloads/verify-KB-AIKB-FAQ-20260822.py)
- [External-source check](https://infinisynapse.com/blog-media/ai-knowledge-base-for-data-analysis/downloads/external-source-check-KB-AIKB-FAQ-20260831.md)
- [Independent-reproduction protocol](https://infinisynapse.com/blog-media/ai-knowledge-base-for-data-analysis/downloads/independent-reproduction-protocol-KB-AIKB-FAQ-20260831.md)

**Independent reproduction:** No qualifying unaffiliated report was known as of 2026-08-31. The verify script checks only published rows and is not a reproduction of the private sanitized composite. **DOI:** none.

## Six-step method (reproducible)

1. Pick one source you are authorized to read. Prefer a replica or sanitized extract.
2. Refuse the help-center export. Write or export ten field notes and one approved report.
3. Upload the pack, then bind it to that source. Binding is a separate click from upload.
4. Ask one analysis goal with the source selected—not a support question.
5. Inspect the artifacts and confirm the retrieved passages match the signed report, not a refund macro.
6. Ask the same goal next week. A successful replay keeps the same definition.

## Source and goal

| Field | Value |
|---|---|
| Source | Sanitized 12,800-row orders extract the desk is authorized to read |
| Source grain | One row per sanitized order record |
| Source version | Fixed desk extract dated 2026-08-22 |
| Collision | Sales FAQ “invoice − COGS” vs finance fee exclusion |
| Pack metadata | First-party sanitized composite; owner: InfiniSynapse Data Team; hashes and source documents are not published |
| Bind mapping | Finance notes + signed report ↔ named orders extract |
| Retrieval configuration | Bound-pack retrieval; configuration values not published |
| Unbound baseline | Same goal and source with FAQ retrieval only |
| Goal asked twice | “What is last-month margin?” |

## Labeled percents

| Retrieval state | Sales FAQ margin | Finance fee-excluded | Labeled pair shown |
|---|---|---|---|
| FAQ retrieval only | 24.0% | 24.0% (one number) | No |
| Bound analysis pack | 24.0% | 18.5% | 18.5% (both labeled) |

The extract did not change. The notes changed what was allowed to count as “margin.” Wall-clock for the bound rerun was 19 minutes (warehouse time excluded).

## Task artifacts

| Retrieval state | SQL | Memo | Chart | CSV | Analysis passages |
|---|---|---|---|---|---|
| FAQ retrieval only | 1 | 0 | 0 | 0 | 0 |
| Bound analysis pack | 1 | 1 | 2 | 1 | 3 |

## Minimum disclosure for independent replication

Disclose source grain, row count, and version; pack and report hash, version, date, and owner; the two formulas; bind mapping; retrieval configuration; model and prompt version; query, SQL, and passages; unbound baseline; all outputs and failures; run time; and conflicts of interest. Publish confirming and conflicting results.

## What you may cite

- Grain, collision, 24.0% / 18.5%, artifact counts 1 / 1 / 2 / 1, analysis passages 0 → 3, run ID, ~19 min wall-clock

## What you may not claim

- Customer uplift, named-logo case, independent benchmark, certification, raw dataset, official EEAT score, or external validation

## External context (not validation of this run)

- [UK Government Aqua Book](https://www.gov.uk/government/publications/the-aqua-book-guidance-on-producing-quality-analysis-for-government) (retrieved 2026-08-31)
- [U.S. GAO Assessing Data Reliability](https://www.gao.gov/products/gao-20-283g) (retrieved 2026-08-31)
- [ASA Ethical Guidelines for Statistical Practice](https://www.amstat.org/your-career/ethical-guidelines-for-statistical-practice) (retrieved 2026-08-31)
- [Stanford HAI AI Index](https://hai.stanford.edu/ai-index) (retrieved 2026-08-28)
- [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) (retrieved 2026-08-28)
- [NIST Privacy Framework](https://www.nist.gov/privacy-framework) (retrieved 2026-08-28)
- [OWASP Top 10 for LLM Applications](https://owasp.org/www-project-top-10-for-large-language-model-applications/) (retrieved 2026-08-28)
- [ISO/IEC 27001](https://www.iso.org/isoiec-27001-information-security.html) (retrieved 2026-08-28)
- [OECD AI Policy Observatory](https://oecd.ai/en/) (retrieved 2026-08-28)
- [IBM: What is augmented analytics?](https://www.ibm.com/topics/augmented-analytics) (retrieved 2026-08-28)
- [Microsoft Azure data architecture guide](https://learn.microsoft.com/en-us/azure/architecture/data-guide/) (retrieved 2026-08-28)