Digital Products — AI Guardrails¶
These guardrails are specific to this venture and are in addition to any company-wide AI guardrails defined at the repository root. They do not replace those; where both apply, follow whichever is stricter.
Prohibited behaviors¶
1. No fabricated market or revenue data in backlog items¶
Backlog entries in ../../portfolio/product-backlog.md and any related research documents must not include invented search-volume figures, revenue estimates, Etsy/Gumroad sales numbers, market-share claims, or competitor performance data. Any market observation that has not been sourced from actual research must be explicitly labeled a hypothesis (e.g., "hypothesized," "unresearched," "assumed"). If a number is needed for illustration only, it must be clearly marked as illustrative and not presented as real.
2. No claiming Excel compatibility without testing¶
Do not state or imply that a workbook works correctly in Excel for Mac, older Excel versions, LibreOffice Calc, Google Sheets (via import), or any environment other than the explicitly tested baseline, unless that claim has actually been verified by testing. See ../standards/compatibility-standard.md.
3. No treating workbook protection as security¶
Sheet or workbook protection (locked cells, protected sheets, protected workbook structure) must never be described as a security control, and must never be relied upon to protect sensitive information. It is a usability safeguard against accidental edits only. See ../standards/protection-and-security-standard.md.
4. No real personal financial data in example workbooks¶
Sample or example data used in workbook demonstrations, screenshots, or documentation must be synthetic. Never use a real person's actual account balances, income, debts, property values, or other personal financial data, including the founder's own real figures.
5. No unreviewed tax/legal/investment conclusions¶
AI-generated content must not present tax, legal, or investment conclusions as authoritative or personalized advice. Educational framing is acceptable when clearly labeled as educational and not personalized; anything that reads as a specific recommendation for a specific person's situation must be flagged for human (and likely professional) review before it ships. See ../standards/financial-product-guardrails.md and ../standards/ai-companion-content-standard.md.
What to do when a guardrail would be violated¶
If completing a task as instructed would require violating one of the above, stop and flag it rather than proceeding with a workaround (e.g., inventing a "placeholder" number that reads as real, or asserting untested compatibility to satisfy a listing requirement). These guardrails exist because this venture sells to real customers making real decisions about their money, homes, and businesses.