# The Hint-Log Spec **A project of the AiGovOps Foundation × NCW AI Expo 2026** *"Keep the receipts" — scaled down for one person.* --- ## What a hint-log is A **hint-log** is one short, structured entry that captures a meaningful AI interaction and records the single most important fact about it: **a human, not the machine, made the final call.** Hint-logs are the small-scale, individual version of what the AiGovOps Foundation's **Beacon** tool does for enterprises. They are the *receipts side* of the three habits we teach at the Expo: 1. **The two-eyes rule** — never ship an AI output a human hasn't reviewed. 2. **Triangulate hard questions** — check important claims against a real source. 3. **Keep the receipts** — this is where the hint-log lives. A hint-log is "policy as code," scaled down for individuals. It is how a teacher proves to a parent that AI didn't grade their kid's IEP — a human did. How a small-business owner shows an auditor they reviewed the AI's draft before it went out. How a community leader proves a public claim came from a verified source, not an AI hallucination. You do not log every AI interaction. You log the ones that matter — the ones where, if someone later asked *"did a person actually check this?"*, you want a clear answer. --- ## The five fields | # | Field | What goes here | |---|-------|----------------| | 1 | **The hint** | The prompt or question you actually asked the AI. | | 2 | **The answer** | A short summary of what the AI gave back. | | 3 | **The catch** | What the AI got wrong, missed, or made up. Be specific. | | 4 | **The human call** | What YOU decided, edited, verified, or did differently. | | 5 | **The receipt** | A link to the source chat, the doc, the email — wherever the artifact lives. | The discipline is in fields 3 and 4. If "the catch" is always blank, you are not reading critically. If "the human call" is always "looks good," you are not really in charge — the AI is. --- ## The blank template ```markdown ## Hint-Log — [date] — [your initials] **1. The hint:** What did you ask the AI? Paste or paraphrase the prompt. **2. The answer:** What did the AI give back? One- or two-sentence summary. **3. The catch:** What did it get wrong, miss, or make up? **4. The human call:** What did YOU decide, edit, verify, or do differently? **5. The receipt:** Link to the chat / doc / email where the artifact lives. ``` --- ## Five worked examples ### 1. A teacher — IEP progress summary - **The hint:** Summarize Maya's reading running-records into 3 IEP goal options at a 6th-grade reading level. - **The answer:** Gave 3 goals plus a fluency target of 120 WPM and a tidy parent-facing paragraph. - **The catch:** The 120 WPM target was invented — not from our assessment data — and it silently dropped her extended-time accommodation. - **The human call:** Replaced the figure with the real 98 WPM from the running record, restored the accommodation, and rewrote one goal in plainer language. - **The receipt:** Chat link + final IEP draft v3 in the SpEd Drive folder, initialed JT, 5/14. ### 2. A small-business owner — refund policy - **The hint:** Write a refund-policy section for our website based on these notes and Washington consumer law. - **The answer:** Produced a clean policy citing a "30-day statutory refund window." - **The catch:** Washington has no blanket 30-day statutory refund right — that was a hallucinated legal claim. - **The human call:** Removed the false legal claim, reframed it as *our* voluntary 30-day policy, and ran the final by our attorney. - **The receipt:** Perplexity search link to the WA Attorney General page + the attorney's approval email, saved 6/2. ### 3. A community leader — campaign claim - **The hint:** Draft talking points claiming our river-cleanup removed "X tons" of trash this year. - **The answer:** Filled in "12 tons" and wrote a confident press paragraph around it. - **The catch:** The 12-ton figure was AI-guessed; our volunteer logs showed 4.6 tons. Publishing it would have been a false public claim. - **The human call:** Swapped in the verified 4.6 tons from our logs and linked the logbook in the press kit. - **The receipt:** Link to the volunteer-log spreadsheet + the AI chat, attached to the board minutes, 6/9. ### 4. A clinic office manager — patient letter - **The hint:** Draft a plain-language letter explaining a new billing process to patients. - **The answer:** Wrote a friendly letter that included a specific copay dollar amount. - **The catch:** The copay amount was fabricated — it pulled a number from nowhere, and the letter implied a policy we don't have. - **The human call:** Removed the dollar figure, replaced it with "contact our billing office," and stripped any content that touched patient specifics (no PHI ever went into the free tool). - **The receipt:** Link to the chat + the approved letter in the practice's shared drive, signed off by the office manager, 6/12. ### 5. A nonprofit grant writer — impact statistic - **The hint:** Help me write the impact section of a grant using these program numbers. - **The answer:** Wrote a compelling section and added "served over 2,000 families" as a headline stat. - **The catch:** Our actual number was 1,340 families; the AI rounded up to a number it found persuasive, not true. - **The human call:** Corrected the figure to 1,340, cited the intake database as the source, and flagged the chat for our ED to double-check before submission. - **The receipt:** Link to the intake-database export + the chat, filed with the grant application packet, 6/18. --- ## A note on free tools and your data Most of the tools you'll log against have a free tier. Remember: when a tool is free, you are often paying with your data. Free AI tools generally train on your prompts, uploads, and edits. **Never paste private data — student records, patient info, IEP drafts, customer details — into a free tool.** Use a paid tier with a "do not train" toggle, your organization's enterprise tenant, or keep the content non-sensitive. The hint-log records *that you made that judgment*, too. --- ## The bridge to Beacon A hint-log is something you keep by hand, for the AI moments that matter to one person. **When you outgrow a personal hint-log, Beacon does this automatically at the enterprise scale.** The AiGovOps Foundation's Beacon captures the same five things — the hint, the answer, the catch, the human call, and the receipt — across an entire organization, turning the habit into governance enforced in code rather than goodwill. It is the same idea: a personal notebook becomes an organizational system of record, so that anyone who later asks *"did a human actually check this?"* always has a verifiable answer. Start with a notebook. Graduate to Beacon when the stakes — and the volume — demand it. --- *AiGovOps Foundation — a 501(c)(3) nonprofit · Apache-2.0 · × NCW Tech Alliance* *Learn more: https://www.aigovops-foundation.com · https://www.ncwtech.org*